Prepare the v0.6.0 release

assets/dnd/combat-turns/schemas/dnd_combat_turns_llm.v1.json

@@ -16,7 +16,8 @@
             "type": "string"
           },
           "turn_kind": {
-            "type": "string"
+            "type": "string",
+            "enum": ["turn", "reaction", "legendary_action", "lair_action", "other"]
           },
           "source_refs": {
             "type": "array",

assets/dnd/enemy-events/schemas/dnd_enemy_events_llm.v1.json

@@ -13,7 +13,10 @@
         "required": ["name", "kind", "source_refs"],
         "properties": {
           "name": {"type": "string"},
-          "kind": {"type": "string"},
+          "kind": {
+            "type": "string",
+            "enum": ["engaged", "killed", "fled", "captured", "incapacitated"]
+          },
           "source_refs": {
             "type": "array",
             "items": {

assets/dnd/item-occurrences/prompts/instructions.md

@@ -18,12 +18,20 @@ when the transcript explicitly describes it being physically destroyed or
 expended as a non-payment component. Use `transferred` only when possession
 moves between two distinct named party members.
 
-Return both `from` and `to` for every occurrence, using `null` when a holder does not
-apply. For `discovered`, set both holders to `null`. For `acquired`, set `from`
-to `null` and provide `to`; for `lost` and `consumed`, provide `from` and set
-`to` to `null`; and for `transferred`, provide both holders. Use `party` only
-for collective or unresolved party possession, never for either side of a
-transfer. Do not emit a transfer for a gift, sale, or payment outside the party.
+Return both `from` and `to` for every occurrence. Use JSON `null`, not an empty
+string, whenever a holder does not apply. Follow this holder matrix exactly:
+
+| `kind` | required `from` | required `to` |
+| --- | --- | --- |
+| `discovered` | `null` | `null` |
+| `acquired` | `null` | `party` or the named party member gaining possession |
+| `lost` | `party` or the named party member losing possession | `null` |
+| `consumed` | `party` or the named party member consuming the item | `null` |
+| `transferred` | one named party member | a different named party member |
+
+Use `party` only for collective or unresolved party possession, never for
+either side of a transfer. Do not emit a transfer for a gift, sale, or payment
+outside the party.
 
 Ordinary non-depleting use is not an occurrence. Do not infer acquisition from a
 discovery, or discovery from an acquisition: emit both only when each is

assets/dnd/item-occurrences/schemas/dnd_item_occurrences_llm.v1.json

@@ -13,7 +13,10 @@
         "required": ["name", "kind", "quantity", "from", "to", "source_refs"],
         "properties": {
           "name": {"type": "string"},
-          "kind": {"type": "string"},
+          "kind": {
+            "type": "string",
+            "enum": ["discovered", "acquired", "lost", "consumed", "transferred"]
+          },
           "quantity": {"type": ["integer", "null"]},
           "from": {"type": ["string", "null"]},
           "to": {"type": ["string", "null"]},

assets/dnd/item-registry/normalize/prompts/instructions.md

@@ -5,5 +5,5 @@ evidence, similar objects, or a shared owner as sufficient.
 Keep currency denominations and materially different item types separate. Keep
 uncertain aliases separate. Do not infer an item property or uniqueness.
 
-When selecting a canonical display name, choose one supplied candidate name
-that is the clearest established designation.
+Set `canonical_candidate_number` to the supplied candidate number whose label
+is the clearest established designation.

assets/dnd/location-registry/normalize/prompts/instructions.md

@@ -5,4 +5,5 @@ nearby evidence, nested places, or generic labels as sufficient.
 Keep parent and child places separate, as well as similarly named places and
 uncertain aliases.
 
-When selecting a canonical display name, prefer the clearest established name.
+Set `canonical_candidate_number` to the supplied candidate number whose label
+is the clearest established name.

assets/dnd/npc-occurrences/schemas/dnd_npc_occurrences_llm.v1.json

@@ -16,7 +16,8 @@
             "type": "string"
           },
           "kind": {
-            "type": "string"
+            "type": "string",
+            "enum": ["mentioned", "noncombat_presence", "dialogue", "combat_ally", "combat_opponent", "other"]
           },
           "source_refs": {
             "type": "array",

assets/dnd/npc-registry/normalize/prompts/instructions.md

@@ -3,7 +3,8 @@ contextual labels and cited transcript windows. Preserve distinct individuals
 even when their names are similar or their contextual descriptions are
 identical.
 
-When selecting a canonical display name, prefer a complete, stable proper name
+Set `canonical_candidate_number` to the supplied candidate number whose label
+is the preferred canonical display name. Prefer a complete, stable proper name
 over an abbreviation. Prefer an unadorned proper name over that name plus a
 contextual class, role, title, or relationship descriptor unless the transcript
 establishes the descriptor as part of the person's name. A longer display name

assets/dnd/scene-descriptions/prompts/instructions.md

@@ -5,9 +5,7 @@ into multiple scenes or use facts that are not supported by it.
 Return one kind, one concise title, and one concise summary. Choose exactly one
 kind:
 
-- combat: active combat materially organizes the scene, including
-  initiative-like exchanges or sustained hostile action. Planning a fight or
-  discussing a completed fight is not combat by itself.
+- combat: a scene classified as combat under the shared combat policy.
 - narrative: current-session in-world play that is not principally active
   combat, a prior-session recap, or sustained out-of-character session
   discussion. This includes exploration, travel, dialogue, investigation,
@@ -20,15 +18,13 @@ kind:
   play.
 
 Narrative is the default for actual current-session gameplay that does not meet
-another definition. When the accepted chunk is mixed:
+another definition. When the accepted chunk has no substantive active combat:
 
-1. use combat when active combat is a substantive central activity, even with
-   brief setup, rules clarification, or immediate aftermath;
-2. otherwise use recap when recounting a previous session is the chunk's
+1. use recap when recounting a previous session is the chunk's
    primary table purpose;
-3. otherwise use meta when sustained out-of-character session discussion is
+2. otherwise use meta when sustained out-of-character session discussion is
    primary and in-world progression is no more than incidental; and
-4. use narrative for all remaining current-session in-world play.
+3. use narrative for all remaining current-session in-world play.
 
 Brief table talk, dice resolution, rules clarification, jokes, or
 administrative comments do not make a gameplay scene meta. A short recollection

assets/dnd/scene-descriptions/prompts/prompt.yaml

@@ -27,6 +27,8 @@ messages:
     content_file: ./sharedassets/common-dnd-transcript-chunk.md
     cache_control:
       type: ephemeral
+  - role: user
+    content_file: ./sharedassets/common-dnd-scene-combat-policy.md
   - role: user
     content_file: ./instructions.md
     cache_control:

assets/dnd/scene-descriptions/schemas/dnd_scene_descriptions_llm.v1.json

@@ -6,7 +6,8 @@
   "required": ["kind", "title", "summary"],
   "properties": {
     "kind": {
-      "type": "string"
+      "type": "string",
+      "enum": ["combat", "narrative", "recap", "meta"]
     },
     "title": {
       "type": "string"

assets/dnd/scene-descriptions/validate/combat-semantics/prompts/instructions.md

@@ -0,0 +1,8 @@
+Classify only whether substantive active combat occurs in the supplied
+transcript chunk under the shared combat policy. Do not judge the scene title,
+summary, non-combat subtype, scene boundary, or any other aspect of a scene
+description.
+
+Return `combat` when the chunk contains substantive active combat and
+`non_combat` otherwise. Give a concise, transcript-grounded explanation for
+the classification.

assets/dnd/scene-descriptions/validate/combat-semantics/prompts/prompt.yaml

@@ -0,0 +1,23 @@
+id: dnd.scene_descriptions.validate_combat
+version: "v1"
+default_profile: dnd-extraction
+inputs:
+  - name: transcript
+    required: true
+    content_type: application/json
+messages:
+  - role: system
+    content_file: ./sharedassets/common-dnd-system.md
+  - role: user
+    content_file: ./sharedassets/common-dnd-scene-combat-policy.md
+  - role: user
+    content_file: ./instructions.md
+  - role: user
+    content_file: ./sharedassets/common-dnd-transcript-chunk.md
+    cache_control:
+      type: ephemeral
+output:
+  format: json
+  validation_mode: json_schema
+  schema_path: dnd_scene_combat_semantics_llm.v1.json
+  repair_attempts: 1

assets/dnd/scene-descriptions/validate/combat-semantics/schemas/dnd_scene_combat_semantics_llm.v1.json

@@ -0,0 +1,18 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "notarius.dnd.scene_descriptions.combat_semantics.llm",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["classification", "explanation"],
+  "properties": {
+    "classification": {
+      "type": "string",
+      "enum": ["combat", "non_combat"]
+    },
+    "explanation": {
+      "type": "string",
+      "minLength": 1,
+      "maxLength": 512
+    }
+  }
+}

assets/dnd/shared/prompts/common-dnd-extraction-evidence.md

@@ -1,6 +1,8 @@
 Transcript units are the only evidence for extracted events and factual claims.
 Every reported factual claim must be supported by cited transcript units. Use
 integer `start_unit_id` and `end_unit_id` values from the transcript.
+Within each range, `start_unit_id` must identify the earlier transcript unit and
+`end_unit_id` the same or a later unit according to transcript order.
 
 When supporting evidence is non-contiguous, use multiple narrow ranges rather
 than a broad range that bridges unrelated conversation.

assets/dnd/shared/prompts/common-dnd-scene-combat-policy.md

@@ -0,0 +1,13 @@
+Use `combat` only when substantive active combat materially organizes the
+scene. Active combat includes initiative or turn exchanges, attacks, combat
+spells, damage, saves, movement, or similarly sustained hostile action.
+
+Do not use `combat` for planning or preparing for a possible fight; threats,
+hostile dialogue, or a tense confrontation; immediate aftermath, looting,
+healing, or discussion of a completed fight; a recap or in-world recollection
+of earlier combat; or out-of-character rules discussion without active
+encounter play.
+
+When a chunk contains substantive active combat alongside brief setup, rules
+clarification, interruption, phase transition, or immediate aftermath, classify
+it as `combat`.

assets/generic/normalize/deduplication/prompts/candidates.md

@@ -1,3 +1,3 @@
-Candidate material:
+Candidate material, including the exact valid candidate-number range:
 
 {{ input "candidates" }}

assets/generic/normalize/deduplication/prompts/protocol.md

@@ -1,7 +1,9 @@
-Use only the positive integer `candidate_id` values supplied in the candidate material.
+Use only the positive integer `candidate_number` values supplied in the candidate material. Valid candidate numbers are exactly the inclusive `first` through `last` values declared in `candidate_number_range`; use the explicit number attached to each candidate.
 
-Return a duplicate group only when the evidence supports that every selected candidate describes the same underlying entity. Each group must contain at least two distinct candidate IDs, and its `canonical_candidate_id` must be one of those IDs. A candidate may appear in at most one group.
+Transcript unit `id` values and evidence `start_unit_id` and `end_unit_id` values identify source positions. They are not candidate numbers and must never be used in `candidate_numbers` or `canonical_candidate_number`.
 
-Omit uncertain matches and candidates that should remain distinct. Do not invent candidates or infer an ID from list position. An empty `duplicate_groups` array is valid.
+Return a duplicate group only when the evidence supports that every selected candidate describes the same underlying entity. Each group must contain at least two distinct candidate numbers, and its `canonical_candidate_number` must be one of those numbers. A candidate may appear in at most one group.
 
-The response must conform exactly to the selected JSON schema. Return IDs only: do not copy candidate names, evidence, transcript text, source identifiers, or source ranges into the response.
+Omit uncertain matches and candidates that should remain distinct. Do not invent candidates or infer a number that is not explicitly supplied. An empty `duplicate_groups` array is valid.
+
+The response must conform exactly to the selected JSON schema. Return candidate numbers only: do not copy candidate names, evidence, transcript text, source identifiers, or source ranges into the response.

assets/generic/normalize/deduplication/schemas/semantic_reconciliation_llm.v1.json

@@ -11,9 +11,9 @@
       "items": {
         "type": "object",
         "additionalProperties": false,
-        "required": ["candidate_ids", "canonical_candidate_id"],
+        "required": ["candidate_numbers", "canonical_candidate_number"],
         "properties": {
-          "candidate_ids": {
+          "candidate_numbers": {
             "type": "array",
             "minItems": 2,
             "items": {
@@ -21,7 +21,7 @@
               "minimum": 1
             }
           },
-          "canonical_candidate_id": {
+          "canonical_candidate_number": {
             "type": "integer",
             "minimum": 1
           }

docs/adr/0013-use-request-local-candidate-handles-for-semantic-reconciliation.md

@@ -83,9 +83,7 @@ entity identity, checkpoint contracts, or cross-request correlation. Changes
 to shared protocol and policy assets must participate in the normal prompt,
 schema, and checkpoint fingerprint mechanisms.
 
-Acceptance of this decision does not imply that the shared mechanism or its
-consumer migrations are implemented. The
-[feature roadmap](../roadmap/semantic-reconciliation.md) owns target behavior
-and status, and the
-[implementation plan](../roadmap/implementation.md) owns delivery sequence
-until the work is complete.
+The shared mechanism and its initial D&D registry consumers are now
+implemented. Current behavior is documented in
+[Module Internals](../internal/modules.md#semantic-reconciliation) and
+[D&D Module Internals](../internal/dnd.md#semantic-registry-reconciliation).

docs/adr/0014-feedback-aware-validation-retries.md

@@ -0,0 +1,59 @@
+# ADR-0014: Use feedback-aware validation retries
+
+**Status:** Accepted
+**Date:** 2026-08-26
+
+## Context
+
+Validation can identify a candidate defect after a producer has returned an
+otherwise well-formed result. Retrying without the validator's deterministic,
+bounded feedback wastes the useful diagnosis, while treating validator
+execution failures as defects would ask a producer to repair conditions it
+cannot control. The mechanism must preserve typed producer ownership,
+checkpoint safety, and the repository's sensitive-data boundaries.
+
+## Decision
+
+The implementation will keep three independent budgets: the producer binding's
+outer `retries` budget, PromptKit's structured-output repair budget, and each
+validator's execution-retry budget. Validators will run sequentially in their
+configured order and aggregate both rejections and execution failures before a
+candidate disposition is selected.
+
+A correction-capable producer will provide the exact single LLM response that
+controlled its candidate using the `single_response_v1` protocol. A correction
+attempt will reconstruct the ordinary request and append exactly two fresh
+messages: that latest response as `assistant`, followed by one deterministic
+aggregate correction request as `user`. Earlier turns will not accumulate.
+
+Validator failures will not recurse into correction. Pipeline policy owns
+terminal disposition, with field-by-field producer overrides over pipeline
+defaults: structural failure and semantic rejection default to `fail_run`, and
+validator execution failure defaults to `warn_continue`. Validators can report
+facts and bounded corrective guidance, but never decide disposition.
+
+Rejected and structurally invalid candidates will not advance. A candidate
+allowed through after a validator execution failure will retain explicit
+incomplete-validation provenance and will not be checkpointed. Exact response
+and correction text remain attempt-local: they are excluded from ordinary
+errors, warnings, manifests, receipts, caches, checkpoints, and default debug
+summaries.
+
+## Alternatives considered
+
+- Retry every producer after any validation outcome. This conflates producer
+  defects with validator operational failures and wastes retry budget.
+- Let validators decide whether to continue. This would distribute pipeline
+  disposition policy across validators and undermine consistent defaults.
+- Reuse the full prior conversation. Accumulated turns introduce unbounded
+  prompt growth and make correction behavior depend on incidental history.
+- Persist raw responses to simplify diagnosis. Raw model output and correction
+  guidance may be sensitive and do not belong in durable pipeline records.
+
+## Consequences
+
+The framework gains transport-neutral correction and candidate contracts,
+producer capability checks, policy resolution, aggregated validation outcomes,
+and conservative checkpoint handling. Prompt construction remains inside the
+LLM adapter, while modules remain responsible for accurately exposing the
+single response that directly controlled a candidate.

docs/adr/0015-separate-process-warnings-from-quality-diagnostics.md

@@ -0,0 +1,67 @@
+# ADR-0015: Separate process warnings from quality diagnostics
+
+**Status:** Accepted
+**Date:** 2026-08-27
+
+## Context
+
+Notarius currently represents process degradation, incomplete validation,
+extraction-quality doubt, and routine normalization with one flat warning
+record. That makes ordinary successful runs noisy, loses the framework context
+needed to explain a finding, and gives `warning_count` no stable operational
+meaning. It also permits output encoders to add a warning after the durable
+warning file has already been written.
+
+The application needs one bounded diagnostic model that preserves exact
+occurrence counts while retaining only safe, representative samples. Fresh and
+resumed logical runs must present the same groups. The model must not alter
+validation decisions, retry budgets, rejected-output behavior, or process exit
+policy.
+
+## Decision
+
+Warnings are reserved for a completed run that advanced under an allowed
+process-level degradation or incomplete-work policy. Extraction-quality signals
+are advisories, and routine accepted transformations are observations. A
+non-degraded successful run therefore has zero actionable warnings.
+
+Modules and validators own a diagnostic's disposition, category, reason code,
+scope, and safe message. The framework adds pipeline origin, including stage,
+step, lane, module, validator, and chunk context where applicable. It then
+aggregates deterministically by disposition, category, reason code, and full
+origin. Chunk context remains on representative samples so equivalent findings
+across chunks aggregate together.
+
+Diagnostics carry exact occurrence counts, at most three distinct samples, and
+numeric omitted-sample metadata. Producers and validators are bounded to 64
+local groups. Final actionable warning groups are bounded without truncation;
+the non-warning collection may truncate represented groups while preserving an
+exact total occurrence count and explicit truncation metadata.
+
+The public contracts will be versioned: grouped actionable warnings use
+`notarius.warnings.v2`, grouped advisories and observations use
+`notarius.diagnostics.v1`, and the run receipt uses
+`notarius.run-result.v2`. Successful output encoders return logical files or
+an error; they do not add post-encoding warnings.
+
+## Alternatives considered
+
+- Keep one warning list and filter only CLI output. This would leave durable
+  consumers with the same semantically mixed, unbounded contract.
+- Map reason codes to severity in a central framework registry. This would
+  split module-owned meaning between synchronized policy tables and make new
+  diagnostic meaning implicit.
+- Preserve local omission warning records. They inflate visible group counts
+  and lose exact occurrence semantics.
+- Keep output-encoder warnings. A one-pass encoder cannot include those
+  records consistently in files it has already serialized; a two-phase encoder
+  protocol is deferred until a demonstrated need exists.
+
+## Consequences
+
+The framework gains validated diagnostic primitives, local collection,
+origin-aware aggregation, and versioned durable presentation. Existing warning
+transport remains temporarily while producers migrate. Current architecture,
+operator, integration, and internal documentation will describe the behavior
+only as each implementation step lands; this accepted decision does not claim
+that the migration is complete.

docs/adr/0016-feedback-aware-module-requested-retries.md

@@ -0,0 +1,76 @@
+# ADR-0016: Use feedback-aware module-requested retries
+
+**Status:** Accepted
+**Date:** 2026-08-28
+
+## Context
+
+An LLM-backed module can discover that a structurally valid model response is
+unsafe while translating it into a typed candidate, before the ordinary
+validator chain runs. Semantic registry reconciliation is the first such case:
+the shared core can identify invalid duplicate-group proposals, and typed
+application can reject a domain-incompatible group while preserving a safe
+partial result. Repeating the original request without the rejected response or
+corrective explanation gives the model no information with which to improve.
+
+The existing feedback-aware validation mechanism already establishes the safe
+correction protocol, but module-owned retry directives also carry internal
+reason codes, operator messages, and fallback diagnostics. Those values are not
+model instructions. Some module retry conditions, including exhausted
+structured-output repair, also have no valid exact response to append.
+
+## Decision
+
+An LLM-backed normalizer may attach optional, bounded model-facing correction
+guidance to a module-requested retry. Guidance is a separate contract field; the
+framework never derives it from a reason code, operator message, diagnostic, or
+error. A feedback-capable directive must include the exact model candidate that
+controlled the safe fallback and must use `single_response_v1`.
+
+The artifact-neutral producer-attempt state machine constructs the correction
+from that exact latest response and the supplied guidance. The producer rebuilds
+its complete ordinary request and appends the response as one assistant message
+and the guidance as one user message. Earlier attempts do not accumulate, and
+the attempt consumes the same configured stage retry budget as operational,
+structural, validation, and feedback-free module retries.
+
+A module retry without guidance remains valid and starts a fresh attempt. This
+is the required behavior when no valid model candidate exists, including after
+structured-output repair is exhausted. If feedback is supplied without a valid
+supported candidate, the framework fails the module contract rather than
+retrying blindly or inventing guidance.
+
+After retry exhaustion, the normalizer's safe fallback continues through its
+validator chain. Safe semantic groups may remain applied, unsafe groups remain
+separate, and bounded fallback diagnostics may explain the process degradation.
+Exact responses and correction text remain attempt-local and are excluded from
+ordinary errors, warnings, manifests, receipts, caches, and checkpoints.
+
+This decision extends, rather than supersedes,
+[ADR-0014](0014-feedback-aware-validation-retries.md): both correction sources
+use the same transport-neutral payload, replacement-request semantics, outer
+retry budget, and sensitive-data boundary.
+
+## Alternatives considered
+
+- Continue blind module retries. This preserves a smaller contract but wastes
+  the module's deterministic diagnosis and commonly repeats the same defect.
+- Convert module safety checks into validators. Typed reconciliation must apply
+  only safe proposal groups and retain a fallback before validation; moving
+  artifact-owned translation and application policy into validators would blur
+  stage ownership.
+- Copy the retry reason or operator message into the model request. Those values
+  are written for provenance and humans, can contain opaque internal labels,
+  and do not reliably describe a correct replacement.
+- Require feedback for every module retry. Structural failures may have no
+  valid exact candidate, so this would either prevent useful fresh retries or
+  fabricate prior-response material.
+
+## Consequences
+
+The normalize retry contract and generic producer-attempt directive gain an
+optional correction-guidance field and candidate-pairing validation. Modules
+that use it must provide semantically meaningful bounded prose and an exact
+candidate. Registry reconciliation maintains separate operator and model
+renderers, and policy fingerprints change so checkpoints created under blind
+retry behavior are not reused.

docs/cli.md

@@ -55,8 +55,8 @@ pipeline ID and **--input** are required.
 | **--session-id id** | Override the generated prompt session identifier with a non-empty value for LLM-backed module calls. |
 | **--reasoning-effort value** | Replace the selected PromptKit profile's reasoning effort for every LLM-backed call in this run. The value must be non-empty and the flag may be specified only once. |
 | **--clear-reasoning-effort** | Clear reasoning effort inherited from the selected PromptKit profile for every LLM-backed call in this run. |
-| **--reference selector=path** | Add or replace a file reference binding. Repeatable. |
-| **--without-reference selector** | Remove a configured optional reference binding. Repeatable. |
+| **--reference selector=path** | Add or replace external file reference bindings at pipeline, lane, chunk, or binding scope. Repeatable. |
+| **--without-reference selector** | Remove matching configured external reference bindings. Repeatable. |
 
 **--chunk_cache** accepts only **auto**, **bypass**, or **refresh**.
 **--debug-dir**, **--output-dir**, **--session-id**, and
@@ -81,33 +81,62 @@ guidance.
 
 ### Reference selectors
 
-Use **--reference** only for a reference slot declared by the selected
-configured target. The accepted selector forms are:
+Use **--reference** only for reference slots declared by the selected
+configured targets. Qualification narrows the scope of an override:
 
 | Form | Target |
 | --- | --- |
-| slot=path | The unique selected target that declares slot. |
+| slot=path | Every selected target that declares slot. |
 | chunk.slot=path | The chunker. |
-| merge.slot=path | The unique selected merger that declares slot. |
-| lane.slot=path | The unique extractor, merger, or normalizer in lane that declares slot. |
+| lane.slot=path | Every extractor, merger, or normalizer in lane that declares slot. |
 | lane.extract.slot=path | The extractor in lane. |
 | lane.merge.slot=path | The merger in lane. |
 | lane.normalize.slot=path | The normalizer in lane. |
 
-**--without-reference** uses the same selector forms without =path. Slot
-names, requiredness, and configured bindings are part of the
+Pipeline- and lane-scoped selectors are expected to match multiple targets and
+fail if they match none. A stage-specific selector fails when its lane is not
+selected or its target does not declare the slot. There is no stage-wide
+`merge.slot` shorthand; name the lane when targeting a merger.
+
+CLI bindings override configured external paths. For overlapping CLI
+selectors, a binding-specific or chunk selector overrides a lane selector, and
+a lane selector overrides a pipeline selector. The last occurrence wins at
+equal scope. Binding and unbinding the same concrete target at equal scope is
+an error; a narrower bind or unbind may create an intentional exception to a
+broader action.
+
+**--without-reference** uses the same selector forms without `=path` and
+removes external bindings only. Neither flag replaces or removes a generated
+artifact handoff; an external/generated collision is a resolution error.
+Required slots are checked after all effective changes. CLI reference paths
+are resolved relative to the process working directory, so subprocess and
+service callers should use absolute paths. Slot names, accepted media types,
+size limits, requiredness, and configured generated bindings are part of the
 [configuration contract](config.md).
 
+For example, one shared campaign reference can reach every compatible target,
+with an optional lane-specific exception:
+
+~~~
+notarius run dnd-session \
+  --input /data/transcript.json \
+  --reference party=/data/references/party.txt \
+  --reference npc-registry.party=/data/references/npc-party-context.txt
+~~~
+
 ### Run output
 
 Without **--json**, standard output contains the completed pipeline ID, counts
 of normalized and rejected outputs, and the output directory. A debug-enabled
 run also prints its debug-bundle path to standard output. A successful run with
-warnings reports the warning count to standard error. The published JSON bundle
+actionable process warnings reports their group and occurrence counts to
+standard error. When the selected output module publishes `warnings.json`, the
+summary also reports that durable file's path. Advisory and observation findings
+do not produce a warning line. The published JSON bundle
 is defined by the [JSON output contract](integrations/json-output.md).
 
 With **--json**, successful standard output is exactly one
-`notarius.run-result.v1` JSON document followed by a newline, with no
+`notarius.run-result.v2` JSON document followed by a newline, with no
 human-oriented status or debug-path line. Its fields and compatibility policy
 are defined by the [run-result contract](integrations/run-result.md). A caller
 must check for exit status 0 before decoding this output; a failed write can
@@ -171,7 +200,7 @@ go run ./cmd/notarius pipelines list \
 Successful commands write their primary result to standard output. Warnings and
 errors are written to standard error.
 
-For **run --json**, warnings remain on standard error and standard output is a
+For **run --json**, actionable process warnings remain on standard error and standard output is a
 machine-readable success result only. Syntax and runtime diagnostics remain on
 standard error. Parse the result only after the process exits with status 0.
 

docs/config.md

@@ -132,7 +132,7 @@ model: example-model
 Keep credentials out of the local-backend object. A PromptKit profile may name
 its credential environment variable through `api_key_env`; set that variable
 only in the run environment. PromptKit owns the
-[pinned profile-file format](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/formats.md),
+[pinned profile-file format](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md),
 including `base_profile` inheritance. Notarius passes profiles through without
 merging them. Filesystem profiles cannot express PromptKit's in-memory
 `APIKeyRequired` setting; an unset `api_key_env` is optional and may reach the
@@ -223,6 +223,7 @@ pipelines:
 | --- | --- | --- | --- |
 | **llm_profile** | string | none | Optional non-empty default PromptKit profile ID for selected LLM-backed bindings and validators. An explicitly present blank value is invalid. |
 | **structured_output_repair_attempts** | integer | prompt-owned (1 in maintained production prompts) | Optional structural-repair limit from 0 through 3 for selected LLM-backed bindings and validators. Omission leaves the prompt's declared policy in control; explicit 0 disables structural repair at that scope. |
+| **validation_policy** | object | see below | Optional terminal policy defaults for producer validation. Its fields inherit independently into chunk, extract, merge, and normalize bindings. |
 | **input** | module binding | none | Required. |
 | **chunk** | module binding | **generic** | Optional. |
 | **output** | module binding | **json** | Optional. |
@@ -251,6 +252,33 @@ explicit `null` and non-integer values are invalid. An explicit value on a
 deterministic binding or validator is invalid, while a pipeline value simply
 does not apply to deterministic selections.
 
+`validation_policy` controls terminal disposition for one complete producer
+attempt and validator chain. It may appear on a pipeline or a **chunk**,
+**extract**, **merge**, or **normalize** module binding; input, output, and
+validator bindings reject it. Every field is optional and resolves in binding,
+pipeline, then application-default order:
+
+| Field | Values | Default |
+| --- | --- | --- |
+| **producer_structural_failure** | **fail_run**, **reject_output** | **fail_run** |
+| **semantic_rejection** | **fail_run**, **reject_output** | **fail_run** |
+| **validator_failure** | **warn_continue**, **fail_run** | **warn_continue** |
+
+The policy object and its fields must be non-null, and unknown fields are
+rejected. A deterministic producer may not explicitly set
+**producer_structural_failure** on its binding, although a pipeline-level
+default remains valid for pipelines that include LLM-backed producers.
+
+After the producer binding's retry budget is exhausted, an invalid structured
+response uses **producer_structural_failure**. One or more semantic validator
+rejections use **semantic_rejection**; rejection takes precedence over an
+exhausted validator failure or skip. With no rejection, an exhausted validator
+failure or skip uses **validator_failure**. `reject_output` records the
+terminal rejection without advancing that candidate. `warn_continue` is valid
+only for validator execution failure: it advances a structurally valid,
+otherwise unrejected result with incomplete-validation provenance and without
+making it reusable checkpoint state.
+
 A lane has these fields:
 
 | Field | Type | Default | Rules |
@@ -289,7 +317,8 @@ extract:
 | **module** | string | none | Required for an object binding. Must be a registered compatible key. |
 | **llm_profile** | string | none | Optional non-empty PromptKit profile ID for an LLM-backed binding. It overrides the pipeline default unless the run supplies **--llm-profile**. |
 | **structured_output_repair_attempts** | integer | pipeline or prompt-owned (1 in maintained production prompts) | Optional structural-repair limit from 0 through 3 for an LLM-backed binding. It overrides the pipeline value; explicit 0 disables structural repair. |
-| **retries** | integer | 0 | Non-negative additional attempts for chunk, extract, merge, and normalize bindings. |
+| **validation_policy** | object | pipeline or application defaults | Optional field-by-field terminal-policy override for a chunk, extract, merge, or normalize binding. |
+| **retries** | integer | 0 | Non-negative additional complete producer attempts for chunk, extract, merge, and normalize bindings. This single budget covers operational errors, invalid structured output, module-requested normalization retry, and semantic correction. |
 | **options** | object | none | Must satisfy the selected module. |
 | **references** | map | none | Valid only on chunk, extract, merge, and normalize bindings. |
 | **validators** | list | production chain | Valid only on chunk, extract, merge, and normalize bindings. |
@@ -297,11 +326,23 @@ extract:
 Omitting **validators** uses the registered chain. **validators: []** selects
 an empty chain; a non-empty list replaces the chain in the listed order.
 Validator bindings accept only **module**, **llm_profile**,
-**structured_output_repair_attempts**, and **options**. They reject
-**references**, **retries**, and nested **validators**. Deterministic validators
-reject explicit **llm_profile** and **structured_output_repair_attempts**.
+**structured_output_repair_attempts**, **retries**, and **options**. Their
+**retries** value is a non-negative additional validator-execution budget and
+is valid only when the selected validator is LLM-backed. A validator retry
+rechecks the same immutable candidate; it never regenerates the producer.
+They reject
+**validation_policy**, **references**, and nested **validators**. Deterministic
+validators reject explicit **llm_profile** and
+**structured_output_repair_attempts**.
 Deterministic module bindings also reject those explicit fields.
 
+An LLM-backed chunk, extract, merge, or normalize producer with both a
+non-empty validator chain and positive **retries** must declare the supported
+single-response correction capability. Preparation rejects a configuration
+that could require semantic correction from a producer that cannot provide an
+exact prior response. A deterministic producer, or an LLM attempt that did
+not make a model call, cannot consume a semantic retry after rejection.
+
 The **json** output module accepts optional **include_chunk_map** and
 **evidence_context** settings:
 
@@ -370,10 +411,15 @@ slot. A generated binding supplies one accepted normalized artifact; it does
 not name a file. A configured generated dependency remains required even when
 that consumer slot is otherwise optional.
 
-Pipeline references are defaults. A matching step-local or binding-local
-external path overrides a pipeline default. Required slots must be bound after
-these configuration values and any CLI reference overrides are applied.
-Reference paths in YAML are resolved relative to the configuration file.
+Pipeline references are configuration defaults. A matching step-local or
+binding-local external path overrides a pipeline default. CLI reference
+bindings are then operational overrides of configured external paths; their
+pipeline, lane, and binding scopes and precedence are defined by the
+[CLI reference](cli.md#reference-selectors). A CLI file reference cannot
+replace a configured generated artifact handoff. Required slots must be bound
+after configuration and CLI reference actions are applied. Reference paths in
+YAML are resolved relative to the configuration file; CLI reference paths are
+resolved relative to the process working directory.
 
 ### D&D Reference Slots
 
@@ -463,11 +509,33 @@ Available validator keys are:
 | Item occurrences | **extract/dnd/item-occurrences/shape**, **extract/dnd/item-occurrences/registry**, **extract/dnd/item-occurrences/source_refs**, **extract/dnd/item-occurrences/source_relatedness**, **normalize/dnd/item-occurrences/invariants** |
 | Item registry | **extract/dnd/item-registry/shape**, **extract/dnd/item-registry/source_refs**, **extract/dnd/item-registry/source_relatedness**, **normalize/dnd/item-registry/identity** |
 | NPC occurrences | **extract/dnd/npc-occurrences/shape**, **extract/dnd/npc-occurrences/registry**, **extract/dnd/npc-occurrences/source_refs**, **extract/dnd/npc-occurrences/source_relatedness**, **normalize/dnd/npc-occurrences/invariants** |
-| Scene descriptions | **extract/dnd/scene-descriptions/shape**, **extract/dnd/scene-descriptions/source_refs**, **extract/dnd/scene-descriptions/source_relatedness**, **normalize/dnd/scene-descriptions/invariants** |
+| Scene descriptions | **extract/dnd/scene-descriptions/shape**, **extract/dnd/scene-descriptions/source_refs**, **extract/dnd/scene-descriptions/source_relatedness**, **extract/dnd/scene-descriptions/combat_semantics** (LLM-backed, opt-in), **normalize/dnd/scene-descriptions/invariants** |
 | Enemy events | **extract/dnd/enemy-events/shape**, **extract/dnd/enemy-events/engagements**, **extract/dnd/enemy-events/source_refs**, **extract/dnd/enemy-events/source_relatedness**, **normalize/dnd/enemy-events/invariants** |
 | Location registry | **extract/dnd/location-registry/shape**, **extract/dnd/location-registry/source_refs**, **extract/dnd/location-registry/source_relatedness**, **normalize/dnd/location-registry/identity** |
 | Location occurrences | **extract/dnd/location-occurrences/shape**, **extract/dnd/location-occurrences/registry**, **extract/dnd/location-occurrences/source_refs**, **extract/dnd/location-occurrences/source_relatedness**, **normalize/dnd/location-occurrences/invariants** |
 
+`extract/dnd/scene-descriptions/combat_semantics` is not in a production default chain. To opt in, replace the scene extractor validator chain with the current ordered chain plus the semantic validator last, and set a positive producer retry budget if a rejection should request a corrected scene:
+
+~~~yaml
+extract:
+  module: dnd/scene-descriptions
+  retries: 1
+  validators:
+    - generic/valid_json
+    - extract/dnd/scene-descriptions/shape
+    - extract/dnd/scene-descriptions/source_refs
+    - generic/valid_json_schema
+    - extract/dnd/scene-descriptions/source_relatedness
+    - module: extract/dnd/scene-descriptions/combat_semantics
+      retries: 1
+~~~
+
+An override replaces, rather than extends, the default chain. See [Module Bindings And Validators](#module-bindings-and-validators) for binding, profile, repair, retry, and failure-policy rules.
+
+The validator retry shown above permits one additional execution against the
+same scene candidate when the LLM-backed validator itself fails; it is separate
+from both the extractor's producer retry and PromptKit structural repair.
+
 When no override is configured, production D&D bindings use the following
 ordered chains. Each row lists extract then normalize; spell chains are the
 same at both stages.

docs/consumers/dnd-pipeline.md

@@ -24,7 +24,9 @@ files. Use absolute paths for service and subprocess deployments. In
 particular, observe these different resolution rules:
 
 - reference paths in YAML are resolved relative to the Notarius configuration
-  file; and
+  file;
+- reference paths passed with `--reference` are resolved relative to the
+  Notarius process working directory; and
 - `promptkit.profile_file` is resolved relative to the Notarius process working
   directory.
 
@@ -73,9 +75,21 @@ notarius run dnd-session \
   --config /absolute/path/to/notarius.yml \
   --input /absolute/path/to/transcripts/final.trimmed.json \
   --output-dir /absolute/path/to/notarius-output \
+  --reference party=/absolute/path/to/references/party.txt \
+  --reference players=/absolute/path/to/references/players.txt \
+  --reference glossary=/absolute/path/to/references/glossary.txt \
+  --reference spell_catalog=/absolute/path/to/references/spells.json \
   --json
 ```
 
+Each unqualified reference is pipeline-scoped: Notarius supplies it to every
+selected D&D target that declares the slot. A deployment may omit an optional
+reference it does not maintain, and may use the lane- or binding-qualified
+forms from the [CLI reference](../cli.md#reference-selectors) for an exceptional
+override. The registry, scene-description, combat-turn, and NPC-occurrence
+references declared between ordered steps in the complete configuration are
+generated artifacts. Do not pass those handoffs on the CLI.
+
 The caller should:
 
 - capture stdout and stderr separately;
@@ -92,7 +106,7 @@ stream and exit-status contract.
 ## Discover The Published Bundle
 
 Decode the successful stdout document as a supported run-result schema. For
-the current contract, `schema_version` is `notarius.run-result.v1`. Tolerate
+the current contract, `schema_version` is `notarius.run-result.v2`. Tolerate
 unknown fields allowed by that version, but reject an unsupported schema
 version.
 
@@ -101,6 +115,12 @@ root. Do not scan the output root for its newest directory, guess a run ID, or
 construct a bundle path. Resolve `index_file` beneath `output_directory` and
 reject an absolute logical path or any result that escapes the bundle root.
 
+The complete configuration uses the application validation defaults. A caller
+that requires fully validated D&D artifacts must also require receipt
+`validation_status: approved`; a successful `incomplete` result reflects the
+configured validator-failure continuation policy and carries its bounded
+validator provenance in `validation_summaries`.
+
 Read `index.json` and locate each requested lane in `output_files` by its exact
 `lane_id`. Do not guess a lane filename. Before decoding a payload:
 
@@ -139,7 +159,8 @@ The JSON encoder always publishes these bundle-management files:
 | `index.json` | Discovery document for lane and pipeline-wide artifacts. |
 | `manifest.json` | Run provenance and result summaries. |
 | `rejected.json` | Rejected pipeline outputs. |
-| `warnings.json` | Accepted-output and run warnings. |
+| `warnings.json` | Actionable process-degradation warnings. |
+| `diagnostics.json` | Advisory and observation findings for accepted artifacts. |
 
 The complete configuration also requests two pipeline-wide artifacts:
 

docs/consumers/subprocess.md

@@ -31,13 +31,36 @@ notarius run pipeline-id \
 ```
 
 Use absolute paths for supplied input, configuration, output-root, and
-reference files. Notarius generates a stable prompt session for the resolved
-input module and exact input bytes. Pass **--session-id** only when intentionally
-grouping different invocations under a different session. Supply credentials
-through Notarius's documented configuration and environment mechanisms, never
-as command-line arguments or generated secret-bearing configuration. In
-particular, a session identifier is provider-visible and is not a credential
-mechanism.
+reference files. Pass each external reference as its own argument-vector pair;
+do not construct and invoke a shell command. An unqualified reference selector
+supplies that file to every compatible selected target. Lane and stage
+qualification are available for exceptional overrides, while generated
+same-run references remain part of configured pipeline composition. The
+[CLI reference](../cli.md#reference-selectors) owns the exact selector and
+precedence contract.
+
+The maintained D&D subprocess workflow uses this facility for campaign context:
+
+```sh
+notarius run dnd-session \
+  --config /absolute/path/to/notarius.yml \
+  --input /absolute/path/to/transcripts/final.trimmed.json \
+  --output-dir /absolute/path/to/notarius-output \
+  --reference party=/absolute/path/to/references/party.txt \
+  --reference players=/absolute/path/to/references/players.txt \
+  --reference glossary=/absolute/path/to/references/glossary.txt \
+  --reference spell_catalog=/absolute/path/to/references/spells.json \
+  --json
+```
+
+Only pass the external references available to and desired by the deployment.
+Notarius generates a stable prompt session for the resolved input module and
+exact input bytes; reference changes do not change it. Pass **--session-id**
+only when intentionally grouping different invocations under a different
+session. Supply credentials through Notarius's documented configuration and
+environment mechanisms, never as command-line arguments or generated
+secret-bearing configuration. In particular, a session identifier is
+provider-visible and is not a credential mechanism.
 
 Wait for the process before interpreting standard output. Only an exit status
 of 0 permits decoding the receipt. On a nonzero exit, retain standard error for
@@ -69,10 +92,17 @@ lanes. The caller decides which lane IDs are required for its own work and
 which are optional; it should make that decision explicitly rather than infer
 failure from the receipt counts alone.
 
+When complete validation is required, also require receipt
+`validation_status: approved` and inspect `validation_summaries`. A successful
+run with `validation_status: incomplete` contains a structurally valid result
+that advanced after validator execution could not complete under the configured
+`warn_continue` policy. It is not reusable checkpoint state and should not be
+silently treated as fully reviewed by the caller.
+
 ## Preserve Provenance And Handle Data Carefully
 
 Keep the receipt with the published `manifest.json`, and retain
-`rejected.json` and `warnings.json` when review or later provenance requires
+`rejected.json`, `warnings.json`, and `diagnostics.json` when review or later provenance requires
 them. Treat the input, output bundle, cache, debug bundle, and captured process
 logs as potentially sensitive data. Apply the caller's access controls and
 retention policy, and avoid copying secrets into arguments, logs, or

docs/integrations/json-output.md

@@ -9,7 +9,7 @@ Output configuration, including chunk-map and evidence-context publication, belo
 ## Bundle Layout
 
 All paths below are logical, relative, slash-separated bundle paths. The
-encoder always emits the first four JSON files below and adds lane or
+encoder always emits the first five JSON files below and adds lane or
 pipeline-wide artifact files when their corresponding artifacts are available:
 
 A subprocess caller first obtains the physical bundle root from the
@@ -21,7 +21,8 @@ root for the logical discovery described here.
 | `index.json` | Entry point that names the other published files and lane payloads. |
 | `manifest.json` | Run provenance and result summaries. |
 | `rejected.json` | Rejected pipeline outputs. |
-| `warnings.json` | Accepted-output and run warnings. |
+| `warnings.json` | Actionable process-degradation warnings. |
+| `diagnostics.json` | Accepted-artifact quality advisories and normalization observations. |
 | `lanes/<safe-lane-id>.json` | One normalized artifact payload for each lane. |
 | `chunk-map.json` | Optional accepted chunk map, when its export is enabled and available. |
 | `evidence-context.json` | Optional selected source-unit excerpt, when evidence publication is enabled. |
@@ -39,7 +40,8 @@ normalized lanes has this valid minimal index:
   "manifest_file": "manifest.json",
   "output_files": [],
   "rejected_file": "rejected.json",
-  "warnings_file": "warnings.json"
+  "warnings_file": "warnings.json",
+  "diagnostics_file": "diagnostics.json"
 }
 ```
 
@@ -49,6 +51,7 @@ normalized lanes has this valid minimal index:
 | `output_files` | Yes | Lane descriptors sorted by `lane_id`. |
 | `rejected_file` | Yes | Always `rejected.json`. |
 | `warnings_file` | Yes | Always `warnings.json`. |
+| `diagnostics_file` | Yes | Always `diagnostics.json`. |
 | `chunk_map` | No | Descriptor for the pipeline-wide `chunk-map.json`; never a lane descriptor. |
 | `evidence_context` | No | Descriptor for the pipeline-wide `evidence-context.json`; never a lane descriptor. |
 
@@ -92,7 +95,7 @@ group into the following externally observable summaries:
 | Run identity and result | `run_id`, `pipeline_id`, `pipeline_digest`, `schema_version`, `validation_status`, `started_at`, `completed_at` |
 | Resolved components | `input_module`, `chunker`, `extractors`, `merger`, `normalizer`, `output_encoder`, `artifact_lanes`, `validator_chains`, `module_metadata` |
 | Source and references | `source_digests`, `references` |
-| Published result summaries | `normalized_outputs`, `rejected_outputs` |
+| Published result summaries | `normalized_outputs`, `rejected_outputs`, `validation_summaries` |
 | Execution summaries | `chunk_plan`, `checkpoint_decisions`, `llm_profiles`, `metadata` |
 
 `references` records provenance such as the target, slot, origin, digest,
@@ -102,6 +105,16 @@ summarize results without embedding lane payload bytes. A chunk-plan summary is
 provenance for the plan used by this run; cache records, debug artifacts, and
 other operational state are not published as bundle files.
 
+Each `validation_summaries` entry is a bounded outcome for one producer result.
+It has required `status`, `producer_attempt_count`, and `terminal_action`;
+the stage and affected step, lane, module, or chunk identity are present when
+applicable. `status` is `complete`, `rejected`, or `incomplete`.
+`rejecting_validators`, `reason_codes`, and `incomplete_validators` preserve
+configured validator order and omit later duplicates. Entries contain no raw
+candidate response, correction guidance, validator diagnostic message, or
+artifact payload. The same shape may appear as `validation` on an affected
+rejection entry.
+
 When present, `metadata.session_id` is the effective non-secret routing
 correlation identifier used for the run. It can be visible to providers and is
 not a substitute for a cache or checkpoint identity. Its generation and
@@ -122,16 +135,57 @@ These values describe observed execution; they are not a backend-registration
 interface. Entries that differ by backend or effective reasoning remain
 distinct even when their profile, provider, and model are otherwise equal.
 
-## Rejections And Warnings
+## Rejections, Warnings, And Diagnostics
 
 `rejected.json` is always an object with a `rejected` array. Each entry has
 required `stage` and `message`; `step_id`, `lane_id`, `module_key`, `chunk_id`,
 `chunk_index`, `validator_name`, `reason_code`, `attempt_count`, and
-`diagnostic_artifact_path` are present only when applicable.
+`diagnostic_artifact_path` are present only when applicable. An entry may also
+contain the bounded `validation` summary described above; the existing singular
+validator and reason fields remain the first configured rejection for
+compatibility.
 
-`warnings.json` is always an object with a `warnings` array. Each warning has
-`reason_code` and `message`; `scope` is optional. Both arrays are empty when
-there is nothing to report.
+`warnings.json` is always the `notarius.warnings.v2` envelope:
+
+```json
+{
+  "schema_version": "notarius.warnings.v2",
+  "group_count": 0,
+  "occurrence_count": 0,
+  "groups": []
+}
+```
+
+It contains only process warnings. `group_count` is exact, and
+`occurrence_count` is the exact sum of its group occurrence counts.
+
+`diagnostics.json` is always the `notarius.diagnostics.v1` envelope:
+
+```json
+{
+  "schema_version": "notarius.diagnostics.v1",
+  "group_count": 0,
+  "occurrence_count": 0,
+  "truncated": false,
+  "unrepresented_occurrence_count": 0,
+  "groups": []
+}
+```
+
+It contains only advisory and observation groups. `group_count` counts groups
+represented in `groups`; `occurrence_count` includes both represented and
+unrepresented occurrences. When `truncated` is true,
+`unrepresented_occurrence_count` is the exact number omitted from group
+representation.
+
+Each group has `disposition`, `category`, `reason_code`, framework-owned
+`origin`, exact `occurrence_count`, bounded `samples`, and
+`omitted_sample_count`. Samples carry safe `scope` and `message`, plus a chunk
+ID and zero-based chunk index when applicable. A group retains at most three
+distinct samples. The framework fails rather than truncating actionable
+warnings beyond 128 groups; it represents at most 256 advisory/observation
+groups and records further occurrences through the diagnostic truncation
+fields above.
 
 ## Compatibility
 

docs/integrations/pkg-promptkit.md

@@ -1,11 +1,11 @@
 # PromptKit Integration
 
 Notarius pins
-[`gitea.maximumdirect.net/eric/promptkit` v0.8.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0)
+[`gitea.maximumdirect.net/eric/promptkit` v0.9.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0)
 as its in-process prompt engine. The upstream
-[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/consumers/pkg-promptkit.md)
+[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/consumers/pkg-promptkit.md)
 owns the public engine API, and the upstream
-[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/formats.md)
+[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md)
 owns prompt, profile, and schema file contracts.
 
 ## Supported Boundary
@@ -15,7 +15,8 @@ Notarius relies on the root `promptkit` package to:
 - construct an `Engine` with filesystem-backed prompt, schema, and optional
   operator and application-fallback profile sources;
 - prepare one frozen execution from a `RunRequest` with named inline artifacts,
-  variables, a direct session ID, prompt identity, and profile selection, then
+  variables, a direct session ID, prompt identity, profile selection, and
+  optional appended rendered messages, then
   record credential-redacted details and run that exact execution;
 - return rendered debug material, validated structured output, selected
   profile, backend, effective model metadata, and token usage;
@@ -26,7 +27,7 @@ Notarius relies on the root `promptkit` package to:
   admission exhaustion through `ErrCapacityExceeded`.
 
 The pinned
-[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/backends.go)
+[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/backends.go)
 owns the registration and backend-capacity contract.
 
 For one completion, the adapter calls `PrepareExecution`, takes a
@@ -52,7 +53,7 @@ Notarius sends one stable effective session through PromptKit's direct session
 field, which is authoritative for provider session behavior. It also retains
 the same value as the `session_id` prompt variable for maintained prompt
 compatibility. The generated identifier is 76 ASCII characters, within
-PromptKit v0.8.0's 256-code-point session limit. Session IDs are non-secret
+PromptKit v0.9.0's 256-code-point session limit. Session IDs are non-secret
 correlation identifiers and may be exposed to providers and provider
 observability. The CLI contract owns generation and override behavior.
 
@@ -85,12 +86,38 @@ configuration and deployment workflow are defined in
 [Operations](../operations.md#promptkit-profile-deployment).
 
 PromptKit owns `base_profile` resolution under its
-[pinned format rules](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/formats.md).
+[pinned format rules](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md).
 Notarius records the selected leaf identity and resolved target without parsing
 or merging inheritance. An unset filesystem `api_key_env` is optional and may
 reach the provider without authorization, which can result in a 401 or 403.
 
-Notarius supports this boundary against PromptKit v0.8.0. Its fallback source,
+PromptKit v0.9.0 accepts only the `developer`, `system`, `user`, and
+`assistant` text-chat roles after normalizing case and surrounding whitespace.
+Maintained Notarius prompt definitions use only `system` and `user`.
+
+For application-owned semantic correction, Notarius uses PromptKit v0.9.0's
+`RunRequest.AppendedMessages` after the ordinary rendered prompt. It supplies
+exactly two messages in order: the latest validated producer response with
+role `assistant`, then deterministic validation guidance with role `user`.
+It never exposes a general caller-selected role API, accumulates earlier
+correction turns, or changes the ordinary prompt prefix. Ordinary requests
+leave appended messages unset.
+
+PromptKit preserves supplied content but does not own Notarius's correction
+bounds. Notarius rejects invalid UTF-8, blank, or oversized assistant material
+(at most 1 MiB), guidance (at most 64 KiB), and combined content (at most
+1,114,112 bytes) before preparing the request. The transport-neutral
+application contract owns defensive copying and these limits. Default request
+and terminal summaries retain only safe counts, digests, identities, and usage;
+complete appended messages remain limited to the explicitly requested detailed
+debug trace.
+
+PromptKit now obtains its maintained OpenRouter and Rakestrawhome backend and
+profile catalogs from independently versioned transitive modules. Notarius
+does not import or register either catalog; PromptKit retains catalog source,
+identity, precedence, credential, and capacity ownership.
+
+Notarius supports this boundary against PromptKit v0.9.0. Its fallback source,
 prepared-execution, inspection, and typed capacity APIs are used as public
 upstream contracts; other PromptKit APIs or file-format behavior are not
 implicitly supported. A dependency upgrade requires reviewing the adapter,

docs/integrations/run-result.md

@@ -9,36 +9,58 @@ Command syntax, streams, and exit statuses are defined in the
 
 ## Schema
 
-The current schema version is `notarius.run-result.v1`.
+The current schema version is `notarius.run-result.v2`.
 
 | Field | Required | Meaning |
 | --- | --- | --- |
-| `schema_version` | Yes | Exactly `notarius.run-result.v1`. |
+| `schema_version` | Yes | Exactly `notarius.run-result.v2`. |
 | `run_id` | Yes | The finalized Notarius run identifier. |
 | `pipeline_id` | Yes | The effective pipeline identifier. |
 | `output_directory` | Yes | Absolute path to the published, run-specific output bundle. |
 | `index_file` | For the production JSON output | Logical path `index.json`; omitted for other output modules. |
 | `normalized_output_count` | Yes | Number of final normalized outputs. |
 | `rejected_output_count` | Yes | Number of recorded rejected outputs. |
-| `warning_count` | Yes | Number of final run warnings. |
+| `warning_group_count` | Yes | Exact number of actionable warning groups. |
+| `warning_occurrence_count` | Yes | Exact occurrences represented by actionable warning groups. |
+| `diagnostic_group_count` | Yes | Number of represented advisory and observation groups. |
+| `diagnostic_occurrence_count` | Yes | Advisory and observation occurrences, including unrepresented occurrences. |
+| `diagnostics_truncated` | Yes | Whether advisory/observation group representation was truncated. |
 | `validation_status` | Yes | The final run manifest validation status. |
+| `validation_summaries` | No | Bounded per-producer validation outcomes; present when producer work ran. |
 | `debug_directory` | No | Absolute path to the run-specific debug bundle when requested debug capture completed. |
 
 For the production `json` output module, `index_file` is present only when the
 completed run returned exactly one logical output file named `index.json`.
 For another output module, its absence does not indicate a failed run.
+`validation_status` is `approved`, `rejected`, or `incomplete`; `incomplete`
+means one or more otherwise accepted results advanced under validator-failure
+`warn_continue`.
 
 ```json
 {
-  "schema_version": "notarius.run-result.v1",
+  "schema_version": "notarius.run-result.v2",
   "run_id": "run-1770000000000000000-0123456789abcdef0123456789abcdef",
   "pipeline_id": "dnd-session",
   "output_directory": "/work/results/run-1770000000000000000-0123456789abcdef0123456789abcdef",
   "index_file": "index.json",
   "normalized_output_count": 6,
   "rejected_output_count": 2,
-  "warning_count": 1,
-  "validation_status": "rejected"
+  "warning_group_count": 1,
+  "warning_occurrence_count": 2,
+  "diagnostic_group_count": 3,
+  "diagnostic_occurrence_count": 5,
+  "diagnostics_truncated": false,
+  "validation_status": "incomplete",
+  "validation_summaries": [
+    {
+      "stage": "extract",
+      "lane_id": "spells",
+      "status": "incomplete",
+      "incomplete_validators": ["dnd/spells/source_refs"],
+      "producer_attempt_count": 1,
+      "terminal_action": "warn_continue"
+    }
+  ]
 }
 ```
 
@@ -48,8 +70,11 @@ For another output module, its absence does not indicate a failed run.
 paths. They identify the paths used by Notarius and do not resolve symlinks.
 `output_directory` is the run-specific bundle, not the configured output root.
 
-The receipt is a summary and discovery document. It does not contain lane
-descriptors, payloads, manifest data, rejections, warnings, or file contents.
+The receipt is a summary and discovery document. Its optional validation
+summaries contain only stable status, identity, validator names, reason codes,
+attempt counts, and terminal actions. It does not contain lane descriptors,
+payloads, manifest payloads, rejection messages, warnings, raw model responses,
+correction guidance, or file contents.
 For the production JSON output, resolve `index_file` beneath
 `output_directory`, reject path escapes, and use the
 [Published JSON Output contract](json-output.md) to discover logical files and

docs/internal/cli.md

@@ -115,6 +115,16 @@ to checkpoint identity and `pipeline.RunInput`. The public flag and stability
 contract are defined by the [CLI reference](../cli.md#run); framework and LLM
 packages only transport the supplied value.
 
+The CLI also owns the scope grammar for reference flags. It enumerates the
+selected chunk and lane targets from registered module specifications, expands
+pipeline- and lane-scoped actions into exact stage-and-lane bindings, and
+resolves overlapping bind and unbind actions by specificity before calling
+configuration resolution. The generic pipeline therefore receives only exact
+`ReferenceBinding` and `ReferenceUnbind` values and has no knowledge of CLI
+selector syntax. Configuration resolution retains ownership of configured
+external/generated conflicts, required slots, and module compatibility; file
+materialization still occurs afterward.
+
 For `run --json`, the CLI constructs and encodes its private run-result receipt
 after a successful runner result is available, before it publishes logical
 output files. It writes the prepared receipt to standard output only after
@@ -166,8 +176,9 @@ is discoverable.
 - **internal/cli/production_contract_test.go** covers registrar composition,
   production catalog contents, assets, and representative configuration
   validation.
-- **internal/cli/reference_contract_test.go** covers CLI reference overrides,
-  origin separation, and materialization boundaries.
+- **internal/cli/reference_contract_test.go** covers scoped CLI reference
+  expansion, specificity, bind/unbind conflicts, generated-reference
+  protection, origin separation, and materialization boundaries.
 - **internal/cli/state_hardening_test.go** covers safe run identity, state
   roots, and failure ordering.
 

docs/internal/configuration.md

@@ -74,10 +74,19 @@ profile-free, and no second inheritance decision occurs during execution. The
 public field definitions and precedence are owned by
 [Configuration](../config.md#pipelines).
 
+The resolver retains configured `validation_policy` overrides and derives one
+detached concrete terminal policy for the chunk producer and every lane's
+extract, merge, and normalize producers. That field-by-field inheritance is
+complete before preparation, and the effective values contribute to pipeline
+and checkpoint identity; execution does not interpret configuration defaults.
+
 The framework resolver supplies defaults, selects lanes, resolves validator
 chains, checks registered module and artifact compatibility, validates module
-options, and returns the fixed ordered pipeline shape. The resulting
-**EffectiveConfig** retains the selected ID, requested selection and reference
+options, and returns the fixed ordered pipeline shape. Positive validator retry
+budgets require an LLM-backed selected validator; deterministic validators are
+rejected during resolution. Eligible LLM-backed producer specifications also
+contribute their declared correction protocol to the resolved metadata. The
+resulting **EffectiveConfig** retains the selected ID, requested selection and reference
 changes, a clone of the input configuration, and the resolved pipeline.
 Callers may therefore retain or modify their input slices and maps without
 changing the resolved result, and later consumers cannot mutate the original
@@ -94,8 +103,8 @@ runtime error class described in the [CLI reference](../cli.md#output-streams-an
 
 The framework assigns the resolved pipeline a deterministic SHA-256 digest
 after defaults, lane selection, module bindings, reference bindings, validator
-chains, effective LLM profiles, and artifact schema identity have been
-resolved. The digest excludes
+chains, selected correction protocols, effective LLM profiles, and artifact
+schema identity have been resolved. The digest excludes
 its own stored value. It identifies resolved composition rather than raw YAML
 bytes, a debug payload, or all runtime state. The CLI records it as invocation
 provenance before execution; cache and checkpoint identity have additional

docs/internal/dnd.md

@@ -29,11 +29,17 @@ The D&D registrar registers the family’s artifact codecs, extractors, typed
 append-order mergers, normalizers, validators, prompt assets, fallback LLM
 profile asset, and default validator chains. Each extractor and normalizer has
 a stable module spec, explicit execution class, strict option decoding, and a
-typed builder. Scene chunking, every extractor, and NPC, location, and item-registry
-normalization are registered as `llm_backed`; the remaining current D&D mergers
-and normalizers are `deterministic`. The metadata is available to catalog inspection and
-resolved-pipeline debug data and determines which selected bindings inherit the
-pipeline profile. Configuration remains the canonical owner of the exact keys,
+typed builder. Scene chunking, every extractor, and NPC, location, and
+item-registry normalization are registered as `llm_backed`; the remaining
+current D&D mergers and normalizers are `deterministic`. The metadata is
+available to catalog inspection and resolved-pipeline debug data and determines
+which selected bindings inherit the pipeline profile. The registry normalizers
+use `single_response_v1`, forwarding corrections to their reconciliation
+completion and retaining the accepted raw proposal only as an owned model
+candidate. When deterministic proposal safety or typed application rejects a
+group, they provide separate model-facing prose for a corrective module retry;
+internal issue categories, reason codes, and operator messages remain
+diagnostic-only. Configuration remains the canonical owner of the exact keys,
 profile precedence, and validator order.
 
 Private structured-LLM response schemas are deliberately minimal. They reject
@@ -42,6 +48,13 @@ unknown fields, while preserving semantic candidates for deterministic
 validation. Do not promote a private response envelope into a durable schema;
 the contracts above define durable data.
 
+A D&D producer that declares `single_response_v1` forwards any supplied
+semantic correction to its structured completion and returns an owned copy of
+that completion's exact validated raw response as its model candidate. It does
+not serialize normalized artifacts to create that candidate, so deterministic
+identity, evidence, warning, and durable-schema behavior remains separate from
+the model transport material.
+
 ## Prompt Construction
 
 D&D LLM-facing content lives beneath `assets/dnd/`. Each module contributes a
@@ -108,14 +121,27 @@ presentation, and final ephemeral generic transcript windows. These orders and
 cache controls are prompt behavior; change them only through the owning
 manifest and prompt declaration.
 
+NPC, item, and location registry reconciliation translate shared proposal
+safety categories into bounded prose that references only the response-local
+duplicate-group ordinals and candidate handles. Item reconciliation appends its
+typed rule that currency may be consolidated only with aliases of the same
+denomination and never with non-currency items. The next normalize attempt
+receives that prose with the exact defective proposal under the shared
+replacement-request protocol. Structurally invalid output has no valid proposal
+candidate and receives a fresh feedback-free attempt instead. If the stage
+budget is exhausted, safe groups stay applied, unsafe groups stay separate, and
+one fallback warning summarizes the final defect without raw model content.
+
 ## Evidence, Candidates, And Normalization
 
 The current transcript is the only durable evidence source. Extractors assign
-the current source identity, preserve candidate evidence ranges for validators,
-and canonically order or remove exact duplicate ranges without asking the
-model to repair semantic errors. Campaign context and generated artifacts may
-ground names or control routing, but they never establish evidence for a D&D
-result.
+the current source identity and losslessly order any reversed range whose two
+endpoints resolve in that source, using transcript position rather than numeric
+unit-ID order. They then canonically order ranges and remove exact duplicates.
+This routine canonicalization does not request a retry or emit a warning.
+Unresolvable or wrong-source ranges remain unchanged for validators to reject.
+Campaign context and generated artifacts may ground names or control routing,
+but they never establish evidence for a D&D result.
 
 Default chains keep responsibilities separate: structural validators assess the
 candidate, source-reference validators resolve cited ranges against the current
@@ -125,15 +151,119 @@ relatedness validators report advisory evidence concerns. The configured order
 is documented in
 [Configuration](../config.md#production-validator-keys-and-default-chains).
 
+The optional `extract/dnd/scene-descriptions/combat_semantics` validator is the
+D&D family's LLM-backed review of only combat versus non-combat classification.
+It selects the shared combat-policy prompt fragment and asks the model to
+classify the current chunk independently as `combat` or `non_combat` without
+receiving the proposed scene kind. Deterministic code compares that
+classification with the proposed kind and either approves it or produces the
+appropriate correction guidance. This keeps every schema-valid classification
+interpretable and avoids anchoring the reviewer on the producer's answer. It
+does not assess titles, summaries, non-combat subtype, or scene boundaries;
+deferred boundary-coherence review remains separate. It is opt-in;
+[Configuration](../config.md) owns selection and retry/failure behavior.
+
+### Combat-semantics provider evaluation
+
+The human-reviewed corpus at
+`internal/modules/dnd/validate/scenedescriptions/combat_semantics/testdata/evaluation_cases.json`
+owns the proposed kind, expected combat classification, and reviewer rationale
+for each synthetic case. Its package test validates the fixture contract only.
+Provider evaluation remains an explicit maintainer operation and must not be
+added to the default offline test suite.
+
+Use the following protocol before proposing default-chain inclusion:
+
+1. Record the Notarius commit, prompt and schema fingerprints, provider, model,
+   profile settings, reasoning effort, structural-repair setting, number of
+   repetitions, and evaluation date before collecting results. Do not revise
+   expected classifications merely to agree with provider output; a substantive
+   corpus correction requires independent human review.
+2. Exercise the production validator construction and prompt assets from an
+   explicitly invoked, disposable evaluation driver or test in the validator
+   package. For each corpus case, construct transcript source units from the
+   listed IDs and text, assign matching per-unit source references, and use
+   `source.MaterializeChunkPlan` with one range spanning those units. Construct
+   exactly one scene whose ID and source range match that chunk and whose kind
+   is the case's `proposed_kind`; title and summary may use fixed placeholders
+   because the validator neither receives nor evaluates them. Invoke the typed
+   validator at extract stage through the production registry and scheduled LLM
+   client. Do not commit provider credentials, generated source material, or an
+   always-on live test.
+3. Compare the model classification with `expected_classification`, then verify
+   that its deterministic comparison with `proposed_kind` yields the expected
+   approval or rejection direction.
+   Record an unexpected approval of an expected rejection as a false
+   acceptance, an unexpected rejection of an expected approval as a false
+   rejection, and any validator execution failure separately from semantic
+   accuracy. Retain per-case results so repeated trials and systematic failure
+   modes remain visible.
+4. Evaluate producer correction separately with representative complete
+   scene-description runs configured as shown in
+   [Configuration](../config.md#production-validator-keys-and-default-chains).
+   For every initial semantic rejection, record whether the next producer
+   attempt returns the requested combat status and is approved. Do not count a
+   PromptKit structural repair as a producer-correction attempt.
+5. Run the correction evaluation with debug capture enabled and without reused
+   extraction checkpoints. Record added validator and producer calls, elapsed
+   latency, and cumulative prompt, completion, cached, and total token usage
+   from the debug attempt records. Compare these values with an otherwise
+   identical run whose scene-description chain omits the semantic validator.
+
+The default-chain review must consider classification error, false acceptance,
+false rejection, execution failure, producer-correction success, added calls,
+latency, and token use together. A structurally successful provider run alone
+is not evidence that the validator should become a default.
+
+Every producer-correctable D&D rejection describes all currently detectable
+corrections in transcript-grounded domain terms, using contextual names,
+model-owned artifact fields, and source segment ranges when useful. Validators
+collect independent record defects in one pass so one retry does not merely
+reveal the next issue. Shared D&D diagnostic helpers keep repeated rules and
+record descriptions stable, de-duplicated, and bounded; each artifact family
+continues to own the semantic rule and its prose.
+
+Operator diagnostics and model guidance are separate products of the same
+assessment. Operator messages may use typed paths, reason details, and opaque
+application identities. Correction guidance must not copy those messages or
+ask the model to reproduce durable entity IDs, hashes, validator module keys,
+reason codes, or Go field paths. A registry-normalization rejection instead
+speaks in terms of the duplicate-group proposal response the normalizer can
+actually revise. Normalization-only deterministic invariants retain useful
+operator detail but do not imply that a model controls derived ordering or
+identity. Only bounded actionable semantic guidance is eligible for a
+correction prompt.
+
+Private LLM schemas use simple enums for closed categorical fields when the
+provider-compatible shape can express the rule directly. Deterministic typed
+validators retain the same checks as defense in depth and for non-LLM
+producers. Private schemas keep every property required and avoid optional
+properties, `uniqueItems`, and conditional cross-field logic.
+
+Item-occurrence shape validation groups repeated holder mistakes by occurrence
+kind and gives the producer the required JSON null/non-null relationship. It
+identifies affected records by contextual item name and cited transcript range,
+never by the deterministically attached durable item ID. Holder mistakes remain
+semantic rejections rather than silent rewrites because changing a holder can
+also change the meaning of the occurrence kind.
+
 Enemy-event extraction additionally rejects a second `engaged` observation for
 the same comparison identity within one scene-scoped result. Normalization may
 combine results from distinct scenes, so it intentionally does not apply that
 rule. Configuration owns the exact validator key and chain position.
 
+Extraction source-reference validators share one full-span chunk-containment
+policy. After ordinary reference validity succeeds, the policy resolves both
+endpoints through document order and requires every source unit in the
+inclusive range to be present in the current chunk. It does not assume numeric
+unit-ID ordering, mutate input, or weaken wrong-source and unresolved-reference
+validation. Scene descriptions remain separate because their validator owns an
+exact one-scene range contract rather than general extraction containment.
+
 Normalizers are deterministic for spells, combat turns, item occurrences, NPC
 occurrences, scene descriptions, enemy events, and location occurrences. They
 canonicalize display values and evidence, use source-document order for stable
-output, and issue bounded warnings for changes or collapsed duplicates. NPC,
+output, and emit bounded normalization observations for changes or collapsed duplicates. NPC,
 item, and location registry normalizers are intentional exceptions: each first
 produces a deterministic candidate set, then may use a bounded structured-LLM
 proposal to reconcile identity groups.
@@ -142,13 +272,16 @@ proposal to reconcile identity groups.
 
 The three registry normalizers instantiate the domain-neutral
 `internal/framework/semanticreconcile` engine with default bounds. Each
-eligible candidate receives a contiguous, one-based `candidate_id` for that
-request. The model sees that handle, the candidate label and source-free
-evidence ranges, plus bounded transcript windows; it returns only duplicate
+eligible candidate receives a contiguous, one-based `candidate_number` for that
+request, and candidate material declares the exact inclusive range. The model
+sees that handle, the candidate label and source-free evidence ranges, plus
+bounded transcript windows; it returns only duplicate
 groups of supplied handles and one supplied canonical handle per group. It
 never returns names, evidence, durable IDs, or replacement records. Identical
 labels and evidence remain independently selectable because their handles are
-distinct.
+distinct. Transcript unit `id` values and evidence `start_unit_id` and
+`end_unit_id` values are source positions in a separate namespace and are
+never valid candidate numbers.
 
 The generic core owns the mandatory handle protocol, candidate and transcript
 presentation, the private response schema, source-reference validation,

docs/internal/llm.md

@@ -42,6 +42,22 @@ observability. The adapter returns PromptKit’s validated raw bytes rather than
 re-encoding the decoded target. An empty optional material is represented as
 one space so its named input is retained by PromptKit.
 
+When a request includes semantic correction material, the adapter validates and
+defensively copies it before preparation, then appends exactly two messages
+after the ordinarily rendered prompt: the prior response as an assistant
+message and the correction guidance as a user message. Requests without a
+correction do not add messages or introduce caller roles. Ordinary request
+summaries record correction byte counts and digests only; complete messages are
+available solely in an explicitly requested debug trace.
+
+The adapter leaves the ordinary rendered message prefix, named inputs,
+variables, session, profile, execution overrides, prepared-execution path, and
+PromptKit repair policy unchanged for a corrected request. It never imports a
+PromptKit message type into a module or pipeline contract. PromptKit reports
+actual structural repair count and cumulative token usage per completion; the
+pipeline's safe terminal debug record projects those values without copying
+message content.
+
 Client construction may also receive a run-wide reasoning-effort override from
 the CLI factory boundary. The adapter copies the caller-owned pointer and
 creates a fresh PromptKit execution override for each request: a nil pointer
@@ -229,10 +245,15 @@ or corrective call remain provider-neutral operational errors with the same
 redaction boundary.
 
 Structural repair does not replace pipeline retry behavior: a binding's
-configured retry count reruns its complete stage attempt after an error or
-rejection. The pipeline owns attempt lifecycle, validation chains, and retry
-diagnostics; see [Pipeline Internals](pipeline.md#validation-retries-and-output)
-and the [binding reference](../config.md#module-bindings-and-validators).
+configured retry count reruns its complete stage attempt after an operational
+or structural error, module-requested retry, or actionable semantic rejection.
+An actionable module-requested retry and a validator rejection both use the
+same correction payload when the producer exposes an exact latest response;
+feedback-free module retries reconstruct the ordinary request without appended
+messages. The pipeline owns attempt lifecycle, validation chains, and retry
+diagnostics; see
+[Pipeline Internals](pipeline.md#validation-retries-and-output) and the
+[binding reference](../config.md#module-bindings-and-validators).
 
 ## Timeout Ownership
 
@@ -262,6 +283,13 @@ surfaced; when the completion already failed, its call error remains the
 result. Debug-bundle location, retention, and handling are operational concerns
 documented in [Operations](../operations.md#debug-bundles).
 
+The attempt-terminal summary is a separate safe trace record: it contains
+attempt kinds, validator outcome counts and reason codes, effective policy,
+terminal action, and repair/usage references. It excludes raw assistant
+responses and correction text. Those values can appear only in the explicitly
+requested detailed prompt and response artifacts, which require sensitive-data
+handling.
+
 Run manifests receive selected profile summaries, including optional effective
 backend and reasoning provenance, and component identities—not prompt, schema,
 source, reference, or response content. The published field semantics belong

docs/internal/modules.md

@@ -22,6 +22,15 @@ to bindings whose declared execution class is `llm_backed` and rejects a
 binding-specific profile on a deterministic module. The user-facing precedence
 contract belongs in [Configuration](../config.md#pipelines).
 
+An eligible LLM-backed chunk, extract, merge, or normalize producer may also
+declare correction protocol `single_response_v1`. That declaration is a
+promise that the implementation accepts one attempt-local semantic correction
+and returns an owned copy of the exact one model response that directly
+controlled the candidate. It must forward correction only to its structured
+completion request; it must not manufacture prior-response material by
+serializing a normalized artifact or expose opaque application IDs. Input,
+output, validator, and deterministic specs cannot declare the protocol.
+
 Implementations that accept options must provide both an option validator and
 a builder. The validator is used while resolving configuration; the builder
 decodes the same options and constructs the implementation from the prepared
@@ -36,6 +45,15 @@ they need, register each leaf implementation, and add any family-owned assets
 or default validator chains. They return contextual errors so production
 composition fails at startup rather than at the first run.
 
+A validator that returns a completed rejection must supply two separate
+bounded values: a stable `ReasonCode` for provenance and actionable
+`CorrectionGuidance` for the producer. Guidance identifies the semantic defect
+and the constraints on one complete corrected replacement. It must not contain
+validator keys, diagnostic paths, opaque application IDs, or other internal
+identifiers. An operator-facing `Message` may explain the same event, but the
+framework never copies it into a model request. Missing or invalid guidance is
+a validator contract failure.
+
 An artifact family can register an optional typed evidence projector alongside
 its codec. The projector returns defensive copies of the artifact's direct
 generic source references and must use the codec's exact Go type. It does not
@@ -77,19 +95,40 @@ source-backed artifact-family normalizer projects its deterministic records
 into contextual candidates and owned typed record envelopes, supplies its
 chosen prompt identity and resolved LLM profile, and constructs an engine with
 explicit limits. The core filters invalid evidence, assigns contiguous
-request-local integer handles, renders bounded candidate and transcript
-materials, invokes the structured-completion boundary, and assesses the
-returned duplicate groups into a stable non-overlapping plan.
+one-based request-local candidate numbers, renders bounded candidate and
+transcript materials, invokes the structured-completion boundary, and assesses
+the returned duplicate groups into a stable non-overlapping plan. Candidate
+material declares the exact inclusive number range for the request. Transcript
+unit IDs and evidence range endpoints remain source positions in a separate
+namespace and are never valid candidate numbers.
 
 The normalizer then applies that plan through a typed `ApplicationPolicy`. The
 core preserves ungrouped records, contribution order, and provenance while the
 artifact family owns group guards, field and evidence consolidation, durable
-ID derivation, retry and fallback presentation, warnings, and postconditions.
-Request-local handles do not enter the typed value or durable artifact. Fewer
-than two eligible candidates skips model invocation; exceeding a candidate or
-combined-material bound preserves the deterministic result under the family's
-fallback policy. Provider, transport, cancellation, and context-construction
-failures remain execution errors.
+ID derivation, retry and fallback presentation, classified diagnostics, and
+postconditions.
+These candidate numbers are the concrete private representation of ADR-0013's
+request-local handles; they do not enter the typed value or durable artifact.
+Fewer than two eligible candidates skips model invocation; exceeding a
+candidate or combined-material bound preserves the deterministic result under
+the family's fallback policy. Provider, transport, cancellation, and
+context-construction failures remain execution errors.
+
+When the engine actually makes a proposal call, its typed result carries the
+owned exact proposal response under the same correction contract as other
+eligible producers. Deterministic skip, limit, and fallback outcomes carry no
+model candidate, so a later rejection applies terminal policy without spending
+an ineffective semantic retry.
+
+When proposal assessment or typed group application rejects a structurally
+valid group, the normalizer may return its safe partial value with a
+module-requested retry. A feedback-capable directive supplies bounded
+model-facing correction guidance separately from operator diagnostics and
+retains the exact proposal response as its candidate. The shared stage retry
+mechanism appends that response and guidance to a fresh complete request. A
+structurally invalid completion has no valid candidate and therefore requests a
+feedback-free fresh attempt. On exhaustion, only the final safe fallback and
+its bounded process diagnostic advance to validation.
 
 The core supplies a conservative generic prompt and the single private
 response schema. A domain prompt may substitute its semantic instructions but
@@ -111,9 +150,13 @@ its domain prompt.
 3. Implement strict option decoding, construction, and the typed stage
    interface. Preserve caller ownership: do not retain mutable request data
    and return defensive copies where an implementation exposes stored data.
+   If declaring correction capability, forward the request correction and
+   retain only the exact validated response that controlled the result.
 4. Register the module through its typed registry helper and add it to the
    owning family registrar. Add a default validator chain only when that
    family owns the behavior; otherwise require an explicit compatible chain.
+   Every rejection path in a validator must provide actionable correction
+   guidance while retaining its stable internal reason code.
 5. Update the selectable-key and chain reference in
    [Configuration](../config.md#production-module-keys), the applicable
    integration contract, and focused tests. Keep the configuration document

docs/internal/pipeline.md

@@ -12,7 +12,7 @@ own durable output shapes. Concrete production extensions are covered by
 The pipeline framework accepts a resolved composition, registries, shared
 dependencies, input bytes, a supplied prompt session, and state/debug
 collaborators. It returns logical output files, normalized artifacts, recorded
-rejections and warnings, manifest provenance, and checkpoint decisions. The
+rejections, grouped diagnostics, manifest provenance, and checkpoint decisions. The
 CLI owns process arguments, configuration discovery, session resolution,
 physical roots, and placement of returned output files.
 
@@ -32,8 +32,14 @@ Resolution turns a configured pipeline profile into a **ResolvedPipeline**.
 It normalizes the pipeline and lane identities, applies stage defaults, selects
 requested lanes where that is supported, resolves validator chains, checks
 module capabilities and typed artifact compatibility, validates options, and
-assigns a deterministic resolved-composition digest. The resolved pipeline
-contains bindings and declared reference targets, not external reference bytes.
+assigns a deterministic resolved-composition digest. A correction protocol is
+selected from each eligible LLM-backed producer specification and becomes part
+of that resolved identity; only `single_response_v1` is currently supported.
+Preparation rejects an LLM-backed producer that combines a non-empty validator
+chain with positive producer retries unless it declares that protocol. Producers
+without validators or without retries remain valid without correction support.
+The resolved pipeline contains bindings and declared reference targets, not
+external reference bytes.
 After selection, the resolver applies command, binding, and pipeline profile
 precedence to LLM-backed bindings and validators only; prompt defaults remain
 an empty resolved binding profile. It resolves structural output repair
@@ -59,9 +65,11 @@ remains declared but has no bytes until its producing step completes.
 
 Preparation is the construction boundary. It validates the resolved shape and
 registry set, clones the resolved data, then constructs the input adapter,
-chunker, stage-local validators, every typed lane, and output encoder. Each
-registered builder receives its own cloned build request immediately before its
-module-owned code runs. Preparation also collects stable checkpoint
+chunker, stage-local validators, every typed lane, and output encoder. The
+prepared producer metadata preserves each selected correction protocol, and
+the resolved digest carrying that metadata participates in checkpoint identity.
+Each registered builder receives its own cloned build request immediately
+before its module-owned code runs. Preparation also collects stable checkpoint
 fingerprints. Missing registrations, incompatible typed entries, nil
 implementations, and constructor failures are reported before source parsing
 or any stage operation begins.
@@ -89,6 +97,13 @@ codec, checks its complete schema and media identity, and records a content
 digest plus bounded producer provenance. A missing, ambiguous, invalid, or
 incompatible producer prevents the consumer step from starting.
 
+Resolution receives only exact stage-and-lane operational reference overrides.
+The CLI may offer broader pipeline- or lane-scoped selectors, but expands and
+arbitrates those before entering the framework. External overrides are applied
+after configured external defaults and local bindings. They cannot coexist
+with a generated binding for the same target and slot, and external unbinds do
+not remove generated handoffs.
+
 ## Execution And Ordering
 
 The runner validates its input, installs no-op state collaborators when none
@@ -128,17 +143,55 @@ their target: chunks, codec-decoded typed candidates, or serialized codec
 bytes. Each typed validator receives a newly decoded value from the one
 candidate serialization for that attempt, while serialized validators receive
 separately owned representation bytes and schema metadata. They may approve,
-approve with warnings, reject, or fail. A rejection is an ordinary pipeline
-result; a validator error is a framework error.
+approve with warnings, reject, fail, or be skipped when a runtime prerequisite
+is unavailable. The shared executor settles every configured validator in
+order. A failed LLM-backed validator retries only itself against the same
+immutable candidate; it does not regenerate the producer or alter the
+validator request. Rejections stop that validator, while other configured
+validators still run. The executor retains ordered results, bounded
+deduplicated correction guidance from rejections, and only the final exhausted
+failure outcome for each validator. The correction builder keeps first
+occurrence order, omits internal reason codes, validator names, and operator
+messages, and requests one complete replacement. Missing guidance or an
+oversized aggregate is a framework contract error; guidance is never inferred
+or truncated.
 
 The runner applies the binding's retry policy around a stage operation and its
-complete validation chain. It preserves warnings only from the final accepted
-or rejected attempt. Cancellation stops retries. Normalizer-specific retry
-directives consume this same budget and validate any final safe fallback through
-the normalizer chain.
+complete validation chain. It preserves terminal diagnostics only from the final accepted
+or rejected attempt, plus one fixed validation-incomplete warning per validator whose execution
+budget was exhausted under `warn_continue`. Cancellation stops retries.
+Normalizer-specific retry directives consume this same budget and validate any
+final safe fallback through the normalizer chain. A directive may carry bounded
+correction guidance only when it also exposes the exact latest
+`single_response_v1` candidate. The state machine then uses the same replacement
+request shape as validator correction. A directive without guidance clears any
+prior correction and starts a fresh attempt, which preserves structural retry
+behavior when no valid response exists.
+
+The artifact-neutral producer-attempt state machine owns that shared budget,
+attempt provenance, semantic-correction material, and terminal-policy
+selection. It accepts producer and complete-validation closures, so artifact
+materialization, cache handling, checkpoints, and debug output stay at the
+operation boundary. It distinguishes operational, structural, module-requested,
+and validator-semantic retries. Model feedback from either semantic source is
+available only for a valid latest `single_response_v1` candidate. A
+deterministic or no-model validator rejection instead settles the semantic
+policy immediately, while a feedback-free module directive remains an ordinary
+fresh retry. Structural-output errors alone use the structural policy, and
+validation failure without rejection settles the validator-failure policy
+without regenerating the producer.
+
+Chunk planning uses this state machine for generated plans. A rejected or
+validation-incomplete automatic cache hit is not model material and therefore
+falls through to a fresh initial generation at producer attempt one; it neither
+receives a correction, consumes retry budget, promotes cached-candidate
+diagnostics, nor overwrites the stored record. An incomplete cache validation
+under `fail_run` terminates instead. Only a newly generated, completely
+validated plan is published to the chunk-plan store. Rejected plans never
+advance, and validation-incomplete plans remain unpublishable.
 
 After terminal lane work, the runner assembles manifest provenance, normalized
-artifacts, rejections, warnings, and an optional accepted chunk map. When an
+artifacts, rejections, final grouped diagnostics, and an optional accepted chunk map. When an
 output policy selected evidence lanes, it decodes accepted serialized normalize
 outputs through their registered codecs and invokes the prepared typed
 projectors. Rejected or absent lanes contribute nothing. This reconstruction is
@@ -149,6 +202,15 @@ The CLI publishes those files only after the runner returns without a framework
 error. Logical file names and schemas are defined by the [output integration
 contracts](../integrations/).
 
+For every completed producer disposition, the runner projects one bounded
+validation summary to the manifest, the affected rejection when present, and
+the CLI result receipt. The summary records status, configured-order rejecting
+validators and reason codes, incomplete validators, producer-attempt count,
+and terminal action. It contains no operator message, correction guidance, or
+model response. `complete`, `rejected`, and `incomplete` describe the final
+candidate disposition; a run-level `incomplete` status indicates at least one
+current-run output advanced under `warn_continue`.
+
 ## Checkpoint And Debug Hooks
 
 The runner receives checkpoint and debug interfaces rather than roots. It
@@ -158,10 +220,26 @@ handoff. Generated-reference dependencies participate in checkpoint decisions.
 Selective recomputation can require a canonical accepted normalized predecessor
 before a dependent lane starts.
 
+The runner writes successful checkpoint artifacts only after complete accepted
+validation. Chunk plans follow the same rule for publication. A rejection,
+invalid structured response, or incomplete validation is never reusable state;
+the current run may still hand off an otherwise valid `warn_continue` result
+according to its terminal policy. The runner carries private reuse eligibility
+through extract, merge, normalize, and generated-reference handoff. Any stage
+derived from incomplete validation skips both checkpoint lookup and all
+checkpoint publication even when that stage's own validation completes.
+External references and fully validated generated references remain eligible.
+Attempt debug records retain safe kind,
+validator, repair-usage, policy, and terminal-decision provenance. Full
+assistant and correction content remains confined to the requested detailed
+LLM trace.
+
 Debug recording is attempt-scoped and application-owned. A failure to persist
 required debug data is a framework error. State roots, persistence, reason-code
 meanings, resume, and cleanup are intentionally owned by
 [Run State Internals](state.md) and [Operations](../operations.md).
+Extract-validator trace scopes include the current chunk ordinal so concurrent
+chunks cannot overwrite one another's validator attempts or LLM artifacts.
 
 ## Invariants To Preserve
 

docs/internal/state.md

@@ -64,12 +64,13 @@ Ordinary resume loads extract, merge, and normalize checkpoints progressively
 and may execute later lane stages after an earlier cache miss. Selective
 recomputation instead asks the loader for the required producer's accepted
 normalize artifact. That lookup reuses the existing normalize files, requires
-workspace schema v3 plus an exact non-empty invocation identity, and deliberately
+workspace schema v4 plus an exact non-empty invocation identity, and deliberately
 does not require extract or merge checkpoint files or dependency fingerprints.
 The runner performs canonical codec and producer-provenance validation before
 cloning the artifact into normal step output. Success restores only stored
-normalize warnings and emits one normalize decision; failure retains the files,
-records the decision, and stops without executing the producer or consumer.
+normalize diagnostics and emits one normalize decision; failure retains the
+files, records the decision, and stops without executing the producer or
+consumer.
 
 The loader assigns a typed category and reason code at each validation site;
 diagnostic prose is not classified after the fact. The runner then applies
@@ -97,7 +98,7 @@ owns the operator workflow and stable reason-code meanings.
 
 `internal/core/debugbundle` allocates an explicitly requested per-run bundle
 with `summary/` and `trace/` roots. `SummaryWriter` persists redacted command,
-resolution, run, warning, and failure artifacts. `internal/framework/debug`
+resolution, run, final grouped diagnostic, and failure artifacts. `internal/framework/debug`
 implements the pipeline-facing trace recorder under the trace root.
 
 The CLI allocates a bundle before pipeline resolution and treats requested

docs/operations.md

@@ -105,10 +105,40 @@ and resolves configuration before module preparation and source parsing. It
 then performs any permitted cache lookup, executes the pipeline, and publishes
 logical output files only after a successful runner result.
 
-On success, the command reports the output bundle path. A warning-bearing run
-still succeeds and reports its warning count on standard error. Errors and
+On success, the command reports the output bundle path. A run with actionable
+process warnings still succeeds and reports warning-group and occurrence counts
+on standard error; advisory and observation findings do not produce a warning
+line. Errors and
 their exit classes are defined in the [CLI reference](cli.md#output-streams-and-exit-statuses).
 
+## Validation Retries And Terminal Outcomes
+
+Each producer binding has one outer **retries** budget. It covers complete
+producer attempts for operational failures, invalid structured output,
+normalizer fallback retry, and semantic correction. It is independent from
+PromptKit's structural-repair calls inside one completion and from an
+LLM-backed validator's own retry budget. A semantic correction rebuilds the
+ordinary producer request and supplies only the latest rejected model response
+plus aggregated validator guidance; it is not a conversation replay.
+
+After the applicable budgets are exhausted, the resolved
+[`validation_policy`](config.md#pipelines) determines the result. Structural
+failure and semantic rejection normally fail the run; an explicit
+`reject_output` records a rejection and allows unrelated work to finish. A
+validator execution failure normally uses `warn_continue`, which keeps an
+otherwise accepted result in the current run with `incomplete` validation
+provenance. It emits one bounded warning for every validator whose execution
+budget was exhausted. A corrected result that later passes validation does not
+retain diagnostics from abandoned attempts.
+
+Treat a successful process exit as a completed run, not as proof that every
+candidate was fully validated. Inspect the receipt's `validation_status`,
+`validation_summaries`, rejection count, and warning group and occurrence
+counts when an orchestrator requires complete validation. The durable fields
+and their meanings are owned by the
+[run-result receipt](integrations/run-result.md) and
+[published JSON output contract](integrations/json-output.md).
+
 ## Output Bundles
 
 Each successful run receives a generated safe run identifier and writes beneath:
@@ -158,7 +188,9 @@ The configured cache mode controls one invocation:
 A reused plan is still materialized and validated against the current source.
 If a prior plan no longer gives acceptable results, use a refresh run rather
 than editing cache files. Deleting a plan is recoverable but can repeat costly
-chunking work.
+chunking work. A plan accepted only under incomplete validation is not
+published, and a rejected cache hit falls through to ordinary generation rather
+than becoming a correction candidate.
 
 ## Checkpoint Recording, Resume, And Recompute
 
@@ -183,6 +215,12 @@ Reasoning-effort inheritance, replacement, and explicit clearing are distinct
 runtime identities, so checkpoints created under one state are not reused by
 either of the others.
 
+Only accepted, completely validated chunk, extract, merge, and normalize
+results are checkpointed for reuse. Rejected, structurally invalid, and
+validation-incomplete producer results remain non-reusable, even when a
+`warn_continue` result advanced during its original run. A resumed invocation
+therefore reruns that producer rather than treating degraded state as accepted.
+
 Checkpoint state is confined below an identity-specific path:
 
 ~~~
@@ -237,13 +275,16 @@ Only a [debug-enabled run](cli.md#run) creates a bundle:
 ~~~
 
 The summary contains redacted invocation and resolution information plus run,
-warning, checkpoint, chunk-plan, and terminal reporting artifacts. The trace
-contains allowlisted application diagnostic records and can include source or
-derived application data. Neither surface is a cache input. Do not treat a
-debug bundle as safe to share merely because its configuration summary is
-redacted. Invocation metadata omits reasoning effort when it is inherited,
-records the replacement value when one is supplied, and records an empty value
-when inherited reasoning was explicitly cleared.
+final grouped diagnostic, checkpoint, chunk-plan, and terminal reporting artifacts. Attempt
+terminal records contain bounded attempt kinds, validator outcomes, policy,
+decision, PromptKit repair count, and usage; they do not contain assistant
+responses or complete correction messages. The trace contains allowlisted
+application diagnostic records and can include source, model, and correction
+content. Neither surface is a cache input. Do not treat a debug bundle as safe
+to share merely because its configuration summary is redacted. Invocation
+metadata omits reasoning effort when it is inherited, records the replacement
+value when one is supplied, and records an empty value when inherited reasoning
+was explicitly cleared.
 
 Notarius never creates debug state without an explicit request and never
 automatically deletes a requested bundle. If allocation succeeds, the command

docs/policy/architecture.md

@@ -146,8 +146,9 @@ lanes, validators, and LLM profile: the canonical source digest selects the
 plan, while the current run still applies its configured chunk validators to
 the materialized chunks.
 
-The framework owns orchestration and handoff provenance. Modules return logical
-results and warnings; they do not own CLI reporting, physical output, cache, or
+The framework owns orchestration, origin enrichment, aggregation, and handoff
+provenance. Modules return logical results and classified diagnostics; they do
+not own CLI reporting, physical output, cache, or
 debug roots, durable file placement, or checkpoint and debug lifecycle.
 
 After pipeline-wide chunking, extraction uses bounded framework concurrency.
@@ -159,26 +160,53 @@ may overlap. The framework must not create unbounded goroutines per lane or
 chunk.
 
 Completion timing does not choose public ordering or errors. The coordinator
-orders accepted artifacts, warnings, rejections, checkpoint events, and
+orders accepted artifacts, grouped diagnostics, rejections, checkpoint events, and
 framework errors by stable pipeline scope. Rejections do not cancel unrelated
 work. A framework error cancels derived work, prevents undispatched work from
 starting, waits for started work, and prevents output encoding.
 
+Warnings are process-only signals: configuration degradation, approved fallback,
+or incomplete configured validation. Quality uncertainty and grounding findings
+are advisories; successful canonicalization and cleanup are observations.
+Modules choose that semantic classification, while the framework attaches
+origin, aggregates groups, enforces bounds, and presents final collections.
+An ordinary successful run therefore has zero warnings. See
+[ADR-0015](../adr/0015-separate-process-warnings-from-quality-diagnostics.md)
+for the decision rationale.
+
 ## Validation
 
 Validation is a framework-managed boundary around outputs from chunk, extract,
-merge, and normalize stages. Validators receive immutable stage output
-and make an explicit whole-output decision: approve, approve with warnings, or
-reject.
+merge, and normalize stages. Validators receive immutable stage output and
+make an explicit whole-output decision: approve, reject, fail, or skip when a
+runtime prerequisite is unavailable.
 
 Typed artifact validators receive the domain value directly. Chunk validators
 receive source-zone chunks, while serialized validators receive immutable
 representation bytes and declared schema metadata. A validator registered for
 one target or artifact kind cannot satisfy an incompatible selection.
 
-Rejection is a recorded pipeline outcome, not a framework execution error.
-Validator execution failures are framework errors. Rejected output does not
-advance to the next stage.
+The framework runs every applicable validator sequentially in configured order.
+It aggregates rejections, exhausted validator failures, and skips before the
+producer policy chooses a disposition. A completed rejection never advances.
+With no rejection, an exhausted validator failure may fail the run or, under
+the configured `warn_continue` policy, advance a structurally valid candidate
+with explicit incomplete-validation provenance. Validators report findings;
+they do not choose candidate disposition.
+
+A completed rejection supplies a stable reason code for internal provenance
+and bounded actionable correction guidance for the candidate producer. Reason
+codes, validator keys, and operator-facing messages remain diagnostic data;
+they are not model instructions. The framework constructs model-facing retry
+text only from the semantic guidance and fails the contract rather than
+inventing or truncating missing guidance.
+
+An LLM-backed module may also request a feedback-aware retry when its own
+deterministic translation or typed safety policy rejects a structurally valid
+model response. It must supply model-facing guidance separately from its
+reason code, operator message, and fallback diagnostics, together with the
+exact `single_response_v1` candidate. A feedback-free module retry remains
+valid when no exact candidate exists.
 
 Default validator chains are production composition policy and are registered
 centrally by stage and module. Configuration may replace a stage-local default,
@@ -199,6 +227,17 @@ PromptKit owns bounded structural correction within one structured completion.
 Notarius owns outer stage attempts, semantic validation, and acceptance policy;
 the two budgets must remain separate.
 
+An LLM-backed producer can participate in semantic correction only when it
+declares `single_response_v1` and returns the exact one response that directly
+controlled its candidate. On an actionable validator rejection or
+feedback-capable module retry, the framework rebuilds the ordinary request and
+appends only the latest defective response as an `assistant` message plus one
+aggregated `user` correction message. This is a fresh replacement request, not
+a growing conversation. The retry budgets, terminal policy, and sensitive-data
+rationale are recorded in
+[ADR-0014](../adr/0014-feedback-aware-validation-retries.md) and
+[ADR-0016](../adr/0016-feedback-aware-module-requested-retries.md).
+
 When a model selects an application entity, callers must supply a contextual
 selection and deterministically attach the opaque application identity whenever
 the selection resolves exactly. Models do not receive or reproduce opaque
@@ -233,7 +272,8 @@ invalid or incompatible.
 
 Run manifests record enough resolved pipeline, module, source, reference, and
 LLM provenance to make a run auditable after configuration changes. Manifests
-record identities and summaries rather than secret or large payload content.
+record identities and bounded validation summaries rather than secret, raw
+model, correction, or large payload content.
 
 ## State, Output, And Safety
 
@@ -253,12 +293,23 @@ an invocation that explicitly requests resume. Debug is never a cache input and
 is never created without an explicit request. Pipeline modules receive
 collaborator interfaces and never physical roots.
 
+Only accepted, completely validated producer output is reusable checkpoint or
+chunk-plan state. Rejected, structurally invalid, and validation-incomplete
+results cannot become cache or checkpoint inputs, even when a
+`warn_continue` result is allowed to advance in the current run. This
+ineligibility follows derived merge and normalize results and generated
+references for the remainder of the run: current-run handoff remains allowed,
+but no dependent cache or checkpoint may be loaded or published.
+
 Writes are atomic where practical. Paths for writes, moves, overwrites, and
 deletion must be narrow and explicit. Notarius never automatically deletes
 output or requested debug bundles; cache cleanup is explicit and recoverable.
 
 Secrets must not appear in errors, logs, output, cache, debug summaries,
-traces, manifests, documentation, examples, or redacted configuration. Debug
+traces, manifests, documentation, examples, or redacted configuration. Raw
+assistant responses and complete correction messages are attempt-local and are
+excluded from ordinary durable records and summaries; the requested detailed
+debug trace is the sole diagnostic surface allowed to retain them. Debug
 collection is allowlisted to application-owned payloads and must not capture
 unrelated process environment values or filesystem content. Trace data may
 contain application data and therefore inherits its sensitivity; operators own

docs/releases/v0.4.0.md

@@ -0,0 +1,83 @@
+# Notarius v0.4.0
+
+This release strengthens LLM reliability and validation throughout the
+configured pipeline, upgrades the PromptKit integration, and establishes the
+source-release and downstream-consumer workflows needed for broader D&D
+pipeline integration.
+
+## Summary
+
+Notarius now distinguishes PromptKit structural-output repair from
+application-owned semantic validation retries. Producer candidates can run
+through complete deterministic validator chains, receive bounded semantic
+correction guidance, and retry under explicit stage policies. Final run
+receipts and manifests preserve bounded validation provenance, while outputs
+that advance with incomplete validation remain available to the current run
+without entering reusable checkpoint state.
+
+The release also adds a maintained complete D&D subprocess-consumer workflow,
+diagnostic build versions, and the source-only release procedure used to
+publish this version.
+
+## Compatibility
+
+- Configuration files must use schema version 4. Version 3 is not decoded or
+  rewritten; rename the top-level `scriptorium` section to `promptkit` when
+  migrating. See [Configuration](../config.md#migrating-version-3-configuration).
+- PromptKit is pinned to v0.9.0. Operator profile files use PromptKit's v0.9.0
+  format and may use its profile-inheritance support. Notarius continues to
+  resolve operator profiles before embedded fallbacks.
+- Structural-output repair and semantic stage retries are separate bounded
+  mechanisms. Maintained production prompts request one structural repair by
+  default; explicit configuration can override the supported repair count.
+- Validation policy can now fail a run, reject an output, or permit an
+  otherwise valid candidate to advance with incomplete-validation provenance.
+  The application defaults are documented in
+  [Configuration](../config.md#pipelines).
+- The `notarius.run-result.v1` receipt remains at schema version 1 and adds
+  optional validation summaries plus a required validation-status field.
+  Consumers of this pre-release contract should follow the current
+  [run-result receipt](../integrations/run-result.md).
+- Existing D&D artifact schema identities remain unchanged. Validation and
+  producer-policy changes can nevertheless cause previously accepted weak
+  candidates to retry, reject, or fail instead.
+
+## Upgrade
+
+1. Migrate every Notarius configuration to version 4 and rename `scriptorium`
+   to `promptkit`.
+2. Review deployed PromptKit profiles against the pinned v0.9.0 profile format
+   and ensure their credential environment variables are available at run
+   time.
+3. Run `notarius config validate --config <path> --pipeline <id>` before the
+   first production invocation.
+4. Review `structured_output_repair_attempts`, producer retry counts, and
+   `validation_policy` wherever the deployment needs behavior different from
+   the documented defaults.
+5. Update subprocess consumers to inspect receipt `validation_status` and to
+   tolerate the optional bounded `validation_summaries` field. A consumer that
+   requires fully validated artifacts should require `approved`.
+
+## Changes
+
+- Upgraded PromptKit from v0.5.0 through v0.9.0 and adopted profile
+  inheritance, structured-output repair, typed error classification, and the
+  correction-aware completion protocol.
+- Added pipeline and binding configuration for structural repair and terminal
+  validation policy, with strict startup validation and effective-setting
+  provenance.
+- Added feedback-aware retries for chunking, extraction, merge, normalize, and
+  semantic reconciliation producers. Retry prompts contain the exact defective
+  response and actionable semantic correction guidance without exposing
+  internal reason codes or opaque entity identifiers.
+- Added complete validator-chain execution, validator retry handling, bounded
+  warnings, terminal dispositions, and durable validation summaries.
+- Prevented validation-incomplete artifacts and all derived lineage from
+  loading or publishing reusable checkpoints while preserving same-run
+  generated-reference handoff.
+- Tightened cached chunk-plan validation so only completely validated plans are
+  reused or replace stored plans.
+- Added a machine-readable subprocess receipt workflow and complete D&D
+  consumer documentation covering all maintained artifacts.
+- Added source-release checks, immutable lightweight-tag guidance, Linux and
+  Darwin build verification, and diagnostic `notarius --version` output.

docs/releases/v0.5.0.md

@@ -0,0 +1,85 @@
+# Notarius v0.5.0
+
+This release separates actionable process warnings from extraction-quality
+advisories and routine normalization observations, giving operators a quiet
+warning channel without discarding durable diagnostic detail.
+
+## Summary
+
+Notarius now carries one validated, origin-aware diagnostic contract from
+producers and validators through retries, reusable state, output publication,
+debug summaries, run receipts, and CLI presentation. Warnings are reserved for
+process degradation or incomplete configured work. Data-quality findings are
+advisories, and successful deterministic cleanup is recorded as observations.
+An ordinary successful run therefore reports zero warnings while retaining
+bounded diagnostic provenance for later review.
+
+The framework aggregates findings deterministically by their stable identity
+and complete pipeline origin, preserves exact occurrence counts, and retains
+bounded representative samples. Warning groups fail rather than truncate;
+advisory and observation representation is bounded with explicit truncation
+metadata and exact unrepresented-occurrence counts.
+
+## Compatibility
+
+- `warnings.json` now uses the incompatible grouped
+  `notarius.warnings.v2` envelope and contains process warnings only. Consumers
+  of the former flat warning payload must migrate to the current
+  [JSON output contract](../integrations/json-output.md).
+- The new `diagnostics.json` file uses `notarius.diagnostics.v1` and contains
+  advisory and observation groups. Production `index.json` files always expose
+  both `warnings_file` and `diagnostics_file`.
+- The machine-readable run receipt is now `notarius.run-result.v2`. It replaces
+  `warning_count` with exact warning group and occurrence counts and adds
+  advisory/observation group, occurrence, and truncation fields. See the
+  current [run-result receipt](../integrations/run-result.md).
+- Custom output modules must return their complete logical file set or an
+  error. The former `OutputResult.Warnings` field has been removed; an output
+  module cannot report a warning after serializing its output.
+- Reusable state now uses `notarius.workspace.v4` and chunk-plan records use
+  `notarius.chunk-plan.v3` so they can preserve structured diagnostics. Older
+  pre-release reusable state is not reused under these contracts; start with
+  clean state when deterministic continuity with an older workspace is not
+  required.
+- Validation acceptance, semantic retry budgets, rejection policy, and D&D
+  artifact schema identities are unchanged by this release.
+
+## Upgrade
+
+1. Update subprocess consumers to require `notarius.run-result.v2` and read
+   `warning_group_count`, `warning_occurrence_count`,
+   `diagnostic_group_count`, `diagnostic_occurrence_count`, and
+   `diagnostics_truncated`.
+2. Update output-bundle consumers to decode `notarius.warnings.v2`, discover
+   `diagnostics.json` through `index.json`, and treat diagnostics as review
+   information rather than process warnings.
+3. Update any custom output module for the removal of
+   `OutputResult.Warnings`; return an error when encoding cannot complete.
+4. Clear pre-release reusable state before the first upgraded production run
+   when deterministic continuity with an older workspace is not required.
+5. Run `notarius config validate --config <path> --pipeline <id>` and perform
+   one representative run before promoting the release in an automated
+   pipeline.
+
+## Changes
+
+- Added validated diagnostic dispositions, categories, origins, stable reason
+  codes, exact occurrence counts, and bounded representative samples.
+- Added deterministic run-level aggregation with separate limits for
+  actionable warning groups and advisory/observation groups.
+- Reclassified D&D source-relatedness and unresolved-identity findings as
+  data-quality advisories and routine normalization changes as observations.
+- Preserved structured diagnostics across producer retries, validation,
+  generated-reference handoff, checkpoints, chunk-plan reuse, and debug
+  summaries while discarding superseded-attempt findings.
+- Added grouped `warnings.json`, a new grouped `diagnostics.json`, and the
+  corresponding production index entries.
+- Upgraded the machine-readable run receipt and human CLI summary to report
+  exact warning and diagnostic counts without allowing advisory volume to
+  create warning output.
+- Removed post-encoding output warnings and hardened diagnostic validation,
+  overflow handling, aggregate memory bounds, and warning-file path
+  presentation.
+- Documented diagnostic ownership, classification, operator interpretation,
+  durable contracts, and architectural invariants in ADR-0015 and the
+  canonical CLI, operations, integration, and internal documentation.

docs/releases/v0.6.0.md

@@ -0,0 +1,84 @@
+# Notarius v0.6.0
+
+This release improves the reliability and ergonomics of unattended,
+subprocess-driven D&D extraction pipelines.
+
+## Summary
+
+Notarius now gives models more precise, semantically useful correction guidance
+when deterministic or model-backed validation rejects an otherwise structured
+candidate. Semantic reconciliation retries identify candidates with compact,
+request-local numbers, preserve valid candidates when a proposal cannot be
+repaired, and report bounded process warnings when reconciliation is
+incomplete. D&D extraction also canonicalizes safely resolvable reversed source
+ranges and applies more consistent schema, prompt, and validator constraints,
+improving successful operation with smaller models without weakening evidence
+validation.
+
+An optional LLM-backed combat-semantics validator can review whether scene
+descriptions correctly consolidate combat and apply the `combat` kind. For
+subprocess callers, command-line references now follow the pipeline-scoped
+configuration model: a shared reference can be supplied once and automatically
+reach every compatible selected target, while lane- and binding-specific forms
+remain available for exceptions.
+
+## Compatibility
+
+- The meaning of an unqualified `--reference slot=path` or
+  `--without-reference slot` selector has changed. It now applies to every
+  selected pipeline target that declares the slot instead of requiring exactly
+  one matching target. Review callers that relied on ambiguity rejection or a
+  unique implicit target; the current selector contract is documented in the
+  [CLI reference](../cli.md).
+- The stage-wide `merge.slot` CLI reference shorthand has been removed. Use
+  `lane.merge.slot` for an exact merge binding, `lane.slot` for all compatible
+  bindings in one lane, or an unqualified `slot` for pipeline scope.
+- Command-line file references cannot replace, remove, or coexist with a
+  generated artifact handoff for the same concrete target and slot. Resolution
+  reports the conflict so the caller can narrow or remove the selector.
+- The private LLM-facing semantic-reconciliation contract now uses contiguous
+  `candidate_number` values rather than application identities. This is not a
+  durable or operator-configurable contract and requires no operator action.
+- Durable D&D artifact schemas, the `notarius.run-result.v2` receipt,
+  `notarius.warnings.v2`, and `notarius.diagnostics.v1` are unchanged from
+  `v0.5.0`.
+
+## Upgrade
+
+1. Update subprocess commands to provide shared inputs once with
+   `--reference slot=path`, and convert any stage-wide `merge.slot` selector to
+   an explicit supported scope. The [subprocess consumer guide](../consumers/subprocess.md)
+   and [complete D&D consumer guide](../consumers/dnd-pipeline.md) show the
+   current invocation pattern.
+2. Review CLI reference overrides that overlap generated same-run references;
+   narrow or remove an external selector rather than attempting to replace the
+   generated handoff.
+3. Run `notarius config validate --config <path> --pipeline <id>` for every
+   deployed pipeline configuration.
+4. Optionally add
+   `extract/dnd/scene-descriptions/combat_semantics` to the scene-description
+   validator chain where the deployment wants LLM-backed combat-scene review;
+   see the [configuration reference](../config.md).
+5. Perform a representative run with the deployed model profile and inspect
+   the machine-readable receipt, warnings, and diagnostics before promotion.
+
+## Changes
+
+- Added feedback-aware, module-requested retry support for normalizers while
+  keeping retry budgets bounded and preserving attempt diagnostics.
+- Reworked semantic reconciliation around compact, contiguous candidate
+  numbers, explicit namespace guidance, complete proposal validation, and
+  model-facing retry feedback that omits opaque internal identities.
+- Preserved candidates separately when an exhausted semantic proposal cannot
+  be safely applied, with a bounded process warning describing the fallback.
+- Canonicalized safely resolvable reversed D&D source-reference endpoints
+  before deterministic coverage validation and clarified the shared evidence
+  prompt.
+- Improved D&D schemas, prompts, and validators with closed-value constraints,
+  contextual correction guidance, shared source-range coverage logic, and
+  clearer item-holder transition rules.
+- Added the optional D&D scene-description combat-semantics validator, shared
+  combat policy assets, evaluation fixtures, and retry-path coverage.
+- Made unqualified CLI references pipeline-scoped, added hierarchical lane and
+  exact-binding selectors, defined deterministic override precedence, and
+  documented first-class subprocess use.

docs/roadmap/archive/warning-signal-and-presentation-audit.md

@@ -0,0 +1,648 @@
+# Warning Signal And Presentation Audit
+
+## Executive Assessment
+
+Notarius warning execution is mechanically stronger than its operator-facing
+presentation. Terminal-attempt promotion, stable ordering after concurrent
+work, checkpoint replay, validation summaries, and debug retention are all
+substantially correct. The audit found no general duplicate-append defect in
+the extract, merge, or normalize handoffs and no leakage of abandoned-attempt
+warnings into a successful result.
+
+The warning channel itself is not coherent. One flat `contracts.Warning` type
+currently represents at least four materially different concepts:
+
+- actionable degradation or incomplete validation;
+- heuristic data-quality doubt;
+- successful but potentially reviewable fallback; and
+- routine canonicalization, ordering, and deduplication observations.
+
+That conflation is the primary reason successful runs produce a count that is
+large but operationally weak. The maintained complete D&D example demonstrates
+the problem without a live provider: an approved run with no rejected outputs
+published 12 warning records, all from three advisory relatedness checks. An
+operator separately reported a successful complete D&D run with 10 outputs,
+one rejection, and 85 warnings. The production bundle for that run was not
+available in this environment, so its reason-code distribution could not be
+measured.
+
+The current implementation also has four correctness or robustness gaps:
+
+1. warning records lose stage, step, lane, module, validator, and chunk
+   provenance when promoted, which makes safe aggregation and diagnosis
+   impossible from `warnings.json` alone;
+2. there is no framework-level validation or aggregate bound, and the NPC- and
+   spell-relatedness validators bypass the D&D warning limiter entirely;
+3. warnings returned by an output encoder are added after `warnings.json` has
+   already been encoded, so the receipt, stderr, debug bundle, and published
+   warning file can disagree; and
+4. a skipped validator contributes to `incomplete` validation but does not
+   receive the warning generated for an exhausted validator failure.
+
+The recommended end state is a structured diagnostic contract with explicit
+disposition, category, origin, occurrence count, and bounded samples. Warnings
+are reserved for process-level degradation or incompleteness. LLM-judged or
+deterministically inferred extraction-quality signals are advisories, never
+warnings, and routine normalization observations remain inspectable without
+being reported as top-level warnings. An ordinary successful run in which all
+configured work completes normally should therefore report zero warnings. This
+is an architectural and durable-contract change, not merely revised CLI prose.
+
+## Evidence And Limits
+
+The audit used:
+
+- a complete static search of production `contracts.Warning` constructors,
+  reason-code constants, result fields, and promotion sites under `internal/`;
+- call-path inspection through producer attempts, validators, lane
+  coordination, chunk-plan reuse, checkpoints, output encoding, debug output,
+  CLI presentation, and run-result construction;
+- the maintained complete and minimal D&D examples with offline fake LLMs;
+- focused deterministic tests for warning bounds, semantic-reconciliation
+  fallback, `warn_continue`, semantic retries, terminal rejection, concurrency
+  ordering, and checkpoint reuse; and
+- the operator-provided observation of an 85-warning complete D&D run.
+
+No provider-backed production run was attempted because this environment has
+no API key. Consequently, the audit can establish warning mechanics, possible
+multiplicity, synthetic volume, and obvious heuristic limitations, but cannot
+estimate production frequency or the real false-positive rate of individual
+D&D advisories. Those measurements are not required to choose the recommended
+architecture; they are required before strengthening any heuristic advisory
+into a rejection or setting a numerical production acceptance target.
+
+## Complete Warning-Producer Inventory
+
+### Framework And Generic Boundaries
+
+| Producer | Reason code | Trigger and consequence | Multiplicity and bound | Current surfaces and coverage |
+| --- | --- | --- | --- | --- |
+| Reference materialization in `internal/framework/pipeline/references.go` | `empty_reference` | A bound external reference is a valid, accepted media type but contains zero bytes. The prompt may receive materially incomplete context. | One per empty bound file; finite by configuration but no shared run-level cap. | Enters `RunInput.Warnings`; reference tests protect contextual scope. |
+| Producer-attempt policy in `internal/framework/pipeline/producer_attempts.go` | `validator_execution_incomplete` | An applicable validator exhausted its execution budget and `warn_continue` accepted the otherwise valid candidate. | One per failed validator on each terminal candidate. An extract chain can multiply this by chunks and lanes. There is no global cap. | Durable warning, receipt count, stderr, debug, and validation summary. `TestWarnContinueRecordsOneWarningForEachExhaustedValidator` covers failures. |
+| Chunk, extract, merge, normalize, and output module result contracts | Module-defined | A module may return arbitrary warnings with its successful candidate. | No contract validation, message limit, per-result cap, or global cap. Current production modules are inventoried below. | Terminal-attempt filtering and concurrency ordering are well tested. |
+| Production JSON output encoder | None | The encoder copies incoming warnings into `warnings.json`; it does not currently create warnings. | Same count as its input. | JSON encoder and assembled-pipeline tests compare the incoming run warnings with the published file. |
+| Output encoder result contract | Module-defined | Any output encoder may return warnings discovered during encoding. | Unbounded by contract. No production encoder currently exercises this capability. | Appended to final `RunOutput` only after logical files were encoded; this is the cross-surface defect described in AUD-WARN-004. |
+
+Input adapters and production mergers do not currently have independent
+warning producers. Chunk-plan and checkpoint decisions are structured manifest
+or debug provenance rather than warnings. Cancellation and hard persistence,
+reference, parsing, serialization, and provider failures remain errors.
+
+### D&D Extraction Gates
+
+| Producer | Reason code | Trigger and consequence | Multiplicity and bound |
+| --- | --- | --- | --- |
+| `dnd/combat-turns` extractor | `scene_classification_unavailable` | The chunk has no exact matching scene-description classification. The extractor returns an empty accepted result and skips the LLM, so combat-turn output may be incomplete. | At most one per chunk for this lane. |
+| `dnd/enemy-events` extractor | `scene_classification_unavailable` | The same missing or mismatched scene gate causes accepted empty enemy-event output. | At most one per chunk for this lane. |
+
+An exact non-combat classification produces an intentional empty result without
+a warning. An exact combat classification proceeds normally. The two producers
+share a code and operator consequence but use different messages; their module
+origins are not retained in the final warning record.
+
+### D&D Source-Relatedness Validators
+
+All ten relatedness validators are deterministic advisories: they approve the
+candidate and warn when contextual prose or an entity name is not lexically
+present in cited text. Shape and source-reference failures are deliberately
+left to blocking validators earlier in the chain. The same relatedness
+validator is registered in both the extract and normalize default chain for
+each artifact family in `internal/modules/dnd/register/chains.go`.
+
+| Artifact family | Warning reason | Per-record trigger | Local bound | Omission reason |
+| --- | --- | --- | --- | --- |
+| Combat turns | `combat_turn_not_near_source` | Actor token sequence absent | 20 per validator invocation | `combat_turn_relatedness_warnings_omitted` |
+| Enemy events | `enemy_event_not_near_source` | Subject token sequence absent | 20 | `enemy_event_relatedness_warnings_omitted` |
+| Item occurrences | `item_occurrence_source_unrelated` | Item name token sequence absent | 20 | `item_occurrence_relatedness_warnings_omitted` |
+| Item registry | `item_not_near_source` | Item name token sequence absent | 20 | `item_relatedness_warnings_omitted` |
+| Location occurrences | `location_occurrence_not_near_source` | Location name token sequence absent | 20 | `location_occurrence_relatedness_warnings_omitted` |
+| Location registry | `location_not_near_source` | Location name token sequence absent | 20 | `location_relatedness_warnings_omitted` |
+| NPC occurrences | `npc_occurrence_not_near_source` | NPC name token sequence absent | 20 | `npc_occurrence_relatedness_warnings_omitted` |
+| NPC registry | `npc_not_near_source` | NPC name token sequence absent | **Unbounded** | None |
+| Scene descriptions | `scene_description_not_near_source` | No significant title or summary token appears; up to two findings per scene | 20 | `scene_description_relatedness_warnings_omitted` |
+| Spells | `spell_not_near_source` | Spell-name token sequence absent | **Unbounded** | None |
+
+The eight limiter-generated omission records are presentation artifacts, not
+new source-relatedness conditions. They occupy a warning slot and make list
+length differ from the actual occurrence count.
+
+### D&D Normalizers
+
+Every production D&D normalizer bounds its returned warning slice to 20 through
+`internal/modules/dnd/shared/diagnostics`, including a final omission record
+when needed. Registry semantic retries reserve space for their fallback
+warning. The following table is complete by semantically distinct condition;
+codes listed together are parallel artifact-family variants.
+
+| Condition | Reason codes | Result impact | Current classification assessment |
+| --- | --- | --- | --- |
+| Display or field whitespace/name canonicalization | `npc_fields_normalized`, `item_fields_normalized`, `location_fields_normalized`, `spell_name_canonicalized`, `combat_actor_canonicalized`, `enemy_event_name_canonicalized`, `item_occurrence_name_canonicalized`, `location_occurrence_name_canonicalized`, `scene_description_prose_normalized` | Deterministic successful mutation. The item-occurrence code can also describe `from`/`to` whitespace, not only the item name. | Routine observation. |
+| Durable ID recomputation | `npc_id_recomputed`, `item_id_recomputed`, `location_id_recomputed` | Restores the deterministic name-derived ID. | Routine observation; invalid identity is separately rejected by default chains. |
+| Source-reference sorting or deduplication | `source_references_normalized` | Sorts and removes exact duplicate references while deliberately preserving invalid references for their validators. | Routine observation. Shared code is useful but ambiguous without producer origin. |
+| Canonical record ordering | `combat_turns_reordered`, `enemy_events_reordered`, `item_occurrences_reordered`, `location_occurrences_reordered`, `npc_occurrences_reordered`, `scene_description_order_normalized` | Deterministic order changes only. | Routine observation. |
+| Exact or approved semantic duplicate consolidation | `duplicate_npc_collapsed`, `duplicate_item_collapsed`, `duplicate_location_collapsed`, `duplicate_spell_cast_collapsed`, `duplicate_combat_turn_collapsed`, `duplicate_enemy_event_collapsed`, `duplicate_item_occurrence_collapsed`, `duplicate_location_occurrence_collapsed`, `duplicate_npc_occurrence_collapsed`, `scene_description_duplicate_collapsed` | Removes duplicate records and preserves or combines canonical evidence according to the artifact policy. Registry codes cover both exact and accepted semantic consolidation. | Durable normalization observation; not normally operator-actionable. |
+| Unresolved external membership | `spell_name_unresolved`, `item_occurrence_unknown_item_id`, `location_occurrence_unknown_location_id` | The value is preserved but is not grounded in the effective catalog or registry. Default chains normally reject the same condition before normalization; it remains reachable with validator overrides or defensive direct use. | Actionable data-quality warning. |
+| Unsafe currency consolidation proposal | `item_semantic_proposal_invalid` | The proposed group is rejected and all records are preserved because denominations or currency/non-currency members are incompatible. The same code is also used internally as a retry reason. | Advisory about model proposal quality; no accepted-data loss. The control and diagnostic meanings should be separated. |
+| Semantic reconciliation unavailable or exhausted | `npc_semantic_reconciliation_exhausted`, `item_semantic_reconciliation_exhausted`, `location_semantic_reconciliation_exhausted` | The safe deterministic result is accepted, but possible semantic duplicates remain. | Actionable fallback warning. |
+| Local warning truncation | `npc_normalization_warnings_omitted`, `item_normalization_warnings_omitted`, `location_normalization_warnings_omitted`, `spell_normalization_warnings_omitted`, `combat_turn_normalization_warnings_omitted`, `enemy_event_normalization_warnings_omitted`, `item_occurrence_normalization_warnings_omitted`, `location_occurrence_normalization_warnings_omitted`, `npc_occurrence_normalization_warnings_omitted`, `scene_description_normalization_warnings_omitted` | Reports that individual records were omitted from presentation. | Group metadata, not an independent warning. |
+
+No production D&D normalization warning exposes raw model responses,
+correction guidance, or provider errors. Most dynamic names are quoted and
+truncated by the shared helper. That local discipline is not enforced by the
+generic warning contract, and the spell relatedness message does not use the
+shared truncation helper.
+
+## Warning Propagation And Surface Map
+
+```text
+external-reference warnings -----------------------------+
+                                                           |
+module candidate warnings -> validation chain warnings     |
+             |                      |                       |
+             +---- producer-attempt terminal policy -------+
+                                  |                        |
+                   accepted / terminal rejection only      |
+                                  |                        |
+              chunk or lane result in canonical order      |
+                                  |                        |
+       checkpoint record/replay and ordered step merge     |
+                                  |                        |
+                            RunOutput.Warnings <------------+
+                                  |
+                   OutputRequest -> output encoder
+                                  |             |
+                         warnings.json     OutputResult.Warnings
+                                                |
+                              appended to final RunOutput only
+                                  |
+                 receipt, stderr, final debug warning summary
+```
+
+### Attempts And Validation
+
+- `runProducerAttempts` promotes only the terminal accepted or terminal
+  rejected candidate's module and completed-validator warnings. Operational,
+  structural, semantic, and module-directed attempts that are superseded are
+  retained in attempt debug artifacts but not in the final collection.
+- A module-directed semantic-reconciliation retry adds its fallback warning
+  only when no retry remains. Earlier attempt warnings are discarded.
+- On `warn_continue`, warnings from the otherwise accepted candidate and
+  completed approved or rejected validators are retained. One fixed,
+  non-sensitive `validator_execution_incomplete` warning is added for every
+  failed validator. Skipped validators affect the validation summary and final
+  `incomplete` status but do not receive such a warning.
+- A semantic terminal rejection retains only warnings from that rejected
+  attempt. Structural rejection after producer failure cannot retain a
+  candidate warning because no valid candidate result exists.
+
+These behaviors are protected by the producer-attempt, extract-handoff,
+rejection-warning, normalize-retry, and attempt-debug tests. They are the right
+foundation for the redesign and should not be replaced with early-exit or
+all-attempt accumulation.
+
+### Concurrency And Ordering
+
+Extract jobs are dispatched chunk-first and lane-second. Results are stored by
+chunk index, finalized in ascending chunk order, and lane continuations are
+merged into a slice indexed by configured lane order. Pipeline steps run in
+configured order. The resulting public order is therefore:
+
+1. pre-run reference warnings;
+2. chunk-stage warnings;
+3. step order;
+4. configured lane order within each step;
+5. chunk order within extract;
+6. merge warnings; then
+7. normalize warnings; followed by any output-result warnings.
+
+`TestRunnerBoundsExtractJobsAndStabilizesReverseCompletion` exercises warning
+order under reversed completion. No completion-order leak was found.
+
+### Chunk Plans And Checkpoints
+
+- A reusable chunk plan stores producer warnings only. Current validators run
+  again, and their current warnings are appended. Warnings from a cached plan
+  candidate that fails current validation are discarded before regeneration.
+- Accepted extract, merge, and normalize checkpoints store the terminal
+  warnings for their stage. Reuse loads and appends those warnings once at the
+  same logical handoff. Tests compare fresh and resumed warning collections and
+  preserve their order.
+- Validation-incomplete accepted outputs are not reusable, preventing a later
+  run from silently treating incomplete validation as complete.
+- Required accepted-normalize hydration replays that normalize checkpoint's
+  warnings; checkpoint decisions separately expose that reuse occurred.
+
+The recommended redesign should keep fresh and resumed logical diagnostics
+equivalent. Whether a result was reused belongs in checkpoint provenance, not
+in the diagnostic grouping key; adding a `reused` distinction would fragment
+groups and make equivalent runs present differently.
+
+### Terminal Surfaces
+
+| Surface | Current content | Audience | Audit result |
+| --- | --- | --- | --- |
+| `RunOutput.Warnings` | Flat final slice | Framework and CLI | Canonical in-memory list, but lacks origin and bounds. |
+| Published `warnings.json` | Object containing the warnings passed into the output encoder | Durable consumers | Exact for the production JSON encoder unless the encoder itself returns warnings. |
+| `index.json` | Path to `warnings.json` | Durable consumers | Stable discovery path; no separate diagnostic-detail path. |
+| Run-result v1 | `warning_count = len(final RunOutput.Warnings)` | Subprocess callers | Count only; no group/occurrence distinction. |
+| Human stderr | `run completed with N warning(s)` | Operators | Count only and no direct detail path. Successful exit remains zero. |
+| Manifest | Validation and rejection summaries, no warning collection | Durable provenance | Correctly avoids duplicating the flat list. |
+| Debug summary `warnings.json` | Raw final warning array | Operators/developers | Includes final output-result warnings and can therefore differ from published `warnings.json`. |
+| Debug run report | Final warning count | Operators/developers | Same final slice length as receipt and stderr. |
+| Attempt/stage debug | Candidate-local warning detail and origin in path/envelope | Forensics | Sufficient to diagnose provenance, but debug capture is optional and is not a durable consumer contract. |
+
+## Empirical Measurements
+
+### Offline And Synthetic Runs
+
+| Scenario | Result | What it establishes |
+| --- | --- | --- |
+| Maintained complete D&D config and transcript with the repository's offline fake LLM | Approved, 10 normalized outputs, 0 rejected outputs, 12 warnings; receipt, stderr, and published file all reported 12 | An ordinary structurally successful workflow can be noisy without fallback or incomplete validation. |
+| Same complete run, grouped after publication | Three reason codes, seven exact `(reason, scope, message)` tuples, maximum exact-tuple repetition of three | The flat count materially overstates distinct operator conditions. Scope resets within chunks and does not identify origin. |
+| Maintained focused scene-description workflow | Approved, one normalized output, 0 warnings | The warning channel can be quiet when synthetic model text is lexically grounded. |
+| Generic warning publication contract | One warning reaches successful stderr, durable output, and debug summary | The ordinary pre-output path is consistent. |
+| NPC semantic-reconciliation candidate-limit fallback | No LLM call, all records preserved, one exhaustion warning, total warnings no greater than 20 | Fallback is bounded and materially different from routine normalization. |
+| `warn_continue` with two failed validators and one skipped validator | Validation status contains all three incomplete validators; warning slice contains two execution-incomplete records | Current warning count does not describe all incomplete validation. |
+| Retrying extract candidate | Two producer attempts; only the accepted attempt's one warning is final | Retry does not amplify abandoned warnings. |
+| Terminal semantic rejection | Only the final rejected attempt's operation and validator warnings are final | Rejection diagnostics are retained without retaining superseded warnings. |
+| Fresh versus reused extract checkpoint | Warning collections are deeply equal | Checkpoint replay does not itself amplify warnings. |
+| Spell normalizer with 21 unresolved entries | 20 records: 19 samples plus one omission record saying two additional warnings were omitted | `warning_count` is neither exact occurrence count nor distinct-condition count. |
+
+The focused audit tests passed in `internal/cli`,
+`internal/framework/pipeline`, the NPC-registry and spell normalizers, and all
+D&D packages.
+
+### Bounded Sample Review
+
+The complete offline D&D run produced:
+
+| Reason | Count | Sample | Review |
+| --- | ---: | --- | --- |
+| `location_not_near_source` | 4 | `Moon Gate` was absent from cited text | Correctly identifies deliberately unsupported fake output. Three records shared the same exact tuple because chunk and stage origin were lost. |
+| `location_occurrence_not_near_source` | 4 | A `Moon Gate` visit was absent from cited text | Correctly identifies the same unsupported registry-driven occurrence, but repeats the same operator concern across extraction and normalization. |
+| `scene_description_not_near_source` | 4 | A title or summary had no significant exact token in cited text | Mixed value. Generic `session scene` prose is ungrounded, while `Arrival` versus transcript `arrive` illustrates an expected lexical false positive. |
+
+This fake workflow is an integration fixture, not a model-quality benchmark.
+It nonetheless proves that the checks carry useful evidence while being too
+imprecise and repetitive to serve as one-warning-per-record operator alerts.
+
+### Production Evidence Still Needed
+
+The reported 85-warning run establishes that high volume occurs in practice,
+but the following remain unknown:
+
+- dominant production reason codes and stage/lane sources;
+- unique group count versus repeated occurrence count;
+- false-positive rate for each relatedness family;
+- how much volume comes from normalization observations versus advisories;
+- whether fresh and resumed production runs remain equivalent; and
+- a defensible numerical acceptance target.
+
+If further data is worthwhile, the operator can supply the v1 receipt,
+`warnings.json`, and manifest validation summaries without supplying transcript
+or lane artifacts. An initial privacy-preserving report should group by reason
+code and normalized scope family, count exact repeated tuples, and omit message
+text. Reviewing heuristic precision requires a separately approved bounded
+sample with its cited source context.
+
+## Classification Of Current Conditions
+
+| Target disposition | Current families | Result impact | Operator action | Durable placement |
+| --- | --- | --- | --- | --- |
+| **Warning** | Empty reference; validator failure or skip accepted under `warn_continue`; unavailable required scene classification; exhausted semantic reconciliation | A configured process completed under an allowed degraded or incomplete policy rather than completing normally | Correct reference/configuration, inspect provider/validator, or rerun | Actionable `warnings.json`, receipt summary, stderr summary, debug |
+| **Advisory** | Source-relatedness heuristics; unresolved spell or registry membership; guarded invalid semantic proposal; any future LLM-judged uncertainty or extraction-quality signal | Uncertain data quality or poor model proposal, but accepted data is structurally valid and deterministic guards prevented unsafe mutation | Optional model/source review; no routine action for every record | Durable diagnostic detail and debug; never a top-level warning |
+| **Observation** | Whitespace/name/ID/source-reference canonicalization; canonical ordering; exact and approved semantic duplicate consolidation | Successful intended normalization | None under normal operation | Durable bounded normalization diagnostics or debug; no stderr warning |
+| **Not a diagnostic** | Rejection, invalid structure, cancellation, persistence error, provider failure under fail-run policy | Candidate or run did not complete according to policy | Inspect rejection/error and retry or correct input/configuration | Existing rejection, validation summary, error, and debug contracts |
+
+Exact and semantic duplicate consolidation should remain distinguishable in
+category or reason metadata even though both are observations. Semantic
+reconciliation exhaustion remains a warning because a capability was not
+applied; successful approved consolidation is an observation because it is the
+normalizer's intended work.
+
+## Ranked Findings
+
+### AUD-WARN-001 — The Flat Warning Type Destroys Signal Quality
+
+- **Priority:** High operator impact; high implementation leverage.
+- **Evidence:** `contracts.Warning` has only scope, reason, and message. Routine
+  normalizer changes, heuristic doubt, fallback, and incomplete validation all
+  enter the same slice and the same CLI count. The offline complete run's 12
+  records were all advisories; the operator observed 85 records in a successful
+  real run.
+- **Impact:** Operators cannot tell whether a warning requires a rerun, a
+  configuration repair, optional review, or no action. Repeated routine output
+  trains them to ignore the channel.
+- **Recommendation:** Replace the flat semantic contract with explicit
+  `warning`, `advisory`, and `observation` dispositions plus a small category
+  vocabulary. Do not infer disposition from message text or require every
+  downstream consumer to maintain a reason-code policy table.
+
+### AUD-WARN-002 — Warning Records Lose The Origin Needed For Diagnosis And Aggregation
+
+- **Priority:** High correctness and usability impact.
+- **Evidence:** The runner knows stage, step, lane, module, validator, chunk ID,
+  and chunk index at promotion time, but `terminalWarnings` flattens module and
+  validator records into `[]contracts.Warning`. Per-chunk scopes such as
+  `locations[0]` and `occurrences[0]` then repeat without identifying their
+  chunk or producer. `source_references_normalized` is intentionally shared
+  across families and is therefore especially ambiguous.
+- **Impact:** `warnings.json` cannot answer which stage or module produced a
+  record. Message- or scope-based deduplication would merge unrelated findings
+  or retain accidental duplicates.
+- **Recommendation:** Keep module findings free of framework context, then have
+  the framework add a structured origin envelope before promotion. Validator
+  findings must retain validator identity instead of passing through
+  `validationReport.Warnings()` as a flat slice.
+
+### AUD-WARN-003 — Warning Volume Is Not End-To-End Bounded Or Validated
+
+- **Priority:** High robustness impact; medium immediate likelihood.
+- **Evidence:** D&D's `LimitWarnings` caps most individual producers at 20, but
+  NPC- and spell-relatedness return one warning per record without the helper.
+  Every extract validator is invoked per chunk, all ten relatedness checks run
+  again after normalization, and there is no run-level collector. The generic
+  contract validates neither disposition nor reason/message size, UTF-8,
+  blankness, or total records.
+- **Impact:** Warning memory and output grow with chunks, lanes, configured
+  validators, and record counts. Local omission records lose exact occurrence
+  semantics while still incrementing `warning_count`.
+- **Recommendation:** Add a generic bounded diagnostic collector that preserves
+  exact occurrence counts and bounded samples. Validate all diagnostic fields
+  at the module/framework boundary. Immediately bring NPC and spell
+  relatedness under the existing cap if the full redesign is staged.
+
+### AUD-WARN-004 — Output Encoder Warnings Make Durable Surfaces Disagree
+
+- **Priority:** Medium current impact; high contract correctness risk.
+- **Evidence:** `Runner.Run` passes existing warnings to `encoder.Encode`, then
+  the production encoder serializes `warnings.json`. Only after encoding does
+  the runner append `OutputResult.Warnings`. The receipt, stderr, debug summary,
+  and debug run report see the final slice; the already-created published file
+  cannot. No production encoder currently returns a warning, so ordinary JSON
+  runs do not trigger the defect.
+- **Impact:** A valid output-module implementation can violate the documented
+  claim that `warning_count` describes the published warning collection.
+- **Recommendation:** Remove successful output warnings from the output-module
+  contract unless a demonstrated use case requires them; encoding failures
+  should be errors and optional encoder observations should be debug data. A
+  two-phase finalize API is the viable but more complex alternative.
+
+### AUD-WARN-005 — Validation Skips Are Incomplete But Not Warned
+
+- **Priority:** Medium operator/correctness impact.
+- **Evidence:** `firstIncompleteValidation` treats failed and skipped validators
+  alike, and validation summaries include both. `incompleteValidationWarnings`
+  emits records only for `validationFailed`. The focused test demonstrates
+  three incomplete validators but two warnings.
+- **Impact:** A successful run can have `validation_status: incomplete` while
+  its warning count understates or even omits the affected validators. A caller
+  that checks only warnings receives a weaker signal than the manifest and
+  receipt status.
+- **Recommendation:** Produce one aggregated incomplete-validation warning
+  group whose occurrences cover both failure and skip, while retaining typed
+  outcome and safe reason metadata in the validation summary. Do not expose
+  provider errors or arbitrary skip prose in model or operator messages.
+
+### AUD-WARN-006 — Relatedness Checks Are Useful But Repetitive And Lexically Weak
+
+- **Priority:** Medium operator impact; low acceptance-policy urgency.
+- **Evidence:** Every family runs the advisory in both extract and normalize
+  chains. The complete fixture contains exact repeated tuples, and the checks
+  rely on exact normalized token sequences or a minimal significant-token
+  overlap. Reason naming drifts between `*_not_near_source` and
+  `*_source_unrelated`.
+- **Impact:** The checks can catch unsupported entities, but aliases, pronouns,
+  inflection, and generic scene prose create predictable false positives. Flat
+  per-record presentation magnifies them.
+- **Recommendation:** Retain the validators and their stage-local execution,
+  but classify and aggregate them as advisories. Normalize reason-code naming
+  when the diagnostic contract changes. Do not strengthen them into rejection
+  rules without a human-reviewed production evaluation.
+
+### AUD-WARN-007 — `warning_count` Has No Stable Operational Meaning
+
+- **Priority:** High downstream-contract impact.
+- **Evidence:** The receipt and CLI use `len(output.Warnings)`. One list element
+  can be an omission summary representing several hidden occurrences; repeated
+  records can represent the same condition; and skipped validators can be
+  absent. A 21-occurrence spell test produces a list length of 20.
+- **Impact:** The value is neither an exact occurrence count nor a distinct
+  warning-group count. Consumers cannot set policy or present a trustworthy
+  summary from it.
+- **Recommendation:** Introduce explicit warning-group and warning-occurrence
+  counts in a versioned receipt. Do not silently redefine the v1 field.
+
+## Recommended Target Contract And Presentation Model
+
+### Diagnostic Model
+
+Use one validated internal diagnostic model with these concepts:
+
+- **disposition:** `warning`, `advisory`, or `observation`;
+- **category:** a small enum such as `configuration`, `degradation`,
+  `validation_incomplete`, `data_quality`, `fallback`, or `normalization`;
+- **reason code:** stable semantic identity owned by the producer;
+- **origin:** framework-added phase/stage, step ID, lane ID, module key,
+  validator name, chunk ID, and chunk index when applicable;
+- **occurrence count:** exact number of matching findings;
+- **samples:** a small deterministic list of bounded scope/message pairs; and
+- **omitted sample count:** `occurrence_count - len(samples)`, represented as
+  metadata rather than another diagnostic record.
+
+Errors and rejected outputs must not become diagnostic dispositions. A warning
+means that the run completed under policy despite a process-level degradation
+or incomplete configured operation. Advisory and observation dispositions can
+describe accepted artifact quality and transformation provenance, but no
+LLM-judged extraction-quality signal may be promoted to a warning. Validation
+status remains authoritative for approval, rejection, and incomplete
+validation.
+
+### Aggregation
+
+The framework runner should own aggregation after it enriches findings with
+origin and before public output construction. Modules and validators retain
+semantic ownership of disposition, category, reason, scope, and message; they
+must not own CLI or file presentation.
+
+The default stable key should be:
+
+```text
+disposition + category + reason_code
++ phase/stage + step_id + lane_id + module_key + validator_name
+```
+
+Chunk, record scope, and message text belong in samples and must not be part of
+the group key. This groups repeated per-chunk findings without merging the same
+code across distinct producers or pipeline locations. Group order should be
+the first occurrence in the runner's existing canonical order; sample order
+should follow the same order. A final canonical sort by the complete origin key
+is also viable, but completion timing must never choose either order.
+
+Aggregation must be incremental and bounded. Producers should use a shared
+collector that counts every occurrence while retaining only bounded samples;
+the framework then merges producer groups without reconstructing counts from
+omission prose. A global maximum group count is also required, with overflow
+represented by structured aggregate metadata and with actionable groups given
+priority over lower dispositions.
+
+### Durable Files
+
+Keep one canonical home for each class:
+
+- `warnings.json` should contain versioned, grouped actionable warnings only;
+- a new `diagnostics.json` should contain versioned advisory and observation
+  groups only, avoiding duplication of warning groups;
+- `index.json` should link both files;
+- `rejected.json` and manifest validation summaries should retain their current
+  separate responsibilities; and
+- debug bundles should retain candidate-attempt detail plus the final grouped
+  projections.
+
+This is preferable to keeping all detail in `warnings.json` and filtering only
+the CLI: downstream consumers would otherwise continue to receive a semantically
+mixed warning contract, and routine observations would still dominate the
+durable file.
+
+### CLI And Receipt
+
+For a successful human run with actionable warnings, print a concise summary
+such as:
+
+```text
+notarius: run completed with 2 warning groups (7 occurrences); details=/.../warnings.json
+```
+
+Advisories and observations should not produce the warning line. Their durable
+path remains discoverable through `index.json`; a concise non-warning count can
+be added to the ordinary success line only if operator testing shows value. An
+ordinary successful run with no process degradation should write nothing to
+the warning stream even when it publishes quality advisories or normalization
+observations.
+
+Create `notarius.run-result.v2` rather than redefining v1. It should expose at
+least:
+
+- `warning_group_count`;
+- `warning_occurrence_count`; and
+- `diagnostic_group_count` for non-warning durable groups.
+
+The receipt should continue to expose validation status, validation summaries,
+and rejected-output count independently. Process exit behavior should not
+change as part of warning presentation reform.
+
+### Checkpoint Semantics
+
+Store the structured terminal diagnostic groups with accepted checkpoints and
+replay them exactly once at their logical stage. Fresh and reused runs should
+produce the same public groups and counts. Checkpoint events and debug records,
+not diagnostic identity, should disclose whether computation was reused.
+
+### Output Encoder Boundary
+
+Prefer removing `OutputResult.Warnings`. A successful output encoder should
+either return the complete logical files or fail. If future encoders genuinely
+need to produce durable post-encoding warnings, introduce an explicit
+two-phase prepare/finalize contract so those warnings can be included in the
+same published collection. Do not retain the current self-inconsistent
+one-pass capability.
+
+## Resolution Of Required Design Questions
+
+| Question | Recommendation | Viable alternative and tradeoff |
+| --- | --- | --- |
+| Explicit severity/disposition or external reason mapping? | Put validated disposition and category in the contract. | A central reason-code registry avoids payload fields but makes new modules depend on a second synchronized policy table and leaves downstream meaning implicit. |
+| Keep all detail in `warnings.json` or separate it? | Separate grouped actionable warnings from grouped advisories/observations in `diagnostics.json`. | Keep the flat durable list and aggregate only CLI output; simpler migration, but it preserves the noisy downstream contract and ambiguous count. |
+| Who owns aggregation? | Framework runner/coordinator after origin enrichment. | Output module aggregation keeps framework types smaller but duplicates policy across encoders and cannot repair missing validator origin. |
+| Stable aggregation key? | Disposition, category, reason code, and full producer origin; exclude chunk/scope/message. | Explicit producer-supplied grouping keys offer flexibility but add another identity that can drift from reason codes. Message-template grouping is brittle and unsafe. |
+| Samples and omissions? | Exact occurrence count plus deterministic bounded samples and numeric omitted-sample count. | Omission warning records preserve the current representation but inflate group counts and require prose parsing. |
+| `warning_count` semantics? | Version receipt and replace ambiguity with group and occurrence counts. | Keep v1 count as published record length and add optional fields; compatible, but two competing warning counts remain easy to misuse. |
+| Which normalization changes remain warnings? | Only exhausted process fallback. Unresolved membership is a data-quality advisory; successful canonicalization, reordering, ID repair, source-ref dedupe, and duplicate consolidation are observations. | Treat unresolved membership or semantic duplicate consolidation as warnings because they affect grounding or cardinality; more conservative, but it violates the process-only warning rule and reports accepted artifact quality as an operational failure. |
+| Source-relatedness disposition? | Grouped advisory by default; preserve current approve behavior. | Retain warning disposition or make rejection configurable. Rejection requires production precision evidence; current lexical rules are not strong enough. |
+| Checkpoint-loaded warnings? | Present the same logical groups as fresh execution and use checkpoint events for reuse provenance. | Mark groups as replayed; aids forensics but fragments aggregation and makes semantically equivalent runs differ. |
+| ADR and schema versions? | Add an ADR and version the run receipt and diagnostic files. | Treat the work as CLI-only presentation and avoid an ADR; insufficient because module contracts, output files, checkpoint payloads, and downstream fields change. |
+
+## Compatibility, Documentation, And ADR Implications
+
+The target alters public and internal contracts enough to require a new ADR.
+It should record:
+
+- the distinction among warnings, advisories, observations, rejections, and
+  errors;
+- the invariant that warnings are process-level signals, LLM-judged extraction
+  quality is never a warning, and ordinary non-degraded success has zero
+  warnings;
+- module semantic ownership versus framework origin/aggregation ownership;
+- bounded group and sample semantics;
+- fresh/checkpoint equivalence; and
+- the output-encoder decision.
+
+Implementation should introduce `notarius.run-result.v2`. The grouped warning
+and diagnostic envelopes should each carry their own schema version. Because
+the content of `warnings.json` changes incompatibly from a flat array wrapper
+to groups, release notes and the published JSON integration contract must call
+out the migration. `index.json` gains the diagnostic file path.
+
+Canonical documentation updates belong in:
+
+- `docs/cli.md` for stderr presentation only;
+- `docs/operations.md` for operator review and debug workflow;
+- `docs/integrations/json-output.md` for warning and diagnostic file schemas;
+- `docs/integrations/run-result.md` for v2 fields and compatibility;
+- `docs/consumers/subprocess.md` and `docs/consumers/dnd-pipeline.md` for
+  downstream policy checks;
+- `docs/internal/pipeline.md` for promotion, aggregation, retry, and checkpoint
+  mechanics;
+- `docs/internal/modules.md` and `docs/internal/dnd.md` for producer rules and
+  the D&D classification matrix; and
+- `docs/policy/architecture.md` for the durable ownership invariant after the
+  ADR is accepted and implemented.
+
+No configuration knob is required for the first implementation. A fixed,
+well-documented taxonomy is easier to reason about than per-reason display
+overrides. Configurable escalation or suppression can be considered only after
+production review demonstrates a concrete operator need.
+
+## Test Coverage Assessment
+
+Existing coverage worth preserving includes:
+
+- accepted-attempt and terminal-rejection warning promotion;
+- validator failure retry exhaustion and `warn_continue`;
+- module semantic retry fallback;
+- deterministic warning order under concurrent lane completion;
+- chunk-plan invalidation and discarded-cache warning behavior;
+- fresh/checkpoint warning equivalence;
+- local D&D warning caps and safe dynamic-message quoting;
+- JSON warning-file publication; and
+- CLI stderr, debug, and receipt counts.
+
+Material gaps are:
+
+- no bound test for NPC- or spell-relatedness warnings;
+- no generic warning-field or result-size validation;
+- no test for output encoder warnings versus published `warnings.json`;
+- no operator-level aggregation or bounded-sample tests;
+- no fresh/resume test for grouped counts because groups do not yet exist; and
+- no production evaluation of advisory precision.
+
+Tests should protect the semantic relationships: exact occurrence counts,
+bounded samples, deterministic group order, actionable-only warning
+presentation, and cross-surface equality. They should not assert one exact
+warning count for every complete D&D run or treat message wording as a public
+API unless the wording itself enforces a security boundary.
+
+## Audit Conclusion
+
+The application is in a good position for warning reform. Its retry,
+validation, checkpoint, and concurrency mechanics provide reliable points at
+which to attach structured diagnostics. The most valuable change is not to
+suppress individual reason codes; it is to replace the semantically flat,
+origin-free collection with bounded typed groups and to reserve the word
+“warning” for conditions that merit operator attention.
+
+Provider-backed runs would improve prioritization and help tune the D&D
+advisories, but they are not necessary to conclude that routine normalization
+and heuristic doubt should not dominate stderr or the durable warning
+contract. They should be gathered before changing heuristic acceptance policy
+or adopting a numerical production warning-volume target.

docs/roadmap/dnd-subprocess-documentation.md

@@ -1,224 +0,0 @@
-# D&D Subprocess Consumer Documentation
-
-## Status
-
-Completed. The target guide is `docs/consumers/dnd-pipeline.md`.
-
-## Purpose
-
-Provide one task-oriented guide for applications that run Notarius as a
-subprocess to execute the maintained complete D&D pipeline and consume its
-published artifacts. The initial concrete consumer is Narratio, but the guide
-must describe the public Notarius workflow rather than depend on Narratio
-internals.
-
-The guide should make the safe integration path obvious without duplicating
-the CLI, input, receipt, output-bundle, or individual artifact contracts that
-already have canonical documentation.
-
-## Current State
-
-The public integration surface is documented accurately but is distributed
-across several documents:
-
-- `docs/consumers/subprocess.md` defines the generic subprocess workflow;
-- `docs/cli.md` owns commands, flags, stream behavior, and exit statuses;
-- `docs/integrations/seriatim.md` owns the accepted transcript input shape;
-- `docs/integrations/run-result.md` owns the machine-readable successful-run
-  receipt;
-- `docs/integrations/json-output.md` owns bundle discovery and logical files;
-- the D&D integration documents own the individual lane payload contracts;
-- `examples/dnd-complete.config.yml` is the maintained complete pipeline.
-
-A consumer can reconstruct the full workflow from those documents, but there
-is no D&D-focused guide that connects the maintained example to its input,
-invocation, complete artifact inventory, discovery procedure, and downstream
-acceptance decisions.
-
-## Target Documentation Set
-
-### Create `docs/consumers/dnd-pipeline.md`
-
-This document should own the end-to-end consumer workflow for the maintained
-complete D&D configuration. It should be useful to Narratio and to another
-subprocess orchestrator with the same needs.
-
-The guide should contain the following sections.
-
-#### Prerequisites And Deployment Configuration
-
-- Link to `examples/dnd-complete.config.yml` rather than embedding a second
-  complete configuration.
-- Explain that a deployment must provide the configured PromptKit profile and
-  campaign reference files.
-- Recommend absolute paths for a service or orchestrator deployment.
-- Call out the path-resolution distinction explicitly: YAML reference paths
-  are relative to the Notarius configuration file, while
-  `promptkit.profile_file` is relative to the Notarius process working
-  directory.
-- Recommend validating the selected configuration and `dnd-session` pipeline
-  before processing sessions.
-
-#### Transcript Input
-
-- State that the complete pipeline consumes a Seriatim JSON document.
-- Link to the canonical Seriatim contract for required fields and validation.
-- Recommend the caller's final trimmed transcript when the caller maintains
-  transcript tiers. For Narratio, identify the implemented source as
-  `narratio.transcript.final_trimmed`, normally stored at
-  `transcripts/final.trimmed.json`.
-- Explain that segment IDs must remain stable because D&D source references
-  cite those units.
-- Explain that Notarius derives its default prompt session from the input
-  module and exact input bytes and that ordinary callers should not supply
-  `--session-id`.
-
-#### Subprocess Invocation
-
-- Show one concise invocation using `notarius run dnd-session`, explicit
-  absolute `--config`, `--input`, and `--output-dir` paths, and `--json`.
-- Direct callers to capture stdout and stderr separately, propagate
-  cancellation, impose an operator-appropriate timeout, and wait for process
-  completion before parsing stdout.
-- State that only exit status zero permits receipt decoding and link to the CLI
-  contract for the complete exit-status definition.
-- Recommend retaining stderr and the invocation context for diagnosis without
-  logging secrets or transcript content.
-
-#### Receipt And Bundle Discovery
-
-- Require callers to accept only supported run-result schema versions while
-  tolerating unknown fields allowed by that version.
-- Direct callers to obtain the exact run-specific bundle from the receipt's
-  absolute `output_directory`; they must not scan for the newest run directory
-  or construct a run ID.
-- Require a confinement check when resolving `index_file` beneath the reported
-  bundle root.
-- Direct callers to discover lane payloads by `lane_id` in `index.json`, then
-  verify descriptor media type and schema identity before decoding them.
-- Explain that descriptor paths are untrusted relative paths and require the
-  same confinement discipline.
-
-#### Complete D&D Artifact Inventory
-
-Include a compact table for the ten lane IDs selected by the maintained
-complete configuration:
-
-- `item-registry`;
-- `npc-registry`;
-- `location-registry`;
-- `scene-descriptions`;
-- `item-occurrences`;
-- `spells`;
-- `combat-turns`;
-- `npc-occurrences`;
-- `location-occurrences`;
-- `enemy-events`.
-
-For each row, give a one-line purpose and link to the corresponding canonical
-D&D artifact contract. Do not copy its fields or schema rules into the
-consumer guide.
-
-Document the four always-published bundle files—`index.json`, `manifest.json`,
-`rejected.json`, and `warnings.json`—and the complete example's configured
-`chunk-map.json` and `evidence-context.json` pipeline-wide artifacts. Link to
-their canonical contracts and distinguish pipeline-wide artifacts from lane
-outputs.
-
-The inventory must say that a file is available only when its corresponding
-artifact was accepted and published. It must not imply that process success
-guarantees every configured lane.
-
-#### Downstream Acceptance And Retention
-
-- Explain that exit status zero can coexist with rejected outputs, warnings,
-  or absent lane descriptors.
-- Require the consumer to define its required lane set explicitly. Recommend
-  treating all ten lanes as required when the caller claims to consume the
-  complete D&D workflow, while allowing another consumer to adopt a narrower
-  documented policy.
-- Recommend retaining the receipt, the complete published bundle, and captured
-  diagnostic streams long enough to support provenance and failure analysis.
-- Explain that `evidence-context.json` is a reading excerpt; authoritative
-  citations remain in lane payloads.
-- Treat transcripts, lane artifacts, evidence context, manifests, and logs as
-  sensitive campaign data.
-
-#### Compatibility Checklist
-
-End with a concise checklist covering process exit, receipt schema, path
-confinement, pipeline identity, index decoding, required descriptors,
-descriptor schema/media compatibility, warnings and rejections, checksums or
-retention, and secure handling. Compatibility should be based on published
-receipt and artifact contracts rather than parsing a human version string.
-
-### Update Existing Navigation
-
-- Add a short link from `docs/consumers/subprocess.md` to the D&D-specific
-  workflow. Keep generic subprocess policy in the existing document.
-- Add the guide to the documentation links in `README.md`.
-- Extend the subprocess-consumer row in `docs/development.md` so maintainers
-  working on the D&D workflow are routed to the new guide and the canonical
-  contracts.
-
-### Verify Canonical Contract Documents
-
-Review the linked integration documents and the complete example while writing
-the guide. Correct an integration document only if repository inspection finds
-an actual stale contract. Do not move schema definitions, field tables, CLI
-flags, or configuration semantics into the new guide.
-
-## Narratio Alignment
-
-The guide may name Narratio as the motivating consumer and identify its current
-final-trimmed transcript source. It must not claim that Narratio already has a
-Notarius adapter or extraction stage. Until that feature is implemented,
-Narratio-specific architecture, configuration, stage behavior, manifest
-records, and artifact source IDs belong in Narratio's roadmap.
-
-Once Narratio implements the integration, its own integration documentation
-should link to this guide and the durable Notarius contracts instead of
-repeating them.
-
-## Validation
-
-Documentation implementation should include:
-
-```sh
-go run ./cmd/notarius config validate \
-  --config examples/dnd-complete.config.yml \
-  --pipeline dnd-session
-go test ./...
-```
-
-Also verify all new and changed relative Markdown links, compare the artifact
-inventory directly with the maintained complete configuration, and confirm
-that commands and path semantics match the CLI and configuration references.
-If the repository still has no automated link checker, record that fact and
-perform a focused manual link review.
-
-## Acceptance Criteria
-
-- A subprocess integrator can follow one D&D-focused guide from a Seriatim
-  transcript through safe discovery of every artifact configured by the
-  complete example.
-- The guide makes stdout, stderr, exit-status, receipt, and path-confinement
-  responsibilities unambiguous.
-- The ten configured D&D lanes and both configured pipeline-wide artifacts are
-  listed and linked to their canonical contracts.
-- The guide distinguishes process success from the caller's required-artifact
-  policy.
-- The profile-path and reference-path resolution rules are clearly stated.
-- Existing navigation makes the guide discoverable.
-- No volatile contract is defined in two places, and no unimplemented Narratio
-  behavior is presented as current.
-
-## Non-Goals
-
-- Implementing or documenting Narratio's future adapter or stage as current
-  Notarius behavior.
-- Adding a new Notarius command, receipt version, output format, or artifact
-  schema.
-- Duplicating the complete configuration or individual D&D payload schemas in
-  prose.
-- Defining a universal partial-result policy for every Notarius consumer.

docs/roadmap/future.md

@@ -7,197 +7,16 @@ not as committed release dates.
 
 ## Near-Term Validation And LLM Reliability
 
-The following work forms one related program but should be promoted into
-separate feature roadmaps and implemented in dependency order. PromptKit owns
-structural output repair within one completion. Notarius owns stage candidates,
-validator chains, semantic rejection policy, and whether another stage attempt
-is warranted.
+PromptKit now owns structural output repair within one completion. Notarius
+owns stage candidates, validator chains, semantic rejection policy, bounded
+feedback-aware stage retries, validation provenance, and reusable-state
+eligibility, and the separation of actionable process warnings from quality
+diagnostics. The remaining near-term work applies those completed foundations
+to domain review and empirical evaluation.
 
-### 1. Upgrade To PromptKit v0.8.0
-
-This item has been promoted to the standalone
-[PromptKit v0.8.0 Upgrade](promptkit-v0.8.md) roadmap. That document owns the
-release-by-release compatibility review, adopted features, structured-repair
-policy, target integration boundary, acceptance criteria, and settled design
-decisions.
-
-### 2. Feedback-Aware Stage Validation Retries
-
-- Model Notarius's corrective stage-retry conversation explicitly after
-  PromptKit v0.8.0. The first attempt sends the ordinary complete initial
-  prompt. If application validation rejects the resulting LLM-produced
-  candidate and another stage attempt is available, reconstruct that complete
-  initial prompt byte-for-byte and append exactly two messages: an assistant
-  message containing the defective response and an application-owned user
-  message detailing every applicable semantic validation error and requesting
-  one corrected, complete replacement response. This is a freshly constructed
-  correction request, not continuation of an accumulating conversation.
-- Use the configured stage `retries` value as the one outer retry budget for
-  this loop. `retries: N` continues to mean at most `N` additional complete
-  chunk, extract, merge, or normalize attempts after the initial attempt,
-  whether an attempt is needed because of a producer error or semantic
-  rejection. Do not add a second semantic-correction count. PromptKit's
-  prompt-level `repair_attempts` budget is independent and internal to each
-  individual LLM completion, and does not consume or replenish the Notarius
-  stage budget.
-- Extend the framework-managed validation boundary for chunk, extract, merge,
-  and normalize stages so a rejected LLM-produced candidate and its exact raw
-  model response remain available to construct the next stage attempt.
-  Deterministic producers cannot improve by repeating the same inputs; a
-  rejection from a deterministic stage is therefore terminal under the
-  configured rejection policy rather than consuming retries mechanically.
-- Preserve the original session ID, selected profile, structured-output
-  contract, prompt inputs, and reusable prompt prefix. Carry only the latest
-  candidate and latest aggregate feedback; do not build an unbounded retry
-  conversation. Keep model-facing corrective guidance separate from
-  operator-facing diagnostics, and apply explicit size, redaction, and debug
-  disclosure rules to both.
-- Run every applicable validator in the configured chain before deciding
-  whether to retry. Do not short-circuit merely because an earlier validator
-  rejected the candidate. Aggregate all semantic rejection reason codes and
-  corrective guidance into the retry message so one retry can address the
-  whole candidate. A validator is applicable only when its declared target and
-  prerequisites can be satisfied; record a deterministic skipped diagnostic
-  rather than invoking a validator on an input it cannot interpret. Initially
-  execute the chain sequentially in configured order so results, diagnostics,
-  costs, and feedback ordering remain deterministic; consider validator
-  concurrency only in response to measured latency.
-- Continue running independent applicable validators after one validator
-  execution failure so the attempt retains as much useful diagnostic
-  information as practical. Do not present validator operational failures as
-  defects in the producer candidate and do not include them in corrective
-  feedback.
-- Distinguish three terminal conditions and make their policies configurable
-  at a coherent pipeline or binding scope:
-  - **producer structural failure:** PromptKit could not return a usable
-    structured candidate after its repair budget. Default to `fail_run`; an
-    allowed alternative may record a terminal stage or lane rejection where
-    execution can safely continue, but may not accept the invalid output;
-  - **semantic rejection:** one or more validators completed and rejected the
-    candidate. Default to `fail_run` after corrective stage retries are
-    exhausted; allow an explicit alternative that records the existing
-    rejected-output outcome without advancing that output;
-  - **validator execution failure:** a validator could not produce a valid
-    decision because of generation, structural-output, transport, or internal
-    failure. Default to a genuine warning and an explicitly recorded
-    `validation_incomplete` or equivalent degraded state while allowing the
-    candidate to continue; allow strict configuration to fail the run instead.
-- An LLM-backed validator uses the same scheduled PromptKit boundary as every
-  other LLM-backed module. Its own response may use PromptKit's bounded
-  structural repair. Distinguish its possible output states:
-  - output rejected by PromptKit's structural contract should consume only the
-    validator prompt's configured PromptKit repair budget;
-  - output that is structurally valid but violates a deterministically
-    checkable validator-result invariant should be classified as a validator
-    execution failure;
-  - output that satisfies the complete validator-result contract is the
-    validator's decision, even though an LLM judgment may remain imperfect.
-    Automatically judging that judgment would require another semantic
-    validator and is outside this feature.
-  If the validator cannot return a contract-valid decision, do not recursively
-  create another Notarius semantic-validation loop around it. Apply the
-  configured validator-failure policy. The default warning must identify the
-  validator and affected stage without exposing sensitive content.
-- Separate validator execution retry from producer correction. A transient
-  validator operational failure must not automatically discard and regenerate
-  an otherwise usable producer candidate. Any bounded retry of the validator
-  itself should reuse that same immutable candidate and remain subordinate to
-  PromptKit and provider retry behavior.
-- Preserve attempt-level provenance, cumulative token usage, validator
-  outcomes, aggregated correction feedback, and terminal policy decisions in
-  the debug and manifest models without copying raw source material into
-  ordinary errors or durable summaries.
-- Define terminal-outcome precedence. A semantic rejection dominates a
-  validator execution failure for the same candidate: use the completed
-  rejections to correct the producer while separately recording incomplete
-  validation. If a later candidate has no semantic rejection but one validator
-  still fails, apply the configured validator-failure policy to that candidate.
-  Never allow a known semantic rejection to become accepted through a
-  warn-and-continue setting, and never accept a structurally invalid producer
-  response. Permissive policy may preserve a rejected-output outcome or accept
-  a structurally valid candidate with explicitly incomplete validation; it may
-  not relabel known-invalid output as approved.
-
-Before implementation, record the generic validation and retry state machine
-in an ADR. The ADR should own the separation between PromptKit repair and
-Notarius correction, use of the existing stage-retry budget, reconstruction of
-correction conversations, all-applicable-validator aggregation, deterministic
-validator ordering, non-recursive validator failure handling, outcome
-precedence, default fail-open/fail-closed choices, configurable terminal
-policies, and provenance and sensitive-data constraints. A dependency-upgrade
-ADR is not needed for PromptKit v0.8.0 itself. Current behavior remains
-authoritative until the validation ADR is implemented and the canonical
-architecture, configuration, operations, and internal documentation are
-updated.
-
-### 3. D&D Combat Scene Semantic Validation
-
-- Add an optional production LLM-backed D&D validator that determines whether
-  proposed scene boundaries and classifications represent substantive active
-  combat correctly. Its central quality goal is that active combat is kept in
-  coherent scenes classified as `combat`, rather than split incorrectly or
-  hidden inside scenes classified as `narrative`, `recap`, or `meta`.
-- Resolve the validator's exact target before implementation. The current
-  `dnd/scenes` chunker owns only complete, gap-free source ranges, while the
-  per-chunk `dnd/scene-descriptions` extractor owns the `combat`, `narrative`,
-  `recap`, and `meta` classification. The preferred initial placement is
-  therefore an extract-stage validator for `dnd/scene-descriptions`, where it
-  can compare one proposed kind with the corresponding transcript chunk.
-- Consider a chunk-stage LLM validator only for a distinct boundary-coherence
-  question that can be answered from the complete transcript and proposed
-  range map, such as whether one continuous combat was fragmented across
-  inappropriate scene boundaries. Do not duplicate the same classification
-  judgment at both stages. Moving classification into chunk-plan annotations
-  would change the deliberately minimal, annotation-free chunk contract and
-  requires an explicit architecture review before it is selected.
-- Validate both false negatives and false positives: a non-combat kind must not
-  omit substantive active combat, and a combat kind must be supported by such
-  combat. Keep the existing deterministic downstream rule that combat-turn
-  extraction runs only for an exact `combat` scene classification; semantic
-  review improves the upstream classification but does not replace that gate.
-- Run the semantic validator through PromptKit, use a minimal required-field
-  structured response schema, and let PromptKit repair structural validator
-  output within its bounded budget. A contract-invalid final validator response
-  is a validator execution failure, not a semantic rejection and not a reason
-  to recursively validate the validator.
-- Evaluate the prompt and decision policy against a small human-reviewed set
-  containing combat setup, active turns, interruptions, multi-phase encounters,
-  brief rules discussion, aftermath, recalled combat, and false-positive
-  hostile dialogue. Measure false acceptance, false rejection, retry success,
-  added calls, latency, and token cost before placing it in the production
-  default chain.
-- An ADR is not required if classification remains owned by
-  `dnd/scene-descriptions` and the validator follows the generic validation ADR.
-  Create or supersede an ADR if the work transfers scene classification into
-  the chunker or otherwise changes stage ownership or the durable chunk-plan
-  contract.
-
-### 4. Warning Signal And Presentation Reform
-
-- Audit every warning producer and representative successful runs. Ordinary
-  success producing dozens of warnings is a failed operator experience: the
-  volume obscures actionable problems and trains operators to ignore the
-  warning channel.
-- Define a small warning taxonomy that distinguishes actionable degradation,
-  incomplete validation, lossy fallback, and data-quality risk from routine
-  normalization observations or informational diagnostics. Preserve detailed
-  traceability in debug or manifest data without promoting every observation
-  to a top-level CLI warning.
-- Consider stable deduplication and aggregation by scope and reason code,
-  bounded samples plus omitted counts, and a concise CLI summary with a path to
-  detailed diagnostics. Do not suppress genuine validator execution failures
-  merely to reduce the count.
-- Decide which warnings affect process status, rejection summaries, durable run
-  receipts, or only debug output. Ensure warning ordering and aggregation are
-  deterministic across concurrent execution.
-- Establish a representative warning-volume acceptance target and human review
-  workflow before changing individual producers piecemeal. The intended result
-  is not zero warnings; it is a small set in which every surfaced warning merits
-  operator attention.
-- This work does not require an ADR unless it changes validation acceptance,
-  failure, or durable contract semantics. CLI presentation and diagnostic
-  taxonomy otherwise belong in a feature roadmap followed by updates to their
-  canonical configuration, operations, integration, and internal documents.
+Near-term reliability work should now be selected from the concrete evaluation
+and extension opportunities below. The retry, validation, and subprocess
+foundations described above are implemented current behavior.
 
 ## Near-Term D&D Pipeline
 

docs/roadmap/implementation.md

@@ -1,782 +0,0 @@
-# PromptKit v0.8.0 Upgrade Implementation Plan
-
-## Purpose
-
-Implement the target state defined by the
-[PromptKit v0.8.0 Upgrade](promptkit-v0.8.md): adopt the useful PromptKit
-v0.6.0, v0.7.0, and v0.8.0 changes; enable one bounded structural correction
-by default; expose pipeline and binding overrides; preserve safe provider
-diagnostics; and keep PromptKit behind Notarius's transport-neutral LLM
-boundary.
-
-This plan is ordered. Each numbered stage is one implementation prompt for a
-gpt-5.6-terra coding agent. Complete and validate one stage before beginning
-the next. Read `docs/development.md` and every policy under `docs/policy/` at
-the start of each stage, inspect the current code and tests named by that
-stage, preserve unrelated worktree changes, and update current-behavior
-documentation in the same stage as the behavior it describes.
-
-Do not retire this plan or `promptkit-v0.8.md` during implementation. Keep both
-until the completed work has passed a separate review. Do not implement the
-future Notarius semantic-validation retry loop, D&D combat-scene validator, or
-warning redesign as part of this plan.
-
-## Decisions Fixed For Implementation
-
-- Pin `gitea.maximumdirect.net/eric/promptkit` v0.8.0 directly, with no
-  `replace`, workspace dependency, or vendored source.
-- Every maintained eligible production prompt defaults to exactly one
-  PromptKit structural repair attempt.
-- Add the exact configuration key
-  `structured_output_repair_attempts` at pipeline scope and on LLM-backed
-  module and validator bindings.
-- Effective precedence is binding value, then pipeline value, then the prompt's
-  declared `repair_attempts` value. Omission inherits; explicit zero disables
-  structural repair at that scope.
-- Accepted values are integers from zero through three. Explicit null and
-  non-integer values are invalid. An explicit binding value on a deterministic
-  module or validator is invalid. A pipeline value is applied only to selected
-  LLM-backed bindings and does not make deterministic bindings invalid.
-- Keep file configuration version 4. This is an additive pre-release field and
-  does not require parallel versioned behavior.
-- Use `StructuredOutputRepairAttempts *int` for presence-aware internal Go
-  fields. Clone pointers at every ownership boundary.
-- A configured override never replaces schema identity, output format, or
-  validation mode. The PromptKit adapter calls `InspectPrompt`, copies the
-  complete normalized prompt-owned output contract, changes only
-  `RepairAttempts`, and supplies the complete replacement on `RunRequest`.
-  Do not add an inspection cache initially.
-- PromptKit repair is internal to one `CompleteStructured` call and does not
-  consume or replenish a binding's existing `retries` budget.
-- Add `RepairAttempts int` to Notarius's structured-completion response. It is
-  the actual corrective-call count reported by PromptKit; token usage remains
-  PromptKit's cumulative usage and must not be summed again.
-- A valid repaired response is successful and produces no warning solely
-  because repair occurred. Exhausted structural validation maps to
-  `ErrInvalidStructuredOutput` with the final candidate and debug material
-  retained.
-- Add an application-owned generation-error sentinel and typed status-bearing
-  error. PromptKit error types must not cross `internal/framework/llm`.
-- HTTP status may appear in the application-owned generation error. Provider
-  code, type, and message are excluded from ordinary errors, warnings,
-  manifests, cache, and checkpoint identity; they may appear only in an
-  explicitly requested debug trace after Notarius redaction.
-- Profile inheritance is owned entirely by PromptKit. Notarius passes sources
-  through, inspects and records the resolved target, and does not parse or merge
-  `base_profile` itself.
-- PromptKit's built-in `rakestrawhome` backend and
-  `rakestrawhome-gemma-4-31b` profile are available generically. Notarius does
-  not register, shadow, or select them by default.
-- Missing optional credential environment values are allowed to reach the
-  provider without `Authorization`; Notarius does not recreate v0.5.0's local
-  failure or add provider-specific authentication logic.
-- No dependency-upgrade ADR is required. Update architecture only with the
-  durable ownership distinction between PromptKit structural repair and
-  Notarius stage/semantic validation policy.
-
-## Stage 1: Upgrade The Dependency And Establish A Clean v0.8.0 Baseline ✅
-
-### Goal
-
-Move the repository to PromptKit v0.8.0, resolve source-compatibility issues,
-and establish a passing baseline before adopting new behavior.
-
-### Implementation
-
-1. Re-read the upstream v0.6.0, v0.7.0, and v0.8.0 release guides and the
-   v0.8.0 package consumer and format documentation. Treat the pinned v0.8.0
-   tag, not the sibling checkout's moving branch, as authoritative.
-2. Update `go.mod` and `go.sum` to PromptKit v0.8.0 and run `go mod tidy` with
-   `GOWORK=off`.
-3. Compile before making compatibility edits. Correct only actual source or
-   behavior incompatibilities. In particular:
-   - convert any positional `promptkit.Profile` or
-     `promptkit.OpenAICompatibleProfileConfig` literals to keyed literals;
-   - confirm Notarius does not register the newly reserved `rakestrawhome`
-     backend ID; and
-   - preserve `PrepareExecution`/`Details`/`RunPrepared` snapshot ownership,
-     `Discard`, session forwarding, reasoning override, profile preflight,
-     and capacity adaptation.
-4. Change `promptKitBuiltinProfileCatalogID` in
-   `internal/framework/llm/promptkit_profile_fingerprint.go` from the v0.5.0
-   catalog marker to an opaque v0.8.0 marker. Do not hash PromptKit internal
-   files or include catalog content in manifests.
-5. Update `docs/integrations/pkg-promptkit.md` to pin and link v0.8.0 and to
-   state that this stage still leaves the production prompt-declared repair
-   budget at its current value. Do not document later configuration or default
-   behavior before it exists.
-6. Update only those existing tests whose public PromptKit types or stable
-   v0.8.0 behavior genuinely changed. Do not rewrite tests merely to match
-   upstream diagnostic wording.
-
-### Tests And Validation
-
-```sh
-GOWORK=off go mod tidy -diff
-GOWORK=off go test ./internal/framework/llm ./internal/cli
-GOWORK=off go test ./...
-GOWORK=off go vet ./...
-GOWORK=off go build ./cmd/notarius
-git diff --check
-```
-
-### Acceptance Criteria
-
-- `go list -m gitea.maximumdirect.net/eric/promptkit` reports v0.8.0.
-- There is no PromptKit `replace`, active Go workspace dependency, or vendor
-  tree.
-- The adapter still uses one frozen prepared execution and all existing LLM
-  tests pass.
-- Checkpoint profile identity includes the v0.8.0 built-in catalog marker.
-- Current integration documentation pins v0.8.0 without claiming that
-  later stages are already active.
-- The full ordinary test suite, vet, and command build pass.
-
-## Stage 2: Verify v0.6.0 Compatibility And Hardening ✅
-
-### Goal
-
-Audit Notarius's assets and boundary values against PromptKit v0.6.0's stricter
-source, path, endpoint, JSON, and cancellation contracts, fixing only concrete
-incompatibilities.
-
-### Implementation
-
-1. Inspect `internal/framework/llm/asset_registry.go`, prompt/profile source
-   composition, all registered asset roots, the conventional local backend,
-   and their focused tests.
-2. Exercise every production asset registry through PromptKit engine
-   construction and the existing production composition tests. Confirm that:
-   - YAML IDs and versions, not filenames, select definitions;
-   - every `content_file` path is exact, relative, contained, and points to a
-     regular embedded file;
-   - every schema and JSON asset is one complete JSON value;
-   - every current output contract is valid under v0.8.0; and
-   - unrelated malformed definitions do not create a second Notarius identity
-     or fallback mechanism.
-3. Review local endpoint parsing and validation. Retain a narrower Notarius
-   rule only if it has independent application value; otherwise rely on
-   PromptKit's absolute HTTP/HTTPS URL contract. Never accept a value that the
-   adapter will later reject.
-4. Review conversion of Notarius variables, inputs, profile extras, and debug
-   values at the adapter boundary for PromptKit's bounded JSON-compatible-value
-   rules. Do not add a second generic JSON walker or duplicate upstream numeric
-   limits.
-5. Verify cancellation and deadline identity through existing adapter tests.
-   Add or refine one focused regression only if Notarius currently destroys an
-   `errors.Is`-relevant context or transport error that the application owns.
-6. Do not add a cross-operation schema cache, artifact cache, provider-body
-   reader, or duplicate JSON framing validation; v0.6.0 owns those mechanisms.
-
-### Tests And Validation
-
-Run the focused asset, profile-source, and adapter packages, then the ordinary
-and race-enabled suites:
-
-```sh
-GOWORK=off go test ./internal/framework/llm ./internal/cli
-GOWORK=off go test ./...
-GOWORK=off go test -race ./...
-git diff --check
-```
-
-Tests must remain offline and should validate Notarius's assembled boundary,
-not reproduce PromptKit's internal path, JSON-depth, or response-size matrices.
-
-### Acceptance Criteria
-
-- Every maintained embedded prompt, schema, and fallback profile can be loaded
-  through the assembled v0.8.0 engine.
-- Current local endpoint and JSON-compatible values either satisfy the stricter
-  upstream contract or fail during preparation with safe diagnostics.
-- No duplicate PromptKit-owned cache, JSON, or response-bound mechanism is
-  introduced.
-- Cancellation and deadline behavior remains discoverable at the Notarius
-  boundary.
-- Ordinary and race-enabled tests pass.
-
-## Stage 3: Adopt Profile Inheritance, Rakestrawhome, And Optional Credentials ✅
-
-### Goal
-
-Make the useful PromptKit v0.7.0 profile and backend behavior work through
-Notarius's existing generic profile boundary without adding provider-specific
-composition logic.
-
-### Implementation
-
-1. Inspect `promptkit_profiles.go`, `asset_registry.go`, profile fingerprinting,
-   CLI profile preflight, profile provenance recording, and their tests before
-   editing.
-2. Add an offline integration test using a temporary operator profile source
-   whose leaf uses `base_profile`. Prove that:
-   - preflight reports the leaf ID;
-   - the effective backend, model, reasoning, and other inherited values match
-     the resolved PromptKit target;
-   - execution uses the same resolved target as inspection; and
-   - a missing parent or cycle fails before provider generation with a safe
-     profile-load diagnostic.
-   Do not duplicate PromptKit's entire field-by-field merge test matrix.
-3. Add a checkpoint-safety test showing that changing a parent definition in
-   an operator profile directory changes Notarius's profile-source fingerprint
-   while profile content and paths remain absent from the fingerprint value.
-   Retain the v0.8.0 catalog marker as coverage for built-in-parent changes.
-4. Verify `rakestrawhome-gemma-4-31b` through the ordinary profile inspector.
-   Assert its selected backend reaches Notarius's application-owned inspection
-   and provenance fields. Use a fake PromptKit client or transport if execution
-   coverage is needed; never contact the live service or require credentials.
-5. Verify that Notarius registers no `rakestrawhome` override and that the
-   existing `local` registration remains independent.
-6. Add one `httptest`-backed adapter integration test for a filesystem profile
-   with a missing optional `api_key_env`. The request must reach the test server
-   without an `Authorization` header. Add a focused in-memory PromptKit profile
-   test for `APIKeyRequired` only if needed to prove Notarius preserves upstream
-   preflight behavior; do not expose a new operator profile API.
-7. Keep `assets/dnd/profiles/dnd-extraction.yaml` standalone and unchanged. No
-   matching v0.8.0 built-in profile owns its `openai/gpt-5.6-luna` target.
-8. Update the current profile-source, deployment, and pinned-integration
-   sections in `docs/config.md`, `docs/operations.md`,
-   `docs/internal/llm.md`, and `docs/integrations/pkg-promptkit.md`. Link to the
-   pinned PromptKit format rules for inheritance. Explain that filesystem
-   profiles cannot express PromptKit's in-memory `APIKeyRequired` field and
-   that an optional missing credential may result in a provider 401/403.
-
-### Tests And Validation
-
-```sh
-GOWORK=off go test ./internal/framework/llm ./internal/core/config ./internal/cli
-GOWORK=off go test ./...
-GOWORK=off go test -race ./internal/framework/llm ./internal/cli
-git diff --check
-```
-
-### Acceptance Criteria
-
-- Inherited operator profiles resolve identically during preflight and
-  execution, with the leaf ID and effective target kept distinct.
-- Parent changes invalidate checkpoint reuse without leaking profile content or
-  paths.
-- Rakestrawhome is available through generic PromptKit profile handling and is
-  not selected by default or registered by Notarius.
-- Missing optional credentials omit authorization and reach the controlled
-  test provider; explicitly required credentials retain upstream behavior.
-- Current documentation accurately describes the implemented profile and
-  credential behavior without duplicating PromptKit's merge algorithm.
-
-## Stage 4: Adapt Structured Generation Errors Safely ✅
-
-### Goal
-
-Use PromptKit v0.7.0's structured generation errors for stable status
-classification and debug-only provider diagnostics without leaking PromptKit
-types or sensitive provider text.
-
-### Implementation
-
-1. In `internal/framework/contracts`, add:
-   - `ErrLLMGeneration` as the provider-neutral generation-failure sentinel;
-   - an application-owned `LLMGenerationError` with private status and safe
-     diagnostic fields, `Error`, `Unwrap`, and `StatusCode` methods; and
-   - a constructor that accepts a nonnegative status and an already-redacted
-     diagnostic. Status zero means no HTTP status was available.
-   Ordinary callers may inspect status with `errors.As` and category with
-   `errors.Is`, but cannot obtain provider code, type, or message from the
-   error.
-2. Add an application-owned `LLMDebugProviderError` with `status_code`, `code`,
-   `type`, and `message` fields, referenced optionally from
-   `LLMDebugResponse`. This is debug material, not a manifest or durable public
-   artifact contract.
-3. In `PromptKitClient.CompleteStructured`, preserve precedence in this order:
-   caller context cancellation/deadline, PromptKit capacity error, structured
-   PromptKit generation error, then other PromptKit generation failures.
-   Map every generation failure to `ErrLLMGeneration`; map
-   `*promptkit.GenerationError` to `LLMGenerationError` with its status.
-   Never wrap or return the PromptKit error value itself.
-4. Keep the ordinary diagnostic limited to PromptKit's safe default error
-   formatting after bearer and known-credential redaction. Do not append
-   `ProviderCode`, `ProviderType`, or `ProviderMessage` to it.
-5. For an explicitly requested debug path, preserve prepared prompt details and
-   attach the PromptKit provider code, type, and message after:
-   - reading only the selected prepared target's `APIKeyEnv`, if any, to obtain
-     the exact known credential solely for redaction;
-   - applying `RedactSecrets` and the existing bearer/key-pattern redaction;
-   - retaining PromptKit's already-normalized bounds; and
-   - discarding the credential value immediately rather than storing it.
-   Do not scan unrelated environment variables.
-6. Return prompt/debug material alongside the error so the existing debug LLM
-   wrapper can persist it only when debug recording is enabled. Confirm that
-   provider fields do not appear in ordinary error text, warnings, manifests,
-   cache, checkpoint data, or a run without debug output.
-7. Refactor error mapping into small helpers if needed to keep
-   `CompleteStructured` readable; do not create provider-specific policy in
-   modules or the pipeline runner.
-8. Update the error and observability sections of `docs/internal/llm.md` and
-   `docs/integrations/pkg-promptkit.md`. Keep operator disclosure rules in
-   `docs/operations.md` concise and link to the internal boundary where useful.
-
-### Tests And Validation
-
-- Use `httptest.Server` to return representative structured 400 and 503
-  responses. Assert `errors.Is(ErrLLMGeneration)`, `errors.As` to the
-  application-owned type, and the exact status without asserting complete
-  human wording.
-- Include a provider message containing the selected test credential and a
-  bearer-shaped value. Verify both are absent from the ordinary error and
-  debug artifact, while a non-sensitive marker appears only in the requested
-  debug trace.
-- Retain existing capacity and context tests to prove their more specific
-  classifications still win.
-
-```sh
-GOWORK=off go test ./internal/framework/contracts ./internal/framework/llm ./internal/framework/pipeline ./internal/cli
-GOWORK=off go test ./...
-GOWORK=off go test -race ./internal/framework/llm ./internal/framework/pipeline
-git diff --check
-```
-
-### Acceptance Criteria
-
-- PromptKit generation errors never escape the adapter error chain.
-- All generation failures match `ErrLLMGeneration`; structured non-success
-  responses expose only application-owned HTTP status to ordinary callers.
-- Provider code, type, and message are available only in an explicitly
-  requested, redacted debug trace.
-- Capacity and context classifications remain unchanged and more specific.
-- Security tests prove selected credentials and bearer tokens are not leaked.
-
-## Stage 5: Add Adapter-Level Structured Repair Support ✅
-
-### Goal
-
-Teach the transport-neutral completion boundary and PromptKit adapter to apply
-an optional repair override and report actual repair behavior, without yet
-exposing the setting in pipeline configuration.
-
-### Implementation
-
-1. Add `StructuredOutputRepairAttempts *int` to
-   `contracts.StructuredCompletionRequest`. Copy the pointed-to value wherever
-   requests are cloned or retained.
-2. Add `RepairAttempts int` to `contracts.StructuredCompletionResponse`. It is
-   the actual number of corrective generation calls, not the configured budget
-   and not the number of total candidates.
-3. Validate a non-nil request value as zero through three at the adapter
-   boundary so programmatic callers cannot bypass later file/config validation.
-4. When the request value is nil, leave `promptkit.RunRequest.Validation` nil
-   so the prompt's complete contract remains authoritative.
-5. When the value is non-nil:
-   - call `Engine.InspectPrompt(ctx, promptID, promptVersion)`;
-   - copy `PromptInspection.OutputContract` by value;
-   - replace only `RepairAttempts`;
-   - pass the complete copied contract as `RunRequest.Validation`; and
-   - prepare and execute exactly as before.
-   Do not infer or hard-code schema paths, validation modes, or formats. Do not
-   cache inspection in this stage.
-6. Map `result.Validation.RepairAttempts` to the response and leave
-   `result.Usage` cumulative values unchanged. The existing debug validation
-   object and prepared output contract should show actual and configured values
-   respectively.
-7. Preserve result semantics:
-   - valid initial and repaired candidates decode normally;
-   - repair exhaustion returns the final raw candidate/debug material with an
-     error matching `ErrInvalidStructuredOutput`;
-   - explicit empty or whitespace-only content follows PromptKit validation;
-   - missing/null/non-string content remains a generation/provider failure;
-   - corrective-call generation errors use Stage 4's application-owned mapping;
-     and
-   - context cancellation wins at every error boundary.
-8. Keep `CompleteStructured` and its helpers provider neutral outside this
-   adapter package. Do not expose PromptKit validation or inspection types.
-9. Update only the adapter-owned repair behavior in `docs/internal/llm.md` and
-   `docs/integrations/pkg-promptkit.md`. State that public pipeline configuration
-   and the production default are added by later stages of this plan.
-
-### Tests And Validation
-
-Add adapter-level behavioral tests using a deterministic fake PromptKit LLM:
-
-- nil override uses the prompt declaration;
-- explicit zero overrides a positive prompt declaration without dropping its
-  JSON Schema contract;
-- explicit one turns an invalid first candidate followed by a valid candidate
-  into one successful response with the final raw bytes, actual repair count
-  one, and cumulative usage;
-- repair exhaustion returns the final candidate and validation diagnostics as
-  `ErrInvalidStructuredOutput`;
-- explicit empty content is eligible for repair;
-- a corrective generation failure maps through Stage 4; and
-- invalid direct values below zero or above three fail before provider work.
-
-Do not assert PromptKit's exact assistant/user correction prose or copy its
-full internal repair matrix.
-
-```sh
-GOWORK=off go test ./internal/framework/contracts ./internal/framework/llm ./internal/framework/pipeline
-GOWORK=off go test -race ./internal/framework/llm ./internal/framework/pipeline
-GOWORK=off go test ./...
-git diff --check
-```
-
-### Acceptance Criteria
-
-- The adapter changes only repair count when applying a request override.
-- Nil and explicit zero remain distinct.
-- Repaired success returns final raw output, cumulative usage, and actual count
-  without a warning.
-- Exhaustion, empty content, corrective generation failure, and cancellation
-  match the target semantics.
-- No PromptKit type crosses the LLM package boundary.
-
-## Stage 6: Propagate Repair Policy Through Framework Requests ✅
-
-### Goal
-
-Carry an optional effective repair budget from each resolved stage or validator
-binding to its module request without changing public file configuration yet.
-
-### Implementation
-
-1. Add `StructuredOutputRepairAttempts *int` alongside `LLMProfile` to every
-   stage request that can belong to an LLM-backed binding:
-   - `ParseRequest`;
-   - `ChunkRequest`;
-   - `TypedExtractionRequest`;
-   - `TypedMergeRequest`;
-   - `TypedNormalizeRequest`;
-   - `OutputRequest`;
-   - `TypedValidationRequest`;
-   - `ChunkValidationRequest`; and
-   - `SerializedValidationRequest`.
-2. Add the same optional field to the erased/internal request carriers used by
-   registry builders, preparation, runner stage attempts, validator targets,
-   retry closures, and debug wrappers. Copy pointer values; never share a
-   mutable pointer owned by configuration.
-3. At every runner stage invocation, obtain the value from the exact resolved
-   producer binding. At every validator invocation, obtain it from that exact
-   resolved validator binding. Do not use the producer's value for a validator
-   or vice versa.
-4. Ensure all retry attempts for the same binding receive the same effective
-   structural-repair value. Do not decrement it in Notarius; PromptKit owns the
-   inner budget independently on each `CompleteStructured` call.
-5. Extend `semanticreconcile.Request` with the optional field and carry it into
-   each generic reconciliation completion. A batched reconciliation may make
-   several completion calls; each call receives the same effective budget.
-6. Update registry erasure/adaptation code for typed merge, normalize, and
-   validation requests so no field is lost. Preserve input/output support even
-   though current production input and output modules are deterministic.
-7. Add focused framework tests for one chunk producer, one extraction
-   producer, one normalizer, and one LLM-backed validator. Verify exact pointer
-   value propagation and separation between producer and validator settings.
-   Do not add repetitive tests for every generic adapter.
-
-### Tests And Validation
-
-```sh
-GOWORK=off go test ./internal/framework/contracts ./internal/framework/pipeline ./internal/framework/semanticreconcile
-GOWORK=off go test -race ./internal/framework/pipeline ./internal/framework/semanticreconcile
-GOWORK=off go test ./...
-git diff --check
-```
-
-### Acceptance Criteria
-
-- Every stage and validator request can carry a detached optional repair value.
-- The runner sources the value from the exact resolved binding.
-- Producer and validator values cannot overwrite one another.
-- Stage retries reuse but do not mutate or consume the inner repair budget.
-- Semantic reconciliation forwards the budget to every one of its completion
-  calls.
-- Existing behavior remains unchanged while all values are nil.
-
-## Stage 7: Forward Repair Policy From Every LLM-Backed Module ✅
-
-### Goal
-
-Complete the internal end-to-end path by having every production LLM-backed
-module forward its stage request value to `CompleteStructured`.
-
-### Implementation
-
-1. Inventory every production `CompleteStructured` call with code search before
-   editing. The expected current owners include:
-   - `dnd/scenes` chunking;
-   - the combat-turn, enemy-event, item-occurrence, item-registry,
-     location-occurrence, location-registry, NPC-occurrence, NPC-registry,
-     scene-description, and spell extractors; and
-   - generic semantic reconciliation used by the item, location, and NPC
-     registry normalizers.
-   Reconcile this list with the actual repository; do not omit a newly added
-   production caller merely because it is not named here.
-2. In each direct caller, set
-   `StructuredCompletionRequest.StructuredOutputRepairAttempts` from the
-   corresponding stage request. Clone the pointer or use a small shared helper
-   if that reduces repeated ownership mistakes without moving domain logic.
-3. Ensure D&D registry normalizers pass their typed normalize request value into
-   `semanticreconcile.Request`, and that the generic engine forwards it as
-   established in Stage 6.
-4. Update existing module prompt-mapping tests that already inspect a captured
-   structured-completion request to assert the new field. Do not create a new
-   one-test-per-module suite solely to memorialize field plumbing; rely on the
-   existing request-contract tests plus a final complete call-site audit.
-5. Search again after editing for production `CompleteStructured` calls and
-   verify each either forwards the field or documents why it cannot receive a
-   pipeline binding. Test-only fakes need only preserve the field when their
-   contract test depends on it.
-6. Do not set a module-specific fallback value. Nil must reach the adapter so
-   the prompt declaration remains authoritative.
-
-### Tests And Validation
-
-Run focused D&D and semantic-reconciliation packages, then the full suite:
-
-```sh
-GOWORK=off go test ./internal/modules/dnd/... ./internal/framework/semanticreconcile
-GOWORK=off go test -race ./internal/modules/dnd/... ./internal/framework/semanticreconcile
-GOWORK=off go test ./...
-git diff --check
-```
-
-### Acceptance Criteria
-
-- Every production LLM-backed completion receives the exact stage or validator
-  repair value.
-- No module invents a default or imports PromptKit.
-- Registry normalizers preserve the value through semantic reconciliation.
-- Existing request-contract tests remain concise and pass.
-- A final call-site audit finds no silent production omission.
-
-## Stage 8: Add The Public Repair Configuration Contract ✅
-
-### Goal
-
-Add presence-aware pipeline and binding configuration for
-`structured_output_repair_attempts` without yet changing runtime resolution.
-
-### Implementation
-
-1. Add `StructuredOutputRepairAttempts *int` to
-   `pipeline.PipelineProfile` and `pipeline.ModuleBinding`, using
-   `json:"structured_output_repair_attempts,omitempty"`.
-2. Add presence-aware YAML support at pipeline and object-binding scope:
-   - exact key `structured_output_repair_attempts`;
-   - integer values zero through three;
-   - explicit null, non-integer, and out-of-range values rejected with scoped
-     diagnostics; and
-   - scalar shorthand bindings continue to omit the binding override.
-   Preserve file configuration version 4.
-3. Update every configuration clone, conversion, redaction, summary, and JSON
-   round-trip carrier. Copy pointers by value into newly allocated storage so
-   parsed, configured, and redacted values do not alias.
-4. Preserve omission versus explicit zero through YAML parsing, profile
-   inheritance, module-binding object form, and JSON round trips. Keep scalar
-   shorthand bindings equivalent to omission.
-5. Do not add a top-level `promptkit.repair_attempts` setting or CLI override.
-6. Add concise parser and ownership tests. Defer execution-class checks,
-   effective precedence, resolved digests, and runtime forwarding to Stage 9,
-   where module metadata is available.
-
-### Tests And Validation
-
-At the parser/config boundary, test omitted, explicit zero, positive bounds,
-negative, above-three, null, non-integer, scalar shorthand, cloning, redaction,
-and JSON round-trip behavior. Use relational boundary tests for the allowed
-range and avoid duplicating the same cases at every layer.
-
-```sh
-GOWORK=off go test ./internal/core/config ./internal/cli
-GOWORK=off go test -race ./internal/core/config
-GOWORK=off go test ./...
-go run ./cmd/notarius config validate \
-  --config examples/dnd-minimal.config.yml \
-  --pipeline dnd-session
-go run ./cmd/notarius config validate \
-  --config examples/dnd-complete.config.yml \
-  --pipeline dnd-session
-git diff --check
-```
-
-### Acceptance Criteria
-
-- The exact public field parses at pipeline and object-binding scope with the
-  fixed range.
-- Nil and explicit zero remain distinguishable through parsing, cloning,
-  inheritance, redaction, summaries, and round trips.
-- Both maintained configurations remain valid without requiring the new field.
-- No runtime or prompt default has changed prematurely.
-
-## Stage 9: Resolve And Apply Repair Configuration ✅
-
-### Goal
-
-Resolve the public field against module execution classes, incorporate the
-effective value into pipeline identity, and connect it to the request plumbing
-completed in Stages 6 and 7.
-
-### Implementation
-
-1. During resolution, compute the effective value for every selected binding:
-   - explicit binding value wins;
-   - otherwise an explicit pipeline value applies to an LLM-backed binding;
-   - otherwise leave nil for prompt-owned policy.
-   Apply the pipeline value to LLM-backed validators as well as producers.
-2. Reject an explicit binding value on a deterministic module or deterministic
-   validator using the same execution-class knowledge used for `llm_profile`.
-   Do not reject a pipeline-level value merely because a selected pipeline also
-   contains deterministic bindings; simply do not apply it to those bindings.
-3. Clone every resolved pointer so the parsed pipeline, resolved profile,
-   redacted summaries, and runner requests have distinct ownership.
-4. Include the effective field in resolved pipeline JSON and digest input. A
-   change between nil, zero, and a positive value must change the resolved
-   digest when it changes an LLM-backed selected binding. Unselected lanes must
-   retain the repository's existing digest and selection semantics.
-5. Pass the resolved value into the Stage 6 request field for every selected
-   input, chunk, extract, merge, normalize, output, and validator binding.
-6. Update `docs/config.md` as the canonical field, range, and precedence
-   contract; `docs/internal/pipeline.md` as the resolution owner; and
-   `docs/operations.md` for the distinction from binding `retries`. The
-   prompt-owned production default remains unchanged until Stage 10.
-7. Add focused resolution and runner tests. Cover representative execution
-   classes rather than repeating the same assertion for every module type.
-
-### Tests And Validation
-
-Test:
-
-- binding over pipeline over nil precedence;
-- inheritance into each selected LLM-backed stage and validator;
-- no inheritance into deterministic bindings;
-- explicit deterministic-binding rejection;
-- detached pointers;
-- runner forwarding for representative producer and validator bindings; and
-- digest changes for execution-relevant nil, zero, and positive changes.
-
-```sh
-GOWORK=off go test ./internal/framework/pipeline ./internal/cli
-GOWORK=off go test -race ./internal/framework/pipeline
-GOWORK=off go test ./...
-go run ./cmd/notarius config validate \
-  --config examples/dnd-minimal.config.yml \
-  --pipeline dnd-session
-go run ./cmd/notarius config validate \
-  --config examples/dnd-complete.config.yml \
-  --pipeline dnd-session
-git diff --check
-```
-
-### Acceptance Criteria
-
-- The exact public field has the fixed binding-over-pipeline-over-prompt
-  precedence for every selected LLM-backed producer and validator.
-- Deterministic binding misuse fails during resolution before execution, while
-  a pipeline value coexists with deterministic bindings.
-- Nil and explicit zero remain distinguishable through resolution, runtime,
-  summaries, and digests.
-- A policy change invalidates checkpoint identity when it changes an effective
-  selected binding.
-- Current configuration, pipeline, and operations documentation matches the
-  implemented behavior.
-
-## Stage 10: Enable The Default, Finish Documentation, And Verify The Feature ✅
-
-### Goal
-
-Set the accepted production default of one repair, reconcile all canonical
-documentation, and run the full repository verification pass.
-
-### Implementation
-
-1. Change `repair_attempts: 0` to `repair_attempts: 1` in every maintained
-   production prompt manifest that produces structured output, including the
-   generic semantic-reconciliation prompt and every D&D chunk, extraction, and
-   registry-normalization prompt. Do not mechanically change unrelated test
-   fixtures whose purpose is to exercise zero.
-2. Inspect every production prompt output contract after the edit. Confirm that
-   each positive budget uses `basic`, `json`, or `json_schema`, remains no
-   greater than three, and retains its existing format and schema path.
-3. Add or refine the smallest durable assembled-assets test that proves the
-   production engine can prepare the maintained prompts with the activated
-   contracts. Do not add a brittle test that asserts an exact prompt count,
-   file count, message prose, correction text, or asset length. The public
-   default may be tested at one canonical assembled boundary because its
-   literal value is an operational contract.
-4. Confirm a successful repair does not create a warning and that exhausted
-   repair remains `ErrInvalidStructuredOutput`. Verify the debug prompt records
-   the configured contract, the debug response records actual repair count,
-   and cumulative usage is not double-counted.
-5. Confirm scheduling behavior with one focused test or existing coverage: the
-   Notarius scheduled client admits one logical `CompleteStructured` operation
-   while PromptKit may make serial corrective provider calls inside it. Do not
-   attempt to reacquire a Notarius permit from inside PromptKit or add a second
-   scheduler.
-6. Reconcile current-state documentation:
-   - `docs/integrations/pkg-promptkit.md` owns the pinned upstream boundary;
-   - `docs/config.md` owns field names, range, default, and precedence;
-   - `docs/operations.md` owns latency/cost, optional credentials, concurrency,
-     timeout, and the upper-bound formula;
-   - `docs/internal/llm.md` owns inspection-based contract replacement,
-     cumulative usage, actual repair count, generation errors, and debug data;
-   - `docs/internal/pipeline.md` owns effective policy propagation and the
-     separation from stage retries; and
-   - `docs/policy/architecture.md` adds only the durable rule that PromptKit
-     owns deterministic structural repair inside one completion while Notarius
-     owns stage attempts and semantic validation.
-7. Remove current-behavior claims that PromptKit is v0.5.0, that every
-   production repair budget is zero, or that PromptKit is always single-pass.
-   Do not alter historical release notes or archived roadmaps.
-8. Keep the maintained minimal and complete examples secret-free and valid.
-   They may omit the new field to demonstrate the default; do not add a
-   redundant complete profile or Rakestrawhome example merely to exercise an
-   upstream catalog entry.
-9. Review `docs/roadmap/future.md` only for consistency. Leave the future
-   feedback-aware stage retry, combat-scene validator, and warning-reform work
-   unimplemented and clearly separate.
-
-### Tests And Validation
-
-Run focused tests first, then all repository checks:
-
-```sh
-GOWORK=off go test ./internal/framework/llm ./internal/framework/pipeline ./internal/framework/semanticreconcile ./internal/modules/dnd/...
-GOWORK=off go test ./...
-GOWORK=off go test -race ./...
-GOWORK=off go vet ./...
-GOWORK=off go build ./cmd/notarius
-GOWORK=off go mod tidy -diff
-go run ./cmd/notarius config validate \
-  --config examples/dnd-minimal.config.yml \
-  --pipeline dnd-session
-go run ./cmd/notarius config validate \
-  --config examples/dnd-complete.config.yml \
-  --pipeline dnd-session
-git diff --check
-```
-
-Also perform focused repository searches that exclude `docs/roadmap/archive/`
-and historical release notes:
-
-- no active v0.5.0 PromptKit pins or links remain;
-- no maintained production prompt still declares `repair_attempts: 0`;
-- every production `CompleteStructured` caller forwards the repair field; and
-- no provider code, type, or message is added to ordinary errors, warnings,
-  manifests, cache, or checkpoint schemas.
-
-If the repository's source-release checker is available and the ordinary
-checks above pass, run `./scripts/check-release-source.sh v0.0.0` as the final
-integrated validation. It must not create a tag, release note, or repository
-artifact.
-
-### Acceptance Criteria
-
-- Every maintained structured prompt defaults to one corrective call and can
-  be overridden to zero through three at pipeline or binding scope.
-- A real assembled Notarius completion follows the PromptKit v0.8.0 repair
-  contract without changing prompt schema identity or cacheable prefix.
-- Actual repair count, cumulative usage, error classification, debug-only
-  provider diagnostics, scheduling, and checkpoint identity match the feature
-  roadmap.
-- Profile inheritance, Rakestrawhome availability, optional credentials, and
-  v0.6.0 hardening remain covered and documented.
-- All canonical documentation describes implemented v0.8.0 behavior in its
-  assigned home and leaves future semantic validation work in the roadmap.
-- Maintained examples validate, all ordinary/race/vet/build/module checks pass,
-  and the worktree contains no generated or sensitive artifacts.

docs/roadmap/promptkit-v0.8.md

@@ -1,520 +0,0 @@
-# PromptKit v0.8.0 Upgrade
-
-## Status
-
-Proposed.
-
-## Purpose
-
-Upgrade Notarius from PromptKit v0.5.0 to v0.8.0 and deliberately adopt the
-useful correctness, profile-composition, provider-diagnostic, backend, and
-structured-output-repair capabilities introduced in PromptKit v0.6.0, v0.7.0,
-and v0.8.0.
-
-The upgrade should improve structured-output reliability without confusing
-PromptKit's bounded deterministic repair with Notarius's existing stage retry
-budget or the future feedback-aware semantic-validation loop. PromptKit types
-and provider behavior must remain behind Notarius's transport-neutral LLM
-boundary.
-
-## Current State
-
-Notarius currently pins PromptKit v0.5.0. Its production adapter prepares one
-frozen execution, records credential-redacted details, and runs that same
-prepared value. It maps PromptKit capacity failures to an application-owned
-error, maps failed structured validation to `ErrInvalidStructuredOutput`, and
-returns PromptKit's raw validated bytes and usage metadata.
-
-Every maintained production prompt uses JSON Schema validation and currently
-declares `repair_attempts: 0`. Notarius stage bindings separately expose
-`retries`, which reruns a complete stage operation after an error or rejected
-candidate. The two mechanisms have different ownership and must remain
-independent.
-
-Notarius also maintains:
-
-- embedded prompt, schema, and fallback-profile filesystems;
-- operator profile-file and profile-directory sources;
-- one optional conventional `local` backend registration;
-- explicit profile preflight through PromptKit inspection;
-- one application-wide scheduled LLM client around the PromptKit adapter;
-- PromptKit profile-source fingerprints for checkpoint safety; and
-- redacted debug and manifest provenance at application-owned boundaries.
-
-The upgrade must preserve those established responsibilities while revising
-the pinned integration contract and any behavior affected by the three
-intervening releases.
-
-This roadmap is based on PromptKit's pinned release guides for
-[v0.6.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/releases/v0.6.0.md),
-[v0.7.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/releases/v0.7.0.md),
-and
-[v0.8.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.8.0/docs/releases/v0.8.0.md),
-plus the public API and format documentation at the v0.8.0 tag.
-
-## Target End State
-
-- `go.mod` and `go.sum` pin PromptKit v0.8.0 without a local replacement or
-  vendored copy.
-- Every maintained PromptKit prompt and profile prepares successfully under
-  v0.8.0's stricter validation and source-loading rules.
-- Eligible Notarius structured completions use one PromptKit corrective call by
-  default after a structurally invalid response. Operators can explicitly set
-  a value from zero through three for a configured pipeline, with a more local
-  LLM-backed binding override where needed.
-- PromptKit repair remains an inner operation within one Notarius stage
-  attempt. It never consumes or replenishes the binding's `retries` budget.
-- A successful repaired result exposes cumulative usage and the actual repair
-  count to Notarius's application-owned response and debug models. A repaired
-  success is not itself a warning.
-- Exhausted PromptKit validation remains an invalid structured-output result,
-  preserving the final candidate and diagnostics for debug and for any
-  applicable outer Notarius stage policy. Invalid structured output is never
-  accepted merely because the repair budget was exhausted.
-- Profile inheritance, the built-in Rakestrawhome backend/profile, optional
-  credential behavior, and structured generation errors work through the
-  existing Notarius PromptKit boundary and are accurately documented.
-- Provider-specific PromptKit types do not escape `internal/framework/llm`.
-- Checkpoint identity, effective configuration, redacted summaries, and debug
-  provenance reflect every execution-affecting repair or profile change.
-- Current documentation pins and describes v0.8.0; future Notarius semantic
-  validation retries remain roadmap behavior rather than being conflated with
-  this dependency upgrade.
-
-## Release-by-Release Adoption
-
-### PromptKit v0.6.0: Correctness, Safety, And Efficiency
-
-PromptKit v0.6.0 adds no public declarations, but intentionally rejects several
-formerly permissive or ambiguous inputs. The upgrade must audit Notarius's
-embedded and operator-facing integration against these rules:
-
-- YAML `id` and `version` metadata, rather than filenames, define prompt and
-  profile identity.
-- Prompt `content_file` paths are exact, relative, contained paths; built-in
-  file artifacts must resolve to regular files.
-- execution controls, output contracts, and repair budgets must be finite and
-  within their documented ranges;
-- provider endpoints must be absolute HTTP or HTTPS URLs with a host and no
-  user information, query, or fragment;
-- JSON documents and successful provider responses contain exactly one value;
-- successful provider responses are bounded to 16 MiB; and
-- JSON-compatible values are bounded for depth and expansion.
-
-Notarius should rely on PromptKit for these rules rather than duplicate its
-parsers or internal limits. Existing Notarius validation may retain a narrower
-application rule where it has independent value, but overlapping validation
-must agree with PromptKit and must not accept a value PromptKit will reject
-later.
-
-The upgrade automatically receives operation-local schema-plan reuse,
-artifact-text memoization, improved cancellation checks, and transport error
-identity preservation. Notarius should verify these changes through its real
-adapter boundary and avoid adding a second cache or response-body layer that
-would duplicate PromptKit's ownership.
-
-### PromptKit v0.7.0: Profiles, Backend Access, And Generation Errors
-
-#### Profile Inheritance
-
-Operator profiles may use `base_profile` to alias or selectively refine a
-built-in, fallback, or higher-precedence operator profile. Notarius must pass
-profile sources through unchanged and let PromptKit own parent lookup, merge
-rules, source precedence, cycle detection, and fully resolved prepared targets.
-
-Preflight inspection must resolve inherited profiles through the same source
-and backend composition used at execution. The selected leaf profile ID remains
-the public profile identity, while effective backend, endpoint, model, and
-reasoning provenance reflect the resolved chain. Notarius must not implement a
-second inheritance parser.
-
-The existing complete `dnd-extraction` fallback remains a standalone profile:
-PromptKit v0.8.0 does not provide a built-in `openai/gpt-5.6-luna` profile that
-would be an appropriate parent. Documentation should nevertheless explain how
-operators can use inheritance for environment-specific workload profiles and
-should link to PromptKit's pinned format contract rather than duplicate its
-field-by-field merge algorithm.
-
-Checkpoint safety must cover inherited behavior. Operator file/directory
-digests already cover changes to definitions in those sources, fallback asset
-digests cover application parents, and the PromptKit built-in catalog marker
-must change from its v0.5.0 identity to v0.8.0 so a changed built-in parent
-cannot reuse an incompatible checkpoint.
-
-#### Rakestrawhome Backend And Profile
-
-PromptKit's reserved `rakestrawhome` backend and
-`rakestrawhome-gemma-4-31b` profile become available without Notarius-specific
-registration. Notarius must not register or shadow the reserved backend ID.
-Profile preflight, backend-capacity reporting, scheduling, generation, and
-provenance should work for it through the same generic paths used by OpenRouter
-and `local`.
-
-The D&D default remains `dnd-extraction`; this upgrade does not silently move a
-production workload to Rakestrawhome. Operator documentation should identify
-the built-in profile as an available selection and link to PromptKit for its
-endpoint, credential environment, model, and capacity defaults.
-
-#### Optional Credentials
-
-An absent or blank optional `APIKeyEnv` now causes PromptKit to omit the
-`Authorization` header and send the request. Notarius must not restore the old
-failure behavior by pre-reading provider credential environment variables or
-by adding provider-specific authentication logic.
-
-Profile inspection may report an explicit `APIKeyRequired` policy without
-reading the credential, and execution remains the boundary at which that
-requirement is enforced. For optional profiles, an authentication-requiring
-provider may instead return a structured 401 or 403 generation failure. The
-configuration and operations documentation must explain this distinction.
-Notarius does not currently expose PromptKit's in-memory profile-registration
-API to operators, and PromptKit's filesystem profile format does not expose
-`APIKeyRequired`; therefore Notarius must not promise that an operator profile
-can force local credential preflight. Operators should provision the named
-environment variable, while Notarius should preserve the provider's structured
-authentication failure when it is absent.
-
-Notarius must continue to document mechanisms and environment-variable names,
-never secret values.
-
-#### Structured Generation Errors
-
-The adapter should recognize `*promptkit.GenerationError` with `errors.As` and
-translate useful information into an immutable, provider-neutral Notarius
-error classification. At minimum, retain the HTTP status code so callers and
-future retry policy can distinguish transport success with provider rejection
-from other generation failures.
-
-PromptKit's provider code, type, and message accessors are bounded but remain
-untrusted and potentially sensitive. They must never appear automatically in
-ordinary CLI output, warnings, manifests, checkpoint identity, or cache data.
-If retained for an explicitly requested debug trace, they must pass through
-Notarius's known-secret and bearer redaction and remain clearly identified as
-untrusted provider diagnostics. Default error formatting should continue to
-use a bounded, redacted application-owned message.
-
-Capacity and cancellation retain their current more specific classifications
-and precedence. This upgrade does not add automatic provider-error retry
-classification; it only preserves safe structured data needed for diagnosis
-and later policy.
-
-### PromptKit v0.8.0: Bounded Structured-Output Repair
-
-#### Default Policy
-
-Every maintained production prompt whose output is consumed as structured data
-should declare one repair attempt. All current production prompts use eligible
-JSON Schema validation, so no current prompt needs a zero default merely
-because of its output mode.
-
-One repair means at most one corrective generation after the initial
-candidate. PromptKit reconstructs the immutable original conversation and
-appends only the latest invalid assistant candidate and latest deterministic
-validation diagnostics. It preserves the selected target, direct session ID,
-provider-native structured-output contract, and backend capacity policy. This
-shape preserves the original cacheable prompt prefix and avoids accumulating
-unbounded failed history.
-
-The default is deliberately small. A single repair captures the common case in
-which a capable model can correct malformed JSON or a schema violation after
-receiving an exact diagnostic, while bounding the extra latency and cost of a
-single structured completion.
-
-#### Configuration Contract
-
-The public configuration is an optional, presence-aware
-`structured_output_repair_attempts` integer at pipeline scope and at each
-LLM-backed module or validator binding. Its effective precedence is:
-
-1. the binding value, when present;
-2. the pipeline value, when present; and
-3. the selected prompt's declared `repair_attempts` value.
-
-The value must be from zero through three. Explicit zero disables PromptKit
-repair at that scope. A deterministic binding must reject the field because it
-cannot perform structured LLM repair. Validator bindings may use it only when
-the selected validator is LLM-backed. Shorthand module bindings continue to
-inherit the pipeline or prompt default.
-
-The long, provider-neutral name is intentional: it distinguishes PromptKit's
-inner structural repair from the existing binding `retries` field, which owns
-complete stage attempts, without exposing a dependency name in generic
-pipeline contracts.
-
-The effective value must survive file parsing, cloning, redacted summaries,
-pipeline resolution, and pipeline digest construction without pointer aliasing
-or loss of presence. It must affect checkpoint identity because it can change
-the selected result, latency, token usage, and provider cost.
-
-#### Adapter Contract
-
-The transport-neutral structured-completion request should carry an optional
-application-owned structural-repair budget. No `promptkit.OutputContract` or
-other PromptKit type may cross the adapter boundary.
-
-PromptKit v0.8.0 request validation replaces the complete prompt output
-contract rather than merging one field. When Notarius has a configured
-override, the adapter must therefore inspect the selected prompt, copy its
-normalized declared format, validation mode, and schema path, change only the
-repair count, and supply that complete contract on the prepared request. A nil
-override continues to use the prompt declaration directly. Inspection and
-preparation must use the same immutable engine sources; a small adapter-local
-cache keyed by normalized prompt ID and version is acceptable but not required
-without measured need.
-
-This approach prevents configuration from accidentally dropping JSON Schema
-validation, avoids duplicating schema paths in pipeline YAML, and keeps prompt
-assets authoritative for every output-contract field other than the explicit
-operator override.
-
-The transport-neutral structured-completion response should report the actual
-number of PromptKit repair calls. PromptKit's returned token usage is already
-cumulative and must be passed through without re-summing it. Debug records
-should distinguish the configured budget from the actual count. Ordinary run
-manifests need not gain raw prompt or response data merely to report repairs;
-any durable aggregate should be added only if it has a clear consumer contract.
-
-#### Result And Failure Semantics
-
-- A valid initial candidate returns normally with zero actual repairs.
-- A valid corrected candidate returns normally with cumulative usage and its
-  positive actual repair count. It does not emit a warning solely because a
-  repair occurred.
-- Exhausting the repair budget returns PromptKit's final candidate and failed
-  validation result. The adapter maps this to
-  `ErrInvalidStructuredOutput`, preserves the response and debug material, and
-  does not decode or accept the candidate.
-- An explicitly empty or whitespace-only candidate participates in the
-  declared structural validation and repair flow. Missing, `null`, or
-  non-string provider content remains a malformed provider response.
-- A generation failure during a corrective call is an operational generation
-  failure and uses the same safe structured-error adaptation as an initial
-  generation failure.
-- Context cancellation remains authoritative throughout the initial and
-  corrective calls.
-
-PromptKit repair happens inside one scheduled `CompleteStructured` operation.
-The Notarius scheduler holds one permit for that logical operation while
-PromptKit performs its initial and serial corrective calls; PromptKit
-reacquires its own selected-backend capacity for each corrective generation.
-Because corrective calls are serial, this cannot expand actual concurrent
-provider work beyond the number of admitted Notarius operations, but
-documentation must stop describing the Notarius permit as a separate admission
-event for every internal repair call.
-
-One `CompleteStructured` invocation with effective PromptKit repair budget `R`
-may make at most `R + 1` provider calls. If one stage attempt makes `C`
-structured-completion invocations, a binding with `retries: N` has an upper
-bound of `(N + 1) * C * (R + 1)` provider calls; `C` may itself be a bounded,
-data-dependent module property, as it is for batched semantic reconciliation.
-LLM-backed validators have their own corresponding invocation counts, budgets,
-and costs. These formulas are upper bounds, not promises that every failure is
-retryable or that every attempt reaches the provider.
-
-## Profile And Prompt Source Compatibility
-
-The upgrade must preserve Notarius's source precedence: an operator source,
-then registered application fallback profiles, then PromptKit built-ins. A
-selected malformed definition remains authoritative and fails rather than
-falling through. Parent resolution introduced by profile inheritance observes
-that same precedence.
-
-All embedded prompt manifests, shared content fragments, response schemas, and
-fallback profiles must be prepared or inspected offline under v0.8.0. The
-review should specifically catch:
-
-- IDs inferred accidentally from filenames;
-- stale or escaping `content_file` paths;
-- missing or non-regular embedded artifacts;
-- repair values outside zero through three or paired with ineligible
-  validation;
-- schemas or examples that are not exact single JSON documents;
-- unsupported endpoint forms; and
-- JSON-compatible variables or profile extras that exceed upstream bounds.
-
-No prompt prose, schema shape, durable D&D artifact contract, or default D&D
-model should change merely to exercise the dependency. Prompt manifests should
-change only as needed to enable the adopted repair default and satisfy v0.8.0
-contracts.
-
-## Provenance, Debugging, And Security
-
-- Update the opaque PromptKit built-in profile-catalog identity from v0.5.0 to
-  v0.8.0. Do not hash or publish PromptKit's internal catalog bytes.
-- Ensure a prompt's repair default remains covered by its existing prompt asset
-  fingerprint and a configured effective override remains covered by the
-  resolved pipeline digest.
-- Preserve selected leaf profile identity while recording the inherited
-  effective target already exposed by PromptKit inspection and prepared
-  details.
-- Add actual structural-repair count and, when useful, the configured budget to
-  application-owned debug material. Token totals remain PromptKit's cumulative
-  values.
-- Do not generate a warning for a successful repair. Repair exhaustion is an
-  invalid-output failure, while provider rejection is a generation failure.
-- Never expose raw provider diagnostic fields without explicit debug capture
-  and application redaction. Do not place them in normal errors or durable
-  summaries.
-- Preserve context and transport error identity sufficiently for
-  `errors.Is`-based cancellation and deadline handling after adapting the
-  external error.
-
-## Documentation And Examples
-
-Implementation should update current-state documentation only when the new
-behavior lands:
-
-- `docs/integrations/pkg-promptkit.md` must pin v0.8.0 and define the revised
-  prepared-execution, repair, profile-inheritance, backend, credential, and
-  error-adaptation boundary.
-- `docs/config.md` must own the repair configuration fields, precedence,
-  allowed range, explicit-zero behavior, profile inheritance availability, and
-  optional credential semantics.
-- `docs/operations.md` must explain structural repair cost, timeout and
-  concurrency effects, credential failures, and its distinction from stage
-  retries.
-- `docs/internal/llm.md` must describe adapter contract replacement, actual
-  repair metadata, error adaptation, source compatibility, and scheduling.
-- `docs/internal/pipeline.md` must describe how effective repair configuration
-  is resolved and how inner repair differs from outer stage attempts.
-- `docs/policy/architecture.md` should receive only the durable ownership rule:
-  PromptKit owns bounded deterministic structural repair within one completion,
-  while Notarius owns stage attempts and semantic validation policy. Detailed
-  fields and retry formulas belong in their canonical configuration and
-  operations documents.
-
-Update maintained configuration examples only if the public Notarius
-configuration contract changes. A short inheritance illustration may remain in
-the configuration reference; do not create a complete example solely to copy
-PromptKit's upstream profile catalog. All upstream links must point to the
-v0.8.0 tag. Historical release or archived roadmap references should remain
-historical.
-
-No ADR is required solely to pin a newer dependency. The durable separation
-between PromptKit structural repair and Notarius semantic stage retries should
-be stated in architecture documentation now; the more extensive future
-validation state machine still warrants the separate ADR already identified in
-`future.md` when that work is promoted.
-
-## Validation And Acceptance Criteria
-
-The implementation is complete when:
-
-- the repository builds and tests against PromptKit v0.8.0 with no replacement
-  directive, workspace dependency, or vendored source;
-- every maintained prompt and profile prepares or inspects successfully under
-  the v0.8.0 source, path, endpoint, output-contract, and JSON-value rules;
-- an invalid first JSON Schema candidate followed by a valid correction returns
-  the valid raw output, cumulative usage, and actual repair count through the
-  Notarius adapter;
-- repair exhaustion returns the final raw candidate and debug material with an
-  error matching `ErrInvalidStructuredOutput`;
-- a corrective generation failure retains safe generation classification and
-  provider status without leaking untrusted provider detail;
-- explicit empty content follows structural validation rather than being
-  misclassified by Notarius;
-- repair configuration is presence-aware, range checked, rejected on
-  deterministic bindings, resolved with documented precedence, and included in
-  effective pipeline identity;
-- inherited profiles resolve consistently during preflight and execution, and
-  changes to any relevant operator, fallback, or built-in parent invalidate
-  checkpoint reuse;
-- the Rakestrawhome built-in profile reaches generic preflight, scheduling, and
-  provenance paths without application-specific registration;
-- optional missing credentials and explicitly required credentials behave as
-  documented without contacting real providers in tests;
-- cancellation, timeout, backend capacity, prepared-execution snapshot,
-  session ID, raw-output, debug-redaction, and existing profile provenance
-  behavior remain intact;
-- maintained examples validate successfully; and
-- canonical documentation contains no active v0.5.0 pin or claim that PromptKit
-  is always single-pass.
-
-Tests should follow `docs/policy/testing.md`: exercise observable Notarius
-contracts with offline fake clients or `httptest` boundaries, and do not copy
-PromptKit's entire internal repair test suite or assert its exact correction
-message prose. The dependency's internal wording is not a Notarius contract.
-
-## Non-Goals
-
-- Implementing Notarius's future feedback-aware semantic stage-retry loop.
-- Adding the D&D combat-scene semantic validator.
-- Redesigning warning policy or treating successful structural repair as a
-  warning.
-- Adding provider transport retries or deciding which HTTP statuses should
-  consume a stage retry.
-- Exposing PromptKit request, response, profile, validation, capacity, or error
-  types outside the LLM adapter.
-- Changing durable artifact schemas, D&D prompt semantics, the D&D default
-  model, or the fixed pipeline shape.
-- Reimplementing PromptKit profile inheritance, schema validation, response
-  bounds, repair conversations, backend admission, or provider parsing inside
-  Notarius.
-
-## Decisions
-
-### 1. Default Structured-Output Repair Budget
-
-**Decision: default to one repair attempt.** Set every maintained
-eligible production prompt to `repair_attempts: 1`. One corrective call is a
-strong fit for Notarius because every current production LLM response has a
-strict JSON Schema contract, smaller cost-effective models are a deliberate
-deployment target, and a precise structural diagnostic often makes one retry
-materially more successful. The budget is paid only after a structurally
-invalid candidate and remains tightly bounded.
-
-**Alternative considered: retain zero by default.** This preserves single-pass
-cost and latency and requires operators to opt in. It is preferable for an
-environment where every additional request is expensive or where upstream
-provider-native schema enforcement already produces negligible invalid output.
-It is less suitable as the Notarius default because one malformed response can
-otherwise discard substantial completed pipeline work.
-
-**Alternative considered: default to two.** This may improve recovery for
-weak models, but it doubles the worst-case corrective cost relative to the
-selected default and compounds with outer stage retries. It should be an
-operator choice supported by configuration, not the initial default, unless
-observational evidence shows that the second correction has a worthwhile
-marginal success rate.
-
-### 2. Repair Override Scope
-
-**Decision: support both pipeline and LLM-backed binding overrides.** Use
-the presence-aware `structured_output_repair_attempts` field and precedence
-defined above. A pipeline value provides the convenient one-line control the
-operator requested, while a binding value permits an expensive normalizer or
-future LLM-backed validator to use a deliberately different budget. This
-mirrors Notarius's established pipeline/binding profile inheritance and scales
-without editing embedded prompts.
-
-**Alternative considered: support only a pipeline override.** This is smaller to
-implement and document and still permits global enablement or disablement for
-one pipeline. Its drawback is that one exceptional prompt cannot opt out or
-request a larger budget without changing an embedded asset for every pipeline.
-
-**Alternative considered: expose one global value under the top-level
-`promptkit` configuration.** This makes client construction simple, but applies
-the same budget to unrelated pipelines and leaks an execution policy into the
-dependency configuration block. It is less compositional than pipeline-owned
-policy and therefore not recommended.
-
-### 3. Retention Of Provider-Supplied Generation Details
-
-**Decision: retain status in the application-owned error contract and
-retain redacted provider code, type, and message only in explicitly requested
-debug traces.** Status is useful for diagnosis and future retry policy without
-usually containing sensitive data. The other fields can materially explain a
-400 response but may echo request or schema content, so they belong only in the
-already-sensitive debug surface after Notarius redaction.
-
-**Alternative considered: retain only HTTP status and discard all provider fields.**
-This is the safest and smallest policy and still improves typed failure
-handling. It sacrifices potentially decisive provider diagnostics, leaving an
-operator with less information when a provider returns a terse status and the
-problem cannot be reproduced easily.
-
-**Alternative considered: include bounded provider code and type in normal
-errors while keeping message debug-only.** Codes and types are often stable and
-less sensitive than messages, but PromptKit explicitly classifies every
-provider field as untrusted. Promoting them to ordinary output creates a
-disclosure and compatibility burden that is not currently justified.

docs/roadmap/source-releases.md

@@ -1,282 +0,0 @@
-# Source-Only Releases
-
-## Status
-
-Implemented. Creating the first release under this procedure remains a
-separate maintainer operation.
-
-## Purpose
-
-Define a repeatable, guarded release process for Notarius without taking on a
-binary-distribution system that its current operator audience does not need.
-The process should make an exact source revision, its compatibility impact,
-and its validation status easy to identify while keeping installation in the
-hands of technically capable operators and deployment automation.
-
-The model is adapted from Weatherreporter's release procedure, but its target
-is deliberately narrower: an immutable source tag and checked-in release note
-are the release. Notarius does not publish executable archives or support
-Windows as part of this work.
-
-## Release Model
-
-Notarius releases come from commits on `main` and use stable semantic-version
-tags in the form `vMAJOR.MINOR.PATCH`. Prerelease tags are not part of the
-initial process.
-
-Every release has one nonempty, version-matched note at
-`docs/releases/<tag>.md`. The note and every affected current-state document
-must be present in the tagged commit. The Git tag and checked-in note together
-are the durable release record; no separately editable release page is
-required.
-
-Published tags are immutable. A maintainer must never move, reuse, or delete a
-published tag. If a published candidate is defective, the correction is made
-on `main` and released under a new patch version. An unpublished local tag may
-be deleted when candidate inspection finds a problem before any remote push.
-
-Before `v1.0.0`, a minor release may intentionally change a documented CLI,
-configuration, durable artifact, integration, or operating contract when its
-release note explains the impact and required operator action. A patch release
-must not intentionally break those documented contracts within its minor
-line.
-
-The existing `v0.1.0`, `v0.2.0`, and `v0.3.0` tags remain unchanged. They
-predate this procedure and do not need retrospective release notes. The first
-release made under this process establishes the release-note series.
-
-## Source-Only Distribution
-
-Notarius does not publish release binaries, archives, installers, container
-images, package-manager entries, checksum files, or signatures. A release tag
-is suitable for Go-native installation and for an operator-controlled build
-from an exact checkout.
-
-The primary installation form is:
-
-```sh
-GOWORK=off go install \
-  gitea.maximumdirect.net/eric/notarius/cmd/notarius@vMAJOR.MINOR.PATCH
-```
-
-Operator documentation should also describe cloning the repository, checking
-out the tag in detached-head state, and building `./cmd/notarius` with the Go
-version declared by `go.mod`. Private-module authentication and `GOPRIVATE`
-configuration belong to the operator environment and must be documented by
-mechanism rather than with real credentials.
-
-Consumers such as Narratio should pin the desired Notarius tag in provisioning
-or deployment configuration. They must continue to decide runtime
-compatibility from Notarius's published receipt and artifact schema contracts,
-not merely from the executable's product version.
-
-Packaged binaries may be reconsidered if distribution demand, installation
-friction, or a broader user audience justifies their build, signing, retention,
-and platform-support costs. They are not a prerequisite for a disciplined
-release process.
-
-## Platform Policy
-
-Linux is the supported deployment platform. Release validation must run the
-test suite and the release build on Linux and must confirm that the command
-builds with `CGO_ENABLED=0` for Linux `amd64` and `arm64`.
-
-macOS is a best-effort development and testing platform. Release validation
-should confirm that the command cross-compiles with `CGO_ENABLED=0` for Darwin
-`amd64` and `arm64`, but the project does not promise packaged artifacts or a
-separate runtime test environment for those targets.
-
-Windows is unsupported. The release process must not require Windows builds,
-Windows-specific compatibility work, or Windows documentation. Platform-
-specific implementation may intentionally use Unix facilities when they are
-important to Notarius's filesystem safety and operational model. Any later
-decision to support Windows requires its own feature scope and validation
-policy.
-
-## Version Reporting
-
-Add a root `notarius --version` interface for deployment diagnostics. It
-prints exactly one line:
-
-```text
-notarius vMAJOR.MINOR.PATCH
-```
-
-when the build has a valid release version, and:
-
-```text
-notarius development
-```
-
-when no release version is available.
-
-The implementation must obtain the main-module version from Go build
-information so `go install ...@vMAJOR.MINOR.PATCH` reports the selected tag. It
-must also accept an optional link-time version override so controlled builds
-and release CI can identify an exact tag from a checkout. The override must be
-validated and must not silently turn arbitrary text into a release version.
-Ordinary unversioned checkout builds remain `development`; the release process
-must not modify a tracked source constant for each release.
-
-Version reporting is an informational product interface. It does not replace
-receipt, configuration, prompt, or artifact schema versioning, and it must not
-be used as the sole downstream compatibility check.
-
-## Release Notes
-
-Each new `docs/releases/<tag>.md` document has this minimum structure:
-
-```markdown
-# Notarius vMAJOR.MINOR.PATCH
-
-This release ...
-
-## Summary
-
-## Compatibility
-
-## Upgrade
-
-## Changes
-```
-
-The note should concisely explain the release's purpose, compatibility with the
-preceding release, operator actions, and material user-visible, operational,
-integration, and maintainer-visible changes. It should link to canonical
-current-state documentation for exact contracts rather than duplicating those
-contracts.
-
-Release notes are durable historical summaries. They must not contain
-credentials, private infrastructure detail, sensitive campaign material, or
-claims that are not true of the tagged candidate. A release note does not
-excuse stale current-state documentation; affected canonical documents are
-updated in the same candidate.
-
-## Candidate Validation
-
-The release procedure must provide copyable POSIX-shell guards that validate
-the release version, release-note filename and heading, required note sections,
-repository state, and module hygiene. Validation must be run from the Notarius
-repository root with Go workspace behavior disabled.
-
-At minimum, a candidate must pass:
-
-- no tracked `go.work` or `go.work.sum`, no vendored tree, and no `replace`
-  directive in `go.mod`;
-- `GOWORK=off go test -count=1 ./...`;
-- `GOWORK=off go test -race -count=1 ./...`;
-- `GOWORK=off go vet ./...`;
-- `GOWORK=off go build ./...`;
-- `GOWORK=off go mod tidy -diff`;
-- `gofmt` verification for every tracked Go file;
-- `git diff --check` and `git diff --cached --check`;
-- validation of both maintained D&D configuration examples with their selected
-  pipeline;
-- Linux `amd64` and `arm64` static command builds;
-- best-effort Darwin `amd64` and `arm64` static command builds; and
-- a focused manual or automated check that every added or changed local
-  Markdown link resolves.
-
-The candidate review also checks for generated binaries, test output,
-credentials, temporary files, module replacements, vendored dependencies, and
-other unintended source-control content. Tests remain offline and do not call
-an LLM provider or require live credentials.
-
-## Candidate Publication
-
-The release procedure must guard the exact commit immediately before tagging.
-It requires:
-
-- the current branch is `main`;
-- the worktree and index are clean;
-- the candidate commit has been pushed and exactly matches `origin/main`;
-- the matching release note exists in that commit;
-- no local or remote tag already uses the selected version; and
-- the substantive release checks have passed for that exact candidate.
-
-The maintainer records the exact candidate commit, creates a lightweight tag
-bound explicitly to that commit, verifies the local tag target, and pushes only
-that tag ref. The procedure must not recommend `git push --tags`.
-
-After publication, the maintainer verifies that the remote tag resolves to the
-guarded commit and that the release note can be read from the tagged tree. A
-fresh temporary checkout or `go install ...@<tag>` must build successfully, and
-the resulting command must report the expected version through `--version`.
-
-## Validation-Only Release Automation
-
-Add a tag-triggered Woodpecker pipeline that validates source releases without
-publishing artifacts. It should:
-
-- accept only stable semantic-version tags;
-- require the version-matched release note;
-- run the same substantive module, test, race, vet, build, formatting, and
-  whitespace checks as the documented local procedure;
-- validate the maintained configuration examples;
-- perform the supported and best-effort cross-build checks; and
-- verify a release-version build's `notarius --version` output on the CI host.
-
-The pipeline must not upload binaries, create archives or checksums, create or
-edit a Gitea release object, or require a release API token. Local guards remain
-authoritative before tag publication because CI begins only after the tag is
-already remote.
-
-If tag validation fails, preserve the published tag, fix the cause on `main`,
-select a new patch version, and repeat the full process. Do not weaken tag
-immutability merely because the release contains source rather than binaries.
-
-## Documentation Ownership
-
-In the target state:
-
-- `docs/release.md` owns the maintainer release procedure, commands, ordering,
-  publication checks, and failure recovery;
-- `docs/releases/` owns one historical summary per release made under the new
-  process;
-- `docs/cli.md` owns the `--version` contract;
-- `README.md` owns the shortest source-installation example and links to the
-  release procedure where useful;
-- `docs/development.md` routes release preparation, tagging, and verification
-  work to `docs/release.md`;
-- `docs/policy/documentation.md` assigns canonical ownership to the release
-  procedure and release notes;
-- `docs/policy/architecture.md` records Linux support, best-effort macOS
-  development, unsupported Windows, and source-only distribution only if those
-  are judged durable development invariants rather than release mechanics; and
-- `docs/operations.md` describes only installation or deployment consequences
-  relevant to operators and links to canonical CLI and release contracts.
-
-Current-state documentation must not describe the new release process,
-`--version`, or automated validation until the corresponding behavior exists.
-
-## Acceptance Criteria
-
-- A maintainer can prepare, validate, tag, publish, and verify a source release
-  by following `docs/release.md` without relying on undocumented knowledge.
-- Every new release has an immutable semantic-version tag and matching
-  checked-in release note in the tagged commit.
-- The guarded candidate is clean, synchronized with `origin/main`, and passes
-  the documented substantive checks before tagging.
-- Tag-triggered CI independently validates the published source and never
-  publishes binary artifacts.
-- `go install` of a tagged version succeeds and `notarius --version` reports
-  that version; ordinary unversioned builds report `development`.
-- Linux is the documented supported deployment platform, macOS has a
-  best-effort development build check, and Windows is explicitly unsupported.
-- Downstream compatibility remains based on durable Notarius contracts rather
-  than the product version alone.
-- Existing pre-procedure tags remain untouched and require no invented release
-  history.
-
-## Non-Goals
-
-- Publishing executable archives, installers, container images, checksums,
-  signatures, or package-manager entries.
-- Supporting or cross-compiling for Windows.
-- Creating or maintaining a mutable Gitea release page.
-- Supporting prerelease tag syntax in the initial procedure.
-- Automating version selection, release-note authorship, commits, or tag
-  creation.
-- Retrospectively creating release notes for `v0.1.0` through `v0.3.0`.
-- Treating a product version as a substitute for receipt, configuration,
-  prompt, or artifact schema compatibility.

go.mod

@@ -3,9 +3,14 @@ module gitea.maximumdirect.net/eric/notarius
 go 1.25.5
 
 require (
-	gitea.maximumdirect.net/eric/promptkit v0.8.0
+	gitea.maximumdirect.net/eric/promptkit v0.9.0
 	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2
 	gopkg.in/yaml.v3 v3.0.1
 )
 
 require golang.org/x/text v0.40.0
+
+require (
+	gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0 // indirect
+	gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0 // indirect
+)

go.sum

@@ -1,5 +1,9 @@
-gitea.maximumdirect.net/eric/promptkit v0.8.0 h1:NGd9hDLu0UMxKbvittMrqM5Ua94eFb+kOE7UIir8l08=
-gitea.maximumdirect.net/eric/promptkit v0.8.0/go.mod h1:R95NM6fbMDGDC0/UomgnSBP6ui2ns+8SZb8bESNvrDQ=
+gitea.maximumdirect.net/eric/promptkit v0.9.0 h1:IpvDRC8L6xRxQ9hpuyKOmMc5b6MeLTKYyx+h1YAjy08=
+gitea.maximumdirect.net/eric/promptkit v0.9.0/go.mod h1:oMJ/WUJImUtwJ5e+6MAGECPYAErAkOaKel0G+3T/b4E=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0 h1:lc062euk2qseO//D762i3JaFyulDNML3eQQX7DkYTho=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0/go.mod h1:AIa7kAu2mfrRQgcspe4L+DW51WqgnALQT60lqkEywJI=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0 h1:j9YY7wsTVjzke2kHH4YAzpU0oUpM+x+nXwl1IeS+2eg=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0/go.mod h1:4RNS+LILDg4JbS4Ts9Lwy1C92wauXJIbeQaalps4Koo=
 github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI=
 github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 h1:KRzFb2m7YtdldCEkzs6KqmJw4nqEVZGK7IN2kJkjTuQ=

internal/cli/assembled_spell_pipeline_contract_test.go

@@ -21,6 +21,10 @@ import (
 
 const assembledSpellExtractorKey = "test/dnd/spell-casts"
 
+const assembledCorrectingSpellExtractorKey = "test/dnd/correcting-spell-casts"
+
+const assembledDirectSpellValidatorKey = "test/dnd/direct-spell-correction"
+
 func TestAssembledSpellPipelineNormalizesMergedCasts(t *testing.T) {
 	registries, resolved, extractor := assembledSpellPipeline(t, assembledSpellPipelineOptions{})
 	prepared, err := pipeline.Prepare(resolved, registries, pipeline.ModuleDependencies{})
@@ -67,28 +71,39 @@ func TestAssembledSpellPipelineNormalizesMergedCasts(t *testing.T) {
 		t.Fatalf("distinct cast = %#v, want separate evidence event", distinct)
 	}
 
-	wantWarningReasons := []string{
+	wantDiagnosticReasons := []string{
 		spellnormalize.ReasonCodeSpellNameCanonicalized,
 		spellnormalize.ReasonCodeSourceReferencesNormalized,
 		spellnormalize.ReasonCodeDuplicateSpellCastCollapsed,
 		"spell_not_near_source",
 	}
-	gotWarningReasons := make([]string, len(output.Warnings))
-	for index, warning := range output.Warnings {
-		gotWarningReasons[index] = warning.ReasonCode
+	gotDiagnosticReasons := make([]string, len(output.Diagnostics.Groups))
+	for index, group := range output.Diagnostics.Groups {
+		gotDiagnosticReasons[index] = group.ReasonCode
 	}
-	if !reflect.DeepEqual(gotWarningReasons, wantWarningReasons) {
-		t.Fatalf("warnings = %#v, want deterministic normalize and validation warnings", output.Warnings)
+	if !reflect.DeepEqual(gotDiagnosticReasons, wantDiagnosticReasons) {
+		t.Fatalf("diagnostics = %#v, want deterministic normalize and validation diagnostics", output.Diagnostics)
 	}
-	if output.Warnings[2].Scope != "spell_casts[0]" || !strings.Contains(output.Warnings[2].Message, "retained input index 0") || !strings.Contains(output.Warnings[2].Message, "removed input indices [1]") {
-		t.Fatalf("duplicate warning = %#v, want retained and removed merged indices", output.Warnings[2])
+	if output.Diagnostics.Groups[2].Samples[0].Scope != "spell_casts[0]" || !strings.Contains(output.Diagnostics.Groups[2].Samples[0].Message, "retained input index 0") || !strings.Contains(output.Diagnostics.Groups[2].Samples[0].Message, "removed input indices [1]") {
+		t.Fatalf("duplicate diagnostic = %#v, want retained and removed merged indices", output.Diagnostics.Groups[2])
 	}
 
 	warningsFile := decodeAssembledOutput[struct {
-		Warnings []contracts.Warning `json:"warnings"`
+		Groups []contracts.DiagnosticGroup `json:"groups"`
 	}](t, output.OutputFiles, "warnings.json")
-	if !reflect.DeepEqual(warningsFile.Warnings, output.Warnings) {
-		t.Fatalf("warnings file = %#v, run warnings = %#v, want manifest output path to preserve warnings", warningsFile.Warnings, output.Warnings)
+	if len(warningsFile.Groups) != 0 {
+		t.Fatalf("warnings file = %#v, want no process warnings for advisory-only diagnostics", warningsFile.Groups)
+	}
+	diagnosticsFile := decodeAssembledOutput[struct {
+		SchemaVersion                string                      `json:"schema_version"`
+		GroupCount                   int                         `json:"group_count"`
+		OccurrenceCount              int                         `json:"occurrence_count"`
+		Truncated                    bool                        `json:"truncated"`
+		UnrepresentedOccurrenceCount int                         `json:"unrepresented_occurrence_count"`
+		Groups                       []contracts.DiagnosticGroup `json:"groups"`
+	}](t, output.OutputFiles, "diagnostics.json")
+	if diagnosticsFile.SchemaVersion != "notarius.diagnostics.v1" || diagnosticsFile.GroupCount != len(output.Diagnostics.Groups) || !reflect.DeepEqual(diagnosticsFile.Groups, output.Diagnostics.Groups) || diagnosticsFile.OccurrenceCount != diagnosticGroupOccurrences(output.Diagnostics.Groups)+output.Diagnostics.UnrepresentedOccurrenceCount || diagnosticsFile.Truncated != output.Diagnostics.Truncated || diagnosticsFile.UnrepresentedOccurrenceCount != output.Diagnostics.UnrepresentedOccurrenceCount {
+		t.Fatalf("diagnostics file = %#v, run diagnostics = %#v", diagnosticsFile, output.Diagnostics)
 	}
 	manifest := decodeAssembledOutput[artifacts.RunManifest](t, output.OutputFiles, "manifest.json")
 	if len(manifest.ArtifactLanes) != 1 || manifest.ArtifactLanes[0].Normalizer != spellnormalize.Key {
@@ -101,6 +116,30 @@ func TestAssembledSpellPipelineNormalizesMergedCasts(t *testing.T) {
 	}
 }
 
+func TestAssembledSpellPipelineCorrectsRejectedDirectExtraction(t *testing.T) {
+	registries, resolved, extractor := assembledCorrectingSpellPipeline(t)
+	prepared, err := pipeline.Prepare(resolved, registries, pipeline.ModuleDependencies{})
+	if err != nil {
+		t.Fatalf("Prepare() error = %v, want nil", err)
+	}
+
+	output, err := pipeline.New().Run(context.Background(), pipeline.RunInput{
+		Prepared:       prepared,
+		RawInput:       readRepositoryFile(t, "examples", "seriatim-minimal-transcript.json"),
+		ChunkCacheMode: pipeline.ChunkCacheBypass,
+	})
+	if err != nil {
+		t.Fatalf("Run() error = %v, want nil", err)
+	}
+	if len(output.Rejected) != 0 || output.Manifest.ValidationStatus != "approved" || len(output.NormalizeOutputs) != 1 {
+		t.Fatalf("run output = %#v, want corrected accepted spell output", output)
+	}
+	correction := extractor.correctionSnapshot()
+	if correction == nil || string(correction.AssistantResponse) != `{"spell":"Mysterious Burst"}` || !strings.Contains(correction.UserGuidance, "use a known spell name") || !strings.Contains(correction.UserGuidance, "complete corrected replacement") || strings.Contains(correction.UserGuidance, "unknown_spell") || strings.Contains(correction.UserGuidance, "spell is not in the catalog") {
+		t.Fatalf("extract correction = %#v, want exact rejected model response and semantic replacement guidance only", correction)
+	}
+}
+
 func TestAssembledSpellPipelineHonorsNormalizeValidatorOverride(t *testing.T) {
 	registries, resolved, _ := assembledSpellPipeline(t, assembledSpellPipelineOptions{normalizeValidatorOverride: true})
 	var normalizeChain *pipeline.ResolvedValidatorChain
@@ -127,15 +166,16 @@ func TestAssembledSpellPipelineHonorsNormalizeValidatorOverride(t *testing.T) {
 	if err != nil || output.Manifest.ValidationStatus != "approved" || len(output.Rejected) != 0 || len(output.NormalizeOutputs) != 1 {
 		t.Fatalf("Run() error = %v output = %#v, want approved override run", err, output)
 	}
-	for _, warning := range output.Warnings {
-		if warning.ReasonCode == "spell_not_near_source" {
-			t.Fatalf("warnings = %#v, want explicit validator override to replace default relatedness chain", output.Warnings)
+	for _, group := range output.Diagnostics.Groups {
+		if group.ReasonCode == "spell_not_near_source" {
+			t.Fatalf("diagnostics = %#v, want explicit validator override to replace default relatedness chain", output.Diagnostics)
 		}
 	}
 }
 
-func TestAssembledSpellPipelinePromotesTerminalUnknownSpellWarning(t *testing.T) {
+func TestAssembledSpellPipelinePromotesTerminalUnknownSpellDiagnostics(t *testing.T) {
 	registries, resolved, _ := assembledSpellPipeline(t, assembledSpellPipelineOptions{unknownSpell: true})
+	resolved.Steps[0].ArtifactLanes[0].NormalizeValidationPolicy.SemanticRejection = pipeline.SemanticRejectionRejectOutput
 	prepared, err := pipeline.Prepare(resolved, registries, pipeline.ModuleDependencies{})
 	if err != nil {
 		t.Fatalf("Prepare() error = %v, want nil", err)
@@ -161,12 +201,12 @@ func TestAssembledSpellPipelinePromotesTerminalUnknownSpellWarning(t *testing.T)
 	if !reflect.DeepEqual(rejectedFile.Rejected, output.Rejected) {
 		t.Fatalf("rejected file = %#v, run rejections = %#v, want durable rejection diagnostic", rejectedFile.Rejected, output.Rejected)
 	}
-	if len(output.Warnings) != 1 || output.Warnings[0].ReasonCode != spellnormalize.ReasonCodeSpellNameUnresolved || output.Warnings[0].Scope != "spell_casts[0]" {
-		t.Fatalf("warnings = %#v, want terminal normalize catalog warning", output.Warnings)
+	if len(output.Diagnostics.Groups) != 2 || output.Diagnostics.Groups[0].ReasonCode != spellnormalize.ReasonCodeSpellNameUnresolved || output.Diagnostics.Groups[0].Samples[0].Scope != "spell_casts[0]" || output.Diagnostics.Groups[1].ReasonCode != "spell_not_near_source" {
+		t.Fatalf("diagnostics = %#v, want complete terminal normalize validation diagnostics", output.Diagnostics)
 	}
 }
 
-func TestAssembledSpellPipelinePromotesUnknownSpellWarningWhenOverrideAccepts(t *testing.T) {
+func TestAssembledSpellPipelinePromotesUnknownSpellAdvisoryWhenOverrideAccepts(t *testing.T) {
 	registries, resolved, _ := assembledSpellPipeline(t, assembledSpellPipelineOptions{normalizeValidatorOverride: true, unknownSpell: true})
 	prepared, err := pipeline.Prepare(resolved, registries, pipeline.ModuleDependencies{})
 	if err != nil {
@@ -190,14 +230,20 @@ func TestAssembledSpellPipelinePromotesUnknownSpellWarningWhenOverrideAccepts(t
 	if len(normalized.SpellCasts) != 1 || normalized.SpellCasts[0].Spell != "Mysterious Burst" {
 		t.Fatalf("normalized casts = %#v, want unresolved name preserved", normalized.SpellCasts)
 	}
-	if len(output.Warnings) != 1 || output.Warnings[0].ReasonCode != spellnormalize.ReasonCodeSpellNameUnresolved || output.Warnings[0].Scope != "spell_casts[0]" {
-		t.Fatalf("warnings = %#v, want promoted scoped unresolved-name warning", output.Warnings)
+	if len(output.Diagnostics.Groups) != 1 || output.Diagnostics.Groups[0].ReasonCode != spellnormalize.ReasonCodeSpellNameUnresolved || output.Diagnostics.Groups[0].Samples[0].Scope != "spell_casts[0]" {
+		t.Fatalf("diagnostics = %#v, want promoted scoped unresolved-name diagnostic", output.Diagnostics)
 	}
 	warningsFile := decodeAssembledOutput[struct {
-		Warnings []contracts.Warning `json:"warnings"`
+		Groups []contracts.DiagnosticGroup `json:"groups"`
 	}](t, output.OutputFiles, "warnings.json")
-	if !reflect.DeepEqual(warningsFile.Warnings, output.Warnings) {
-		t.Fatalf("warnings file = %#v, run warnings = %#v, want durable unresolved-name warning", warningsFile.Warnings, output.Warnings)
+	if len(warningsFile.Groups) != 0 {
+		t.Fatalf("warnings file = %#v, want no process warnings for an advisory diagnostic", warningsFile.Groups)
+	}
+	diagnosticsFile := decodeAssembledOutput[struct {
+		Groups []contracts.DiagnosticGroup `json:"groups"`
+	}](t, output.OutputFiles, "diagnostics.json")
+	if !reflect.DeepEqual(diagnosticsFile.Groups, output.Diagnostics.Groups) {
+		t.Fatalf("diagnostics file = %#v, run diagnostics = %#v", diagnosticsFile.Groups, output.Diagnostics)
 	}
 }
 
@@ -206,6 +252,47 @@ type assembledSpellPipelineOptions struct {
 	unknownSpell               bool
 }
 
+func assembledCorrectingSpellPipeline(t *testing.T) (pipeline.Registries, pipeline.ResolvedPipeline, *assembledCorrectingSpellExtractor) {
+	t.Helper()
+	components := productionTestComponents(t)
+	extractor := &assembledCorrectingSpellExtractor{}
+	if err := pipeline.RegisterExtractor[dnd.SpellList](components.registries.Extractors, pipeline.ModuleSpec{
+		Key:                assembledCorrectingSpellExtractorKey,
+		Stage:              pipeline.StageExtract,
+		ExecutionClass:     contracts.ExecutionClassLLMBacked,
+		CorrectionProtocol: contracts.CorrectionProtocolSingleResponseV1,
+		Requires:           []string{"chunks", "source.transcript"},
+		Provides:           []string{"dnd.spell_casts"},
+		ArtifactKind:       dnd.SpellListKind,
+	}, func() (contracts.Extractor[dnd.SpellList], error) {
+		return extractor, nil
+	}); err != nil {
+		t.Fatalf("register correcting extractor: %v", err)
+	}
+	if err := pipeline.RegisterTypedValidator[dnd.SpellList](components.registries.Validators, dnd.SpellListKind, pipeline.ValidatorSpec{Key: assembledDirectSpellValidatorKey, ExecutionClass: contracts.ExecutionClassDeterministic}, func() (contracts.TypedValidator[dnd.SpellList], error) {
+		return assembledDirectSpellValidator{}, nil
+	}); err != nil {
+		t.Fatalf("register direct spell validator: %v", err)
+	}
+
+	extract := pipeline.Binding(assembledCorrectingSpellExtractorKey)
+	extract.Retries = 1
+	extract.Validators = pipeline.ValidatorOverride{Set: true, Validators: []pipeline.ModuleBinding{{Module: assembledDirectSpellValidatorKey}}}
+	resolved, err := pipeline.ResolvePipeline(pipeline.PipelineProfile{
+		ID:    "assembled-dnd-correcting-spells",
+		Input: pipeline.Binding("seriatim"),
+		Chunk: pipeline.ModuleBinding{Module: "generic", Options: map[string]any{"max_units": 1}},
+		Artifacts: map[string]pipeline.ArtifactLaneProfile{
+			"spells": {Extract: extract, Normalize: pipeline.Binding(spellnormalize.Key)},
+		},
+		Output: pipeline.Binding("json"),
+	}, pipeline.ResolveOptions{}, catalogFromRegistries(components.registries))
+	if err != nil {
+		t.Fatalf("ResolvePipeline() error = %v, want nil", err)
+	}
+	return components.registries, resolved, extractor
+}
+
 func assembledSpellPipeline(t *testing.T, options assembledSpellPipelineOptions) (pipeline.Registries, pipeline.ResolvedPipeline, *assembledSpellExtractor) {
 	t.Helper()
 	components := productionTestComponents(t)
@@ -251,6 +338,69 @@ type assembledSpellExtractor struct {
 	unknownSpell bool
 }
 
+type assembledCorrectingSpellExtractor struct {
+	mu         sync.Mutex
+	correction *contracts.SemanticCorrection
+}
+
+func (*assembledCorrectingSpellExtractor) Key() string { return assembledCorrectingSpellExtractorKey }
+
+func (*assembledCorrectingSpellExtractor) ReferenceSlots() []contracts.ReferenceSlot { return nil }
+
+func (e *assembledCorrectingSpellExtractor) Extract(_ context.Context, req contracts.TypedExtractionRequest) (contracts.TypedExtractionResult[dnd.SpellList], error) {
+	if req.Source == nil || req.Chunk == nil {
+		return contracts.TypedExtractionResult[dnd.SpellList]{}, fmt.Errorf("correcting assembled extractor requires source and chunk")
+	}
+	response := `{"spell":"accepted"}`
+	value := dnd.SpellList{SpellCasts: []dnd.SpellCast{}}
+	if req.Chunk.Index == 0 && req.Correction == nil {
+		response = `{"spell":"Mysterious Burst"}`
+		value.SpellCasts = []dnd.SpellCast{{Caster: "Aria", Spell: "Mysterious Burst", SourceRefs: []source.SourceRef{{SourceID: req.Source.ID, StartUnitID: 1, EndUnitID: 1}}}}
+	}
+	if req.Chunk.Index == 0 && req.Correction != nil {
+		correction, err := contracts.CloneSemanticCorrection(req.Correction)
+		if err != nil {
+			return contracts.TypedExtractionResult[dnd.SpellList]{}, err
+		}
+		e.mu.Lock()
+		e.correction = correction
+		e.mu.Unlock()
+		value.SpellCasts = []dnd.SpellCast{{Caster: "Aria", Spell: "Cure Wounds", SourceRefs: []source.SourceRef{{SourceID: req.Source.ID, StartUnitID: 1, EndUnitID: 1}}}}
+	}
+	candidate, err := contracts.NewModelCandidate([]byte(response), contracts.CorrectionProtocolSingleResponseV1)
+	if err != nil {
+		return contracts.TypedExtractionResult[dnd.SpellList]{}, err
+	}
+	return contracts.TypedExtractionResult[dnd.SpellList]{Value: value, ModelCandidate: candidate}, nil
+}
+
+func (e *assembledCorrectingSpellExtractor) correctionSnapshot() *contracts.SemanticCorrection {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	correction, err := contracts.CloneSemanticCorrection(e.correction)
+	if err != nil {
+		return nil
+	}
+	return correction
+}
+
+type assembledDirectSpellValidator struct{}
+
+func (assembledDirectSpellValidator) Name() string { return assembledDirectSpellValidatorKey }
+
+func (assembledDirectSpellValidator) ExecutionClass() contracts.ExecutionClass {
+	return contracts.ExecutionClassDeterministic
+}
+
+func (assembledDirectSpellValidator) Validate(_ context.Context, req contracts.TypedValidationRequest[dnd.SpellList]) (contracts.ValidationResult, error) {
+	for _, cast := range req.Value.SpellCasts {
+		if cast.Spell == "Mysterious Burst" {
+			return contracts.ValidationResult{Approved: false, ReasonCode: "unknown_spell", Message: "spell is not in the catalog", CorrectionGuidance: "use a known spell name"}, nil
+		}
+	}
+	return contracts.ValidationResult{Approved: true}, nil
+}
+
 func (e *assembledSpellExtractor) Key() string { return assembledSpellExtractorKey }
 
 func (*assembledSpellExtractor) ReferenceSlots() []contracts.ReferenceSlot { return nil }
@@ -296,6 +446,14 @@ func (e *assembledSpellExtractor) chunkIndexesSnapshot() []int {
 	return append([]int(nil), e.chunkIndexes...)
 }
 
+func diagnosticGroupOccurrences(groups []contracts.DiagnosticGroup) int {
+	count := 0
+	for _, group := range groups {
+		count += group.OccurrenceCount
+	}
+	return count
+}
+
 func decodeAssembledOutput[T any](t *testing.T, files []contracts.OutputFile, name string) T {
 	t.Helper()
 	for _, file := range files {

internal/cli/dnd_enemy_events_contract_test.go

@@ -396,8 +396,12 @@ func (client *enemyEventLLMClient) CompleteStructured(ctx context.Context, reque
 	if err := json.Unmarshal(content, output); err != nil {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("populate fake structured target: %w", err)
 	}
+	snapshot, err := contracts.CloneStructuredCompletionRequest(request)
+	if err != nil {
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("clone fake request: %w", err)
+	}
 	client.mu.Lock()
-	client.requests = append(client.requests, request)
+	client.requests = append(client.requests, snapshot)
 	client.mu.Unlock()
 	return contracts.StructuredCompletionResponse{Content: content, Provider: "test", Model: "deterministic", ProfileID: request.ProfileID}, nil
 }
@@ -408,7 +412,11 @@ func (client *enemyEventLLMClient) requestsFor(promptID string) []contracts.Stru
 	var requests []contracts.StructuredCompletionRequest
 	for _, request := range client.requests {
 		if request.PromptID == promptID {
-			requests = append(requests, request)
+			snapshot, err := contracts.CloneStructuredCompletionRequest(request)
+			if err != nil {
+				panic(err)
+			}
+			requests = append(requests, snapshot)
 		}
 	}
 	return requests

internal/cli/dnd_scene_descriptions_contract_test.go

@@ -91,8 +91,8 @@ func TestProductionSceneDescriptionWorkflow(t *testing.T) {
 	if !reflect.DeepEqual(durable, want) {
 		t.Fatalf("durable output payload = %#v, want %#v", durable, want)
 	}
-	if len(output.Warnings) != 0 {
-		t.Fatalf("warnings = %#v, want grounded descriptions without warnings", output.Warnings)
+	if len(output.Diagnostics.Groups) != 0 {
+		t.Fatalf("diagnostics = %#v, want grounded descriptions without diagnostics", output.Diagnostics)
 	}
 }
 

internal/cli/example_contract_test.go

@@ -239,10 +239,10 @@ func TestMaintainedMinimalInvocationProducesJSONBundle(t *testing.T) {
 		t.Fatalf("rejected = %#v, want empty rejection list", rejected.Rejected)
 	}
 	warnings := readProductionJSON[struct {
-		Warnings []json.RawMessage `json:"warnings"`
+		Groups []json.RawMessage `json:"groups"`
 	}](t, filepath.Join(runRoot, "warnings.json"))
-	if len(warnings.Warnings) != 0 {
-		t.Fatalf("warnings = %#v, want empty warning list", warnings.Warnings)
+	if len(warnings.Groups) != 0 {
+		t.Fatalf("warnings = %#v, want empty warning list", warnings.Groups)
 	}
 }
 

internal/cli/production_contract_test.go

@@ -343,6 +343,9 @@ func TestProductionPromptAssetsPrepareWithoutProviderCredentials(t *testing.T) {
 		Inputs  []struct {
 			Name string `yaml:"name"`
 		} `yaml:"inputs"`
+		Messages []struct {
+			Role string `yaml:"role"`
+		} `yaml:"messages"`
 	}
 	preparedPrompts := 0
 	if err := fs.WalkDir(promptFS, ".", func(path string, entry fs.DirEntry, walkErr error) error {
@@ -360,6 +363,11 @@ func TestProductionPromptAssetsPrepareWithoutProviderCredentials(t *testing.T) {
 		if err := yaml.Unmarshal(data, &prompt); err != nil {
 			return err
 		}
+		for _, message := range prompt.Messages {
+			if message.Role != promptkit.RoleSystem && message.Role != promptkit.RoleUser {
+				return fmt.Errorf("production prompt %q uses role %q, want system or user", prompt.ID, message.Role)
+			}
+		}
 		inputs := make(map[string]promptkit.ArtifactRef, len(prompt.Inputs))
 		for _, input := range prompt.Inputs {
 			inputs[input.Name] = promptkit.Inline(`{}`)
@@ -844,10 +852,10 @@ func TestProductionSceneRunRecordsAnnotationFreeChunkPlanAndProvenance(t *testin
 		t.Fatalf("chunk map range annotations = %#v, want none", chunkMap.Chunks[0].Annotations)
 	}
 	warnings := readProductionJSON[struct {
-		Warnings []contracts.Warning `json:"warnings"`
+		Groups []contracts.DiagnosticGroup `json:"groups"`
 	}](t, filepath.Join(outputRoot, productionRunID, "warnings.json"))
-	if len(warnings.Warnings) != 0 {
-		t.Fatalf("warnings = %#v, want none", warnings.Warnings)
+	if len(warnings.Groups) != 0 {
+		t.Fatalf("warnings = %#v, want none", warnings.Groups)
 	}
 	if len(fake.requestsFor(scenes.PromptID)) != 1 || len(fake.requestsFor(spells.PromptID)) != 1 || len(fake.requestsFor(itemoccurrenceextract.PromptID)) != 1 {
 		t.Fatalf("fake prompt requests = %#v, want one scene, spell, and item-occurrence request", fake.requestPrompts())
@@ -1091,8 +1099,12 @@ func (client *productionFakeLLMClient) CompleteStructured(ctx context.Context, r
 	if err := json.Unmarshal(content, out); err != nil {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("populate fake structured target: %w", err)
 	}
+	snapshot, err := contracts.CloneStructuredCompletionRequest(req)
+	if err != nil {
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("clone fake request: %w", err)
+	}
 	client.mu.Lock()
-	client.requests = append(client.requests, req)
+	client.requests = append(client.requests, snapshot)
 	client.mu.Unlock()
 	return contracts.StructuredCompletionResponse{Content: content, Provider: "test", Model: "deterministic", ProfileID: req.ProfileID}, nil
 }
@@ -1103,7 +1115,11 @@ func (client *productionFakeLLMClient) requestsFor(promptID string) []contracts.
 	var requests []contracts.StructuredCompletionRequest
 	for _, req := range client.requests {
 		if req.PromptID == promptID {
-			requests = append(requests, req)
+			snapshot, err := contracts.CloneStructuredCompletionRequest(req)
+			if err != nil {
+				panic(err)
+			}
+			requests = append(requests, snapshot)
 		}
 	}
 	return requests

internal/cli/reference_contract_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"os"
 	"path/filepath"
+	"slices"
 	"strings"
 	"testing"
 
@@ -14,20 +15,16 @@ import (
 
 func TestReferenceSelectorsParseAndApplyAllDocumentedForms(t *testing.T) {
 	tests := []struct {
-		name      string
-		selector  string
-		only      []string
-		wantStage pipeline.ModuleStage
-		wantLane  string
-		wantSlot  string
+		name     string
+		selector string
+		want     []string
 	}{
-		{name: "flat", selector: "alpha-slot", wantStage: pipeline.StageExtract, wantLane: "alpha", wantSlot: "alpha-slot"},
-		{name: "chunk", selector: "chunk.chunk-slot", wantStage: pipeline.StageChunk, wantSlot: "chunk-slot"},
-		{name: "merge", selector: "merge.alpha-merge", only: []string{"alpha"}, wantStage: pipeline.StageMerge, wantLane: "alpha", wantSlot: "alpha-merge"},
-		{name: "lane", selector: "alpha.alpha-slot", wantStage: pipeline.StageExtract, wantLane: "alpha", wantSlot: "alpha-slot"},
-		{name: "lane extract", selector: "alpha.extract.alpha-slot", wantStage: pipeline.StageExtract, wantLane: "alpha", wantSlot: "alpha-slot"},
-		{name: "lane merge", selector: "alpha.merge.alpha-merge", wantStage: pipeline.StageMerge, wantLane: "alpha", wantSlot: "alpha-merge"},
-		{name: "lane normalize", selector: "alpha.normalize.alpha-normalize", wantStage: pipeline.StageNormalize, wantLane: "alpha", wantSlot: "alpha-normalize"},
+		{name: "pipeline", selector: "shared", want: []string{"alpha.extract.shared", "alpha.merge.shared", "alpha.normalize.shared", "beta.extract.shared", "beta.merge.shared", "beta.normalize.shared"}},
+		{name: "chunk", selector: "chunk.chunk-slot", want: []string{"chunk.chunk-slot"}},
+		{name: "lane", selector: "alpha.shared", want: []string{"alpha.extract.shared", "alpha.merge.shared", "alpha.normalize.shared"}},
+		{name: "lane extract", selector: "alpha.extract.alpha-slot", want: []string{"alpha.extract.alpha-slot"}},
+		{name: "lane merge", selector: "alpha.merge.alpha-merge", want: []string{"alpha.merge.alpha-merge"}},
+		{name: "lane normalize", selector: "alpha.normalize.alpha-normalize", want: []string{"alpha.normalize.alpha-normalize"}},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
@@ -37,70 +34,135 @@ func TestReferenceSelectorsParseAndApplyAllDocumentedForms(t *testing.T) {
 			if err != nil {
 				t.Fatal(err)
 			}
-			overrides, _, err := resolveCLIReferenceRequests(cfg, "demo", tt.only, catalog, []cliReferenceRequest{{Selector: selector, Source: "reference.txt"}}, nil)
+			overrides, _, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog, []cliReferenceRequest{{Selector: selector, Source: "reference.txt"}}, nil)
 			if err != nil {
 				t.Fatalf("resolve selector: %v", err)
 			}
-			if len(overrides) != 1 {
-				t.Fatalf("overrides = %#v, want one binding", overrides)
+			if got := referenceContractBindingLabels(overrides); !slices.Equal(got, tt.want) {
+				t.Fatalf("binding targets = %#v, want %#v", got, tt.want)
 			}
-			got := overrides[0]
-			if got.Stage != tt.wantStage || got.LaneID != tt.wantLane || got.SlotName != tt.wantSlot || got.BindingSource != contracts.ReferenceBindingSourceCLI {
-				t.Fatalf("binding = %#v, want %s/%s/%s from CLI", got, tt.wantStage, tt.wantLane, tt.wantSlot)
-			}
-		})
-	}
-}
-
-func TestReferenceSelectorsRejectAmbiguityWithSpecificSuggestions(t *testing.T) {
-	cfg := referenceContractConfig()
-	catalog := referenceContractCatalog(t, true, true)
-	for _, tt := range []struct {
-		name     string
-		selector string
-		want     []string
-	}{
-		{name: "flat shared slot", selector: "shared", want: []string{"alpha.extract.shared", "beta.extract.shared"}},
-		{name: "lane shared slot", selector: "alpha.shared", want: []string{"alpha.extract.shared", "alpha.merge.shared", "alpha.normalize.shared"}},
-		{name: "all mergers", selector: "merge.shared", want: []string{"alpha.merge.shared", "beta.merge.shared"}},
-	} {
-		t.Run(tt.name, func(t *testing.T) {
-			selector, err := parseReferenceSelector(tt.selector, "--reference")
-			if err != nil {
-				t.Fatal(err)
-			}
-			_, _, err = resolveCLIReferenceRequests(cfg, "demo", nil, catalog, []cliReferenceRequest{{Selector: selector, Source: "reference.txt"}}, nil)
-			if err == nil {
-				t.Fatal("resolve selector succeeded, want ambiguity error")
-			}
-			for _, fragment := range tt.want {
-				if !strings.Contains(err.Error(), fragment) {
-					t.Fatalf("error = %q, want suggestion %q", err, fragment)
+			for _, binding := range overrides {
+				if binding.Source != "reference.txt" || binding.BindingSource != contracts.ReferenceBindingSourceCLI {
+					t.Fatalf("binding = %#v, want CLI source", binding)
 				}
 			}
 		})
 	}
 }
 
-func TestReferenceSelectorsRespectSelectedLanesBeforeMaterialization(t *testing.T) {
+func TestReferenceSelectorSpecificityAndFinalOccurrenceChooseConcreteBindings(t *testing.T) {
+	cfg := referenceContractConfig()
+	catalog := referenceContractCatalog(t, true, true)
+	requests := []cliReferenceRequest{
+		{Selector: mustParseReferenceSelector(t, "shared", "--reference"), Source: "pipeline-first.txt"},
+		{Selector: mustParseReferenceSelector(t, "shared", "--reference"), Source: "pipeline-final.txt"},
+		{Selector: mustParseReferenceSelector(t, "alpha.shared", "--reference"), Source: "lane.txt"},
+		{Selector: mustParseReferenceSelector(t, "alpha.extract.shared", "--reference"), Source: "binding.txt"},
+	}
+	overrides, unbinds, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog, requests, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(unbinds) != 0 {
+		t.Fatalf("unbinds = %#v, want none", unbinds)
+	}
+	want := map[string]string{
+		"alpha.extract.shared":   "binding.txt",
+		"alpha.merge.shared":     "lane.txt",
+		"alpha.normalize.shared": "lane.txt",
+		"beta.extract.shared":    "pipeline-final.txt",
+		"beta.merge.shared":      "pipeline-final.txt",
+		"beta.normalize.shared":  "pipeline-final.txt",
+	}
+	for _, binding := range overrides {
+		label := referenceContractBindingLabel(binding)
+		if binding.Source != want[label] {
+			t.Fatalf("binding %s source = %q, want %q", label, binding.Source, want[label])
+		}
+		delete(want, label)
+	}
+	if len(want) != 0 {
+		t.Fatalf("missing bindings: %#v", want)
+	}
+}
+
+func TestCompleteDNDSharedCLIReferencesExpandAcrossCompatibleTargets(t *testing.T) {
+	cfg := loadMaintainedExample(t, repositoryPath("examples", "dnd-complete.config.yml"))
+	catalog := catalogFromRegistries(productionTestComponents(t).registries)
+	sources := map[string]string{
+		"party":         "/references/party.txt",
+		"players":       "/references/players.txt",
+		"glossary":      "/references/glossary.txt",
+		"spell_catalog": "/references/spells.json",
+	}
+	requests := make([]cliReferenceRequest, 0, len(sources))
+	for _, slot := range []string{"party", "players", "glossary", "spell_catalog"} {
+		requests = append(requests, cliReferenceRequest{
+			Selector: mustParseReferenceSelector(t, slot, "--reference"),
+			Source:   sources[slot],
+		})
+	}
+	overrides, unbinds, err := resolveCLIReferenceRequests(cfg, "dnd-session", nil, catalog, requests, nil)
+	if err != nil {
+		t.Fatalf("expand complete D&D references: %v", err)
+	}
+	if len(unbinds) != 0 {
+		t.Fatalf("unbinds = %#v, want none", unbinds)
+	}
+
+	actual := make(map[string]pipeline.ReferenceBinding, len(overrides))
+	for _, binding := range overrides {
+		actual[referenceContractBindingLabel(binding)] = binding
+	}
+	targets, err := selectedReferenceTargets(cfg, "dnd-session", nil, catalog)
+	if err != nil {
+		t.Fatal(err)
+	}
+	matched := make(map[string]int, len(sources))
+	for _, target := range targets {
+		for slot, sourcePath := range sources {
+			if _, ok := target.slots[slot]; !ok {
+				continue
+			}
+			matched[slot]++
+			label := targetLabel(target) + "." + slot
+			binding, ok := actual[label]
+			if !ok || binding.Source != sourcePath || binding.BindingSource != contracts.ReferenceBindingSourceCLI {
+				t.Fatalf("binding %q = %#v, want CLI source %q", label, binding, sourcePath)
+			}
+		}
+	}
+	for slot := range sources {
+		if matched[slot] < 2 {
+			t.Fatalf("reference %q matched %d target(s), want a shared D&D reference", slot, matched[slot])
+		}
+	}
+	if _, err := cfg.Resolve(config.ResolveInput{PipelineID: "dnd-session", Catalog: catalog, ReferenceOverrides: overrides}); err != nil {
+		t.Fatalf("resolve complete D&D CLI references: %v", err)
+	}
+}
+
+func TestReferenceSelectorsRejectInvalidOrUnselectedScopesBeforeMaterialization(t *testing.T) {
 	cfg := referenceContractConfig()
 	catalog := referenceContractCatalog(t, true, true)
 	for _, tt := range []struct {
 		name     string
 		selector string
+		only     []string
 		want     string
 	}{
-		{name: "unselected lane", selector: "beta.extract.beta-slot", want: `reference lane "beta" is not selected`},
+		{name: "pipeline slot", selector: "missing", want: `reference slot "missing" is not declared by any selected target`},
+		{name: "lane slot", selector: "alpha.missing", want: `reference slot "missing" is not declared by selected lane "alpha"`},
+		{name: "binding slot", selector: "alpha.extract.missing", want: `reference slot "missing" is not declared`},
+		{name: "former merge shorthand", selector: "merge.shared", want: `reference lane "merge" is not selected`},
+		{name: "unselected lane", selector: "beta.extract.beta-slot", only: []string{"alpha"}, want: `reference lane "beta" is not selected`},
 		{name: "unknown lane", selector: "missing.extract.beta-slot", want: `reference lane "missing" is not selected`},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
-			selector, err := parseReferenceSelector(tt.selector, "--reference")
-			if err != nil {
-				t.Fatal(err)
-			}
-			_, _, err = resolveCLIReferenceRequests(cfg, "demo", []string{"alpha"}, catalog, []cliReferenceRequest{{Selector: selector, Source: filepath.Join(t.TempDir(), "missing.txt")}}, nil)
+			selector := mustParseReferenceSelector(t, tt.selector, "--reference")
+			_, _, err := resolveCLIReferenceRequests(cfg, "demo", tt.only, catalog, []cliReferenceRequest{{Selector: selector, Source: filepath.Join(t.TempDir(), "missing.txt")}}, nil)
 			if err == nil || !strings.Contains(err.Error(), tt.want) || strings.Contains(err.Error(), "missing.txt") {
-				t.Fatalf("error = %v, want selection failure before file access", err)
+				t.Fatalf("error = %v, want selection failure containing %q before file access", err, tt.want)
 			}
 		})
 	}
@@ -131,40 +193,50 @@ func TestReferenceSyntaxErrorsReturnTwo(t *testing.T) {
 	}
 }
 
-func TestReferenceOverridesUseFinalExactTargetBinding(t *testing.T) {
+func TestReferenceBindAndUnbindSpecificity(t *testing.T) {
 	cfg := referenceContractConfig()
 	catalog := referenceContractCatalog(t, true, true)
-	alphaShared, err := parseReferenceSelector("alpha.extract.shared", "--reference")
-	if err != nil {
-		t.Fatal(err)
-	}
-	betaShared, err := parseReferenceSelector("beta.extract.shared", "--reference")
-	if err != nil {
-		t.Fatal(err)
-	}
-	overrides, unbinds, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog, []cliReferenceRequest{
-		{Selector: alphaShared, Source: "alpha-first.txt"},
-		{Selector: alphaShared, Source: "alpha-final.txt"},
-		{Selector: betaShared, Source: "beta-only.txt"},
-	}, nil)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(unbinds) != 0 {
-		t.Fatalf("unbinds = %#v, want none", unbinds)
-	}
-	effective, err := cfg.Resolve(config.ResolveInput{PipelineID: "demo", Catalog: catalog, ReferenceOverrides: overrides})
-	if err != nil {
-		t.Fatalf("resolve pipeline: %v", err)
-	}
-	alpha := referenceContractLane(t, effective.ResolvedPipeline, "alpha")
-	beta := referenceContractLane(t, effective.ResolvedPipeline, "beta")
-	if source := referenceContractBindingSource(alpha.ExtractReferences.Bindings, "shared"); source != "alpha-final.txt" {
-		t.Fatalf("alpha shared source = %q, want final exact-target override", source)
-	}
-	if source := referenceContractBindingSource(beta.ExtractReferences.Bindings, "shared"); source != "beta-only.txt" {
-		t.Fatalf("beta shared source = %q, want target-specific override", source)
-	}
+	t.Run("specific unbind carves out broad binding", func(t *testing.T) {
+		overrides, unbinds, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog,
+			[]cliReferenceRequest{{Selector: mustParseReferenceSelector(t, "shared", "--reference"), Source: "shared.txt"}},
+			[]cliReferenceUnbindRequest{{Selector: mustParseReferenceSelector(t, "alpha.extract.shared", "--without-reference")}},
+		)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if got := referenceContractBindingLabels(overrides); slices.Contains(got, "alpha.extract.shared") || len(got) != 5 {
+			t.Fatalf("overrides = %#v, want all shared targets except alpha extract", got)
+		}
+		if got := referenceContractUnbindLabels(unbinds); !slices.Equal(got, []string{"alpha.extract.shared"}) {
+			t.Fatalf("unbinds = %#v, want alpha extract", got)
+		}
+	})
+
+	t.Run("specific binding restores broad unbind", func(t *testing.T) {
+		overrides, unbinds, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog,
+			[]cliReferenceRequest{{Selector: mustParseReferenceSelector(t, "alpha.extract.shared", "--reference"), Source: "alpha.txt"}},
+			[]cliReferenceUnbindRequest{{Selector: mustParseReferenceSelector(t, "shared", "--without-reference")}},
+		)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if got := referenceContractBindingLabels(overrides); !slices.Equal(got, []string{"alpha.extract.shared"}) {
+			t.Fatalf("overrides = %#v, want alpha extract", got)
+		}
+		if got := referenceContractUnbindLabels(unbinds); slices.Contains(got, "alpha.extract.shared") || len(got) != 5 {
+			t.Fatalf("unbinds = %#v, want all shared targets except alpha extract", got)
+		}
+	})
+
+	t.Run("same specificity conflicts", func(t *testing.T) {
+		_, _, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog,
+			[]cliReferenceRequest{{Selector: mustParseReferenceSelector(t, "alpha.shared", "--reference"), Source: "alpha.txt"}},
+			[]cliReferenceUnbindRequest{{Selector: mustParseReferenceSelector(t, "alpha.shared", "--without-reference")}},
+		)
+		if err == nil || !strings.Contains(err.Error(), "same specificity") {
+			t.Fatalf("error = %v, want same-specificity conflict", err)
+		}
+	})
 }
 
 func TestReferenceUnbindsRemoveOptionalAndProtectRequiredSlots(t *testing.T) {
@@ -257,6 +329,52 @@ func TestReferenceMaterializationSeparatesCLIAndConfigPathOrigins(t *testing.T)
 	}
 }
 
+func TestPipelineScopedCLIReferenceProtectsGeneratedHandoff(t *testing.T) {
+	cfg := referenceContractConfig()
+	profile := cfg.Pipelines["demo"]
+	alpha := profile.Artifacts["alpha"]
+	beta := profile.Artifacts["beta"]
+	alpha.Extract.References["shared"] = pipeline.GeneratedReference("produce", "beta")
+	profile.Artifacts = nil
+	profile.Steps = []pipeline.PipelineStepProfile{
+		{ID: "produce", Artifacts: map[string]pipeline.ArtifactLaneProfile{"beta": beta}},
+		{ID: "consume", Artifacts: map[string]pipeline.ArtifactLaneProfile{"alpha": alpha}},
+	}
+	cfg.Pipelines["demo"] = profile
+	catalog := referenceContractCatalog(t, true, true)
+
+	t.Run("binding conflicts before file access", func(t *testing.T) {
+		overrides, _, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog, []cliReferenceRequest{{
+			Selector: mustParseReferenceSelector(t, "shared", "--reference"),
+			Source:   filepath.Join(t.TempDir(), "never-read.json"),
+		}}, nil)
+		if err != nil {
+			t.Fatalf("expand CLI reference: %v", err)
+		}
+		_, err = cfg.Resolve(config.ResolveInput{PipelineID: "demo", Catalog: catalog, ReferenceOverrides: overrides})
+		if err == nil || !strings.Contains(err.Error(), "conflicting generated and external bindings") || strings.Contains(err.Error(), "never-read.json") {
+			t.Fatalf("resolve error = %v, want generated/external conflict before file access", err)
+		}
+	})
+
+	t.Run("unbind leaves generated source intact", func(t *testing.T) {
+		_, unbinds, err := resolveCLIReferenceRequests(cfg, "demo", nil, catalog, nil, []cliReferenceUnbindRequest{{
+			Selector: mustParseReferenceSelector(t, "shared", "--without-reference"),
+		}})
+		if err != nil {
+			t.Fatalf("expand CLI unbind: %v", err)
+		}
+		effective, err := cfg.Resolve(config.ResolveInput{PipelineID: "demo", Catalog: catalog, ReferenceUnbinds: unbinds})
+		if err != nil {
+			t.Fatalf("resolve generated reference with CLI unbind: %v", err)
+		}
+		binding := referenceContractFindBinding(referenceContractLane(t, effective.ResolvedPipeline, "alpha").ExtractReferences.Bindings, "shared")
+		if binding == nil || binding.Artifact == nil || binding.Artifact.Step != "produce" || binding.Artifact.Lane != "beta" {
+			t.Fatalf("generated binding = %#v, want preserved produce/beta handoff", binding)
+		}
+	})
+}
+
 func TestReferenceTargetLookupUsesArtifactVariantsAndReportsMissingContext(t *testing.T) {
 	cfg := referenceContractConfig()
 	full := referenceContractCatalog(t, true, true)
@@ -368,7 +486,7 @@ func referenceContractCatalog(t *testing.T, includeBetaMerger, includeBetaNormal
 	register(registries.Chunkers.RegisterWithSpec(pipeline.ModuleSpec{Key: "reference/chunk", Stage: pipeline.StageChunk, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"source"}, Provides: []string{"chunks"}, ReferenceSlots: []contracts.ReferenceSlot{{Name: "chunk-slot"}, {Name: "required-chunk", Required: true}}}, func() (contracts.Chunker, error) { return stateTestChunker{}, nil }))
 	register(pipeline.RegisterArtifactCodec(registries.ArtifactCodecs, referenceContractCodecA{}))
 	register(pipeline.RegisterArtifactCodec(registries.ArtifactCodecs, referenceContractCodecB{}))
-	register(pipeline.RegisterExtractor(registries.Extractors, pipeline.ModuleSpec{Key: "reference/extract-alpha", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: referenceContractKindAlpha, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "alpha-slot"}, {Name: "required-extract", Required: true}}}, func() (contracts.Extractor[stateTestArtifact], error) { return stateTestExtractor{}, nil }))
+	register(pipeline.RegisterExtractor(registries.Extractors, pipeline.ModuleSpec{Key: "reference/extract-alpha", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: referenceContractKindAlpha, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared", AcceptedArtifactKinds: []contracts.ArtifactKind{referenceContractKindBeta}, AcceptedMediaTypes: []string{"application/json"}}, {Name: "alpha-slot"}, {Name: "required-extract", Required: true}}}, func() (contracts.Extractor[stateTestArtifact], error) { return stateTestExtractor{}, nil }))
 	register(pipeline.RegisterExtractor(registries.Extractors, pipeline.ModuleSpec{Key: "reference/extract-beta", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: referenceContractKindBeta, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "beta-slot"}, {Name: "required-extract", Required: true}}}, func() (contracts.Extractor[stateTestArtifact], error) { return stateTestExtractor{}, nil }))
 	register(pipeline.RegisterMerger(registries.Mergers, pipeline.ModuleSpec{Key: "reference/shared-merge", Stage: pipeline.StageMerge, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"artifact"}, Provides: []string{"merged"}, ArtifactKind: referenceContractKindAlpha, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "alpha-merge"}, {Name: "required-merge", Required: true}}}, func() (contracts.Merger[stateTestArtifact], error) { return stateTestMerger{}, nil }))
 	if includeBetaMerger {
@@ -444,6 +562,42 @@ func referenceContractBindingSource(bindings []pipeline.ReferenceBinding, slot s
 	return ""
 }
 
+func mustParseReferenceSelector(t *testing.T, value, flagName string) cliReferenceSelector {
+	t.Helper()
+	selector, err := parseReferenceSelector(value, flagName)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return selector
+}
+
+func referenceContractBindingLabels(bindings []pipeline.ReferenceBinding) []string {
+	labels := make([]string, 0, len(bindings))
+	for _, binding := range bindings {
+		labels = append(labels, referenceContractBindingLabel(binding))
+	}
+	return labels
+}
+
+func referenceContractBindingLabel(binding pipeline.ReferenceBinding) string {
+	if binding.Stage == pipeline.StageChunk {
+		return "chunk." + binding.SlotName
+	}
+	return binding.LaneID + "." + string(binding.Stage) + "." + binding.SlotName
+}
+
+func referenceContractUnbindLabels(unbinds []pipeline.ReferenceUnbind) []string {
+	labels := make([]string, 0, len(unbinds))
+	for _, unbind := range unbinds {
+		labels = append(labels, referenceContractBindingLabel(pipeline.ReferenceBinding{
+			Stage:    unbind.Stage,
+			LaneID:   unbind.LaneID,
+			SlotName: unbind.SlotName,
+		}))
+	}
+	return labels
+}
+
 func referenceContractFindBinding(bindings []pipeline.ReferenceBinding, slot string) *pipeline.ReferenceBinding {
 	for i := range bindings {
 		if bindings[i].SlotName == slot {

internal/cli/run.go

@@ -177,7 +177,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	fs.Var(&llmProfile, "llm-profile", "LLM profile override")
 	fs.Var(&reasoningEffort, "reasoning-effort", "reasoning effort override")
 	fs.Var(&chunkCache, "chunk_cache", "chunk plan cache mode: auto, bypass, or refresh")
-	fs.Var(&referenceFlags, "reference", "reference binding, as slot=path, chunk.slot=path, merge.slot=path, lane.slot=path, lane.extract.slot=path, lane.merge.slot=path, or lane.normalize.slot=path")
+	fs.Var(&referenceFlags, "reference", "reference binding, as slot=path, chunk.slot=path, lane.slot=path, lane.extract.slot=path, lane.merge.slot=path, or lane.normalize.slot=path")
 	fs.Var(&withoutReferenceFlags, "without-reference", "unbind a reference, using the same selector forms as --reference")
 	fs.Var(&recomputeStep, "recompute-step", "recompute one ordered pipeline step and dependent lanes")
 	if err := validateRunFlagValues(args); err != nil {
@@ -375,7 +375,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	if err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("resolve working directory: %w", err))
 	}
-	materialized, referenceWarnings, err := pipeline.MaterializeReferences(effective.ResolvedPipeline, catalog, pipeline.ReferenceMaterializationOptions{
+	materialized, referenceDiagnostics, err := pipeline.MaterializeReferences(effective.ResolvedPipeline, catalog, pipeline.ReferenceMaterializationOptions{
 		ConfigPath: loadedConfigPath,
 		WorkingDir: workingDir,
 	})
@@ -462,7 +462,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		StartedAt:        startedAt,
 		LLMProfiles:      llmProfiles,
 		Metadata:         runMetadata(effective.Config.Output.Directory, debugPath),
-		Warnings:         referenceWarnings,
+		Diagnostics:      referenceDiagnostics,
 		ChunkCacheMode:   effective.Config.Cache.ChunkPlans.Mode,
 		ChunkPlans:       chunkPlans,
 		Checkpoints:      checkpointRecorder,
@@ -471,7 +471,10 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		Debug:            debugRecorder,
 		ExtractWorkers:   cfg.Concurrency.StageWorkers["extract"],
 	})
-	commandState.observeOutput(output)
+	diagnosticProjection, diagnosticErr := contracts.ProjectDiagnosticCollection(output.Diagnostics)
+	if diagnosticErr == nil {
+		commandState.observeOutput(output, diagnosticProjection)
+	}
 	if err != nil {
 		primaryErr := fmt.Errorf("run pipeline %q: %w", pipelineID, err)
 		if output.Manifest.PipelineID != "" {
@@ -479,15 +482,21 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 				return failPipelineCommand(stderr, commandState, terminalWriter, primaryErr, fmt.Errorf("write debug summary: %w", summaryErr))
 			}
 		}
+		if diagnosticErr != nil {
+			return failPipelineCommand(stderr, commandState, terminalWriter, primaryErr, fmt.Errorf("summarize run diagnostics: %w", diagnosticErr))
+		}
 		return failPipelineCommand(stderr, commandState, terminalWriter, primaryErr)
 	}
+	if diagnosticErr != nil {
+		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("summarize run diagnostics: %w", diagnosticErr))
+	}
 
 	if err := writePartialSummary(summary, output); err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("write debug summary: %w", err))
 	}
 	var encodedResult []byte
 	if *machineOutput {
-		result, err := newRunResult(effective.ResolvedPipeline, output, runOutputDir, debugPath)
+		result, err := newRunResultWithDiagnostics(effective.ResolvedPipeline, output, runOutputDir, debugPath, diagnosticProjection)
 		if err != nil {
 			return failPipelineCommand(stderr, commandState, terminalWriter, err)
 		}
@@ -513,12 +522,25 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 			fmt.Fprintf(stdout, "debug=%s\n", debugPath)
 		}
 	}
-	if len(output.Warnings) > 0 {
-		fmt.Fprintf(stderr, "notarius: run completed with %d warning(s)\n", len(output.Warnings))
+	if warningGroups := len(diagnosticProjection.Warnings); warningGroups > 0 {
+		fmt.Fprintf(stderr, "notarius: run completed with %d warning group(s), %d occurrence(s)", warningGroups, diagnosticProjection.WarningOccurrenceCount)
+		if warningFile, ok := logicalOutputFile(output.OutputFiles, "warnings.json"); ok {
+			fmt.Fprintf(stderr, "; details=%s", filepath.Join(runOutputDir, warningFile))
+		}
+		fmt.Fprintln(stderr)
 	}
 	return 0
 }
 
+func logicalOutputFile(files []contracts.OutputFile, name string) (string, bool) {
+	for _, file := range files {
+		if file.Name == name {
+			return file.Name, true
+		}
+	}
+	return "", false
+}
+
 func writeSummary(summary *debugbundle.SummaryWriter, write func() error) error {
 	if summary == nil {
 		return nil
@@ -538,7 +560,7 @@ func writePartialSummary(summary *debugbundle.SummaryWriter, output pipeline.Run
 			return err
 		}
 	}
-	if err := summary.WriteWarnings(output.Warnings); err != nil {
+	if err := summary.WriteDiagnostics(output.Diagnostics); err != nil {
 		return err
 	}
 	return summary.WriteCheckpointEvents(output.CheckpointEvents)
@@ -1265,11 +1287,21 @@ type cliReferenceUnbindRequest struct {
 }
 
 type cliReferenceSelector struct {
+	Scope    cliReferenceSelectorScope
 	LaneID   string
 	Stage    pipeline.ModuleStage
 	SlotName string
 }
 
+type cliReferenceSelectorScope uint8
+
+const (
+	cliReferenceScopePipeline cliReferenceSelectorScope = iota
+	cliReferenceScopeLane
+	cliReferenceScopeChunk
+	cliReferenceScopeBinding
+)
+
 func parseReferenceFlags(values []string) ([]cliReferenceRequest, error) {
 	if len(values) == 0 {
 		return nil, nil
@@ -1278,7 +1310,7 @@ func parseReferenceFlags(values []string) ([]cliReferenceRequest, error) {
 	for _, raw := range values {
 		name, source, ok := strings.Cut(raw, "=")
 		if !ok {
-			return nil, fmt.Errorf("--reference must use slot=path or lane.slot=path")
+			return nil, fmt.Errorf("--reference must use slot=path, lane.slot=path, or lane.stage.slot=path")
 		}
 		if strings.TrimSpace(source) == "" {
 			return nil, fmt.Errorf("--reference path must not be empty; use --without-reference to unbind")
@@ -1328,17 +1360,14 @@ func parseReferenceSelector(raw string, flagName string) (cliReferenceSelector,
 	}
 	switch len(parts) {
 	case 1:
-		return cliReferenceSelector{SlotName: strings.TrimSpace(parts[0])}, nil
+		return cliReferenceSelector{Scope: cliReferenceScopePipeline, SlotName: strings.TrimSpace(parts[0])}, nil
 	case 2:
 		first := strings.TrimSpace(parts[0])
 		slotName := strings.TrimSpace(parts[1])
 		if first == string(pipeline.StageChunk) {
-			return cliReferenceSelector{Stage: pipeline.StageChunk, SlotName: slotName}, nil
+			return cliReferenceSelector{Scope: cliReferenceScopeChunk, Stage: pipeline.StageChunk, SlotName: slotName}, nil
 		}
-		if first == string(pipeline.StageMerge) {
-			return cliReferenceSelector{Stage: pipeline.StageMerge, SlotName: slotName}, nil
-		}
-		return cliReferenceSelector{LaneID: first, SlotName: slotName}, nil
+		return cliReferenceSelector{Scope: cliReferenceScopeLane, LaneID: first, SlotName: slotName}, nil
 	case 3:
 		laneID := strings.TrimSpace(parts[0])
 		stage := pipeline.ModuleStage(strings.TrimSpace(parts[1]))
@@ -1346,9 +1375,9 @@ func parseReferenceSelector(raw string, flagName string) (cliReferenceSelector,
 		if stage != pipeline.StageExtract && stage != pipeline.StageMerge && stage != pipeline.StageNormalize {
 			return cliReferenceSelector{}, fmt.Errorf("%s lane-qualified selector must use lane.extract.slot, lane.merge.slot, or lane.normalize.slot", flagName)
 		}
-		return cliReferenceSelector{LaneID: laneID, Stage: stage, SlotName: slotName}, nil
+		return cliReferenceSelector{Scope: cliReferenceScopeBinding, LaneID: laneID, Stage: stage, SlotName: slotName}, nil
 	default:
-		return cliReferenceSelector{}, fmt.Errorf("%s must use slot, chunk.slot, merge.slot, lane.slot, lane.extract.slot, lane.merge.slot, or lane.normalize.slot", flagName)
+		return cliReferenceSelector{}, fmt.Errorf("%s must use slot, chunk.slot, lane.slot, lane.extract.slot, lane.merge.slot, or lane.normalize.slot", flagName)
 	}
 }
 
@@ -1369,37 +1398,153 @@ func resolveCLIReferenceRequests(
 		return nil, nil, err
 	}
 
-	overrides := make([]pipeline.ReferenceBinding, 0, len(referenceRequests))
+	// Broad CLI selectors are only presentation syntax. Collapse them into one
+	// highest-specificity action per concrete framework target before pipeline
+	// resolution so the generic reference contract stays stage-and-lane exact.
+	actions := make(map[cliReferenceTargetKey]resolvedCLIReferenceAction)
 	for _, request := range referenceRequests {
-		target, err := resolveCLIReferenceTarget(targets, request.Selector)
+		matches, err := resolveCLIReferenceTargets(targets, request.Selector)
 		if err != nil {
 			return nil, nil, err
 		}
-		overrides = append(overrides, pipeline.ReferenceBinding{
-			Stage:         target.stage,
-			LaneID:        target.laneID,
-			SlotName:      request.Selector.SlotName,
-			Source:        request.Source,
-			BindingSource: contracts.ReferenceBindingSourceCLI,
-		})
+		for _, target := range matches {
+			candidate := resolvedCLIReferenceAction{
+				kind:        cliReferenceActionBind,
+				selector:    request.Selector,
+				target:      target,
+				slotName:    request.Selector.SlotName,
+				source:      request.Source,
+				specificity: request.Selector.specificity(),
+			}
+			if err := mergeCLIReferenceAction(actions, candidate); err != nil {
+				return nil, nil, err
+			}
+		}
 	}
-
-	unbinds := make([]pipeline.ReferenceUnbind, 0, len(unbindRequests))
 	for _, request := range unbindRequests {
-		target, err := resolveCLIReferenceTarget(targets, request.Selector)
+		matches, err := resolveCLIReferenceTargets(targets, request.Selector)
 		if err != nil {
 			return nil, nil, err
 		}
-		unbinds = append(unbinds, pipeline.ReferenceUnbind{
-			Stage:    target.stage,
-			LaneID:   target.laneID,
-			SlotName: request.Selector.SlotName,
-		})
+		for _, target := range matches {
+			candidate := resolvedCLIReferenceAction{
+				kind:        cliReferenceActionUnbind,
+				selector:    request.Selector,
+				target:      target,
+				slotName:    request.Selector.SlotName,
+				specificity: request.Selector.specificity(),
+			}
+			if err := mergeCLIReferenceAction(actions, candidate); err != nil {
+				return nil, nil, err
+			}
+		}
 	}
 
+	resolved := make([]resolvedCLIReferenceAction, 0, len(actions))
+	for _, action := range actions {
+		resolved = append(resolved, action)
+	}
+	sort.Slice(resolved, func(i, j int) bool {
+		left, right := resolved[i], resolved[j]
+		if left.target.laneID != right.target.laneID {
+			return left.target.laneID < right.target.laneID
+		}
+		if left.target.stage != right.target.stage {
+			return referenceStageOrder(left.target.stage) < referenceStageOrder(right.target.stage)
+		}
+		return left.slotName < right.slotName
+	})
+
+	overrides := make([]pipeline.ReferenceBinding, 0, len(resolved))
+	unbinds := make([]pipeline.ReferenceUnbind, 0, len(resolved))
+	for _, action := range resolved {
+		switch action.kind {
+		case cliReferenceActionBind:
+			overrides = append(overrides, pipeline.ReferenceBinding{
+				Stage:         action.target.stage,
+				LaneID:        action.target.laneID,
+				SlotName:      action.slotName,
+				Source:        action.source,
+				BindingSource: contracts.ReferenceBindingSourceCLI,
+			})
+		case cliReferenceActionUnbind:
+			unbinds = append(unbinds, pipeline.ReferenceUnbind{
+				Stage:    action.target.stage,
+				LaneID:   action.target.laneID,
+				SlotName: action.slotName,
+			})
+		}
+	}
 	return overrides, unbinds, nil
 }
 
+type cliReferenceActionKind uint8
+
+const (
+	cliReferenceActionBind cliReferenceActionKind = iota
+	cliReferenceActionUnbind
+)
+
+type cliReferenceTargetKey struct {
+	stage    pipeline.ModuleStage
+	laneID   string
+	slotName string
+}
+
+type resolvedCLIReferenceAction struct {
+	kind        cliReferenceActionKind
+	selector    cliReferenceSelector
+	target      selectedReferenceTarget
+	slotName    string
+	source      string
+	specificity int
+}
+
+func mergeCLIReferenceAction(actions map[cliReferenceTargetKey]resolvedCLIReferenceAction, candidate resolvedCLIReferenceAction) error {
+	key := cliReferenceTargetKey{stage: candidate.target.stage, laneID: candidate.target.laneID, slotName: candidate.slotName}
+	current, ok := actions[key]
+	if !ok || candidate.specificity > current.specificity {
+		actions[key] = candidate
+		return nil
+	}
+	if candidate.specificity < current.specificity {
+		return nil
+	}
+	if candidate.kind != current.kind {
+		return fmt.Errorf("reference target %q slot %q is both bound by %q and unbound by %q at the same specificity", targetLabel(candidate.target), candidate.slotName, current.selector.String(), candidate.selector.String())
+	}
+	actions[key] = candidate
+	return nil
+}
+
+func (selector cliReferenceSelector) specificity() int {
+	switch selector.Scope {
+	case cliReferenceScopePipeline:
+		return 0
+	case cliReferenceScopeLane:
+		return 1
+	case cliReferenceScopeChunk, cliReferenceScopeBinding:
+		return 2
+	default:
+		return -1
+	}
+}
+
+func (selector cliReferenceSelector) String() string {
+	switch selector.Scope {
+	case cliReferenceScopePipeline:
+		return selector.SlotName
+	case cliReferenceScopeLane:
+		return selector.LaneID + "." + selector.SlotName
+	case cliReferenceScopeChunk:
+		return "chunk." + selector.SlotName
+	case cliReferenceScopeBinding:
+		return selector.LaneID + "." + string(selector.Stage) + "." + selector.SlotName
+	default:
+		return selector.SlotName
+	}
+}
+
 type selectedReferenceTarget struct {
 	laneID string
 	stage  pipeline.ModuleStage
@@ -1616,114 +1761,77 @@ func referenceSlotSet(slots []contracts.ReferenceSlot) map[string]struct{} {
 	return slotSet
 }
 
-func resolveCLIReferenceTarget(targets []selectedReferenceTarget, selector cliReferenceSelector) (selectedReferenceTarget, error) {
+func resolveCLIReferenceTargets(targets []selectedReferenceTarget, selector cliReferenceSelector) ([]selectedReferenceTarget, error) {
 	slotName := strings.TrimSpace(selector.SlotName)
 	if slotName == "" {
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot must not be empty")
+		return nil, fmt.Errorf("reference slot must not be empty")
 	}
-	if selector.Stage == pipeline.StageChunk {
+	switch selector.Scope {
+	case cliReferenceScopePipeline:
+		matches := make([]selectedReferenceTarget, 0, len(targets))
+		for _, target := range targets {
+			if _, ok := target.slots[slotName]; ok {
+				matches = append(matches, target)
+			}
+		}
+		if len(matches) == 0 {
+			return nil, fmt.Errorf("reference slot %q is not declared by any selected target", slotName)
+		}
+		return matches, nil
+	case cliReferenceScopeChunk:
 		for _, target := range targets {
 			if target.stage != pipeline.StageChunk {
 				continue
 			}
 			if _, ok := target.slots[slotName]; !ok {
-				return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is not declared by chunk module %q", slotName, target.module)
+				return nil, fmt.Errorf("reference slot %q is not declared by chunk module %q", slotName, target.module)
 			}
-			return target, nil
-		}
-		return selectedReferenceTarget{}, fmt.Errorf("reference chunk target is not selected")
-	}
-	if selector.Stage == pipeline.StageExtract || selector.Stage == pipeline.StageMerge || selector.Stage == pipeline.StageNormalize {
-		if selector.LaneID == "" && selector.Stage == pipeline.StageMerge {
-			return resolveCLIReferenceStageTarget(targets, selector.Stage, slotName)
+			return []selectedReferenceTarget{target}, nil
 		}
+		return nil, fmt.Errorf("reference chunk target is not selected")
+	case cliReferenceScopeLane:
+		laneSelected := false
+		matches := make([]selectedReferenceTarget, 0, 3)
 		for _, target := range targets {
-			if target.laneID == selector.LaneID && target.stage == selector.Stage {
-				if _, ok := target.slots[slotName]; !ok {
-					return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is not declared by selected %s target %q", slotName, selector.Stage, targetLabel(target))
-				}
-				return target, nil
+			if target.laneID != selector.LaneID {
+				continue
+			}
+			laneSelected = true
+			if _, ok := target.slots[slotName]; ok {
+				matches = append(matches, target)
 			}
 		}
-		return selectedReferenceTarget{}, fmt.Errorf("reference lane %q is not selected", selector.LaneID)
-	}
-	if strings.TrimSpace(selector.LaneID) != "" {
-		return resolveCLIReferenceLaneTarget(targets, strings.TrimSpace(selector.LaneID), slotName)
-	}
-	return resolveCLIReferenceFlatTarget(targets, slotName)
-}
-
-func resolveCLIReferenceStageTarget(targets []selectedReferenceTarget, stage pipeline.ModuleStage, slotName string) (selectedReferenceTarget, error) {
-	matches := make([]selectedReferenceTarget, 0, 2)
-	for _, target := range targets {
-		if target.stage != stage {
-			continue
+		if !laneSelected {
+			return nil, fmt.Errorf("reference lane %q is not selected", selector.LaneID)
 		}
-		if _, ok := target.slots[slotName]; ok {
-			matches = append(matches, target)
+		if len(matches) == 0 {
+			return nil, fmt.Errorf("reference slot %q is not declared by selected lane %q", slotName, selector.LaneID)
 		}
-	}
-	switch len(matches) {
-	case 0:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is not declared by any selected %s target", slotName, stage)
-	case 1:
-		return matches[0], nil
+		return matches, nil
+	case cliReferenceScopeBinding:
+		laneSelected := false
+		for _, target := range targets {
+			if target.laneID != selector.LaneID {
+				continue
+			}
+			laneSelected = true
+			if target.stage != selector.Stage {
+				continue
+			}
+			if _, ok := target.slots[slotName]; !ok {
+				return nil, fmt.Errorf("reference slot %q is not declared by selected %s target %q", slotName, selector.Stage, targetLabel(target))
+			}
+			return []selectedReferenceTarget{target}, nil
+		}
+		if !laneSelected {
+			return nil, fmt.Errorf("reference lane %q is not selected", selector.LaneID)
+		}
+		return nil, fmt.Errorf("reference %s target is not selected for lane %q", selector.Stage, selector.LaneID)
 	default:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is declared by multiple selected %s targets (%s); use a more specific selector such as %s", slotName, stage, targetList(matches), selectorSuggestions(matches, slotName))
+		return nil, fmt.Errorf("reference selector has unknown scope")
 	}
 }
 
-func resolveCLIReferenceLaneTarget(targets []selectedReferenceTarget, laneID string, slotName string) (selectedReferenceTarget, error) {
-	laneSelected := false
-	matches := make([]selectedReferenceTarget, 0, 2)
-	for _, target := range targets {
-		if target.laneID != laneID {
-			continue
-		}
-		laneSelected = true
-		if _, ok := target.slots[slotName]; ok {
-			matches = append(matches, target)
-		}
-	}
-	if !laneSelected {
-		return selectedReferenceTarget{}, fmt.Errorf("reference lane %q is not selected", laneID)
-	}
-	switch len(matches) {
-	case 0:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is not declared by selected lane %q", slotName, laneID)
-	case 1:
-		return matches[0], nil
-	default:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is declared by multiple selected targets in lane %q (%s); use a more specific selector such as %s", slotName, laneID, targetList(matches), selectorSuggestions(matches, slotName))
-	}
-}
-
-func resolveCLIReferenceFlatTarget(targets []selectedReferenceTarget, slotName string) (selectedReferenceTarget, error) {
-	matches := make([]selectedReferenceTarget, 0, 2)
-	for _, target := range targets {
-		if _, ok := target.slots[slotName]; ok {
-			matches = append(matches, target)
-		}
-	}
-	switch len(matches) {
-	case 0:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is not declared by any selected reference target", slotName)
-	case 1:
-		return matches[0], nil
-	default:
-		return selectedReferenceTarget{}, fmt.Errorf("reference slot %q is declared by multiple selected targets (%s); use a more specific selector such as %s", slotName, targetList(matches), selectorSuggestions(matches, slotName))
-	}
-}
-
-func targetList(targets []selectedReferenceTarget) string {
-	labels := make([]string, 0, len(targets))
-	for _, target := range targets {
-		labels = append(labels, targetLabel(target))
-	}
-	sort.Strings(labels)
-	return strings.Join(labels, ", ")
-}
-
 func targetLabel(target selectedReferenceTarget) string {
 	if target.stage == pipeline.StageChunk {
 		return "chunk"
@@ -1731,17 +1839,19 @@ func targetLabel(target selectedReferenceTarget) string {
 	return target.laneID + "." + string(target.stage)
 }
 
-func selectorSuggestions(targets []selectedReferenceTarget, slotName string) string {
-	suggestions := make([]string, 0, len(targets))
-	for _, target := range targets {
-		if target.stage == pipeline.StageChunk {
-			suggestions = append(suggestions, "chunk."+slotName)
-			continue
-		}
-		suggestions = append(suggestions, target.laneID+"."+string(target.stage)+"."+slotName)
+func referenceStageOrder(stage pipeline.ModuleStage) int {
+	switch stage {
+	case pipeline.StageChunk:
+		return 0
+	case pipeline.StageExtract:
+		return 1
+	case pipeline.StageMerge:
+		return 2
+	case pipeline.StageNormalize:
+		return 3
+	default:
+		return 4
 	}
-	sort.Strings(suggestions)
-	return strings.Join(suggestions, " or ")
 }
 
 func sortedPipelineIDs(cfg config.Config) []string {

internal/cli/run_contract_test.go

@@ -590,10 +590,11 @@ func TestRunWarningsRemainSuccessfulAndReachDurableSurfaces(t *testing.T) {
 	roots := newStateTestRoots(t)
 	harness := newStateTestHarness()
 	harness.includeWarnings = true
-	harness.chunkWarnings = []contracts.Warning{{Scope: "chunk", ReasonCode: "contract-warning", Message: "warning retained"}}
+	harness.includeWarningFile = true
+	harness.chunkDiagnostics = []contracts.ProducerDiagnostic{stateTestDiagnostic("chunk", "contract-warning", "warning retained")}
 	var stdout, stderr bytes.Buffer
 	code := RunWithOptions([]string{"run", "sample", "--config", roots.config, "--input", roots.input, "--chunk_cache", "bypass", "--debug"}, &stdout, &stderr, harness.options())
-	if code != 0 || !strings.Contains(stdout.String(), "outputs=1") || !strings.Contains(stderr.String(), "1 warning(s)") {
+	if code != 0 || !strings.Contains(stdout.String(), "outputs=1") || !strings.Contains(stderr.String(), "1 warning group(s), 1 occurrence(s)") || !strings.Contains(stderr.String(), "warnings.json") {
 		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 	}
 	outputPath := filepath.Join(onlyChildDir(t, roots.output), "result.json")
@@ -601,11 +602,12 @@ func TestRunWarningsRemainSuccessfulAndReachDurableSurfaces(t *testing.T) {
 	if err != nil || !strings.Contains(string(output), "contract-warning") {
 		t.Fatalf("durable output = %q, %v", output, err)
 	}
+	assertFile(t, filepath.Join(filepath.Dir(outputPath), "warnings.json"))
 	bundle := onlyChildDir(t, roots.debug)
-	var warnings []contracts.Warning
-	readStateTestSummaryJSON(t, bundle, "warnings.json", &warnings)
-	if len(warnings) != 1 || warnings[0].ReasonCode != "contract-warning" {
-		t.Fatalf("debug warnings = %#v", warnings)
+	var diagnostics contracts.DiagnosticCollection
+	readStateTestSummaryJSON(t, bundle, "final-diagnostics.json", &diagnostics)
+	if len(diagnostics.Groups) != 1 || diagnostics.Groups[0].ReasonCode != "contract-warning" {
+		t.Fatalf("debug diagnostics = %#v", diagnostics)
 	}
 }
 

internal/cli/run_result.go

@@ -7,25 +7,40 @@ import (
 	"path/filepath"
 	"strings"
 
+	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
-const runResultSchemaVersion = "notarius.run-result.v1"
+const runResultSchemaVersion = "notarius.run-result.v2"
 
 type runResult struct {
-	SchemaVersion         string `json:"schema_version"`
-	RunID                 string `json:"run_id"`
-	PipelineID            string `json:"pipeline_id"`
-	OutputDirectory       string `json:"output_directory"`
-	IndexFile             string `json:"index_file,omitempty"`
-	NormalizedOutputCount int    `json:"normalized_output_count"`
-	RejectedOutputCount   int    `json:"rejected_output_count"`
-	WarningCount          int    `json:"warning_count"`
-	ValidationStatus      string `json:"validation_status"`
-	DebugDirectory        string `json:"debug_directory,omitempty"`
+	SchemaVersion             string                        `json:"schema_version"`
+	RunID                     string                        `json:"run_id"`
+	PipelineID                string                        `json:"pipeline_id"`
+	OutputDirectory           string                        `json:"output_directory"`
+	IndexFile                 string                        `json:"index_file,omitempty"`
+	NormalizedOutputCount     int                           `json:"normalized_output_count"`
+	RejectedOutputCount       int                           `json:"rejected_output_count"`
+	WarningGroupCount         int                           `json:"warning_group_count"`
+	WarningOccurrenceCount    int                           `json:"warning_occurrence_count"`
+	DiagnosticGroupCount      int                           `json:"diagnostic_group_count"`
+	DiagnosticOccurrenceCount int                           `json:"diagnostic_occurrence_count"`
+	DiagnosticsTruncated      bool                          `json:"diagnostics_truncated"`
+	ValidationStatus          string                        `json:"validation_status"`
+	ValidationSummaries       []artifacts.ValidationSummary `json:"validation_summaries,omitempty"`
+	DebugDirectory            string                        `json:"debug_directory,omitempty"`
 }
 
 func newRunResult(resolved pipeline.ResolvedPipeline, output pipeline.RunOutput, outputDirectory, debugDirectory string) (runResult, error) {
+	diagnosticProjection, err := contracts.ProjectDiagnosticCollection(output.Diagnostics)
+	if err != nil {
+		return runResult{}, fmt.Errorf("summarize run diagnostics: %w", err)
+	}
+	return newRunResultWithDiagnostics(resolved, output, outputDirectory, debugDirectory, diagnosticProjection)
+}
+
+func newRunResultWithDiagnostics(resolved pipeline.ResolvedPipeline, output pipeline.RunOutput, outputDirectory, debugDirectory string, diagnosticProjection contracts.DiagnosticProjection) (runResult, error) {
 	if strings.TrimSpace(output.Manifest.RunID) == "" {
 		return runResult{}, fmt.Errorf("run result requires a run ID")
 	}
@@ -51,14 +66,19 @@ func newRunResult(resolved pipeline.ResolvedPipeline, output pipeline.RunOutput,
 	}
 
 	result := runResult{
-		SchemaVersion:         runResultSchemaVersion,
-		RunID:                 output.Manifest.RunID,
-		PipelineID:            output.Manifest.PipelineID,
-		OutputDirectory:       absOutputDirectory,
-		NormalizedOutputCount: len(output.NormalizeOutputs),
-		RejectedOutputCount:   len(output.Rejected),
-		WarningCount:          len(output.Warnings),
-		ValidationStatus:      output.Manifest.ValidationStatus,
+		SchemaVersion:             runResultSchemaVersion,
+		RunID:                     output.Manifest.RunID,
+		PipelineID:                output.Manifest.PipelineID,
+		OutputDirectory:           absOutputDirectory,
+		NormalizedOutputCount:     len(output.NormalizeOutputs),
+		RejectedOutputCount:       len(output.Rejected),
+		WarningGroupCount:         len(diagnosticProjection.Warnings),
+		WarningOccurrenceCount:    diagnosticProjection.WarningOccurrenceCount,
+		DiagnosticGroupCount:      len(diagnosticProjection.Diagnostics),
+		DiagnosticOccurrenceCount: diagnosticProjection.DiagnosticOccurrenceCount,
+		DiagnosticsTruncated:      output.Diagnostics.Truncated,
+		ValidationStatus:          output.Manifest.ValidationStatus,
+		ValidationSummaries:       cloneValidationSummaries(output.Manifest.ValidationSummaries),
 	}
 
 	if strings.TrimSpace(debugDirectory) != "" {
@@ -85,6 +105,17 @@ func newRunResult(resolved pipeline.ResolvedPipeline, output pipeline.RunOutput,
 	return result, nil
 }
 
+func cloneValidationSummaries(summaries []artifacts.ValidationSummary) []artifacts.ValidationSummary {
+	if len(summaries) == 0 {
+		return nil
+	}
+	cloned := make([]artifacts.ValidationSummary, len(summaries))
+	for index, summary := range summaries {
+		cloned[index] = artifacts.CloneValidationSummary(summary)
+	}
+	return cloned
+}
+
 func encodeRunResult(result runResult) ([]byte, error) {
 	encoded, err := json.Marshal(result)
 	if err != nil {

internal/cli/run_result_command_test.go

@@ -27,7 +27,7 @@ func TestMaintainedMinimalInvocationEmitsRunResult(t *testing.T) {
 	}
 
 	receipt := decodeRunResultDocument(t, stdout.String())
-	if got := receipt["schema_version"]; got != "notarius.run-result.v1" {
+	if got := receipt["schema_version"]; got != "notarius.run-result.v2" {
 		t.Fatalf("schema_version = %q", got)
 	}
 	if got := receipt["run_id"]; got != productionRunID {
@@ -45,8 +45,8 @@ func TestMaintainedMinimalInvocationEmitsRunResult(t *testing.T) {
 	if got := receipt["rejected_output_count"]; got != float64(0) {
 		t.Fatalf("rejected_output_count = %v", got)
 	}
-	if got := receipt["warning_count"]; got != float64(0) {
-		t.Fatalf("warning_count = %v", got)
+	if got := receipt["warning_group_count"]; got != float64(0) || receipt["warning_occurrence_count"] != float64(0) || receipt["diagnostic_group_count"] != float64(0) || receipt["diagnostic_occurrence_count"] != float64(0) || receipt["diagnostics_truncated"] != false {
+		t.Fatalf("diagnostic counts = %#v", receipt)
 	}
 	if got := receipt["validation_status"]; got != "approved" {
 		t.Fatalf("validation_status = %q", got)
@@ -63,19 +63,19 @@ func TestMaintainedMinimalInvocationEmitsRunResult(t *testing.T) {
 func TestRunResultReportsWarningsAndDebugBundle(t *testing.T) {
 	roots := newStateTestRoots(t)
 	harness := newStateTestHarness()
-	harness.chunkWarnings = []contracts.Warning{{Scope: "chunk", ReasonCode: "contract-warning", Message: "warning retained"}}
+	harness.chunkDiagnostics = []contracts.ProducerDiagnostic{stateTestDiagnostic("chunk", "contract-warning", "warning retained")}
 	var stdout, stderr bytes.Buffer
 	code := RunWithOptions([]string{
 		"run", "sample", "--config", roots.config, "--input", roots.input,
 		"--chunk_cache", "bypass", "--debug", "--json",
 	}, &stdout, &stderr, harness.options())
-	if code != 0 || !strings.Contains(stderr.String(), "1 warning(s)") {
+	if code != 0 || !strings.Contains(stderr.String(), "1 warning group(s), 1 occurrence(s)") || strings.Contains(stderr.String(), "warnings.json") {
 		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 	}
 
 	receipt := decodeRunResultDocument(t, stdout.String())
-	if got := receipt["warning_count"]; got != float64(1) {
-		t.Fatalf("warning_count = %v", got)
+	if got := receipt["warning_group_count"]; got != float64(1) || receipt["warning_occurrence_count"] != float64(1) || receipt["diagnostic_group_count"] != float64(0) || receipt["diagnostic_occurrence_count"] != float64(0) || receipt["diagnostics_truncated"] != false {
+		t.Fatalf("diagnostic counts = %#v", receipt)
 	}
 	debugDirectory, ok := receipt["debug_directory"].(string)
 	if !ok || !filepath.IsAbs(debugDirectory) || debugDirectory != onlyChildDir(t, roots.debug) {
@@ -92,7 +92,7 @@ func TestRunResultReportsSuccessfulRejection(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	configBytes = []byte(replaceRequiredOnce(t, string(configBytes), "        normalize: test/normalize\n", "        normalize:\n          module: test/normalize\n          validators:\n            - generic/always_reject\n"))
+	configBytes = []byte(replaceRequiredOnce(t, string(configBytes), "        normalize: test/normalize\n", "        normalize:\n          module: test/normalize\n          validators:\n            - generic/always_reject\n          validation_policy:\n            semantic_rejection: reject_output\n"))
 	if err := os.WriteFile(roots.config, configBytes, 0o600); err != nil {
 		t.Fatal(err)
 	}

internal/cli/run_result_test.go

@@ -52,8 +52,11 @@ func TestRunResultEncodesRequiredFieldsAndCounts(t *testing.T) {
 	if got := decoded["rejected_output_count"]; got != float64(1) {
 		t.Fatalf("rejected_output_count = %v", got)
 	}
-	if got := decoded["warning_count"]; got != float64(1) {
-		t.Fatalf("warning_count = %v", got)
+	if got := decoded["warning_group_count"]; got != float64(1) || decoded["warning_occurrence_count"] != float64(1) || decoded["diagnostic_group_count"] != float64(0) || decoded["diagnostic_occurrence_count"] != float64(0) || decoded["diagnostics_truncated"] != false {
+		t.Fatalf("diagnostic counts = %#v", decoded)
+	}
+	if got := decoded["validation_summaries"]; got != nil {
+		t.Fatalf("validation_summaries = %#v, want omitted when empty", got)
 	}
 	if got := decoded["output_directory"]; got != filepath.Join(mustWorkingDirectory(t), "relative-output") {
 		t.Fatalf("output_directory = %q", got)
@@ -112,6 +115,26 @@ func TestRunResultOmitsIndexFileForOtherOutputModules(t *testing.T) {
 	}
 }
 
+func TestRunResultProjectsOwnedValidationSummaries(t *testing.T) {
+	output := testRunOutput()
+	output.Manifest.ValidationSummaries = []artifacts.ValidationSummary{{Status: "incomplete", IncompleteValidators: []string{"validator"}, ProducerAttemptCount: 1, TerminalAction: "warn_continue"}}
+	result, err := newRunResult(testResolvedPipeline(pipeline.DefaultOutputModule), output, "output", "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	output.Manifest.ValidationSummaries[0].IncompleteValidators[0] = "caller mutation"
+	if got := result.ValidationSummaries[0].IncompleteValidators; len(got) != 1 || got[0] != "validator" {
+		t.Fatalf("result validation summaries = %#v", result.ValidationSummaries)
+	}
+	encoded, err := encodeRunResult(result)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !bytes.Contains(encoded, []byte(`"validation_summaries":[{"status":"incomplete","incomplete_validators":["validator"],"producer_attempt_count":1,"terminal_action":"warn_continue"}]`)) {
+		t.Fatalf("encoded result = %s", encoded)
+	}
+}
+
 func TestRunResultRequiresOneProductionIndexFile(t *testing.T) {
 	tests := []struct {
 		name  string
@@ -159,8 +182,15 @@ func testRunOutput() pipeline.RunOutput {
 		Manifest:         artifacts.RunManifest{RunID: "run-123", PipelineID: "sample", ValidationStatus: "rejected"},
 		NormalizeOutputs: []contracts.SerializedOutput{{}, {}},
 		Rejected:         []contracts.RejectedOutput{{}},
-		Warnings:         []contracts.Warning{{}},
-		OutputFiles:      []contracts.OutputFile{{Name: "index.json"}},
+		Diagnostics: contracts.DiagnosticCollection{Groups: []contracts.DiagnosticGroup{{
+			Disposition:     contracts.DiagnosticDispositionWarning,
+			Category:        contracts.DiagnosticCategoryFallback,
+			ReasonCode:      "fallback",
+			Origin:          contracts.DiagnosticOrigin{Stage: contracts.DiagnosticOriginStageNormalize, StepID: "step", LaneID: "lane", ModuleKey: "module"},
+			OccurrenceCount: 1,
+			Samples:         []contracts.DiagnosticSample{{Scope: "scope", Message: "message"}},
+		}}},
+		OutputFiles: []contracts.OutputFile{{Name: "index.json"}},
 	}
 }
 

internal/cli/run_terminal.go

@@ -6,6 +6,7 @@ import (
 	"io"
 
 	"gitea.maximumdirect.net/eric/notarius/internal/core/debugbundle"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
@@ -33,13 +34,17 @@ func (s *pipelineCommandState) setDebugPath(debugPath string) {
 	}
 }
 
-func (s *pipelineCommandState) observeOutput(output pipeline.RunOutput) {
+func (s *pipelineCommandState) observeOutput(output pipeline.RunOutput, diagnostics contracts.DiagnosticProjection) {
 	if s == nil {
 		return
 	}
 	s.report.OutputCount = len(output.NormalizeOutputs)
 	s.report.RejectedCount = len(output.Rejected)
-	s.report.WarningCount = len(output.Warnings)
+	s.report.WarningGroupCount = len(diagnostics.Warnings)
+	s.report.WarningOccurrenceCount = diagnostics.WarningOccurrenceCount
+	s.report.DiagnosticGroupCount = len(diagnostics.Diagnostics)
+	s.report.DiagnosticOccurrenceCount = diagnostics.DiagnosticOccurrenceCount
+	s.report.DiagnosticsTruncated = output.Diagnostics.Truncated
 	s.report.ValidationStatus = output.Manifest.ValidationStatus
 }
 

internal/cli/spell_catalog_identity_contract_test.go

@@ -238,7 +238,7 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 			Kind: dnd.SpellListKind, Schema: normalizeSchema, MediaType: "application/json", Content: []byte(`{"spell_casts":[]}`),
 		},
 	}
-	if err := recorder.NormalizeSucceeded("spells", spellnormalize.Key, normalizeDependencies, normalizeArtifact, nil); err != nil {
+	if err := recorder.NormalizeSucceeded("spells", spellnormalize.Key, normalizeDependencies, normalizeArtifact); err != nil {
 		t.Fatal(err)
 	}
 
@@ -264,7 +264,7 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 	if _, decision := changedLoader.Normalize("spells", spellnormalize.Key, normalizeDependencies); decision.Reused {
 		t.Fatalf("changed normalize fingerprint decision = %#v, want normalize checkpoint cold miss", decision)
 	}
-	changedMapping := replaceCheckpointFingerprintValue(t, fingerprints, extractSpellMappingFingerprintName(), "dnd.spells.extract_mapping.v3")
+	changedMapping := replaceCheckpointFingerprintValue(t, fingerprints, extractSpellMappingFingerprintName(), "dnd.spells.extract_mapping.changed")
 	assertOnlyCheckpointFingerprintChanged(t, fingerprints, changedMapping, extractSpellMappingFingerprintName())
 	_, mappingLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, changedMapping, llmFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, true)
 	if err != nil {

internal/cli/spell_catalog_retry_contract_test.go

@@ -19,12 +19,12 @@ import (
 func TestProductionSpellCatalogValidationRetries(t *testing.T) {
 	const retries = 2
 	tests := []struct {
-		name            string
-		responses       []string
-		wantCalls       int
-		wantRejected    bool
-		wantSpell       string
-		wantWarningCode string
+		name             string
+		responses        []string
+		wantCalls        int
+		wantRejected     bool
+		wantSpell        string
+		wantAdvisoryCode string
 	}{
 		{
 			name: "unknown spell remains rejected after exhaustion",
@@ -42,9 +42,9 @@ func TestProductionSpellCatalogValidationRetries(t *testing.T) {
 				productionSpellResponse("Unknown Spell"),
 				productionSpellResponse("Aegis of Emberfall"),
 			},
-			wantCalls:       2,
-			wantSpell:       "Aegis of Emberfall",
-			wantWarningCode: "spell_not_near_source",
+			wantCalls:        2,
+			wantSpell:        "Aegis of Emberfall",
+			wantAdvisoryCode: "spell_not_near_source",
 		},
 	}
 
@@ -65,6 +65,7 @@ func TestProductionSpellCatalogValidationRetries(t *testing.T) {
 				t.Fatalf("materialize production references: %v", err)
 			}
 			materialized.Steps[0].ArtifactLanes[0].Extract.Retries = retries
+			materialized.Steps[0].ArtifactLanes[0].ExtractValidationPolicy.SemanticRejection = pipeline.SemanticRejectionRejectOutput
 
 			llmClient := &catalogRetryLLMClient{responses: tt.responses}
 			prepared, err := pipeline.Prepare(materialized, components.registries, pipeline.ModuleDependencies{LLM: llmClient})
@@ -91,8 +92,8 @@ func TestProductionSpellCatalogValidationRetries(t *testing.T) {
 				if rejection.ReasonCode != "unknown_spell" || rejection.AttemptCount != retries+1 {
 					t.Fatalf("rejection = %#v, want exhausted unknown-spell rejection", rejection)
 				}
-				if len(output.Warnings) != 0 {
-					t.Fatalf("warnings = %#v, want no emitted warnings from rejected attempts", output.Warnings)
+				if len(output.Diagnostics.Groups) != 1 || output.Diagnostics.Groups[0].ReasonCode != "spell_not_near_source" {
+					t.Fatalf("diagnostics = %#v, want complete terminal validation diagnostic", output.Diagnostics)
 				}
 				return
 			}
@@ -107,8 +108,16 @@ func TestProductionSpellCatalogValidationRetries(t *testing.T) {
 			if len(value.SpellCasts) != 1 || value.SpellCasts[0].Spell != tt.wantSpell {
 				t.Fatalf("normalized spell list = %#v, want accepted overlay spell", value)
 			}
-			if len(output.Warnings) != 2 || output.Warnings[0].ReasonCode != tt.wantWarningCode || output.Warnings[1].ReasonCode != tt.wantWarningCode {
-				t.Fatalf("warnings = %#v, want accepted-attempt warnings from extract and normalize validation", output.Warnings)
+			if len(output.Diagnostics.Groups) != 2 || output.Diagnostics.Groups[0].ReasonCode != tt.wantAdvisoryCode {
+				t.Fatalf("diagnostics = %#v, want terminal extract and normalize diagnostics", output.Diagnostics)
+			}
+			if len(output.Diagnostics.Groups) != 2 {
+				t.Fatalf("diagnostics = %#v, want extract and normalize data-quality advisories", output.Diagnostics)
+			}
+			for _, diagnostic := range output.Diagnostics.Groups {
+				if diagnostic.Disposition != contracts.DiagnosticDispositionAdvisory || diagnostic.ReasonCode != tt.wantAdvisoryCode {
+					t.Fatalf("diagnostics = %#v, want only data-quality advisories", output.Diagnostics)
+				}
 			}
 		})
 	}

internal/cli/state_hardening_test.go

@@ -349,7 +349,7 @@ func TestRunUsesOneInjectedIdentityForDebugOutputAndManifest(t *testing.T) {
 		t.Fatalf("debug invocation session = %q, want %q", invocation.SessionID, wantSessionID)
 	}
 	report := readStateTestRunReport(t, debugPath)
-	if !report.Succeeded || report.RunID != runID || report.PipelineID != "sample" || report.OutputPath != outputPath || report.DebugPath != debugPath || report.OutputCount != 1 || report.RejectedCount != 0 || report.WarningCount != 0 || report.ValidationStatus != "approved" {
+	if !report.Succeeded || report.RunID != runID || report.PipelineID != "sample" || report.OutputPath != outputPath || report.DebugPath != debugPath || report.OutputCount != 1 || report.RejectedCount != 0 || report.WarningGroupCount != 0 || report.WarningOccurrenceCount != 0 || report.DiagnosticGroupCount != 0 || report.DiagnosticOccurrenceCount != 0 || report.DiagnosticsTruncated || report.ValidationStatus != "approved" {
 		t.Fatalf("success report = %#v", report)
 	}
 	if !strings.Contains(result.stdout, "outputs=1 rejected=0") {
@@ -431,7 +431,7 @@ func TestRunWritesTerminalArtifactsForResolutionPipelineAndOutputFailures(t *tes
 			bundlePath := onlyChildDir(t, roots.debug)
 			runID := filepath.Base(bundlePath)
 			report := readStateTestRunReport(t, bundlePath)
-			if report.Succeeded || report.RunID != runID || report.PipelineID != tc.pipelineID || report.OutputPath != filepath.Join(roots.output, runID) || report.DebugPath != bundlePath || report.OutputCount != tc.wantOutputs || report.RejectedCount != 0 || report.WarningCount != 0 || report.ValidationStatus != tc.wantValidation {
+			if report.Succeeded || report.RunID != runID || report.PipelineID != tc.pipelineID || report.OutputPath != filepath.Join(roots.output, runID) || report.DebugPath != bundlePath || report.OutputCount != tc.wantOutputs || report.RejectedCount != 0 || report.WarningGroupCount != 0 || report.WarningOccurrenceCount != 0 || report.DiagnosticGroupCount != 0 || report.DiagnosticOccurrenceCount != 0 || report.DiagnosticsTruncated || report.ValidationStatus != tc.wantValidation {
 				t.Fatalf("failure report = %#v", report)
 			}
 			errorLog, err := os.ReadFile(filepath.Join(bundlePath, "summary", "error.log"))
@@ -445,7 +445,7 @@ func TestRunWritesTerminalArtifactsForResolutionPipelineAndOutputFailures(t *tes
 func TestRunRetainsPartialPipelineOutcomeInFailureSummary(t *testing.T) {
 	roots := newStateTestRoots(t)
 	harness := newStateTestHarness()
-	harness.chunkWarnings = []contracts.Warning{{Scope: "chunk", ReasonCode: "partial-warning", Message: "warning retained before failure"}}
+	harness.chunkDiagnostics = []contracts.ProducerDiagnostic{stateTestDiagnostic("chunk", "partial-warning", "warning retained before failure")}
 	harness.extractErr = errors.New("synthetic partial pipeline failure")
 
 	result := runStateTest(t, roots, harness.options(), true, true, "bypass")
@@ -454,7 +454,7 @@ func TestRunRetainsPartialPipelineOutcomeInFailureSummary(t *testing.T) {
 	}
 	bundlePath := onlyChildDir(t, roots.debug)
 	report := readStateTestRunReport(t, bundlePath)
-	if report.Succeeded || report.OutputCount != 0 || report.RejectedCount != 0 || report.WarningCount != 1 || report.ValidationStatus != "failed" {
+	if report.Succeeded || report.OutputCount != 0 || report.RejectedCount != 0 || report.WarningGroupCount != 1 || report.WarningOccurrenceCount != 1 || report.DiagnosticGroupCount != 0 || report.DiagnosticOccurrenceCount != 0 || report.DiagnosticsTruncated || report.ValidationStatus != "failed" {
 		t.Fatalf("partial failure report = %#v", report)
 	}
 
@@ -463,10 +463,10 @@ func TestRunRetainsPartialPipelineOutcomeInFailureSummary(t *testing.T) {
 	if manifest.RunID != report.RunID || manifest.PipelineID != "sample" || manifest.ValidationStatus != "failed" {
 		t.Fatalf("partial manifest = %#v", manifest)
 	}
-	var warnings []contracts.Warning
-	readStateTestSummaryJSON(t, bundlePath, "warnings.json", &warnings)
-	if len(warnings) != 1 || warnings[0].ReasonCode != "partial-warning" {
-		t.Fatalf("partial warnings = %#v", warnings)
+	var diagnostics contracts.DiagnosticCollection
+	readStateTestSummaryJSON(t, bundlePath, "final-diagnostics.json", &diagnostics)
+	if len(diagnostics.Groups) != 1 || diagnostics.Groups[0].ReasonCode != "partial-warning" {
+		t.Fatalf("partial diagnostics = %#v", diagnostics)
 	}
 	var events []pipeline.CheckpointEvent
 	readStateTestSummaryJSON(t, bundlePath, "checkpoint-events.json", &events)
@@ -863,11 +863,12 @@ type stateTestHarness struct {
 	chunkCalls, extractCalls int
 	runIDCalls               uint64
 	extractErr               error
-	chunkWarnings            []contracts.Warning
+	chunkDiagnostics         []contracts.ProducerDiagnostic
 	moduleProfiles           []string
 	sessionIDs               []string
-	outputWarnings           []contracts.Warning
+	outputDiagnostics        contracts.DiagnosticCollection
 	includeWarnings          bool
+	includeWarningFile       bool
 }
 
 func newStateTestHarness() *stateTestHarness { return &stateTestHarness{} }
@@ -892,7 +893,7 @@ func (h *stateTestHarness) options() Options {
 		panic(err)
 	}
 	if err := registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "test/output", Stage: pipeline.StageOutput, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) {
-		return stateTestOutput{harness: h, includeWarnings: h.includeWarnings}, nil
+		return stateTestOutput{harness: h, includeDiagnostics: h.includeWarnings}, nil
 	}); err != nil {
 		panic(err)
 	}
@@ -931,7 +932,7 @@ func (c stateTestChunker) Plan(_ context.Context, req contracts.ChunkRequest) (c
 	c.harness.mu.Lock()
 	c.harness.chunkCalls++
 	c.harness.mu.Unlock()
-	return contracts.ChunkPlanResult{Plan: source.ChunkPlan{SourceDigest: req.Source.Digest, Ranges: []source.ChunkRange{{StartUnitID: 1, EndUnitID: 1}}}, Warnings: append([]contracts.Warning(nil), c.harness.chunkWarnings...)}, nil
+	return contracts.ChunkPlanResult{Plan: source.ChunkPlan{SourceDigest: req.Source.Digest, Ranges: []source.ChunkRange{{StartUnitID: 1, EndUnitID: 1}}}, Diagnostics: contracts.CloneProducerDiagnostics(c.harness.chunkDiagnostics)}, nil
 }
 
 const stateTestArtifactKind contracts.ArtifactKind = "test/artifact"
@@ -999,20 +1000,28 @@ func (n stateTestNormalizer) Normalize(_ context.Context, req contracts.TypedNor
 }
 
 type stateTestOutput struct {
-	harness         *stateTestHarness
-	includeWarnings bool
+	harness            *stateTestHarness
+	includeDiagnostics bool
 }
 
 func (o stateTestOutput) Key() string { return "test/output" }
 func (o stateTestOutput) Encode(_ context.Context, req contracts.OutputRequest) (contracts.OutputResult, error) {
 	o.harness.mu.Lock()
-	o.harness.outputWarnings = append([]contracts.Warning(nil), req.Warnings...)
+	o.harness.outputDiagnostics = contracts.CloneDiagnosticCollection(req.Diagnostics)
 	o.harness.mu.Unlock()
 	data := []byte("{\"ok\":true}\n")
-	if o.includeWarnings && len(req.Warnings) > 0 {
-		data = []byte(fmt.Sprintf("{\"ok\":true,\"warnings\":%q}\n", req.Warnings[0].ReasonCode))
+	if o.includeDiagnostics && len(req.Diagnostics.Groups) > 0 {
+		data = []byte(fmt.Sprintf("{\"ok\":true,\"diagnostics\":%q}\n", req.Diagnostics.Groups[0].ReasonCode))
 	}
-	return contracts.OutputResult{Files: []contracts.OutputFile{{Name: "result.json", Bytes: data}}}, nil
+	files := []contracts.OutputFile{{Name: "result.json", Bytes: data}}
+	if o.harness.includeWarningFile {
+		files = append(files, contracts.OutputFile{Name: "warnings.json", Bytes: []byte("{\"warnings\":true}\n")})
+	}
+	return contracts.OutputResult{Files: files}, nil
+}
+
+func stateTestDiagnostic(scope, reasonCode, message string) contracts.ProducerDiagnostic {
+	return contracts.ProducerDiagnostic{Disposition: contracts.DiagnosticDispositionWarning, Category: contracts.DiagnosticCategoryDegradation, ReasonCode: reasonCode, OccurrenceCount: 1, Samples: []contracts.DiagnosticSample{{Scope: scope, Message: message}}}
 }
 
 type failingDebugRecorder struct{}

internal/core/artifacts/artifacts.go

@@ -93,17 +93,44 @@ type NormalizedOutputManifest struct {
 }
 
 type RejectedOutputManifest struct {
-	Stage                  string `json:"stage"`
-	StepID                 string `json:"step_id,omitempty"`
-	LaneID                 string `json:"lane_id,omitempty"`
-	ModuleKey              string `json:"module_key,omitempty"`
-	ChunkID                string `json:"chunk_id,omitempty"`
-	ChunkIndex             int    `json:"chunk_index,omitempty"`
-	ValidatorName          string `json:"validator_name,omitempty"`
-	ReasonCode             string `json:"reason_code,omitempty"`
-	Message                string `json:"message,omitempty"`
-	AttemptCount           int    `json:"attempt_count,omitempty"`
-	DiagnosticArtifactPath string `json:"diagnostic_artifact_path,omitempty"`
+	Stage                  string             `json:"stage"`
+	StepID                 string             `json:"step_id,omitempty"`
+	LaneID                 string             `json:"lane_id,omitempty"`
+	ModuleKey              string             `json:"module_key,omitempty"`
+	ChunkID                string             `json:"chunk_id,omitempty"`
+	ChunkIndex             int                `json:"chunk_index,omitempty"`
+	ValidatorName          string             `json:"validator_name,omitempty"`
+	ReasonCode             string             `json:"reason_code,omitempty"`
+	Message                string             `json:"message,omitempty"`
+	AttemptCount           int                `json:"attempt_count,omitempty"`
+	DiagnosticArtifactPath string             `json:"diagnostic_artifact_path,omitempty"`
+	Validation             *ValidationSummary `json:"validation,omitempty"`
+}
+
+// ValidationSummary is the bounded, durable outcome of validating one
+// producer result. It deliberately contains identities and stable codes, not
+// model responses, corrective guidance, validator diagnostics, or payloads.
+type ValidationSummary struct {
+	Stage                string   `json:"stage,omitempty"`
+	StepID               string   `json:"step_id,omitempty"`
+	LaneID               string   `json:"lane_id,omitempty"`
+	ModuleKey            string   `json:"module_key,omitempty"`
+	ChunkID              string   `json:"chunk_id,omitempty"`
+	ChunkIndex           int      `json:"chunk_index,omitempty"`
+	Status               string   `json:"status"`
+	RejectingValidators  []string `json:"rejecting_validators,omitempty"`
+	ReasonCodes          []string `json:"reason_codes,omitempty"`
+	IncompleteValidators []string `json:"incomplete_validators,omitempty"`
+	ProducerAttemptCount int      `json:"producer_attempt_count"`
+	TerminalAction       string   `json:"terminal_action"`
+}
+
+// CloneValidationSummary returns an independently owned durable summary.
+func CloneValidationSummary(summary ValidationSummary) ValidationSummary {
+	summary.RejectingValidators = append([]string(nil), summary.RejectingValidators...)
+	summary.ReasonCodes = append([]string(nil), summary.ReasonCodes...)
+	summary.IncompleteValidators = append([]string(nil), summary.IncompleteValidators...)
+	return summary
 }
 
 type CheckpointDecisionManifest struct {
@@ -163,6 +190,7 @@ type RunManifest struct {
 	References          []ReferenceProvenance        `json:"references,omitempty"`
 	NormalizedOutputs   []NormalizedOutputManifest   `json:"normalized_outputs,omitempty"`
 	RejectedOutputs     []RejectedOutputManifest     `json:"rejected_outputs,omitempty"`
+	ValidationSummaries []ValidationSummary          `json:"validation_summaries,omitempty"`
 	CheckpointDecisions []CheckpointDecisionManifest `json:"checkpoint_decisions,omitempty"`
 	LLMProfiles         []LLMProfileManifest         `json:"llm_profiles,omitempty"`
 	Metadata            map[string]any               `json:"metadata,omitempty"`

internal/core/config/config.go

@@ -116,6 +116,7 @@ func (c *ConcurrencyConfig) recomputeStageWorkerDefaults() {
 
 func clonePipelineProfile(in pipeline.PipelineProfile) pipeline.PipelineProfile {
 	out := in
+	out.ValidationPolicy = cloneValidationPolicyOverride(in.ValidationPolicy)
 	if in.StructuredOutputRepairAttempts != nil {
 		value := *in.StructuredOutputRepairAttempts
 		out.StructuredOutputRepairAttempts = &value
@@ -200,6 +201,7 @@ func cloneReferenceSource(in pipeline.ReferenceSource) pipeline.ReferenceSource
 
 func cloneModuleBinding(in pipeline.ModuleBinding) pipeline.ModuleBinding {
 	out := in
+	out.ValidationPolicy = cloneValidationPolicyOverride(in.ValidationPolicy)
 	if in.StructuredOutputRepairAttempts != nil {
 		value := *in.StructuredOutputRepairAttempts
 		out.StructuredOutputRepairAttempts = &value
@@ -212,6 +214,26 @@ func cloneModuleBinding(in pipeline.ModuleBinding) pipeline.ModuleBinding {
 	return out
 }
 
+func cloneValidationPolicyOverride(in *pipeline.ValidationPolicyOverride) *pipeline.ValidationPolicyOverride {
+	if in == nil {
+		return nil
+	}
+	out := *in
+	if in.ProducerStructuralFailure != nil {
+		value := *in.ProducerStructuralFailure
+		out.ProducerStructuralFailure = &value
+	}
+	if in.SemanticRejection != nil {
+		value := *in.SemanticRejection
+		out.SemanticRejection = &value
+	}
+	if in.ValidatorFailure != nil {
+		value := *in.ValidatorFailure
+		out.ValidatorFailure = &value
+	}
+	return &out
+}
+
 func cloneValidatorOverride(in pipeline.ValidatorOverride) pipeline.ValidatorOverride {
 	out := pipeline.ValidatorOverride{Set: in.Set}
 	if len(in.Validators) > 0 {

internal/core/config/file_config.go

@@ -37,6 +37,7 @@ type FilePromptKitLocalBackendConfig struct {
 type FilePipelineProfile struct {
 	LLMProfile                     *string                            `yaml:"llm_profile,omitempty"`
 	StructuredOutputRepairAttempts *int                               `yaml:"structured_output_repair_attempts,omitempty"`
+	ValidationPolicy               *pipeline.ValidationPolicyOverride `yaml:"validation_policy,omitempty"`
 	Input                          fileModuleBinding                  `yaml:"input"`
 	Chunk                          *fileModuleBinding                 `yaml:"chunk,omitempty"`
 	Artifacts                      map[string]FileArtifactLaneProfile `yaml:"artifacts,omitempty"`
@@ -55,12 +56,19 @@ func (p *FilePipelineProfile) UnmarshalYAML(node *yaml.Node) error {
 	type plainFilePipelineProfile FilePipelineProfile
 	var decoded plainFilePipelineProfile
 	seen, err := decodeKnownMapping(node, &decoded, map[string]struct{}{
-		"llm_profile": {}, "structured_output_repair_attempts": {}, "input": {}, "chunk": {}, "artifacts": {}, "steps": {}, "output": {}, "references": {},
+		"llm_profile": {}, "structured_output_repair_attempts": {}, "validation_policy": {}, "input": {}, "chunk": {}, "artifacts": {}, "steps": {}, "output": {}, "references": {},
 	}, "pipeline profile")
 	if err != nil {
 		return err
 	}
 	*p = FilePipelineProfile(decoded)
+	if validationPolicyNode, ok := mappingValue(node, "validation_policy"); ok {
+		policy, err := parseValidationPolicy(validationPolicyNode, "pipeline profile")
+		if err != nil {
+			return err
+		}
+		p.ValidationPolicy = policy
+	}
 	_, p.artifactsSet = seen["artifacts"]
 	_, p.stepsSet = seen["steps"]
 	_, p.llmProfileSet = seen["llm_profile"]
@@ -154,6 +162,7 @@ type fileModuleBinding struct {
 	Module                         string
 	LLMProfile                     string
 	StructuredOutputRepairAttempts *int
+	ValidationPolicy               *pipeline.ValidationPolicyOverride
 	Retries                        int
 	Options                        map[string]any
 	References                     map[string]fileReferenceSource
@@ -250,9 +259,14 @@ func (b *fileModuleBinding) UnmarshalYAML(node *yaml.Node) error {
 		b.Module = strings.TrimSpace(module)
 		return nil
 	case yaml.MappingNode:
+		seen := make(map[string]struct{}, len(node.Content)/2)
 		for i := 0; i < len(node.Content); i += 2 {
 			keyNode := node.Content[i]
 			valueNode := node.Content[i+1]
+			if _, exists := seen[keyNode.Value]; exists {
+				return fmt.Errorf("module binding field %q is duplicated", keyNode.Value)
+			}
+			seen[keyNode.Value] = struct{}{}
 			switch keyNode.Value {
 			case "module":
 				var module string
@@ -275,6 +289,12 @@ func (b *fileModuleBinding) UnmarshalYAML(node *yaml.Node) error {
 					return err
 				}
 				b.StructuredOutputRepairAttempts = attempts
+			case "validation_policy":
+				policy, err := parseValidationPolicy(valueNode, "module binding")
+				if err != nil {
+					return err
+				}
+				b.ValidationPolicy = policy
 			case "retries":
 				var retries int
 				if err := valueNode.Decode(&retries); err != nil {
@@ -318,6 +338,7 @@ func (b fileModuleBinding) toPipelineBinding() pipeline.ModuleBinding {
 		Module:                         strings.TrimSpace(b.Module),
 		LLMProfile:                     strings.TrimSpace(b.LLMProfile),
 		StructuredOutputRepairAttempts: cloneStructuredOutputRepairAttempts(b.StructuredOutputRepairAttempts),
+		ValidationPolicy:               cloneValidationPolicyOverride(b.ValidationPolicy),
 		Retries:                        b.Retries,
 		Options:                        cloneOptions(b.Options),
 		References:                     fileReferenceSourcesToPipeline(b.References),
@@ -325,6 +346,51 @@ func (b fileModuleBinding) toPipelineBinding() pipeline.ModuleBinding {
 	}
 }
 
+func mappingValue(node *yaml.Node, key string) (*yaml.Node, bool) {
+	for i := 0; i < len(node.Content); i += 2 {
+		if node.Content[i].Value == key {
+			return node.Content[i+1], true
+		}
+	}
+	return nil, false
+}
+
+func parseValidationPolicy(node *yaml.Node, context string) (*pipeline.ValidationPolicyOverride, error) {
+	if node == nil || node.Tag == "!!null" || node.Kind != yaml.MappingNode {
+		return nil, fmt.Errorf("%s validation_policy must be an object", context)
+	}
+	policy := &pipeline.ValidationPolicyOverride{}
+	seen := make(map[string]struct{}, len(node.Content)/2)
+	for i := 0; i < len(node.Content); i += 2 {
+		key := node.Content[i].Value
+		value := node.Content[i+1]
+		if _, exists := seen[key]; exists {
+			return nil, fmt.Errorf("%s validation_policy field %q is duplicated", context, key)
+		}
+		seen[key] = struct{}{}
+		if value.Tag == "!!null" || value.Kind != yaml.ScalarNode || value.Tag != "!!str" {
+			return nil, fmt.Errorf("%s validation_policy.%s must be a string", context, key)
+		}
+		switch key {
+		case "producer_structural_failure":
+			value := pipeline.ProducerStructuralFailureAction(value.Value)
+			policy.ProducerStructuralFailure = &value
+		case "semantic_rejection":
+			value := pipeline.SemanticRejectionAction(value.Value)
+			policy.SemanticRejection = &value
+		case "validator_failure":
+			value := pipeline.ValidatorFailureAction(value.Value)
+			policy.ValidatorFailure = &value
+		default:
+			return nil, fmt.Errorf("field %s not found in %s validation_policy", key, context)
+		}
+	}
+	if err := policy.Validate(); err != nil {
+		return nil, fmt.Errorf("%s validation_policy: %w", context, err)
+	}
+	return policy, nil
+}
+
 func validateStructuredOutputRepairAttemptsNode(node *yaml.Node, context string) error {
 	if node.Kind != yaml.MappingNode {
 		return fmt.Errorf("%s must be an object", context)
@@ -573,6 +639,7 @@ func (c *Config) applyFileConfigWithLookup(fileCfg FileConfig, lookup func(strin
 			ID:                             pipelineID,
 			LLMProfile:                     llmProfile,
 			StructuredOutputRepairAttempts: cloneStructuredOutputRepairAttempts(filePipeline.StructuredOutputRepairAttempts),
+			ValidationPolicy:               cloneValidationPolicyOverride(filePipeline.ValidationPolicy),
 			Input:                          filePipeline.Input.toPipelineBinding(),
 			Artifacts:                      make(map[string]pipeline.ArtifactLaneProfile, len(filePipeline.Artifacts)),
 			References:                     fileReferenceSourcesToPipeline(filePipeline.References),

internal/core/config/redaction.go

@@ -32,6 +32,7 @@ func (e EffectiveConfig) RedactedResolvedPipelinePayload() pipeline.ResolvedPipe
 
 func cloneResolvedPipeline(in pipeline.ResolvedPipeline) pipeline.ResolvedPipeline {
 	out := in
+	out.ConfiguredValidationPolicy = cloneValidationPolicyOverride(in.ConfiguredValidationPolicy)
 	out.Input = redactBinding(cloneModuleBinding(in.Input))
 	out.Chunk = redactBinding(cloneModuleBinding(in.Chunk))
 	out.ChunkReferences = pipeline.CloneReferenceTarget(in.ChunkReferences)

internal/core/config/redaction_test.go

@@ -225,6 +225,39 @@ func TestRedactedResolvedPipelinePayloadHandlesTypedOptionContainers(t *testing.
 	}
 }
 
+func TestRedactedEffectiveConfigPayloadOwnsValidationPolicies(t *testing.T) {
+	semantic := pipeline.SemanticRejectionRejectOutput
+	validator := pipeline.ValidatorFailureFailRun
+	configured := &pipeline.ValidationPolicyOverride{SemanticRejection: &semantic, ValidatorFailure: &validator}
+	effective := EffectiveConfig{
+		Config: Config{Pipelines: map[string]pipeline.PipelineProfile{
+			"main": {ValidationPolicy: configured},
+		}},
+		ResolvedPipeline: pipeline.ResolvedPipeline{
+			ConfiguredValidationPolicy: configured,
+			ChunkValidationPolicy: pipeline.ValidationPolicy{
+				ProducerStructuralFailure: pipeline.ProducerStructuralFailureFailRun,
+				SemanticRejection:         pipeline.SemanticRejectionRejectOutput,
+				ValidatorFailure:          pipeline.ValidatorFailureFailRun,
+			},
+		},
+	}
+
+	payload := effective.RedactedSummaryPayload().(EffectiveConfig)
+	encoded, err := json.Marshal(payload)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(encoded), `"configured_validation_policy":{"semantic_rejection":"reject_output","validator_failure":"fail_run"}`) || !strings.Contains(string(encoded), `"chunk_validation_policy":{"producer_structural_failure":"fail_run","semantic_rejection":"reject_output","validator_failure":"fail_run"}`) {
+		t.Fatalf("redacted payload omitted validation policy: %s", encoded)
+	}
+	*payload.Config.Pipelines["main"].ValidationPolicy.SemanticRejection = pipeline.SemanticRejectionFailRun
+	*payload.ResolvedPipeline.ConfiguredValidationPolicy.ValidatorFailure = pipeline.ValidatorFailureWarnContinue
+	if *effective.Config.Pipelines["main"].ValidationPolicy.SemanticRejection != pipeline.SemanticRejectionRejectOutput || *effective.ResolvedPipeline.ConfiguredValidationPolicy.ValidatorFailure != pipeline.ValidatorFailureFailRun {
+		t.Fatal("redacted payload aliases validation policy")
+	}
+}
+
 func redactionTestBinding(name string) pipeline.ModuleBinding {
 	return pipeline.ModuleBinding{
 		Module: "safe-" + name,

internal/core/config/validation.go

@@ -119,6 +119,11 @@ func validatePipelineProfiles(profiles map[string]pipeline.PipelineProfile) erro
 		if err := validateStructuredOutputRepairAttempts(fmt.Sprintf("pipeline %q", id), profile.StructuredOutputRepairAttempts); err != nil {
 			return err
 		}
+		if profile.ValidationPolicy != nil {
+			if err := profile.ValidationPolicy.Validate(); err != nil {
+				return fmt.Errorf("pipeline %q validation_policy: %w", id, err)
+			}
+		}
 		if err := validateBinding(id, "", "input", profile.Input, false); err != nil {
 			return err
 		}
@@ -198,6 +203,16 @@ func validateBinding(
 	binding pipeline.ModuleBinding,
 	referencesAllowed bool,
 ) error {
+	if binding.ValidationPolicy != nil {
+		switch slot {
+		case "chunk", "extract", "merge", "normalize":
+		default:
+			return fmt.Errorf("%s validation_policy is not supported", referenceContext(pipelineID, laneID, slot))
+		}
+		if err := binding.ValidationPolicy.Validate(); err != nil {
+			return fmt.Errorf("%s validation_policy: %w", referenceContext(pipelineID, laneID, slot), err)
+		}
+	}
 	if err := validateStructuredOutputRepairAttempts(referenceContext(pipelineID, laneID, slot), binding.StructuredOutputRepairAttempts); err != nil {
 		return err
 	}
@@ -255,8 +270,11 @@ func validateValidatorOverride(pipelineID string, laneID string, slot string, ov
 		if validator.Validators.Set {
 			return fmt.Errorf("%s nested validators are not supported", context)
 		}
-		if validator.Retries != 0 {
-			return fmt.Errorf("%s retries are not supported", context)
+		if validator.ValidationPolicy != nil {
+			return fmt.Errorf("%s validation_policy is not supported", context)
+		}
+		if validator.Retries < 0 {
+			return fmt.Errorf("%s retries must be greater than or equal to zero", context)
 		}
 		if validator.LLMProfile != "" && strings.TrimSpace(validator.LLMProfile) == "" {
 			return fmt.Errorf("%s llm_profile must not be empty when set", context)

internal/core/config/validation_contract_test.go

@@ -374,17 +374,17 @@ func TestValidateValidatorBindingRules(t *testing.T) {
 			want: "chunk validators[0] module must not be empty",
 		},
 		{
-			name: "validator retries",
+			name: "negative validator retries",
 			setup: func(profile *pipeline.PipelineProfile) {
 				profile.Chunk.Validators = pipeline.ValidatorOverride{
 					Set: true,
 					Validators: []pipeline.ModuleBinding{{
 						Module:  "validator",
-						Retries: 1,
+						Retries: -1,
 					}},
 				}
 			},
-			want: "chunk validators[0] retries are not supported",
+			want: "chunk validators[0] retries must be greater than or equal to zero",
 		},
 		{
 			name: "validator references",

internal/core/config/validation_policy_contract_test.go

@@ -0,0 +1,97 @@
+package config
+
+import (
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
+)
+
+func TestValidationPolicyFileConfigurationIsStrictAndPresenceAware(t *testing.T) {
+	const valid = `version: 4
+pipelines:
+  main:
+    validation_policy:
+      producer_structural_failure: reject_output
+      semantic_rejection: fail_run
+    input: seriatim
+    chunk:
+      module: generic
+      validation_policy:
+        validator_failure: fail_run
+    artifacts:
+      lane:
+        extract:
+          module: extract
+          validation_policy:
+            semantic_rejection: reject_output
+`
+	cfg := applyFileConfig(t, valid)
+	profile := cfg.Pipelines["main"]
+	if profile.ValidationPolicy == nil || profile.ValidationPolicy.ProducerStructuralFailure == nil || *profile.ValidationPolicy.ProducerStructuralFailure != pipeline.ProducerStructuralFailureRejectOutput || profile.ValidationPolicy.SemanticRejection == nil || *profile.ValidationPolicy.SemanticRejection != pipeline.SemanticRejectionFailRun || profile.ValidationPolicy.ValidatorFailure != nil {
+		t.Fatalf("pipeline validation policy = %#v", profile.ValidationPolicy)
+	}
+	if profile.Chunk.ValidationPolicy == nil || profile.Chunk.ValidationPolicy.ValidatorFailure == nil || *profile.Chunk.ValidationPolicy.ValidatorFailure != pipeline.ValidatorFailureFailRun {
+		t.Fatalf("chunk validation policy = %#v", profile.Chunk.ValidationPolicy)
+	}
+	lane := profile.Artifacts["lane"]
+	if lane.Extract.ValidationPolicy == nil || lane.Extract.ValidationPolicy.SemanticRejection == nil || *lane.Extract.ValidationPolicy.SemanticRejection != pipeline.SemanticRejectionRejectOutput {
+		t.Fatalf("extract validation policy = %#v", lane.Extract.ValidationPolicy)
+	}
+
+	for _, test := range []struct {
+		name string
+		yaml string
+	}{
+		{"null object", strings.Replace(valid, "validation_policy:\n      producer_structural_failure: reject_output\n      semantic_rejection: fail_run", "validation_policy: null", 1)},
+		{"null field", strings.Replace(valid, "semantic_rejection: fail_run", "semantic_rejection: null", 1)},
+		{"unknown field", strings.Replace(valid, "semantic_rejection: fail_run", "unknown: fail_run", 1)},
+		{"duplicate field", strings.Replace(valid, "semantic_rejection: fail_run", "semantic_rejection: fail_run\n      semantic_rejection: reject_output", 1)},
+		{"invalid enum", strings.Replace(valid, "semantic_rejection: fail_run", "semantic_rejection: continue", 1)},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			if _, err := ParseFileConfigYAML([]byte(test.yaml)); err == nil {
+				t.Fatal("ParseFileConfigYAML() error = nil, want strict validation-policy rejection")
+			}
+		})
+	}
+}
+
+func TestValidationPolicyPlacementRules(t *testing.T) {
+	policy := &pipeline.ValidationPolicyOverride{}
+	semantic := pipeline.SemanticRejectionRejectOutput
+	policy.SemanticRejection = &semantic
+	base := pipeline.PipelineProfile{
+		ID:    "main",
+		Input: pipeline.Binding("input"),
+		Artifacts: map[string]pipeline.ArtifactLaneProfile{
+			"lane": {Extract: pipeline.Binding("extract")},
+		},
+	}
+	for _, test := range []struct {
+		name   string
+		mutate func(*pipeline.PipelineProfile)
+	}{
+		{"input", func(profile *pipeline.PipelineProfile) { profile.Input.ValidationPolicy = policy }},
+		{"output", func(profile *pipeline.PipelineProfile) {
+			profile.Output = pipeline.Binding("output")
+			profile.Output.ValidationPolicy = policy
+		}},
+		{"validator", func(profile *pipeline.PipelineProfile) {
+			lane := profile.Artifacts["lane"]
+			lane.Extract.Validators = pipeline.ValidatorOverride{Set: true, Validators: []pipeline.ModuleBinding{{Module: "validator", ValidationPolicy: policy}}}
+			profile.Artifacts["lane"] = lane
+		}},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			profile := base
+			profile.Artifacts = map[string]pipeline.ArtifactLaneProfile{"lane": base.Artifacts["lane"]}
+			test.mutate(&profile)
+			cfg := Default()
+			cfg.Pipelines = map[string]pipeline.PipelineProfile{"main": profile}
+			if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "validation_policy") {
+				t.Fatalf("Config.Validate() error = %v, want placement rejection", err)
+			}
+		})
+	}
+}

internal/core/debugbundle/bundle_test.go

@@ -110,7 +110,7 @@ func TestSummaryWriterWritesEverySummaryArtifact(t *testing.T) {
 	if err := summary.WriteRunReport(RunReport{RunID: bundle.RunID(), PipelineID: "test"}); err != nil {
 		t.Fatal(err)
 	}
-	if err := summary.WriteWarnings([]contracts.Warning{{ReasonCode: "test"}}); err != nil {
+	if err := summary.WriteDiagnostics(contracts.DiagnosticCollection{}); err != nil {
 		t.Fatal(err)
 	}
 	if err := summary.WriteError("failed"); err != nil {
@@ -126,7 +126,7 @@ func TestSummaryWriterWritesEverySummaryArtifact(t *testing.T) {
 		ArtifactRunManifest,
 		ArtifactChunkPlan,
 		ArtifactRunReport,
-		ArtifactWarnings,
+		ArtifactDiagnostics,
 		ArtifactErrorLog,
 	} {
 		info, err := os.Stat(filepath.Join(bundle.SummaryRoot(), name))

internal/core/debugbundle/summary.go

@@ -20,7 +20,7 @@ const (
 	ArtifactRunManifest        = "run-manifest.json"
 	ArtifactChunkPlan          = "chunk-plan.json"
 	ArtifactRunReport          = "run-report.json"
-	ArtifactWarnings           = "warnings.json"
+	ArtifactDiagnostics        = "final-diagnostics.json"
 	ArtifactErrorLog           = "error.log"
 )
 
@@ -45,15 +45,19 @@ type Invocation struct {
 	StartedAt               time.Time `json:"started_at"`
 }
 type RunReport struct {
-	RunID            string `json:"run_id"`
-	PipelineID       string `json:"pipeline_id"`
-	OutputPath       string `json:"output_path,omitempty"`
-	DebugPath        string `json:"debug_path,omitempty"`
-	Succeeded        bool   `json:"succeeded"`
-	OutputCount      int    `json:"output_count"`
-	RejectedCount    int    `json:"rejected_count"`
-	WarningCount     int    `json:"warning_count"`
-	ValidationStatus string `json:"validation_status,omitempty"`
+	RunID                     string `json:"run_id"`
+	PipelineID                string `json:"pipeline_id"`
+	OutputPath                string `json:"output_path,omitempty"`
+	DebugPath                 string `json:"debug_path,omitempty"`
+	Succeeded                 bool   `json:"succeeded"`
+	OutputCount               int    `json:"output_count"`
+	RejectedCount             int    `json:"rejected_count"`
+	WarningGroupCount         int    `json:"warning_group_count"`
+	WarningOccurrenceCount    int    `json:"warning_occurrence_count"`
+	DiagnosticGroupCount      int    `json:"diagnostic_group_count"`
+	DiagnosticOccurrenceCount int    `json:"diagnostic_occurrence_count"`
+	DiagnosticsTruncated      bool   `json:"diagnostics_truncated"`
+	ValidationStatus          string `json:"validation_status,omitempty"`
 }
 type SummaryWriter struct {
 	root, runID string
@@ -101,8 +105,8 @@ func (w *SummaryWriter) WriteChunkPlan(v artifacts.ChunkPlanSummary) error {
 	return w.writeJSON(ArtifactChunkPlan, v)
 }
 func (w *SummaryWriter) WriteRunReport(v RunReport) error { return w.writeJSON(ArtifactRunReport, v) }
-func (w *SummaryWriter) WriteWarnings(v []contracts.Warning) error {
-	return w.writeJSON(ArtifactWarnings, v)
+func (w *SummaryWriter) WriteDiagnostics(v contracts.DiagnosticCollection) error {
+	return w.writeJSON(ArtifactDiagnostics, contracts.CloneDiagnosticCollection(v))
 }
 func (w *SummaryWriter) WriteError(message string) error {
 	return w.writeBytes(ArtifactErrorLog, []byte(message+"\n"))

internal/framework/checkpoint/filesystem_test.go

@@ -24,7 +24,6 @@ type filesystemCheckpointFixture struct {
 	merge        pipeline.CheckpointArtifact
 	normalize    pipeline.CheckpointArtifact
 	dependencies []pipeline.CheckpointFingerprint
-	warnings     []contracts.Warning
 	rejected     []contracts.RejectedOutput
 }
 
@@ -36,9 +35,9 @@ func TestFilesystemCheckpointRoundTripsAllStages(t *testing.T) {
 	fixture.doc.Metadata["owner"] = "caller mutation"
 	fixture.extract.Artifact.Content[0] = 'x'
 	fixture.extract.Artifact.Metadata["content"] = "caller mutation"
+	fixture.extract.Diagnostics[0].Diagnostic.Samples[0].Message = "caller mutation"
 	fixture.merge.Artifact.Content[0] = 'x'
 	fixture.normalize.Artifact.Content[0] = 'x'
-	fixture.warnings[0].Message = "caller mutation"
 	fixture.rejected[0].Message = "caller mutation"
 
 	t.Run("source", func(t *testing.T) {
@@ -60,11 +59,11 @@ func TestFilesystemCheckpointRoundTripsAllStages(t *testing.T) {
 
 	t.Run("extract", func(t *testing.T) {
 		got, decision := fixture.loader.Extract("lane-a", "extract-module", fixture.dependencies)
-		if !decision.Reused || len(got.Outputs) != 1 || len(got.Rejected) != 1 || len(got.Warnings) != 1 {
+		if !decision.Reused || len(got.Outputs) != 1 || len(got.Rejected) != 1 || len(got.Outputs[0].Diagnostics) != 1 {
 			t.Fatalf("extract result=%#v decision=%#v", got, decision)
 		}
 		output := got.Outputs[0]
-		if !bytes.Equal(output.Artifact.Content, []byte(`{"spell":"fire"}`)) || output.Artifact.Kind != "spell" || output.Artifact.Schema.ID != "spell-schema" || output.Artifact.Schema.Version != "1" || output.Artifact.MediaType != "application/json" || output.Artifact.Metadata["chunk"] != "chunk-a" || output.ChunkRef.StartUnitID != 1 || got.Warnings[0].ReasonCode != "partial" || got.Rejected[0].ReasonCode != "invalid_source" {
+		if !bytes.Equal(output.Artifact.Content, []byte(`{"spell":"fire"}`)) || output.Artifact.Kind != "spell" || output.Artifact.Schema.ID != "spell-schema" || output.Artifact.Schema.Version != "1" || output.Artifact.MediaType != "application/json" || output.Artifact.Metadata["chunk"] != "chunk-a" || output.ChunkRef.StartUnitID != 1 || output.Diagnostics[0].Diagnostic.ReasonCode != "normalized_record" || got.Rejected[0].ReasonCode != "invalid_source" {
 			t.Fatalf("extract values were not restored: %#v", got)
 		}
 		manifest := readManifest[ExtractLaneManifest](t, filepath.Join(fixture.root, mustRelativePath(t, fixture.identity), "extract", "lane-a", "manifest.json"))
@@ -73,37 +72,36 @@ func TestFilesystemCheckpointRoundTripsAllStages(t *testing.T) {
 		}
 
 		got.Outputs[0].Artifact.Content[0] = 'y'
-		got.Warnings[0].Message = "loaded mutation"
 		reloaded, decision := fixture.loader.Extract("lane-a", "extract-module", fixture.dependencies)
-		if !decision.Reused || !bytes.Equal(reloaded.Outputs[0].Artifact.Content, []byte(`{"spell":"fire"}`)) || reloaded.Warnings[0].Message != "partial output" {
+		if !decision.Reused || !bytes.Equal(reloaded.Outputs[0].Artifact.Content, []byte(`{"spell":"fire"}`)) {
 			t.Fatalf("extract reload changed after loaded mutation: %#v decision=%#v", reloaded, decision)
 		}
 	})
 
 	for _, tt := range []struct {
 		name string
-		load func() (pipeline.CheckpointArtifact, []contracts.Warning, pipeline.CheckpointDecision)
+		load func() (pipeline.CheckpointArtifact, pipeline.CheckpointDecision)
 		want []byte
 	}{
-		{name: "merge", load: func() (pipeline.CheckpointArtifact, []contracts.Warning, pipeline.CheckpointDecision) {
+		{name: "merge", load: func() (pipeline.CheckpointArtifact, pipeline.CheckpointDecision) {
 			got, decision := fixture.loader.Merge("lane-a", "merge-module", fixture.dependencies)
-			return got.Output, got.Warnings, decision
+			return got.Output, decision
 		}, want: []byte(`{"spells":["fire"]}`)},
-		{name: "normalize", load: func() (pipeline.CheckpointArtifact, []contracts.Warning, pipeline.CheckpointDecision) {
+		{name: "normalize", load: func() (pipeline.CheckpointArtifact, pipeline.CheckpointDecision) {
 			got, decision := fixture.loader.Normalize("lane-a", "normalize-module", fixture.dependencies)
-			return got.Output, got.Warnings, decision
+			return got.Output, decision
 		}, want: []byte(`{"spells":["fire"],"normalized":true}`)},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
-			got, warnings, decision := tt.load()
-			if !decision.Reused || !bytes.Equal(got.Artifact.Content, tt.want) || got.Artifact.Kind != "spell" || got.Artifact.Schema.ID != "spell-schema" || got.Artifact.Schema.Version != "1" || got.Artifact.Metadata["lane"] != "lane-a" || len(warnings) != 1 || warnings[0].ReasonCode != "review" {
-				t.Fatalf("%s result=%#v warnings=%#v decision=%#v", tt.name, got, warnings, decision)
+			got, decision := tt.load()
+			if !decision.Reused || !bytes.Equal(got.Artifact.Content, tt.want) || got.Artifact.Kind != "spell" || got.Artifact.Schema.ID != "spell-schema" || got.Artifact.Schema.Version != "1" || got.Artifact.Metadata["lane"] != "lane-a" || len(got.Diagnostics) != 1 || got.Diagnostics[0].Diagnostic.ReasonCode != "normalized_record" {
+				t.Fatalf("%s result=%#v decision=%#v", tt.name, got, decision)
 			}
 
 			got.Artifact.Content[0] = 'z'
-			reloaded, warnings, decision := tt.load()
-			if !decision.Reused || !bytes.Equal(reloaded.Artifact.Content, tt.want) || warnings[0].Message != "review manually" {
-				t.Fatalf("%s reload changed after loaded mutation: %#v warnings=%#v decision=%#v", tt.name, reloaded, warnings, decision)
+			reloaded, decision := tt.load()
+			if !decision.Reused || !bytes.Equal(reloaded.Artifact.Content, tt.want) {
+				t.Fatalf("%s reload changed after loaded mutation: %#v decision=%#v", tt.name, reloaded, decision)
 			}
 		})
 	}
@@ -159,6 +157,7 @@ func TestFilesystemCheckpointRejectsIncompatibleManifests(t *testing.T) {
 	}{
 		{"v1 schema", func(m map[string]any) { m["workspace_schema_version"] = WorkspaceSchemaVersionV1 }, pipeline.CheckpointReasonWorkspaceSchemaIncompatible},
 		{"v2 schema", func(m map[string]any) { m["workspace_schema_version"] = WorkspaceSchemaVersionV2 }, pipeline.CheckpointReasonWorkspaceSchemaIncompatible},
+		{"v3 schema", func(m map[string]any) { m["workspace_schema_version"] = WorkspaceSchemaVersionV3 }, pipeline.CheckpointReasonWorkspaceSchemaIncompatible},
 		{"unknown schema", func(m map[string]any) { m["workspace_schema_version"] = "notarius.workspace.future" }, pipeline.CheckpointReasonWorkspaceSchemaIncompatible},
 		{"identity", func(m map[string]any) { m["metadata"].(map[string]any)["checkpoint_identity_digest"] = "sha256:other" }, pipeline.CheckpointReasonIdentityMismatch},
 		{"stage", func(m map[string]any) { m["stage"] = string(StageMerge) }, pipeline.CheckpointReasonStageMismatch},
@@ -204,6 +203,11 @@ func TestFilesystemCheckpointRejectsIncompleteArtifactsAndContent(t *testing.T)
 		{"content digest", func(m map[string]any) {
 			m["outputs"].([]any)[0].(map[string]any)["content"].(map[string]any)["content_digest"] = "sha256:other"
 		}, pipeline.CheckpointReasonArtifactDigestMismatch},
+		{"diagnostics", func(m map[string]any) {
+			m["outputs"].([]any)[0].(map[string]any)["diagnostics"] = []any{map[string]any{
+				"diagnostic": map[string]any{"disposition": "warning", "category": "configuration", "reason_code": "invalid", "occurrence_count": float64(0)},
+			}}
+		}, pipeline.CheckpointReasonArtifactPayloadInvalid},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
 			fixture := seedFilesystemCheckpoints(t)
@@ -337,9 +341,6 @@ func TestFilesystemLoaderReadsAcceptedNormalizeWithoutStageDependencies(t *testi
 	if checkpoint.Output.Artifact.Content == nil || string(checkpoint.Output.Artifact.Content) != string(fixture.normalize.Artifact.Content) {
 		t.Fatalf("accepted normalize output = %#v, want recorded artifact", checkpoint.Output)
 	}
-	if len(checkpoint.Warnings) != 1 || checkpoint.Warnings[0].ReasonCode != "normalized" {
-		t.Fatalf("accepted normalize warnings = %#v", checkpoint.Warnings)
-	}
 }
 
 func TestFilesystemLoaderRejectsInvalidAcceptedNormalize(t *testing.T) {
@@ -402,8 +403,7 @@ func seedAcceptedNormalizeCheckpoint(t *testing.T) filesystemCheckpointFixture {
 		t.Fatal(err)
 	}
 	stepRecorder := recorder.(pipeline.StepCheckpointRecorder)
-	warnings := []contracts.Warning{{Scope: "normalize", ReasonCode: "normalized", Message: "normalized warning"}}
-	if err := stepRecorder.NormalizeSucceededForStep("step-1", "lane-a", "normalize-module", []pipeline.CheckpointFingerprint{{Name: "merge", Value: "sha256:unavailable"}}, artifact, warnings); err != nil {
+	if err := stepRecorder.NormalizeSucceededForStep("step-1", "lane-a", "normalize-module", []pipeline.CheckpointFingerprint{{Name: "merge", Value: "sha256:unavailable"}}, artifact); err != nil {
 		t.Fatal(err)
 	}
 	loader, err := NewFilesystemLoader(root, identity)
@@ -493,20 +493,18 @@ func seedFilesystemCheckpoints(t *testing.T) filesystemCheckpointFixture {
 		merge:        checkpointArtifact("merge", `{"spells":["fire"]}`),
 		normalize:    checkpointArtifact("normalize", `{"spells":["fire"],"normalized":true}`),
 		dependencies: []pipeline.CheckpointFingerprint{{Name: "source", Value: "sha256:source"}, {Name: "chunk-plan", Value: "sha256:plan"}},
-		warnings:     []contracts.Warning{{Scope: "extract", ReasonCode: "partial", Message: "partial output"}},
 		rejected:     []contracts.RejectedOutput{{Stage: "extract", LaneID: "lane-a", ModuleKey: "extract-module", ChunkID: "chunk-a", ValidatorName: "source_refs", ReasonCode: "invalid_source", Message: "source reference is invalid", AttemptCount: 1}},
 	}
 	if err := recorder.SourceSucceeded("source-module", &fixture.doc); err != nil {
 		t.Fatal(err)
 	}
-	if err := recorder.ExtractSucceeded("lane-a", "extract-module", fixture.dependencies, []pipeline.CheckpointArtifact{fixture.extract}, fixture.rejected, fixture.warnings); err != nil {
+	if err := recorder.ExtractSucceeded("lane-a", "extract-module", fixture.dependencies, []pipeline.CheckpointArtifact{fixture.extract}, fixture.rejected); err != nil {
 		t.Fatal(err)
 	}
-	mergeWarnings := []contracts.Warning{{Scope: "merge", ReasonCode: "review", Message: "review manually"}}
-	if err := recorder.MergeSucceeded("lane-a", "merge-module", fixture.dependencies, fixture.merge, mergeWarnings); err != nil {
+	if err := recorder.MergeSucceeded("lane-a", "merge-module", fixture.dependencies, fixture.merge); err != nil {
 		t.Fatal(err)
 	}
-	if err := recorder.NormalizeSucceeded("lane-a", "normalize-module", fixture.dependencies, fixture.normalize, mergeWarnings); err != nil {
+	if err := recorder.NormalizeSucceeded("lane-a", "normalize-module", fixture.dependencies, fixture.normalize); err != nil {
 		t.Fatal(err)
 	}
 	fixture.loader, err = NewFilesystemLoader(root, identity)
@@ -527,7 +525,8 @@ func checkpointArtifact(module, content string) pipeline.CheckpointArtifact {
 	return pipeline.CheckpointArtifact{
 		LaneID: "lane-a", ModuleKey: module, SourceID: "document-1", ChunkID: "chunk-a", ChunkIndex: 0,
 		ChunkRef: source.SourceRef{SourceID: "document-1", StartUnitID: 1, EndUnitID: 1}, SchemaDigest: "sha256:schema",
-		Artifact: contracts.SerializedArtifact{Kind: "spell", Schema: contracts.ArtifactSchema{ID: "spell-schema", Name: "Spell", Version: "1", JSONSchema: []byte(`{"type":"object"}`)}, MediaType: "application/json", Content: []byte(content), Metadata: map[string]any{"chunk": "chunk-a", "lane": "lane-a"}},
+		Artifact:    contracts.SerializedArtifact{Kind: "spell", Schema: contracts.ArtifactSchema{ID: "spell-schema", Name: "Spell", Version: "1", JSONSchema: []byte(`{"type":"object"}`)}, MediaType: "application/json", Content: []byte(content), Metadata: map[string]any{"chunk": "chunk-a", "lane": "lane-a"}},
+		Diagnostics: []pipeline.CheckpointDiagnostic{{Diagnostic: contracts.ProducerDiagnostic{Disposition: contracts.DiagnosticDispositionObservation, Category: contracts.DiagnosticCategoryNormalization, ReasonCode: "normalized_record", OccurrenceCount: 1, Samples: []contracts.DiagnosticSample{{Scope: "fixture", Message: "record normalized"}}}}},
 	}
 }
 

internal/framework/checkpoint/loader.go

@@ -86,7 +86,7 @@ func (l *FilesystemLoader) ExtractForStep(stepID, laneID, moduleKey string, depe
 	if !fingerprintsEqual(checkpointToPipelineFingerprints(manifest.OutputDigests), artifactOutputDigests(outputs)) {
 		return pipeline.ExtractCheckpoint{}, decision(pipeline.CheckpointDecisionExecuted, pipeline.CheckpointReasonArtifactDigestMismatch)
 	}
-	return pipeline.ExtractCheckpoint{Outputs: outputs, Rejected: cloneRejectedOutputs(payload.Rejected), Warnings: cloneWarnings(payload.Warnings)}, reusedDecision()
+	return pipeline.ExtractCheckpoint{Outputs: outputs, Rejected: cloneRejectedOutputs(payload.Rejected)}, reusedDecision()
 }
 
 func (l *FilesystemLoader) Merge(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint) (pipeline.MergeCheckpoint, pipeline.CheckpointDecision) {
@@ -112,7 +112,7 @@ func (l *FilesystemLoader) MergeForStep(stepID, laneID, moduleKey string, depend
 	if !fingerprintsEqual(checkpointToPipelineFingerprints(manifest.OutputDigests), artifactOutputDigests(values)) {
 		return pipeline.MergeCheckpoint{}, decision(pipeline.CheckpointDecisionExecuted, pipeline.CheckpointReasonArtifactDigestMismatch)
 	}
-	return pipeline.MergeCheckpoint{Output: values[0], Warnings: cloneWarnings(payload.Warnings)}, reusedDecision()
+	return pipeline.MergeCheckpoint{Output: values[0]}, reusedDecision()
 }
 
 func (l *FilesystemLoader) Normalize(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint) (pipeline.NormalizeCheckpoint, pipeline.CheckpointDecision) {
@@ -138,7 +138,7 @@ func (l *FilesystemLoader) NormalizeForStep(stepID, laneID, moduleKey string, de
 	if !fingerprintsEqual(checkpointToPipelineFingerprints(manifest.OutputDigests), artifactOutputDigests(values)) {
 		return pipeline.NormalizeCheckpoint{}, decision(pipeline.CheckpointDecisionExecuted, pipeline.CheckpointReasonArtifactDigestMismatch)
 	}
-	return pipeline.NormalizeCheckpoint{Output: values[0], Warnings: cloneWarnings(payload.Warnings)}, reusedDecision()
+	return pipeline.NormalizeCheckpoint{Output: values[0]}, reusedDecision()
 }
 
 func (l *FilesystemLoader) AcceptedNormalize(stepID, laneID, moduleKey string) (pipeline.NormalizeCheckpoint, pipeline.CheckpointDecision) {
@@ -163,7 +163,7 @@ func (l *FilesystemLoader) AcceptedNormalize(stepID, laneID, moduleKey string) (
 	if !fingerprintsEqual(checkpointToPipelineFingerprints(manifest.OutputDigests), artifactOutputDigests(values)) {
 		return pipeline.NormalizeCheckpoint{}, decision(pipeline.CheckpointDecisionExecuted, pipeline.CheckpointReasonArtifactDigestMismatch)
 	}
-	return pipeline.NormalizeCheckpoint{Output: values[0], Warnings: cloneWarnings(payload.Warnings)}, decision(pipeline.CheckpointDecisionReused, pipeline.CheckpointReasonAcceptedArtifactReused)
+	return pipeline.NormalizeCheckpoint{Output: values[0]}, decision(pipeline.CheckpointDecisionReused, pipeline.CheckpointReasonAcceptedArtifactReused)
 }
 
 func (l *FilesystemLoader) validateAcceptedNormalizeManifest(manifest StageManifest, stepID, laneID, moduleKey string) pipeline.CheckpointDecision {
@@ -205,11 +205,33 @@ func artifactCheckpointOutputs(values []artifactCheckpointEnvelope) ([]pipeline.
 		if strings.TrimSpace(string(v.Kind)) == "" || strings.TrimSpace(v.Schema.ID) == "" || strings.TrimSpace(v.Schema.Version) == "" || strings.TrimSpace(v.SchemaDigest) == "" {
 			return nil, &artifactPayloadError{code: pipeline.CheckpointReasonArtifactCodecIncompatible, err: fmt.Errorf("artifact codec identity is incomplete")}
 		}
-		out = append(out, pipeline.CheckpointArtifact{LaneID: v.LaneID, ModuleKey: v.ModuleKey, SourceID: v.SourceID, ChunkID: v.ChunkID, ChunkIndex: v.ChunkIndex, ChunkRef: v.ChunkRef, SchemaDigest: v.SchemaDigest, Artifact: contracts.SerializedArtifact{Kind: v.Kind, Schema: v.Schema, MediaType: v.Content.MediaType, Content: content, Metadata: cloneMetadata(v.Content.Metadata)}})
+		diagnostics, err := cloneAndValidateCheckpointDiagnostics(v.Diagnostics)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, pipeline.CheckpointArtifact{LaneID: v.LaneID, ModuleKey: v.ModuleKey, SourceID: v.SourceID, ChunkID: v.ChunkID, ChunkIndex: v.ChunkIndex, ChunkRef: v.ChunkRef, SchemaDigest: v.SchemaDigest, Artifact: contracts.SerializedArtifact{Kind: v.Kind, Schema: v.Schema, MediaType: v.Content.MediaType, Content: content, Metadata: cloneMetadata(v.Content.Metadata)}, Diagnostics: diagnostics})
 	}
 	return out, nil
 }
 
+func cloneAndValidateCheckpointDiagnostics(values []pipeline.CheckpointDiagnostic) ([]pipeline.CheckpointDiagnostic, error) {
+	if len(values) == 0 {
+		return nil, nil
+	}
+	diagnostics := make([]contracts.ProducerDiagnostic, len(values))
+	for index, value := range values {
+		diagnostics[index] = value.Diagnostic
+	}
+	if err := contracts.ValidateProducerDiagnostics(diagnostics); err != nil {
+		return nil, &artifactPayloadError{code: pipeline.CheckpointReasonArtifactPayloadInvalid, err: fmt.Errorf("checkpoint diagnostics: %w", err)}
+	}
+	cloned := make([]pipeline.CheckpointDiagnostic, len(values))
+	for index, value := range values {
+		cloned[index] = pipeline.CheckpointDiagnostic{Diagnostic: contracts.CloneProducerDiagnostics([]contracts.ProducerDiagnostic{value.Diagnostic})[0], ValidatorKey: value.ValidatorKey}
+	}
+	return cloned, nil
+}
+
 func (l *FilesystemLoader) readJSON(name string, out any) pipeline.CheckpointDecision {
 	if !l.Enabled() {
 		return decision(pipeline.CheckpointDecisionExecuted, pipeline.CheckpointReasonLoadingDisabled)

internal/framework/checkpoint/manifest.go

@@ -3,7 +3,8 @@ package checkpoint
 import "time"
 
 const (
-	WorkspaceSchemaVersion   = "notarius.workspace.v3"
+	WorkspaceSchemaVersion   = "notarius.workspace.v4"
+	WorkspaceSchemaVersionV3 = "notarius.workspace.v3"
 	WorkspaceSchemaVersionV2 = "notarius.workspace.v2"
 	WorkspaceSchemaVersionV1 = "notarius.workspace.v1"
 )

internal/framework/checkpoint/recorder.go

@@ -10,6 +10,7 @@ import (
 	"strings"
 	"time"
 
+	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/fileio"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
@@ -80,18 +81,18 @@ func (r *FilesystemRecorder) ExtractRunningForStep(stepID, laneID string, module
 	return r.writeManifest(laneManifestPath("extract", stepID, laneID), ExtractLaneManifest{StageManifest: manifest})
 }
 
-func (r *FilesystemRecorder) ExtractSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, outputs []pipeline.CheckpointArtifact, rejected []contracts.RejectedOutput, warnings []contracts.Warning) error {
-	return r.ExtractSucceededForStep("", laneID, moduleKey, dependencies, outputs, rejected, warnings)
+func (r *FilesystemRecorder) ExtractSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, outputs []pipeline.CheckpointArtifact, rejected []contracts.RejectedOutput) error {
+	return r.ExtractSucceededForStep("", laneID, moduleKey, dependencies, outputs, rejected)
 }
 
-func (r *FilesystemRecorder) ExtractSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, outputs []pipeline.CheckpointArtifact, rejected []contracts.RejectedOutput, warnings []contracts.Warning) error {
-	payload := artifactExtractEnvelope{Outputs: artifactCheckpointEnvelopes(outputs), Rejected: cloneRejectedOutputs(rejected), Warnings: cloneWarnings(warnings)}
+func (r *FilesystemRecorder) ExtractSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, outputs []pipeline.CheckpointArtifact, rejected []contracts.RejectedOutput) error {
+	payload := artifactExtractEnvelope{Outputs: artifactCheckpointEnvelopes(outputs), Rejected: cloneRejectedOutputs(rejected)}
 	if err := r.writePayload(lanePayloadPath("extract", stepID, laneID, "outputs.json"), payload); err != nil {
 		return err
 	}
 	manifest := r.laneManifest(StageExtract, statusForRejected(rejected), stepID, laneID, moduleKey, dependencies)
 	manifest.OutputDigests = checkpointFingerprints(artifactOutputDigests(outputs))
-	manifest.ValidationStatus = validationStatusString(warnings, rejected)
+	manifest.ValidationStatus = validationStatusString(rejected)
 	manifest.Rejections = rejectionSummaries(rejected)
 	manifest.CompletedAt = timePtr(r.timestamp())
 	return r.writeManifest(laneManifestPath("extract", stepID, laneID), ExtractLaneManifest{StageManifest: manifest, ChunkCount: len(outputs) + len(rejected), OutputCount: len(outputs)})
@@ -118,17 +119,17 @@ func (r *FilesystemRecorder) MergeRunningForStep(stepID, laneID string, moduleKe
 	return r.writeManifest(laneManifestPath("merge", stepID, laneID), MergeLaneManifest{StageManifest: manifest})
 }
 
-func (r *FilesystemRecorder) MergeSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact, warnings []contracts.Warning) error {
-	return r.MergeSucceededForStep("", laneID, moduleKey, dependencies, output, warnings)
+func (r *FilesystemRecorder) MergeSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact) error {
+	return r.MergeSucceededForStep("", laneID, moduleKey, dependencies, output)
 }
 
-func (r *FilesystemRecorder) MergeSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact, warnings []contracts.Warning) error {
-	if err := r.writePayload(lanePayloadPath("merge", stepID, laneID, "output.json"), artifactSingleEnvelope{Output: artifactCheckpointEnvelopeFromOutput(output), Warnings: cloneWarnings(warnings)}); err != nil {
+func (r *FilesystemRecorder) MergeSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact) error {
+	if err := r.writePayload(lanePayloadPath("merge", stepID, laneID, "output.json"), artifactSingleEnvelope{Output: artifactCheckpointEnvelopeFromOutput(output)}); err != nil {
 		return err
 	}
 	manifest := r.laneManifest(StageMerge, StatusSucceeded, stepID, laneID, moduleKey, dependencies)
 	manifest.OutputDigests = checkpointFingerprints(artifactOutputDigests([]pipeline.CheckpointArtifact{output}))
-	manifest.ValidationStatus = validationStatusString(warnings, nil)
+	manifest.ValidationStatus = validationStatusString(nil)
 	manifest.CompletedAt = timePtr(r.timestamp())
 	return r.writeManifest(laneManifestPath("merge", stepID, laneID), MergeLaneManifest{StageManifest: manifest, InputCount: len(dependencies)})
 }
@@ -166,17 +167,17 @@ func (r *FilesystemRecorder) NormalizeRunningForStep(stepID, laneID string, modu
 	return r.writeManifest(laneManifestPath("normalize", stepID, laneID), NormalizeLaneManifest{StageManifest: manifest})
 }
 
-func (r *FilesystemRecorder) NormalizeSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact, warnings []contracts.Warning) error {
-	return r.NormalizeSucceededForStep("", laneID, moduleKey, dependencies, output, warnings)
+func (r *FilesystemRecorder) NormalizeSucceeded(laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact) error {
+	return r.NormalizeSucceededForStep("", laneID, moduleKey, dependencies, output)
 }
 
-func (r *FilesystemRecorder) NormalizeSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact, warnings []contracts.Warning) error {
-	if err := r.writePayload(lanePayloadPath("normalize", stepID, laneID, "output.json"), artifactSingleEnvelope{Output: artifactCheckpointEnvelopeFromOutput(output), Warnings: cloneWarnings(warnings)}); err != nil {
+func (r *FilesystemRecorder) NormalizeSucceededForStep(stepID, laneID, moduleKey string, dependencies []pipeline.CheckpointFingerprint, output pipeline.CheckpointArtifact) error {
+	if err := r.writePayload(lanePayloadPath("normalize", stepID, laneID, "output.json"), artifactSingleEnvelope{Output: artifactCheckpointEnvelopeFromOutput(output)}); err != nil {
 		return err
 	}
 	manifest := r.laneManifest(StageNormalize, StatusSucceeded, stepID, laneID, moduleKey, dependencies)
 	manifest.OutputDigests = checkpointFingerprints(artifactOutputDigests([]pipeline.CheckpointArtifact{output}))
-	manifest.ValidationStatus = validationStatusString(warnings, nil)
+	manifest.ValidationStatus = validationStatusString(nil)
 	manifest.CompletedAt = timePtr(r.timestamp())
 	return r.writeManifest(laneManifestPath("normalize", stepID, laneID), NormalizeLaneManifest{StageManifest: manifest, InputCount: len(dependencies)})
 }
@@ -248,39 +249,37 @@ type sourceDocumentEnvelope struct {
 }
 
 type binaryEnvelope struct {
-	ContentBase64 string              `json:"content_base64,omitempty"`
-	ContentDigest string              `json:"content_digest,omitempty"`
-	MediaType     string              `json:"media_type,omitempty"`
-	Metadata      map[string]any      `json:"metadata,omitempty"`
-	Warnings      []contracts.Warning `json:"warnings,omitempty"`
+	ContentBase64 string         `json:"content_base64,omitempty"`
+	ContentDigest string         `json:"content_digest,omitempty"`
+	MediaType     string         `json:"media_type,omitempty"`
+	Metadata      map[string]any `json:"metadata,omitempty"`
 }
 
 type artifactCheckpointEnvelope struct {
-	LaneID       string                   `json:"lane_id"`
-	ModuleKey    string                   `json:"module_key"`
-	SourceID     string                   `json:"source_id,omitempty"`
-	ChunkID      string                   `json:"chunk_id,omitempty"`
-	ChunkIndex   int                      `json:"chunk_index,omitempty"`
-	ChunkRef     source.SourceRef         `json:"chunk_ref,omitempty"`
-	Kind         contracts.ArtifactKind   `json:"artifact_kind"`
-	Schema       contracts.ArtifactSchema `json:"schema"`
-	SchemaDigest string                   `json:"schema_digest"`
-	Content      binaryEnvelope           `json:"content"`
+	LaneID       string                          `json:"lane_id"`
+	ModuleKey    string                          `json:"module_key"`
+	SourceID     string                          `json:"source_id,omitempty"`
+	ChunkID      string                          `json:"chunk_id,omitempty"`
+	ChunkIndex   int                             `json:"chunk_index,omitempty"`
+	ChunkRef     source.SourceRef                `json:"chunk_ref,omitempty"`
+	Kind         contracts.ArtifactKind          `json:"artifact_kind"`
+	Schema       contracts.ArtifactSchema        `json:"schema"`
+	SchemaDigest string                          `json:"schema_digest"`
+	Content      binaryEnvelope                  `json:"content"`
+	Diagnostics  []pipeline.CheckpointDiagnostic `json:"diagnostics,omitempty"`
 }
 type artifactExtractEnvelope struct {
 	Outputs  []artifactCheckpointEnvelope `json:"outputs"`
 	Rejected []contracts.RejectedOutput   `json:"rejected,omitempty"`
-	Warnings []contracts.Warning          `json:"warnings,omitempty"`
 }
 type artifactSingleEnvelope struct {
-	Output   artifactCheckpointEnvelope `json:"output"`
-	Warnings []contracts.Warning        `json:"warnings,omitempty"`
+	Output artifactCheckpointEnvelope `json:"output"`
 }
 
 func artifactCheckpointEnvelopeFromOutput(output pipeline.CheckpointArtifact) artifactCheckpointEnvelope {
 	schema := contracts.CloneArtifactSchema(output.Artifact.Schema)
 	schema.JSONSchema = nil
-	return artifactCheckpointEnvelope{LaneID: output.LaneID, ModuleKey: output.ModuleKey, SourceID: output.SourceID, ChunkID: output.ChunkID, ChunkIndex: output.ChunkIndex, ChunkRef: output.ChunkRef, Kind: output.Artifact.Kind, Schema: schema, SchemaDigest: output.SchemaDigest, Content: binaryEnvelopeFromContent(output.Artifact.Content, output.Artifact.MediaType, output.Artifact.Metadata, nil)}
+	return artifactCheckpointEnvelope{LaneID: output.LaneID, ModuleKey: output.ModuleKey, SourceID: output.SourceID, ChunkID: output.ChunkID, ChunkIndex: output.ChunkIndex, ChunkRef: output.ChunkRef, Kind: output.Artifact.Kind, Schema: schema, SchemaDigest: output.SchemaDigest, Content: binaryEnvelopeFromContent(output.Artifact.Content, output.Artifact.MediaType, output.Artifact.Metadata), Diagnostics: cloneCheckpointDiagnostics(output.Diagnostics)}
 }
 func artifactCheckpointEnvelopes(outputs []pipeline.CheckpointArtifact) []artifactCheckpointEnvelope {
 	if len(outputs) == 0 {
@@ -292,6 +291,20 @@ func artifactCheckpointEnvelopes(outputs []pipeline.CheckpointArtifact) []artifa
 	}
 	return out
 }
+
+func cloneCheckpointDiagnostics(values []pipeline.CheckpointDiagnostic) []pipeline.CheckpointDiagnostic {
+	if len(values) == 0 {
+		return nil
+	}
+	cloned := make([]pipeline.CheckpointDiagnostic, len(values))
+	for index, value := range values {
+		cloned[index] = pipeline.CheckpointDiagnostic{
+			Diagnostic:   contracts.CloneProducerDiagnostics([]contracts.ProducerDiagnostic{value.Diagnostic})[0],
+			ValidatorKey: value.ValidatorKey,
+		}
+	}
+	return cloned
+}
 func artifactOutputDigests(outputs []pipeline.CheckpointArtifact) []pipeline.CheckpointFingerprint {
 	values := make([]pipeline.CheckpointFingerprint, 0, len(outputs))
 	for i, v := range outputs {
@@ -300,13 +313,12 @@ func artifactOutputDigests(outputs []pipeline.CheckpointArtifact) []pipeline.Che
 	return normalizeFingerprints(values)
 }
 
-func binaryEnvelopeFromContent(content []byte, mediaType string, metadata map[string]any, warnings []contracts.Warning) binaryEnvelope {
+func binaryEnvelopeFromContent(content []byte, mediaType string, metadata map[string]any) binaryEnvelope {
 	return binaryEnvelope{
 		ContentBase64: base64.StdEncoding.EncodeToString(content),
 		ContentDigest: contentDigest(content),
 		MediaType:     mediaType,
 		Metadata:      cloneMetadata(metadata),
-		Warnings:      cloneWarnings(warnings),
 	}
 }
 
@@ -333,18 +345,19 @@ func cloneSourceUnits(units []source.SourceUnit) []source.SourceUnit {
 	return out
 }
 
-func cloneWarnings(warnings []contracts.Warning) []contracts.Warning {
-	if len(warnings) == 0 {
-		return nil
-	}
-	return append([]contracts.Warning(nil), warnings...)
-}
-
 func cloneRejectedOutputs(rejected []contracts.RejectedOutput) []contracts.RejectedOutput {
 	if len(rejected) == 0 {
 		return nil
 	}
-	return append([]contracts.RejectedOutput(nil), rejected...)
+	cloned := make([]contracts.RejectedOutput, len(rejected))
+	for index, item := range rejected {
+		cloned[index] = item
+		if item.Validation != nil {
+			summary := artifacts.CloneValidationSummary(*item.Validation)
+			cloned[index].Validation = &summary
+		}
+	}
+	return cloned
 }
 
 func cloneMetadata(metadata map[string]any) map[string]any {
@@ -452,13 +465,10 @@ func statusForRejected(rejected []contracts.RejectedOutput) StageStatus {
 	return StatusSucceeded
 }
 
-func validationStatusString(warnings []contracts.Warning, rejected []contracts.RejectedOutput) string {
+func validationStatusString(rejected []contracts.RejectedOutput) string {
 	if len(rejected) > 0 {
 		return "rejected"
 	}
-	if len(warnings) > 0 {
-		return "approved_with_warnings"
-	}
 	return "approved"
 }
 

internal/framework/checkpoint/recorder_test.go

@@ -17,7 +17,7 @@ func TestRootBasedRecorderOutputIsReusable(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if err := recorder.ExtractSucceeded("lane", "module", nil, nil, nil, nil); err != nil {
+	if err := recorder.ExtractSucceeded("lane", "module", nil, nil, nil); err != nil {
 		t.Fatal(err)
 	}
 	loader, err := NewFilesystemLoader(root, identity)
@@ -44,7 +44,7 @@ func TestStepAwareRecorderAndLoaderIsolateLaneState(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if err := recorder.(pipeline.StepCheckpointRecorder).ExtractSucceededForStep("step-a", "lane", "module", nil, nil, nil, nil); err != nil {
+	if err := recorder.(pipeline.StepCheckpointRecorder).ExtractSucceededForStep("step-a", "lane", "module", nil, nil, nil); err != nil {
 		t.Fatal(err)
 	}
 	loader, err := NewFilesystemLoader(root, identity)
@@ -87,7 +87,7 @@ func TestStepAwareCheckpointPreservesDistinctDotIdentities(t *testing.T) {
 			LaneID: "lane", ModuleKey: "normalize-module", SourceID: "source", ChunkID: "chunk", ChunkRef: source.SourceRef{SourceID: "source", StartUnitID: 1, EndUnitID: 1}, SchemaDigest: "sha256:schema",
 			Artifact: contracts.SerializedArtifact{Kind: "kind", Schema: contracts.ArtifactSchema{ID: "schema", Name: "Schema", Version: "1"}, MediaType: "application/json", Content: []byte(test.content)},
 		}
-		if err := stepRecorder.NormalizeSucceededForStep(test.stepID, "lane", "normalize-module", nil, artifact, nil); err != nil {
+		if err := stepRecorder.NormalizeSucceededForStep(test.stepID, "lane", "normalize-module", nil, artifact); err != nil {
 			t.Fatalf("record %q: %v", test.stepID, err)
 		}
 	}
@@ -119,7 +119,7 @@ func TestStepAwareCheckpointPreservesDistinctDotIdentities(t *testing.T) {
 }
 
 func TestCheckpointSchemaCompatibilityIdentifiers(t *testing.T) {
-	if WorkspaceSchemaVersion != "notarius.workspace.v3" || WorkspaceSchemaVersionV2 != "notarius.workspace.v2" || WorkspaceSchemaVersionV1 != "notarius.workspace.v1" {
+	if WorkspaceSchemaVersion != "notarius.workspace.v4" || WorkspaceSchemaVersionV3 != "notarius.workspace.v3" || WorkspaceSchemaVersionV2 != "notarius.workspace.v2" || WorkspaceSchemaVersionV1 != "notarius.workspace.v1" {
 		t.Fatal("checkpoint schema identifiers are incorrect")
 	}
 }

internal/framework/chunkplan/store.go

@@ -14,6 +14,7 @@ import (
 	"syscall"
 
 	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
@@ -340,6 +341,9 @@ func validateRecord(record pipeline.ChunkPlanRecord, requestedDigest string) err
 	if record.SchemaVersion != SchemaVersion {
 		return fmt.Errorf("schema_version %q is not supported", record.SchemaVersion)
 	}
+	if err := contracts.ValidateProducerDiagnostics(record.Diagnostics); err != nil {
+		return fmt.Errorf("diagnostics: %w", err)
+	}
 	if _, err := digestPathSegment(requestedDigest); err != nil {
 		return err
 	}

internal/framework/chunkplan/store_test.go

@@ -281,7 +281,7 @@ func TestFilesystemStoreReportsInvalidRecordsAsRecoverable(t *testing.T) {
 			return bytes.Replace(data, []byte(`{"schema_version"`), []byte(`{"SENTINEL_UNKNOWN_FIELD":true,"schema_version"`), 1)
 		}},
 		{name: "truncated JSON", mutate: func(data []byte) []byte { return data[:len(data)/2] }},
-		{name: "legacy v1 record", mutate: replaceJSON(`notarius.chunk-plan.v2`, `notarius.chunk-plan.v1`)},
+		{name: "legacy v2 record", mutate: replaceJSON(`notarius.chunk-plan.v3`, `notarius.chunk-plan.v2`)},
 		{name: "source mismatch", mutate: replaceJSON(testSourceDigest, "sha256:"+strings.Repeat("b", 64))},
 		{name: "plan digest mismatch", mutate: func(data []byte) []byte {
 			prefix := []byte(`"plan_digest":"sha256:`)
@@ -498,7 +498,13 @@ func testRecord(t *testing.T, value int) pipeline.ChunkPlanRecord {
 			References: []artifacts.ReferenceProvenance{{Stage: "chunk", SlotName: "guide", OriginType: "file", OriginURI: "file:///guide.txt", Digest: "sha256:reference"}},
 			Metadata:   map[string]any{"prompt_id": "test/prompt", "enabled": true},
 		},
-		Warnings:  []contracts.Warning{{Scope: "chunk/test", ReasonCode: "observed", Message: "warning"}},
+		Diagnostics: []contracts.ProducerDiagnostic{{
+			Disposition:     contracts.DiagnosticDispositionWarning,
+			Category:        contracts.DiagnosticCategoryConfiguration,
+			ReasonCode:      "empty_reference",
+			OccurrenceCount: 1,
+			Samples:         []contracts.DiagnosticSample{{Scope: "reference", Message: "Reference was empty."}},
+		}},
 		CreatedAt: time.Date(2026, 7, 18, 12, 0, 0, 0, time.UTC),
 	}
 }

internal/framework/contracts/completion_request_debug.go

@@ -0,0 +1,84 @@
+package contracts
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+)
+
+// DebugStructuredCompletionRequest is the content-safe representation of a
+// structured completion request for ordinary diagnostics and summaries.
+// Detailed prompt material remains available only through the explicitly
+// requested LLM debug trace.
+type DebugStructuredCompletionRequest struct {
+	StageName                      string                   `json:"stage_name,omitempty"`
+	PromptID                       string                   `json:"prompt_id,omitempty"`
+	PromptVersion                  string                   `json:"prompt_version,omitempty"`
+	ProfileID                      string                   `json:"profile_id,omitempty"`
+	SessionID                      string                   `json:"session_id,omitempty"`
+	InputCount                     int                      `json:"input_count"`
+	VariableCount                  int                      `json:"variable_count"`
+	StructuredOutputRepairAttempts *int                     `json:"structured_output_repair_attempts,omitempty"`
+	Correction                     *DebugSemanticCorrection `json:"correction,omitempty"`
+}
+
+// DebugSemanticCorrection records only safe correction metadata. It never
+// exposes the assistant response or user guidance text.
+type DebugSemanticCorrection struct {
+	AssistantResponseBytes  int    `json:"assistant_response_bytes"`
+	AssistantResponseDigest string `json:"assistant_response_digest"`
+	UserGuidanceBytes       int    `json:"user_guidance_bytes"`
+	UserGuidanceDigest      string `json:"user_guidance_digest"`
+}
+
+// DebugSummary returns a content-safe representation suitable for ordinary
+// diagnostics. It does not validate or retain correction content.
+func (request StructuredCompletionRequest) DebugSummary() DebugStructuredCompletionRequest {
+	summary := DebugStructuredCompletionRequest{
+		StageName:     request.StageName,
+		PromptID:      request.PromptID,
+		PromptVersion: request.PromptVersion,
+		ProfileID:     request.ProfileID,
+		SessionID:     request.SessionID,
+		InputCount:    len(request.Inputs),
+		VariableCount: len(request.Vars),
+	}
+	if request.StructuredOutputRepairAttempts != nil {
+		attempts := *request.StructuredOutputRepairAttempts
+		summary.StructuredOutputRepairAttempts = &attempts
+	}
+	if request.Correction != nil {
+		summary.Correction = request.Correction.DebugSummary()
+	}
+	return summary
+}
+
+// DebugSummary returns content-safe correction metadata suitable for ordinary
+// diagnostics.
+func (correction *SemanticCorrection) DebugSummary() *DebugSemanticCorrection {
+	if correction == nil {
+		return nil
+	}
+	return &DebugSemanticCorrection{
+		AssistantResponseBytes:  len(correction.AssistantResponse),
+		AssistantResponseDigest: debugContentDigest(correction.AssistantResponse),
+		UserGuidanceBytes:       len(correction.UserGuidance),
+		UserGuidanceDigest:      debugContentDigest([]byte(correction.UserGuidance)),
+	}
+}
+
+// String prevents ordinary request formatting from exposing correction
+// content. Use the explicitly requested debug trace for complete messages.
+func (request StructuredCompletionRequest) String() string {
+	return fmt.Sprintf("%+v", request.DebugSummary())
+}
+
+// GoString gives %#v formatting the same content-safe behavior as String.
+func (request StructuredCompletionRequest) GoString() string {
+	return request.String()
+}
+
+func debugContentDigest(content []byte) string {
+	sum := sha256.Sum256(content)
+	return "sha256:" + hex.EncodeToString(sum[:])
+}

internal/framework/contracts/contracts.go

@@ -9,14 +9,15 @@ import (
 )
 
 type StructuredCompletionRequest struct {
-	StageName                      string         `json:"stage_name"`
-	PromptID                       string         `json:"prompt_id,omitempty"`
-	PromptVersion                  string         `json:"prompt_version,omitempty"`
-	ProfileID                      string         `json:"profile_id,omitempty"`
-	SessionID                      string         `json:"session_id,omitempty"`
-	Inputs                         LLMInputSet    `json:"inputs,omitempty"`
-	Vars                           map[string]any `json:"vars,omitempty"`
-	StructuredOutputRepairAttempts *int           `json:"structured_output_repair_attempts,omitempty"`
+	StageName                      string              `json:"stage_name"`
+	PromptID                       string              `json:"prompt_id,omitempty"`
+	PromptVersion                  string              `json:"prompt_version,omitempty"`
+	ProfileID                      string              `json:"profile_id,omitempty"`
+	SessionID                      string              `json:"session_id,omitempty"`
+	Inputs                         LLMInputSet         `json:"inputs,omitempty"`
+	Vars                           map[string]any      `json:"vars,omitempty"`
+	StructuredOutputRepairAttempts *int                `json:"structured_output_repair_attempts,omitempty"`
+	Correction                     *SemanticCorrection `json:"-"`
 }
 
 type StructuredCompletionResponse struct {
@@ -156,12 +157,14 @@ type ChunkRequest struct {
 	References                     ReferenceSet           `json:"references,omitempty"`
 	LLMProfile                     string                 `json:"llm_profile,omitempty"`
 	StructuredOutputRepairAttempts *int                   `json:"structured_output_repair_attempts,omitempty"`
+	Correction                     *SemanticCorrection    `json:"-"`
 	Metadata                       map[string]any         `json:"metadata,omitempty"`
 }
 
 type ChunkPlanResult struct {
-	Plan     source.ChunkPlan `json:"plan"`
-	Warnings []Warning        `json:"warnings,omitempty"`
+	Plan           source.ChunkPlan     `json:"plan"`
+	Diagnostics    []ProducerDiagnostic `json:"diagnostics,omitempty"`
+	ModelCandidate *ModelCandidate      `json:"-"`
 }
 
 type Chunker interface {
@@ -280,24 +283,19 @@ const (
 )
 
 type ValidationResult struct {
-	Approved               bool      `json:"approved"`
-	ReasonCode             string    `json:"reason_code,omitempty"`
-	Message                string    `json:"message,omitempty"`
-	DiagnosticArtifactPath string    `json:"diagnostic_artifact_path,omitempty"`
-	Warnings               []Warning `json:"warnings,omitempty"`
-}
-
-type Warning struct {
-	Scope      string `json:"scope,omitempty"`
-	ReasonCode string `json:"reason_code"`
-	Message    string `json:"message"`
+	Approved               bool                 `json:"approved"`
+	ReasonCode             string               `json:"reason_code,omitempty"`
+	Message                string               `json:"message,omitempty"`
+	CorrectionGuidance     string               `json:"-"`
+	DiagnosticArtifactPath string               `json:"diagnostic_artifact_path,omitempty"`
+	Diagnostics            []ProducerDiagnostic `json:"diagnostics,omitempty"`
 }
 
 type OutputRequest struct {
 	Manifest                       artifacts.RunManifest `json:"manifest"`
 	NormalizeOutputs               []SerializedOutput    `json:"normalize_outputs,omitempty"`
 	Rejected                       []RejectedOutput      `json:"rejected,omitempty"`
-	Warnings                       []Warning             `json:"warnings,omitempty"`
+	Diagnostics                    DiagnosticCollection  `json:"diagnostics,omitempty"`
 	LLMProfile                     string                `json:"llm_profile,omitempty"`
 	StructuredOutputRepairAttempts *int                  `json:"structured_output_repair_attempts,omitempty"`
 	Metadata                       map[string]any        `json:"metadata,omitempty"`
@@ -312,8 +310,7 @@ type OutputFile struct {
 }
 
 type OutputResult struct {
-	Files    []OutputFile `json:"files,omitempty"`
-	Warnings []Warning    `json:"warnings,omitempty"`
+	Files []OutputFile `json:"files,omitempty"`
 }
 
 type OutputEncoder interface {
@@ -322,17 +319,18 @@ type OutputEncoder interface {
 }
 
 type RejectedOutput struct {
-	Stage                  string `json:"stage"`
-	StepID                 string `json:"step_id,omitempty"`
-	LaneID                 string `json:"lane_id,omitempty"`
-	ModuleKey              string `json:"module_key,omitempty"`
-	ChunkID                string `json:"chunk_id,omitempty"`
-	ChunkIndex             int    `json:"chunk_index,omitempty"`
-	ValidatorName          string `json:"validator_name,omitempty"`
-	ReasonCode             string `json:"reason_code,omitempty"`
-	Message                string `json:"message"`
-	AttemptCount           int    `json:"attempt_count,omitempty"`
-	DiagnosticArtifactPath string `json:"diagnostic_artifact_path,omitempty"`
+	Stage                  string                       `json:"stage"`
+	StepID                 string                       `json:"step_id,omitempty"`
+	LaneID                 string                       `json:"lane_id,omitempty"`
+	ModuleKey              string                       `json:"module_key,omitempty"`
+	ChunkID                string                       `json:"chunk_id,omitempty"`
+	ChunkIndex             int                          `json:"chunk_index,omitempty"`
+	ValidatorName          string                       `json:"validator_name,omitempty"`
+	ReasonCode             string                       `json:"reason_code,omitempty"`
+	Message                string                       `json:"message"`
+	AttemptCount           int                          `json:"attempt_count,omitempty"`
+	DiagnosticArtifactPath string                       `json:"diagnostic_artifact_path,omitempty"`
+	Validation             *artifacts.ValidationSummary `json:"validation,omitempty"`
 }
 
 type ManifestMetadataProvider interface {

internal/framework/contracts/correction.go

@@ -0,0 +1,178 @@
+package contracts
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+	"unicode/utf8"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+)
+
+// CorrectionProtocol identifies how a producer can represent the model output
+// that directly controlled a candidate.
+type CorrectionProtocol string
+
+const (
+	CorrectionProtocolSingleResponseV1 CorrectionProtocol = "single_response_v1"
+)
+
+const (
+	MaxValidationReasonCodeBytes         = 128
+	MaxValidationCorrectionGuidanceBytes = 4 * 1024
+	MaxAssistantResponseBytes            = 1 << 20
+	MaxCorrectionGuidanceBytes           = 64 * 1024
+	MaxCorrectionContentBytes            = MaxAssistantResponseBytes + MaxCorrectionGuidanceBytes
+)
+
+func (protocol CorrectionProtocol) Validate() error {
+	if protocol == CorrectionProtocolSingleResponseV1 {
+		return nil
+	}
+	return fmt.Errorf("unsupported correction protocol %q", protocol)
+}
+
+// SemanticCorrection carries the latest model response and application-owned
+// guidance for one fresh corrected request. Its content is sensitive and is
+// deliberately excluded from ordinary JSON serialization.
+type SemanticCorrection struct {
+	AssistantResponse []byte `json:"-"`
+	UserGuidance      string `json:"-"`
+}
+
+func NewSemanticCorrection(assistantResponse []byte, userGuidance string) (*SemanticCorrection, error) {
+	correction := &SemanticCorrection{
+		AssistantResponse: append([]byte(nil), assistantResponse...),
+		UserGuidance:      userGuidance,
+	}
+	if err := correction.Validate(); err != nil {
+		return nil, err
+	}
+	return correction, nil
+}
+
+func CloneSemanticCorrection(correction *SemanticCorrection) (*SemanticCorrection, error) {
+	if correction == nil {
+		return nil, nil
+	}
+	return NewSemanticCorrection(correction.AssistantResponse, correction.UserGuidance)
+}
+
+// CloneStructuredCompletionRequest returns a request whose mutable values are
+// owned by the caller. It is suitable for clients that retain requests after
+// CompleteStructured returns.
+func CloneStructuredCompletionRequest(request StructuredCompletionRequest) (StructuredCompletionRequest, error) {
+	correction, err := CloneSemanticCorrection(request.Correction)
+	if err != nil {
+		return StructuredCompletionRequest{}, fmt.Errorf("clone correction: %w", err)
+	}
+	vars, err := source.CloneMetadata(request.Vars)
+	if err != nil {
+		return StructuredCompletionRequest{}, fmt.Errorf("clone variables: %w", err)
+	}
+	request.Inputs = request.Inputs.Clone()
+	request.Vars = vars
+	request.Correction = correction
+	if request.StructuredOutputRepairAttempts != nil {
+		attempts := *request.StructuredOutputRepairAttempts
+		request.StructuredOutputRepairAttempts = &attempts
+	}
+	return request, nil
+}
+
+func (correction SemanticCorrection) Validate() error {
+	if err := validateAssistantResponse(correction.AssistantResponse); err != nil {
+		return fmt.Errorf("semantic correction assistant response: %w", err)
+	}
+	if err := validateBoundedText(correction.UserGuidance, MaxCorrectionGuidanceBytes, "semantic correction user guidance", false); err != nil {
+		return err
+	}
+	if len(correction.AssistantResponse)+len(correction.UserGuidance) > MaxCorrectionContentBytes {
+		return errors.New("semantic correction content exceeds maximum length")
+	}
+	return nil
+}
+
+// ModelCandidate preserves the exact single model response that directly
+// controlled a producer result. It is attempt-local and never durable data.
+type ModelCandidate struct {
+	Response []byte             `json:"-"`
+	Protocol CorrectionProtocol `json:"-"`
+}
+
+func NewModelCandidate(response []byte, protocol CorrectionProtocol) (*ModelCandidate, error) {
+	candidate := &ModelCandidate{Response: append([]byte(nil), response...), Protocol: protocol}
+	if err := candidate.Validate(); err != nil {
+		return nil, err
+	}
+	return candidate, nil
+}
+
+func CloneModelCandidate(candidate *ModelCandidate) (*ModelCandidate, error) {
+	if candidate == nil {
+		return nil, nil
+	}
+	return NewModelCandidate(candidate.Response, candidate.Protocol)
+}
+
+func (candidate ModelCandidate) Validate() error {
+	if err := candidate.Protocol.Validate(); err != nil {
+		return err
+	}
+	if err := validateAssistantResponse(candidate.Response); err != nil {
+		return fmt.Errorf("model candidate response: %w", err)
+	}
+	return nil
+}
+
+func ValidateValidationResult(result ValidationResult) error {
+	if err := ValidateProducerDiagnostics(result.Diagnostics); err != nil {
+		return fmt.Errorf("validation diagnostics: %w", err)
+	}
+	if !result.Approved && result.ReasonCode == "" {
+		return errors.New("validation rejection reason code must not be empty")
+	}
+	if result.ReasonCode != "" {
+		if err := validateBoundedText(result.ReasonCode, MaxValidationReasonCodeBytes, "validation reason code", false); err != nil {
+			return err
+		}
+	}
+	if !result.Approved && result.CorrectionGuidance == "" {
+		return errors.New("validation rejection correction guidance must not be empty")
+	}
+	if result.CorrectionGuidance != "" {
+		if err := validateBoundedText(result.CorrectionGuidance, MaxValidationCorrectionGuidanceBytes, "validation correction guidance", false); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func validateAssistantResponse(response []byte) error {
+	if len(response) > MaxAssistantResponseBytes {
+		return errors.New("exceeds maximum length")
+	}
+	if !utf8.Valid(response) {
+		return errors.New("must be valid UTF-8")
+	}
+	if len(strings.TrimSpace(string(response))) == 0 {
+		return errors.New("must not be blank")
+	}
+	return nil
+}
+
+func validateBoundedText(value string, maximum int, name string, optional bool) error {
+	if value == "" && optional {
+		return nil
+	}
+	if !utf8.ValidString(value) {
+		return fmt.Errorf("%s must be valid UTF-8", name)
+	}
+	if strings.TrimSpace(value) == "" {
+		return fmt.Errorf("%s must not be blank", name)
+	}
+	if len(value) > maximum {
+		return fmt.Errorf("%s exceeds maximum length", name)
+	}
+	return nil
+}

internal/framework/contracts/correction_test.go

@@ -0,0 +1,187 @@
+package contracts
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"testing"
+)
+
+func TestSemanticCorrectionOwnsValidatedContent(t *testing.T) {
+	assistant := []byte(`{"items":["original"]}`)
+	correction, err := NewSemanticCorrection(assistant, "Return one corrected replacement.")
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
+	assistant[0] = '['
+	if got := string(correction.AssistantResponse); got != `{"items":["original"]}` {
+		t.Fatalf("assistant response = %q, want owned original content", got)
+	}
+
+	clone, err := CloneSemanticCorrection(correction)
+	if err != nil {
+		t.Fatalf("CloneSemanticCorrection() error = %v", err)
+	}
+	clone.AssistantResponse[0] = '['
+	if got := string(correction.AssistantResponse); got != `{"items":["original"]}` {
+		t.Fatalf("source correction changed through clone = %q", got)
+	}
+	if nilClone, err := CloneSemanticCorrection(nil); err != nil || nilClone != nil {
+		t.Fatalf("CloneSemanticCorrection(nil) = %#v, %v; want nil, nil", nilClone, err)
+	}
+
+	encoded, err := json.Marshal(correction)
+	if err != nil {
+		t.Fatalf("marshal correction: %v", err)
+	}
+	if string(encoded) != "{}" {
+		t.Fatalf("correction JSON = %s, want no sensitive content", encoded)
+	}
+}
+
+func TestCorrectionContractsRejectInvalidContent(t *testing.T) {
+	tooLongAssistant := bytes.Repeat([]byte("a"), MaxAssistantResponseBytes+1)
+	tooLongGuidance := strings.Repeat("a", MaxCorrectionGuidanceBytes+1)
+	tooLongReason := strings.Repeat("a", MaxValidationReasonCodeBytes+1)
+	tooLongValidationGuidance := strings.Repeat("a", MaxValidationCorrectionGuidanceBytes+1)
+
+	for _, test := range []struct {
+		name string
+		call func() error
+	}{
+		{"blank assistant", func() error { _, err := NewSemanticCorrection([]byte(" \n"), "guidance"); return err }},
+		{"invalid assistant utf8", func() error { _, err := NewSemanticCorrection([]byte{0xff}, "guidance"); return err }},
+		{"oversized assistant", func() error { _, err := NewSemanticCorrection(tooLongAssistant, "guidance"); return err }},
+		{"blank guidance", func() error { _, err := NewSemanticCorrection([]byte("response"), " \t"); return err }},
+		{"invalid guidance utf8", func() error { _, err := NewSemanticCorrection([]byte("response"), string([]byte{0xff})); return err }},
+		{"oversized guidance", func() error { _, err := NewSemanticCorrection([]byte("response"), tooLongGuidance); return err }},
+		{"unsupported protocol", func() error { _, err := NewModelCandidate([]byte("response"), "multiple_responses"); return err }},
+		{"missing candidate protocol", func() error { _, err := NewModelCandidate([]byte("response"), ""); return err }},
+		{"blank candidate response", func() error { _, err := NewModelCandidate([]byte(" "), CorrectionProtocolSingleResponseV1); return err }},
+		{"missing rejection reason code", func() error {
+			return ValidateValidationResult(ValidationResult{CorrectionGuidance: "Correct the response."})
+		}},
+		{"missing rejection guidance", func() error { return ValidateValidationResult(ValidationResult{ReasonCode: "invalid"}) }},
+		{"oversized reason code", func() error {
+			return ValidateValidationResult(ValidationResult{ReasonCode: tooLongReason, CorrectionGuidance: "Correct the response."})
+		}},
+		{"blank reason code", func() error {
+			return ValidateValidationResult(ValidationResult{ReasonCode: " \t", CorrectionGuidance: "Correct the response."})
+		}},
+		{"invalid correction guidance utf8", func() error {
+			return ValidateValidationResult(ValidationResult{ReasonCode: "invalid", CorrectionGuidance: string([]byte{0xff})})
+		}},
+		{"oversized correction guidance", func() error {
+			return ValidateValidationResult(ValidationResult{ReasonCode: "invalid", CorrectionGuidance: tooLongValidationGuidance})
+		}},
+		{"invalid diagnostics", func() error {
+			return ValidateValidationResult(ValidationResult{Approved: true, Diagnostics: []ProducerDiagnostic{{}}})
+		}},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			if err := test.call(); err == nil {
+				t.Fatal("validation error = nil, want error")
+			}
+		})
+	}
+}
+
+func TestModelCandidateOwnsValidatedResponse(t *testing.T) {
+	response := []byte(`{"items":["original"]}`)
+	candidate, err := NewModelCandidate(response, CorrectionProtocolSingleResponseV1)
+	if err != nil {
+		t.Fatalf("NewModelCandidate() error = %v", err)
+	}
+	response[0] = '['
+	if got := string(candidate.Response); got != `{"items":["original"]}` {
+		t.Fatalf("candidate response = %q, want owned original content", got)
+	}
+	clone, err := CloneModelCandidate(candidate)
+	if err != nil {
+		t.Fatalf("CloneModelCandidate() error = %v", err)
+	}
+	clone.Response[0] = '['
+	if got := string(candidate.Response); got != `{"items":["original"]}` {
+		t.Fatalf("source candidate changed through clone = %q", got)
+	}
+	if nilClone, err := CloneModelCandidate(nil); err != nil || nilClone != nil {
+		t.Fatalf("CloneModelCandidate(nil) = %#v, %v; want nil, nil", nilClone, err)
+	}
+}
+
+func TestValidationResultAllowsAbsentOptionalCorrectionFields(t *testing.T) {
+	if err := ValidateValidationResult(ValidationResult{Approved: true}); err != nil {
+		t.Fatalf("ValidateValidationResult() error = %v, want nil", err)
+	}
+	if err := ValidateValidationResult(ValidationResult{ReasonCode: "invalid-evidence", CorrectionGuidance: "Provide source-backed evidence."}); err != nil {
+		t.Fatalf("ValidateValidationResult() error = %v, want nil", err)
+	}
+	if err := CorrectionProtocol("").Validate(); err == nil {
+		t.Fatal("empty correction protocol validation error = nil, want error")
+	}
+}
+
+func TestCloneStructuredCompletionRequestOwnsCorrection(t *testing.T) {
+	correction, err := NewSemanticCorrection([]byte(`{"value":"original"}`), "Correct the value.")
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
+	attempts := 2
+	request := StructuredCompletionRequest{
+		Inputs:                         LLMInputSet{"source": NewLLMInputMaterial("source", "application/json", []byte(`{"source":true}`), "", "")},
+		Vars:                           map[string]any{"labels": []string{"original"}},
+		StructuredOutputRepairAttempts: &attempts,
+		Correction:                     correction,
+	}
+	clone, err := CloneStructuredCompletionRequest(request)
+	if err != nil {
+		t.Fatalf("CloneStructuredCompletionRequest() error = %v", err)
+	}
+	correction.AssistantResponse[0] = '['
+	request.Inputs["source"] = NewLLMInputMaterial("source", "application/json", []byte(`{"source":false}`), "", "")
+	*request.StructuredOutputRepairAttempts = 7
+	if got := string(clone.Correction.AssistantResponse); got != `{"value":"original"}` {
+		t.Fatalf("cloned correction response = %q, want owned original content", got)
+	}
+	if got := string(clone.Inputs["source"].Content); got != `{"source":true}` {
+		t.Fatalf("cloned input = %q, want owned original content", got)
+	}
+	if clone.StructuredOutputRepairAttempts == nil || *clone.StructuredOutputRepairAttempts != 2 {
+		t.Fatalf("cloned repair attempts = %v, want 2", clone.StructuredOutputRepairAttempts)
+	}
+}
+
+func TestStructuredCompletionRequestDebugSummaryOmitsCorrectionContent(t *testing.T) {
+	const assistantResponse = `{"secret":"assistant response"}`
+	const userGuidance = "secret user guidance"
+	correction, err := NewSemanticCorrection([]byte(assistantResponse), userGuidance)
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
+	request := StructuredCompletionRequest{
+		Inputs:     LLMInputSet{"source": NewLLMInputMaterial("source", "application/json", []byte(`{"source":true}`), "", "")},
+		Vars:       map[string]any{"custom": "value"},
+		Correction: correction,
+	}
+
+	summary := request.DebugSummary()
+	if summary.InputCount != 1 || summary.VariableCount != 1 || summary.Correction == nil {
+		t.Fatalf("debug summary = %#v, want input, variable, and correction metadata", summary)
+	}
+	if summary.Correction.AssistantResponseBytes != len(assistantResponse) || summary.Correction.UserGuidanceBytes != len(userGuidance) ||
+		summary.Correction.AssistantResponseDigest == "" || summary.Correction.UserGuidanceDigest == "" {
+		t.Fatalf("correction summary = %#v, want byte counts and digests", summary.Correction)
+	}
+	encoded, err := json.Marshal(summary)
+	if err != nil {
+		t.Fatalf("marshal debug summary: %v", err)
+	}
+	for _, rendered := range []string{string(encoded), fmt.Sprintf("%+v", request), fmt.Sprintf("%#v", request)} {
+		for _, secret := range []string{assistantResponse, userGuidance} {
+			if strings.Contains(rendered, secret) {
+				t.Fatalf("content-safe request rendering leaked %q: %s", secret, rendered)
+			}
+		}
+	}
+}

internal/framework/contracts/diagnostics.go

@@ -0,0 +1,401 @@
+package contracts
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+	"unicode/utf8"
+)
+
+const (
+	MaxDiagnosticReasonCodeBytes = 128
+	MaxDiagnosticScopeBytes      = 512
+	MaxDiagnosticMessageBytes    = 4 * 1024
+	MaxDiagnosticSamples         = 3
+	MaxProducerDiagnosticGroups  = 64
+)
+
+// DiagnosticDisposition identifies the operator significance of a producer
+// finding. Warnings are reserved for process-level degradation or incomplete
+// configured work.
+type DiagnosticDisposition string
+
+const (
+	DiagnosticDispositionWarning     DiagnosticDisposition = "warning"
+	DiagnosticDispositionAdvisory    DiagnosticDisposition = "advisory"
+	DiagnosticDispositionObservation DiagnosticDisposition = "observation"
+)
+
+// DiagnosticCategory gives a stable, bounded classification for a producer
+// finding.
+type DiagnosticCategory string
+
+const (
+	DiagnosticCategoryConfiguration        DiagnosticCategory = "configuration"
+	DiagnosticCategoryDegradation          DiagnosticCategory = "degradation"
+	DiagnosticCategoryValidationIncomplete DiagnosticCategory = "validation_incomplete"
+	DiagnosticCategoryFallback             DiagnosticCategory = "fallback"
+	DiagnosticCategoryDataQuality          DiagnosticCategory = "data_quality"
+	DiagnosticCategoryNormalization        DiagnosticCategory = "normalization"
+)
+
+// DiagnosticOriginStage identifies the framework operation that promoted a
+// diagnostic. It is framework-owned rather than producer-owned.
+type DiagnosticOriginStage string
+
+const (
+	DiagnosticOriginStageReferences DiagnosticOriginStage = "references"
+	DiagnosticOriginStageChunk      DiagnosticOriginStage = "chunk"
+	DiagnosticOriginStageExtract    DiagnosticOriginStage = "extract"
+	DiagnosticOriginStageMerge      DiagnosticOriginStage = "merge"
+	DiagnosticOriginStageNormalize  DiagnosticOriginStage = "normalize"
+)
+
+// DiagnosticSample is a bounded, safe example of a diagnostic occurrence.
+// Chunk identity is attached by the framework when it promotes a producer
+// diagnostic into a final group.
+type DiagnosticSample struct {
+	Scope      string `json:"scope"`
+	Message    string `json:"message"`
+	ChunkID    string `json:"chunk_id,omitempty"`
+	ChunkIndex *int   `json:"chunk_index,omitempty"`
+}
+
+// ProducerDiagnostic is the locally grouped form returned by one producer or
+// validator. It intentionally has no pipeline origin.
+type ProducerDiagnostic struct {
+	Disposition        DiagnosticDisposition `json:"disposition"`
+	Category           DiagnosticCategory    `json:"category"`
+	ReasonCode         string                `json:"reason_code"`
+	OccurrenceCount    int                   `json:"occurrence_count"`
+	Samples            []DiagnosticSample    `json:"samples"`
+	OmittedSampleCount int                   `json:"omitted_sample_count"`
+}
+
+// DiagnosticOrigin is framework-owned context used to distinguish findings
+// from different pipeline locations during final aggregation.
+type DiagnosticOrigin struct {
+	Stage        DiagnosticOriginStage `json:"stage"`
+	StepID       string                `json:"step_id,omitempty"`
+	LaneID       string                `json:"lane_id,omitempty"`
+	ModuleKey    string                `json:"module_key,omitempty"`
+	ValidatorKey string                `json:"validator_key,omitempty"`
+}
+
+// DiagnosticGroup is a producer diagnostic after framework origin enrichment.
+type DiagnosticGroup struct {
+	Disposition        DiagnosticDisposition `json:"disposition"`
+	Category           DiagnosticCategory    `json:"category"`
+	ReasonCode         string                `json:"reason_code"`
+	Origin             DiagnosticOrigin      `json:"origin"`
+	OccurrenceCount    int                   `json:"occurrence_count"`
+	Samples            []DiagnosticSample    `json:"samples"`
+	OmittedSampleCount int                   `json:"omitted_sample_count"`
+}
+
+// DiagnosticCollection is the grouped collection supplied to later durable
+// and presentation boundaries. Global aggregation policy is applied by the
+// framework before it reaches those boundaries.
+type DiagnosticCollection struct {
+	Groups                       []DiagnosticGroup `json:"groups"`
+	Truncated                    bool              `json:"truncated"`
+	UnrepresentedOccurrenceCount int               `json:"unrepresented_occurrence_count"`
+}
+
+// DiagnosticProjection is the validated warning/non-warning view used by
+// durable and presentation boundaries. Occurrence totals are checked before
+// they leave the framework contract.
+type DiagnosticProjection struct {
+	Warnings                  []DiagnosticGroup
+	Diagnostics               []DiagnosticGroup
+	WarningOccurrenceCount    int
+	DiagnosticOccurrenceCount int
+}
+
+// Validate checks a producer-local diagnostic against the public safety and
+// classification contract.
+func (diagnostic ProducerDiagnostic) Validate() error {
+	return validateDiagnostic(
+		diagnostic.Disposition,
+		diagnostic.Category,
+		diagnostic.ReasonCode,
+		diagnostic.OccurrenceCount,
+		diagnostic.Samples,
+		diagnostic.OmittedSampleCount,
+		false,
+	)
+}
+
+// ValidateProducerDiagnostics validates the complete set returned by one
+// producer or validator result.
+func ValidateProducerDiagnostics(diagnostics []ProducerDiagnostic) error {
+	if len(diagnostics) > MaxProducerDiagnosticGroups {
+		return errors.New("producer diagnostics exceed maximum group count")
+	}
+	for index, diagnostic := range diagnostics {
+		if err := diagnostic.Validate(); err != nil {
+			return fmt.Errorf("producer diagnostic %d: %w", index, err)
+		}
+	}
+	return nil
+}
+
+// Validate checks framework-owned origin fields.
+func (origin DiagnosticOrigin) Validate() error {
+	switch origin.Stage {
+	case DiagnosticOriginStageReferences, DiagnosticOriginStageChunk, DiagnosticOriginStageExtract, DiagnosticOriginStageMerge, DiagnosticOriginStageNormalize:
+	default:
+		return errors.New("diagnostic origin stage is invalid")
+	}
+	for _, field := range []struct {
+		name  string
+		value string
+	}{
+		{name: "diagnostic origin step ID", value: origin.StepID},
+		{name: "diagnostic origin lane ID", value: origin.LaneID},
+		{name: "diagnostic origin module key", value: origin.ModuleKey},
+		{name: "diagnostic origin validator key", value: origin.ValidatorKey},
+	} {
+		if field.value != "" && (!utf8.ValidString(field.value) || strings.TrimSpace(field.value) == "") {
+			return fmt.Errorf("%s must be valid nonblank UTF-8 when present", field.name)
+		}
+	}
+	return nil
+}
+
+// Validate checks a final origin-enriched group.
+func (group DiagnosticGroup) Validate() error {
+	if err := group.Origin.Validate(); err != nil {
+		return err
+	}
+	return validateDiagnostic(
+		group.Disposition,
+		group.Category,
+		group.ReasonCode,
+		group.OccurrenceCount,
+		group.Samples,
+		group.OmittedSampleCount,
+		true,
+	)
+}
+
+// Validate checks the collection shape without imposing later global
+// aggregation limits.
+func (collection DiagnosticCollection) Validate() error {
+	if collection.UnrepresentedOccurrenceCount < 0 {
+		return errors.New("diagnostic collection unrepresented occurrence count must not be negative")
+	}
+	if !collection.Truncated && collection.UnrepresentedOccurrenceCount != 0 {
+		return errors.New("diagnostic collection has unrepresented occurrences without truncation")
+	}
+	seen := make(map[diagnosticGroupKey]struct{}, len(collection.Groups))
+	for index, group := range collection.Groups {
+		if err := group.Validate(); err != nil {
+			return fmt.Errorf("diagnostic group %d: %w", index, err)
+		}
+		key := diagnosticGroupKeyFromGroup(group)
+		if _, exists := seen[key]; exists {
+			return errors.New("diagnostic collection contains duplicate group identity")
+		}
+		seen[key] = struct{}{}
+	}
+	return nil
+}
+
+// ProjectDiagnosticCollection validates, partitions, and totals one finalized
+// collection. Unrepresented occurrences belong to the non-warning projection.
+func ProjectDiagnosticCollection(collection DiagnosticCollection) (DiagnosticProjection, error) {
+	if err := collection.Validate(); err != nil {
+		return DiagnosticProjection{}, err
+	}
+	projection := DiagnosticProjection{
+		Warnings:    make([]DiagnosticGroup, 0),
+		Diagnostics: make([]DiagnosticGroup, 0),
+	}
+	for _, group := range collection.Groups {
+		if group.Disposition == DiagnosticDispositionWarning {
+			count, err := addDiagnosticOccurrences(projection.WarningOccurrenceCount, group.OccurrenceCount)
+			if err != nil {
+				return DiagnosticProjection{}, fmt.Errorf("warning occurrences: %w", err)
+			}
+			projection.WarningOccurrenceCount = count
+			projection.Warnings = append(projection.Warnings, cloneDiagnosticGroup(group))
+			continue
+		}
+		count, err := addDiagnosticOccurrences(projection.DiagnosticOccurrenceCount, group.OccurrenceCount)
+		if err != nil {
+			return DiagnosticProjection{}, fmt.Errorf("diagnostic occurrences: %w", err)
+		}
+		projection.DiagnosticOccurrenceCount = count
+		projection.Diagnostics = append(projection.Diagnostics, cloneDiagnosticGroup(group))
+	}
+	count, err := addDiagnosticOccurrences(projection.DiagnosticOccurrenceCount, collection.UnrepresentedOccurrenceCount)
+	if err != nil {
+		return DiagnosticProjection{}, fmt.Errorf("diagnostic occurrences: %w", err)
+	}
+	projection.DiagnosticOccurrenceCount = count
+	return projection, nil
+}
+
+// CloneProducerDiagnostics returns independent diagnostic slice ownership.
+func CloneProducerDiagnostics(diagnostics []ProducerDiagnostic) []ProducerDiagnostic {
+	if len(diagnostics) == 0 {
+		return nil
+	}
+	cloned := make([]ProducerDiagnostic, len(diagnostics))
+	for index, diagnostic := range diagnostics {
+		cloned[index] = cloneProducerDiagnostic(diagnostic)
+	}
+	return cloned
+}
+
+// CloneDiagnosticCollection returns independent collection ownership.
+func CloneDiagnosticCollection(collection DiagnosticCollection) DiagnosticCollection {
+	groups := collection.Groups
+	collection.Groups = make([]DiagnosticGroup, len(groups))
+	for index, group := range groups {
+		collection.Groups[index] = cloneDiagnosticGroup(group)
+	}
+	return collection
+}
+
+func validateDiagnostic(disposition DiagnosticDisposition, category DiagnosticCategory, reasonCode string, occurrenceCount int, samples []DiagnosticSample, omittedSampleCount int, allowChunkContext bool) error {
+	if !diagnosticCategoryAllowed(disposition, category) {
+		return errors.New("diagnostic disposition and category combination is invalid")
+	}
+	if err := validateDiagnosticText(reasonCode, MaxDiagnosticReasonCodeBytes, "diagnostic reason code"); err != nil {
+		return err
+	}
+	if occurrenceCount <= 0 {
+		return errors.New("diagnostic occurrence count must be positive")
+	}
+	if len(samples) == 0 {
+		return errors.New("diagnostic samples must not be empty")
+	}
+	if len(samples) > MaxDiagnosticSamples {
+		return errors.New("diagnostic samples exceed maximum count")
+	}
+	seen := make(map[diagnosticSampleKey]struct{}, len(samples))
+	for index, sample := range samples {
+		if err := validateDiagnosticSample(sample, allowChunkContext); err != nil {
+			return fmt.Errorf("diagnostic sample %d: %w", index, err)
+		}
+		key := diagnosticSampleKeyFromSample(sample)
+		if _, exists := seen[key]; exists {
+			return errors.New("diagnostic samples must be distinct")
+		}
+		seen[key] = struct{}{}
+	}
+	if occurrenceCount < len(samples) {
+		return errors.New("diagnostic occurrence count is smaller than sample count")
+	}
+	if omittedSampleCount != occurrenceCount-len(samples) {
+		return errors.New("diagnostic omitted sample count is inconsistent")
+	}
+	return nil
+}
+
+func diagnosticCategoryAllowed(disposition DiagnosticDisposition, category DiagnosticCategory) bool {
+	switch disposition {
+	case DiagnosticDispositionWarning:
+		return category == DiagnosticCategoryConfiguration || category == DiagnosticCategoryDegradation || category == DiagnosticCategoryValidationIncomplete || category == DiagnosticCategoryFallback
+	case DiagnosticDispositionAdvisory:
+		return category == DiagnosticCategoryDataQuality
+	case DiagnosticDispositionObservation:
+		return category == DiagnosticCategoryNormalization
+	default:
+		return false
+	}
+}
+
+func validateDiagnosticSample(sample DiagnosticSample, allowChunkContext bool) error {
+	if err := validateDiagnosticText(sample.Scope, MaxDiagnosticScopeBytes, "diagnostic sample scope"); err != nil {
+		return err
+	}
+	if err := validateDiagnosticText(sample.Message, MaxDiagnosticMessageBytes, "diagnostic sample message"); err != nil {
+		return err
+	}
+	if !allowChunkContext && (sample.ChunkID != "" || sample.ChunkIndex != nil) {
+		return errors.New("producer diagnostic sample must not include framework chunk context")
+	}
+	if sample.ChunkID != "" && (!utf8.ValidString(sample.ChunkID) || strings.TrimSpace(sample.ChunkID) == "") {
+		return errors.New("diagnostic sample chunk ID must be valid nonblank UTF-8 when present")
+	}
+	if sample.ChunkIndex != nil && *sample.ChunkIndex < 0 {
+		return errors.New("diagnostic sample chunk index must not be negative")
+	}
+	return nil
+}
+
+func validateDiagnosticText(value string, maximum int, name string) error {
+	if !utf8.ValidString(value) {
+		return fmt.Errorf("%s must be valid UTF-8", name)
+	}
+	if strings.TrimSpace(value) == "" {
+		return fmt.Errorf("%s must not be blank", name)
+	}
+	if len(value) > maximum {
+		return fmt.Errorf("%s exceeds maximum length", name)
+	}
+	return nil
+}
+
+func cloneProducerDiagnostic(diagnostic ProducerDiagnostic) ProducerDiagnostic {
+	diagnostic.Samples = cloneDiagnosticSamples(diagnostic.Samples)
+	return diagnostic
+}
+
+func cloneDiagnosticGroup(group DiagnosticGroup) DiagnosticGroup {
+	group.Samples = cloneDiagnosticSamples(group.Samples)
+	return group
+}
+
+func addDiagnosticOccurrences(current, incoming int) (int, error) {
+	if incoming > int(^uint(0)>>1)-current {
+		return 0, errors.New("occurrence count overflow")
+	}
+	return current + incoming, nil
+}
+
+func cloneDiagnosticSamples(samples []DiagnosticSample) []DiagnosticSample {
+	if len(samples) == 0 {
+		return nil
+	}
+	cloned := make([]DiagnosticSample, len(samples))
+	for index, sample := range samples {
+		if sample.ChunkIndex != nil {
+			chunkIndex := *sample.ChunkIndex
+			sample.ChunkIndex = &chunkIndex
+		}
+		cloned[index] = sample
+	}
+	return cloned
+}
+
+type diagnosticSampleKey struct {
+	scope         string
+	message       string
+	chunkID       string
+	chunkIndex    int
+	hasChunkIndex bool
+}
+
+func diagnosticSampleKeyFromSample(sample DiagnosticSample) diagnosticSampleKey {
+	key := diagnosticSampleKey{scope: sample.Scope, message: sample.Message, chunkID: sample.ChunkID}
+	if sample.ChunkIndex != nil {
+		key.chunkIndex = *sample.ChunkIndex
+		key.hasChunkIndex = true
+	}
+	return key
+}
+
+type diagnosticGroupKey struct {
+	disposition DiagnosticDisposition
+	category    DiagnosticCategory
+	reasonCode  string
+	origin      DiagnosticOrigin
+}
+
+func diagnosticGroupKeyFromGroup(group DiagnosticGroup) diagnosticGroupKey {
+	return diagnosticGroupKey{disposition: group.Disposition, category: group.Category, reasonCode: group.ReasonCode, origin: group.Origin}
+}

internal/framework/contracts/diagnostics_test.go

@@ -0,0 +1,198 @@
+package contracts
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestProducerDiagnosticValidationAcceptsClassificationMatrix(t *testing.T) {
+	for _, test := range []struct {
+		name        string
+		disposition DiagnosticDisposition
+		category    DiagnosticCategory
+	}{
+		{name: "configuration warning", disposition: DiagnosticDispositionWarning, category: DiagnosticCategoryConfiguration},
+		{name: "degradation warning", disposition: DiagnosticDispositionWarning, category: DiagnosticCategoryDegradation},
+		{name: "incomplete validation warning", disposition: DiagnosticDispositionWarning, category: DiagnosticCategoryValidationIncomplete},
+		{name: "fallback warning", disposition: DiagnosticDispositionWarning, category: DiagnosticCategoryFallback},
+		{name: "quality advisory", disposition: DiagnosticDispositionAdvisory, category: DiagnosticCategoryDataQuality},
+		{name: "normalization observation", disposition: DiagnosticDispositionObservation, category: DiagnosticCategoryNormalization},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			diagnostic := validProducerDiagnostic()
+			diagnostic.Disposition = test.disposition
+			diagnostic.Category = test.category
+			if err := diagnostic.Validate(); err != nil {
+				t.Fatalf("Validate() error = %v", err)
+			}
+		})
+	}
+}
+
+func TestProducerDiagnosticValidationRejectsInvalidFieldsAndCounts(t *testing.T) {
+	tooLongReason := strings.Repeat("r", MaxDiagnosticReasonCodeBytes+1)
+	tooLongScope := strings.Repeat("s", MaxDiagnosticScopeBytes+1)
+	tooLongMessage := strings.Repeat("m", MaxDiagnosticMessageBytes+1)
+	for _, test := range []struct {
+		name   string
+		mutate func(*ProducerDiagnostic)
+	}{
+		{name: "invalid classification", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Category = DiagnosticCategoryDataQuality }},
+		{name: "blank reason", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.ReasonCode = " \t" }},
+		{name: "invalid reason UTF-8", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.ReasonCode = string([]byte{0xff}) }},
+		{name: "oversized reason", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.ReasonCode = tooLongReason }},
+		{name: "blank scope", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Scope = "\n" }},
+		{name: "invalid scope UTF-8", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Scope = string([]byte{0xff}) }},
+		{name: "oversized scope", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Scope = tooLongScope }},
+		{name: "blank message", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Message = " " }},
+		{name: "invalid message UTF-8", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Message = string([]byte{0xff}) }},
+		{name: "oversized message", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.Samples[0].Message = tooLongMessage }},
+		{name: "zero occurrences", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.OccurrenceCount = 0 }},
+		{name: "missing samples", mutate: func(diagnostic *ProducerDiagnostic) {
+			diagnostic.Samples = nil
+			diagnostic.OmittedSampleCount = diagnostic.OccurrenceCount
+		}},
+		{name: "too many samples", mutate: func(diagnostic *ProducerDiagnostic) {
+			diagnostic.OccurrenceCount = 4
+			diagnostic.Samples = []DiagnosticSample{{Scope: "one", Message: "one"}, {Scope: "two", Message: "two"}, {Scope: "three", Message: "three"}, {Scope: "four", Message: "four"}}
+			diagnostic.OmittedSampleCount = 0
+		}},
+		{name: "duplicate samples", mutate: func(diagnostic *ProducerDiagnostic) {
+			diagnostic.OccurrenceCount = 2
+			diagnostic.Samples = []DiagnosticSample{{Scope: "scope", Message: "message"}, {Scope: "scope", Message: "message"}}
+			diagnostic.OmittedSampleCount = 0
+		}},
+		{name: "inconsistent omission", mutate: func(diagnostic *ProducerDiagnostic) { diagnostic.OmittedSampleCount = 1 }},
+		{name: "producer chunk context", mutate: func(diagnostic *ProducerDiagnostic) { chunkIndex := 0; diagnostic.Samples[0].ChunkIndex = &chunkIndex }},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			diagnostic := validProducerDiagnostic()
+			test.mutate(&diagnostic)
+			if err := diagnostic.Validate(); err == nil {
+				t.Fatal("Validate() error = nil, want invalid diagnostic error")
+			}
+		})
+	}
+}
+
+func TestValidateProducerDiagnosticsEnforcesLocalGroupBound(t *testing.T) {
+	diagnostics := make([]ProducerDiagnostic, MaxProducerDiagnosticGroups)
+	for index := range diagnostics {
+		diagnostics[index] = validProducerDiagnostic()
+		diagnostics[index].ReasonCode = "reason-" + string(rune('a'+index))
+	}
+	if err := ValidateProducerDiagnostics(diagnostics); err != nil {
+		t.Fatalf("ValidateProducerDiagnostics() error = %v", err)
+	}
+	diagnostics = append(diagnostics, validProducerDiagnostic())
+	if err := ValidateProducerDiagnostics(diagnostics); err == nil {
+		t.Fatal("ValidateProducerDiagnostics() error = nil, want excessive-group error")
+	}
+}
+
+func TestDiagnosticGroupValidationPreservesChunkIndexZero(t *testing.T) {
+	chunkIndex := 0
+	group := DiagnosticGroup{
+		Disposition:     DiagnosticDispositionAdvisory,
+		Category:        DiagnosticCategoryDataQuality,
+		ReasonCode:      "unresolved",
+		Origin:          DiagnosticOrigin{Stage: DiagnosticOriginStageExtract, StepID: "extract", LaneID: "spells", ModuleKey: "dnd/spells", ValidatorKey: "dnd/spells/source-relatedness"},
+		OccurrenceCount: 1,
+		Samples:         []DiagnosticSample{{Scope: "spells[0]", Message: "Spell was not found", ChunkID: "chunk-1", ChunkIndex: &chunkIndex}},
+	}
+	if err := group.Validate(); err != nil {
+		t.Fatalf("Validate() error = %v", err)
+	}
+	collection := DiagnosticCollection{Groups: []DiagnosticGroup{group}}
+	if err := collection.Validate(); err != nil {
+		t.Fatalf("DiagnosticCollection.Validate() error = %v", err)
+	}
+}
+
+func TestDiagnosticGroupRejectsRepeatedSampleWithEqualChunkIndex(t *testing.T) {
+	firstIndex := 0
+	secondIndex := 0
+	group := DiagnosticGroup{
+		Disposition:     DiagnosticDispositionAdvisory,
+		Category:        DiagnosticCategoryDataQuality,
+		ReasonCode:      "unresolved",
+		Origin:          DiagnosticOrigin{Stage: DiagnosticOriginStageExtract, StepID: "extract", LaneID: "spells", ModuleKey: "dnd/spells"},
+		OccurrenceCount: 2,
+		Samples: []DiagnosticSample{
+			{Scope: "spells[0]", Message: "Spell was not found", ChunkID: "chunk-1", ChunkIndex: &firstIndex},
+			{Scope: "spells[0]", Message: "Spell was not found", ChunkID: "chunk-1", ChunkIndex: &secondIndex},
+		},
+	}
+	if err := group.Validate(); err == nil {
+		t.Fatal("Validate() error = nil, want duplicate sample error")
+	}
+}
+
+func TestCloneDiagnosticCollectionOwnsGroupsAndChunkIndex(t *testing.T) {
+	chunkIndex := 0
+	collection := DiagnosticCollection{Groups: []DiagnosticGroup{{
+		Disposition:     DiagnosticDispositionAdvisory,
+		Category:        DiagnosticCategoryDataQuality,
+		ReasonCode:      "unresolved",
+		Origin:          DiagnosticOrigin{Stage: DiagnosticOriginStageExtract, StepID: "extract", LaneID: "spells", ModuleKey: "dnd/spells"},
+		OccurrenceCount: 1,
+		Samples:         []DiagnosticSample{{Scope: "spells[0]", Message: "Spell was not found", ChunkIndex: &chunkIndex}},
+	}}}
+	cloned := CloneDiagnosticCollection(collection)
+	collection.Groups[0].Samples[0].Message = "changed"
+	*collection.Groups[0].Samples[0].ChunkIndex = 1
+	if got := cloned.Groups[0].Samples[0]; got.Message != "Spell was not found" || got.ChunkIndex == nil || *got.ChunkIndex != 0 {
+		t.Fatalf("cloned sample = %#v, want independently owned original", got)
+	}
+}
+
+func TestProjectDiagnosticCollectionPartitionsAndChecksTotals(t *testing.T) {
+	warning := validDiagnosticGroup(DiagnosticDispositionWarning, DiagnosticCategoryFallback, "fallback", 2)
+	diagnostic := validDiagnosticGroup(DiagnosticDispositionAdvisory, DiagnosticCategoryDataQuality, "quality", 3)
+	collection := DiagnosticCollection{
+		Groups:                       []DiagnosticGroup{warning, diagnostic},
+		Truncated:                    true,
+		UnrepresentedOccurrenceCount: 4,
+	}
+	projection, err := ProjectDiagnosticCollection(collection)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(projection.Warnings) != 1 || len(projection.Diagnostics) != 1 || projection.WarningOccurrenceCount != 2 || projection.DiagnosticOccurrenceCount != 7 {
+		t.Fatalf("projection = %#v, want partitioned exact totals", projection)
+	}
+	collection.Groups[0].Samples[0].Message = "mutated"
+	if projection.Warnings[0].Samples[0].Message != "message" {
+		t.Fatal("projection retained caller-owned sample storage")
+	}
+
+	overflow := DiagnosticCollection{Groups: []DiagnosticGroup{
+		validDiagnosticGroup(DiagnosticDispositionWarning, DiagnosticCategoryFallback, "first", int(^uint(0)>>1)),
+		validDiagnosticGroup(DiagnosticDispositionWarning, DiagnosticCategoryFallback, "second", 1),
+	}}
+	if _, err := ProjectDiagnosticCollection(overflow); err == nil {
+		t.Fatal("ProjectDiagnosticCollection() overflow error = nil")
+	}
+}
+
+func validDiagnosticGroup(disposition DiagnosticDisposition, category DiagnosticCategory, reason string, occurrences int) DiagnosticGroup {
+	return DiagnosticGroup{
+		Disposition:        disposition,
+		Category:           category,
+		ReasonCode:         reason,
+		Origin:             DiagnosticOrigin{Stage: DiagnosticOriginStageNormalize, StepID: "step", LaneID: "lane", ModuleKey: "module"},
+		OccurrenceCount:    occurrences,
+		Samples:            []DiagnosticSample{{Scope: "scope", Message: "message"}},
+		OmittedSampleCount: occurrences - 1,
+	}
+}
+
+func validProducerDiagnostic() ProducerDiagnostic {
+	return ProducerDiagnostic{
+		Disposition:     DiagnosticDispositionWarning,
+		Category:        DiagnosticCategoryConfiguration,
+		ReasonCode:      "empty_reference",
+		OccurrenceCount: 1,
+		Samples:         []DiagnosticSample{{Scope: "references.glossary", Message: "Reference is empty"}},
+	}
+}

internal/framework/contracts/typed_pipeline.go

@@ -42,12 +42,14 @@ type TypedExtractionRequest struct {
 	References                     ReferenceSet
 	LLMProfile                     string
 	StructuredOutputRepairAttempts *int
+	Correction                     *SemanticCorrection
 	Metadata                       map[string]any
 }
 
 type TypedExtractionResult[T any] struct {
-	Value    T
-	Warnings []Warning
+	Value          T
+	Diagnostics    []ProducerDiagnostic
+	ModelCandidate *ModelCandidate
 }
 
 type Extractor[T any] interface {
@@ -65,12 +67,14 @@ type TypedMergeRequest[T any] struct {
 	References                     ReferenceSet
 	LLMProfile                     string
 	StructuredOutputRepairAttempts *int
+	Correction                     *SemanticCorrection
 	Metadata                       map[string]any
 }
 
 type TypedMergeResult[T any] struct {
-	Value    T
-	Warnings []Warning
+	Value          T
+	Diagnostics    []ProducerDiagnostic
+	ModelCandidate *ModelCandidate
 }
 
 type Merger[T any] interface {
@@ -87,28 +91,32 @@ type TypedNormalizeRequest[T any] struct {
 	References                     ReferenceSet
 	LLMProfile                     string
 	StructuredOutputRepairAttempts *int
+	Correction                     *SemanticCorrection
 	Metadata                       map[string]any
 }
 
 type TypedNormalizeResult[T any] struct {
-	Value    T
-	Warnings []Warning
-	Retry    *NormalizeRetry
+	Value          T
+	Diagnostics    []ProducerDiagnostic
+	Retry          *NormalizeRetry
+	ModelCandidate *ModelCandidate
 }
 
-// Normalize retry diagnostic limits bound module-provided values before the
-// framework persists them in debug artifacts.
+// Normalize retry limits bound module-provided control and diagnostic text
+// before the framework consumes or records it.
 const (
-	MaxNormalizeRetryReasonCodeBytes = 128
-	MaxNormalizeRetryMessageBytes    = 4096
+	MaxNormalizeRetryReasonCodeBytes         = 128
+	MaxNormalizeRetryMessageBytes            = 4096
+	MaxNormalizeRetryCorrectionGuidanceBytes = 4096
 )
 
 // NormalizeRetry asks the framework to retry normalization while retaining a
 // safe candidate for acceptance if the retry budget is exhausted.
 type NormalizeRetry struct {
-	ReasonCode       string
-	Message          string
-	FallbackWarnings []Warning
+	ReasonCode          string
+	Message             string
+	CorrectionGuidance  string
+	FallbackDiagnostics []ProducerDiagnostic
 }
 
 type Normalizer[T any] interface {

internal/framework/diagnostics/aggregator.go

@@ -0,0 +1,164 @@
+package diagnostics
+
+import (
+	"errors"
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+const (
+	MaxWarningGroups    = 128
+	MaxNonWarningGroups = 256
+)
+
+// Aggregator merges origin-enriched diagnostics in caller-supplied canonical
+// order. Its zero value is ready for use.
+type Aggregator struct {
+	groups                   []contracts.DiagnosticGroup
+	indices                  map[groupKey]int
+	warningGroups            int
+	nonWarningGroups         int
+	warningOccurrences       int
+	nonWarningOccurrences    int
+	unrepresentedOccurrences int
+}
+
+// Add validates and incorporates one final diagnostic group. Actionable
+// warnings cannot overflow; later non-warning groups are represented by exact
+// unrepresented-occurrence metadata once their fixed bound is reached.
+func (aggregator *Aggregator) Add(group contracts.DiagnosticGroup) error {
+	if err := group.Validate(); err != nil {
+		return fmt.Errorf("diagnostic group: %w", err)
+	}
+	if aggregator.indices == nil {
+		aggregator.indices = make(map[groupKey]int)
+	}
+	key := groupKeyFromGroup(group)
+	if index, exists := aggregator.indices[key]; exists {
+		if err := aggregator.checkOccurrenceTotal(group.Disposition, group.OccurrenceCount); err != nil {
+			return err
+		}
+		if err := aggregator.merge(index, group); err != nil {
+			return err
+		}
+		aggregator.addOccurrenceTotal(group.Disposition, group.OccurrenceCount)
+		return nil
+	}
+	if group.Disposition == contracts.DiagnosticDispositionWarning {
+		if aggregator.warningGroups >= MaxWarningGroups {
+			return errors.New("diagnostic warning groups exceed maximum count")
+		}
+	} else if aggregator.nonWarningGroups >= MaxNonWarningGroups {
+		if err := aggregator.checkOccurrenceTotal(group.Disposition, group.OccurrenceCount); err != nil {
+			return err
+		}
+		aggregator.addOccurrenceTotal(group.Disposition, group.OccurrenceCount)
+		return aggregator.addUnrepresented(group.OccurrenceCount)
+	}
+	if err := aggregator.checkOccurrenceTotal(group.Disposition, group.OccurrenceCount); err != nil {
+		return err
+	}
+	if group.Disposition == contracts.DiagnosticDispositionWarning {
+		aggregator.warningGroups++
+	} else {
+		aggregator.nonWarningGroups++
+	}
+	aggregator.addOccurrenceTotal(group.Disposition, group.OccurrenceCount)
+	aggregator.indices[key] = len(aggregator.groups)
+	aggregator.groups = append(aggregator.groups, contracts.CloneDiagnosticCollection(contracts.DiagnosticCollection{Groups: []contracts.DiagnosticGroup{group}}).Groups[0])
+	return nil
+}
+
+func (aggregator *Aggregator) checkOccurrenceTotal(disposition contracts.DiagnosticDisposition, count int) error {
+	current := aggregator.nonWarningOccurrences
+	if disposition == contracts.DiagnosticDispositionWarning {
+		current = aggregator.warningOccurrences
+	}
+	if count > maximumInt()-current {
+		return errors.New("diagnostic occurrence count overflow")
+	}
+	return nil
+}
+
+func (aggregator *Aggregator) addOccurrenceTotal(disposition contracts.DiagnosticDisposition, count int) {
+	if disposition == contracts.DiagnosticDispositionWarning {
+		aggregator.warningOccurrences += count
+		return
+	}
+	aggregator.nonWarningOccurrences += count
+}
+
+// Collection returns an independently owned grouped result in first-occurrence
+// order.
+func (aggregator *Aggregator) Collection() contracts.DiagnosticCollection {
+	if aggregator == nil {
+		return contracts.DiagnosticCollection{}
+	}
+	return contracts.CloneDiagnosticCollection(contracts.DiagnosticCollection{
+		Groups:                       aggregator.groups,
+		Truncated:                    aggregator.unrepresentedOccurrences > 0,
+		UnrepresentedOccurrenceCount: aggregator.unrepresentedOccurrences,
+	})
+}
+
+func (aggregator *Aggregator) merge(index int, incoming contracts.DiagnosticGroup) error {
+	current := &aggregator.groups[index]
+	if incoming.OccurrenceCount > maximumInt()-current.OccurrenceCount {
+		return errors.New("diagnostic occurrence count overflow")
+	}
+	current.OccurrenceCount += incoming.OccurrenceCount
+	for _, sample := range incoming.Samples {
+		if len(current.Samples) == contracts.MaxDiagnosticSamples || containsGroupSample(current.Samples, sample) {
+			continue
+		}
+		current.Samples = append(current.Samples, cloneGroupSample(sample))
+	}
+	current.OmittedSampleCount = current.OccurrenceCount - len(current.Samples)
+	return nil
+}
+
+func (aggregator *Aggregator) addUnrepresented(count int) error {
+	if count > maximumInt()-aggregator.unrepresentedOccurrences {
+		return errors.New("diagnostic unrepresented occurrence count overflow")
+	}
+	aggregator.unrepresentedOccurrences += count
+	return nil
+}
+
+func containsGroupSample(samples []contracts.DiagnosticSample, candidate contracts.DiagnosticSample) bool {
+	for _, sample := range samples {
+		if sample.Scope != candidate.Scope || sample.Message != candidate.Message || sample.ChunkID != candidate.ChunkID {
+			continue
+		}
+		if sample.ChunkIndex == nil || candidate.ChunkIndex == nil {
+			if sample.ChunkIndex == candidate.ChunkIndex {
+				return true
+			}
+			continue
+		}
+		if *sample.ChunkIndex == *candidate.ChunkIndex {
+			return true
+		}
+	}
+	return false
+}
+
+func cloneGroupSample(sample contracts.DiagnosticSample) contracts.DiagnosticSample {
+	if sample.ChunkIndex != nil {
+		value := *sample.ChunkIndex
+		sample.ChunkIndex = &value
+	}
+	return sample
+}
+
+type groupKey struct {
+	disposition contracts.DiagnosticDisposition
+	category    contracts.DiagnosticCategory
+	reasonCode  string
+	origin      contracts.DiagnosticOrigin
+}
+
+func groupKeyFromGroup(group contracts.DiagnosticGroup) groupKey {
+	return groupKey{disposition: group.Disposition, category: group.Category, reasonCode: group.ReasonCode, origin: group.Origin}
+}

internal/framework/diagnostics/aggregator_test.go

@@ -0,0 +1,89 @@
+package diagnostics
+
+import (
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+func TestAggregatorMergesByOriginAndPreservesFirstOccurrenceOrder(t *testing.T) {
+	aggregator := Aggregator{}
+	for _, group := range []contracts.DiagnosticGroup{
+		groupForAggregation("first", "message one", contracts.DiagnosticDispositionAdvisory, contracts.DiagnosticCategoryDataQuality, "dnd/spells"),
+		groupForAggregation("second", "message two", contracts.DiagnosticDispositionWarning, contracts.DiagnosticCategoryFallback, "dnd/items"),
+		groupForAggregation("third", "message three", contracts.DiagnosticDispositionAdvisory, contracts.DiagnosticCategoryDataQuality, "dnd/spells"),
+	} {
+		if err := aggregator.Add(group); err != nil {
+			t.Fatalf("Add() error = %v", err)
+		}
+	}
+	collection := aggregator.Collection()
+	if len(collection.Groups) != 2 {
+		t.Fatalf("group count = %d, want 2", len(collection.Groups))
+	}
+	if got := []string{collection.Groups[0].Origin.ModuleKey, collection.Groups[1].Origin.ModuleKey}; !equalStrings(got, []string{"dnd/spells", "dnd/items"}) {
+		t.Fatalf("group order = %#v, want first occurrence order", got)
+	}
+	if collection.Groups[0].OccurrenceCount != 2 || collection.Groups[0].OmittedSampleCount != 0 {
+		t.Fatalf("merged group = %#v, want two represented occurrences", collection.Groups[0])
+	}
+	if got := []string{collection.Groups[0].Samples[0].Scope, collection.Groups[0].Samples[1].Scope}; !equalStrings(got, []string{"first", "third"}) {
+		t.Fatalf("merged samples = %#v, want first occurrence order", got)
+	}
+}
+
+func TestAggregatorEnforcesWarningBoundAndTruncatesOnlyNonWarnings(t *testing.T) {
+	warnings := Aggregator{}
+	for index := 0; index < MaxWarningGroups; index++ {
+		group := groupForAggregation("scope", "message", contracts.DiagnosticDispositionWarning, contracts.DiagnosticCategoryFallback, "module")
+		group.ReasonCode = "warning-" + string(rune('a'+index))
+		if err := warnings.Add(group); err != nil {
+			t.Fatalf("warning Add(%d) error = %v", index, err)
+		}
+	}
+	if err := warnings.Add(groupForAggregation("overflow", "overflow", contracts.DiagnosticDispositionWarning, contracts.DiagnosticCategoryFallback, "overflow")); err == nil {
+		t.Fatal("warning overflow error = nil, want error")
+	}
+
+	nonWarnings := Aggregator{}
+	for index := 0; index < MaxNonWarningGroups+3; index++ {
+		group := groupForAggregation("scope", "message", contracts.DiagnosticDispositionAdvisory, contracts.DiagnosticCategoryDataQuality, "module")
+		group.ReasonCode = "advisory-" + string(rune('a'+index))
+		if err := nonWarnings.Add(group); err != nil {
+			t.Fatalf("non-warning Add(%d) error = %v", index, err)
+		}
+	}
+	collection := nonWarnings.Collection()
+	if len(collection.Groups) != MaxNonWarningGroups || !collection.Truncated || collection.UnrepresentedOccurrenceCount != 3 {
+		t.Fatalf("collection = %#v, want bounded non-warning groups and three unrepresented occurrences", collection)
+	}
+}
+
+func TestAggregatorRejectsOccurrenceTotalOverflow(t *testing.T) {
+	aggregator := Aggregator{}
+	first := groupForAggregation("first", "first", contracts.DiagnosticDispositionAdvisory, contracts.DiagnosticCategoryDataQuality, "first")
+	first.OccurrenceCount = maximumInt()
+	first.OmittedSampleCount = maximumInt() - 1
+	if err := aggregator.Add(first); err != nil {
+		t.Fatal(err)
+	}
+	second := groupForAggregation("second", "second", contracts.DiagnosticDispositionAdvisory, contracts.DiagnosticCategoryDataQuality, "second")
+	if err := aggregator.Add(second); err == nil {
+		t.Fatal("Add() occurrence total overflow error = nil")
+	}
+	collection := aggregator.Collection()
+	if len(collection.Groups) != 1 || collection.Groups[0].OccurrenceCount != maximumInt() {
+		t.Fatalf("collection changed after rejected overflow = %#v", collection)
+	}
+}
+
+func groupForAggregation(scope string, message string, disposition contracts.DiagnosticDisposition, category contracts.DiagnosticCategory, module string) contracts.DiagnosticGroup {
+	return contracts.DiagnosticGroup{
+		Disposition:     disposition,
+		Category:        category,
+		ReasonCode:      "source_unrelated",
+		Origin:          contracts.DiagnosticOrigin{Stage: contracts.DiagnosticOriginStageExtract, StepID: "extract", LaneID: "spells", ModuleKey: module},
+		OccurrenceCount: 1,
+		Samples:         []contracts.DiagnosticSample{{Scope: scope, Message: message}},
+	}
+}

internal/framework/diagnostics/collector.go

@@ -0,0 +1,93 @@
+// Package diagnostics provides bounded local grouping for producer and
+// validator diagnostic results.
+package diagnostics
+
+import (
+	"errors"
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+// Collector merges local producer diagnostics by their semantic identity. Its
+// zero value is ready for use.
+type Collector struct {
+	diagnostics []contracts.ProducerDiagnostic
+	indices     map[key]int
+}
+
+// NewCollector returns an empty local diagnostic collector.
+func NewCollector() *Collector {
+	return &Collector{}
+}
+
+// Add validates and merges one producer diagnostic. Every occurrence remains
+// counted, while the first three distinct samples in input order are retained.
+func (collector *Collector) Add(diagnostic contracts.ProducerDiagnostic) error {
+	if err := diagnostic.Validate(); err != nil {
+		return fmt.Errorf("producer diagnostic: %w", err)
+	}
+	if collector.indices == nil {
+		collector.indices = make(map[key]int)
+	}
+	diagnosticKey := key{disposition: diagnostic.Disposition, category: diagnostic.Category, reasonCode: diagnostic.ReasonCode}
+	index, exists := collector.indices[diagnosticKey]
+	if !exists {
+		if len(collector.diagnostics) >= contracts.MaxProducerDiagnosticGroups {
+			return errors.New("producer diagnostics exceed maximum group count")
+		}
+		collector.indices[diagnosticKey] = len(collector.diagnostics)
+		collector.diagnostics = append(collector.diagnostics, contracts.CloneProducerDiagnostics([]contracts.ProducerDiagnostic{diagnostic})[0])
+		return nil
+	}
+
+	current := &collector.diagnostics[index]
+	if diagnostic.OccurrenceCount > maximumInt()-current.OccurrenceCount {
+		return errors.New("producer diagnostic occurrence count overflow")
+	}
+	current.OccurrenceCount += diagnostic.OccurrenceCount
+	for _, sample := range diagnostic.Samples {
+		if len(current.Samples) == contracts.MaxDiagnosticSamples || containsSample(current.Samples, sample) {
+			continue
+		}
+		current.Samples = append(current.Samples, cloneSample(sample))
+	}
+	current.OmittedSampleCount = current.OccurrenceCount - len(current.Samples)
+	return nil
+}
+
+// Diagnostics returns an independently owned snapshot in first-occurrence
+// order.
+func (collector *Collector) Diagnostics() []contracts.ProducerDiagnostic {
+	if collector == nil {
+		return nil
+	}
+	return contracts.CloneProducerDiagnostics(collector.diagnostics)
+}
+
+func containsSample(samples []contracts.DiagnosticSample, candidate contracts.DiagnosticSample) bool {
+	for _, sample := range samples {
+		if sample.Scope == candidate.Scope && sample.Message == candidate.Message {
+			return true
+		}
+	}
+	return false
+}
+
+func cloneSample(sample contracts.DiagnosticSample) contracts.DiagnosticSample {
+	if sample.ChunkIndex != nil {
+		chunkIndex := *sample.ChunkIndex
+		sample.ChunkIndex = &chunkIndex
+	}
+	return sample
+}
+
+func maximumInt() int {
+	return int(^uint(0) >> 1)
+}
+
+type key struct {
+	disposition contracts.DiagnosticDisposition
+	category    contracts.DiagnosticCategory
+	reasonCode  string
+}

internal/framework/diagnostics/collector_test.go

@@ -0,0 +1,103 @@
+package diagnostics
+
+import (
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+func TestCollectorCountsOccurrencesAndRetainsDistinctSamplesInOrder(t *testing.T) {
+	collector := NewCollector()
+	for _, diagnostic := range []contracts.ProducerDiagnostic{
+		advisory("one", "first"),
+		advisory("one", "first"),
+		advisory("two", "second"),
+		advisory("three", "third"),
+		advisory("four", "fourth"),
+	} {
+		if err := collector.Add(diagnostic); err != nil {
+			t.Fatalf("Add() error = %v", err)
+		}
+	}
+
+	diagnostics := collector.Diagnostics()
+	if len(diagnostics) != 1 {
+		t.Fatalf("group count = %d, want 1", len(diagnostics))
+	}
+	group := diagnostics[0]
+	if group.OccurrenceCount != 5 || group.OmittedSampleCount != 2 {
+		t.Fatalf("group counts = %#v, want five occurrences and two omitted samples", group)
+	}
+	if got := []string{group.Samples[0].Scope, group.Samples[1].Scope, group.Samples[2].Scope}; !equalStrings(got, []string{"one", "two", "three"}) {
+		t.Fatalf("sample order = %#v, want first three distinct samples", got)
+	}
+}
+
+func TestCollectorSeparatesGroupsAndRejectsInvalidOrExcessiveGroups(t *testing.T) {
+	collector := NewCollector()
+	if err := collector.Add(advisory("one", "first")); err != nil {
+		t.Fatalf("Add() error = %v", err)
+	}
+	warning := advisory("two", "second")
+	warning.Disposition = contracts.DiagnosticDispositionWarning
+	warning.Category = contracts.DiagnosticCategoryFallback
+	if err := collector.Add(warning); err != nil {
+		t.Fatalf("Add() error = %v", err)
+	}
+	if got := len(collector.Diagnostics()); got != 2 {
+		t.Fatalf("group count = %d, want 2", got)
+	}
+
+	invalid := advisory("bad", "bad")
+	invalid.ReasonCode = ""
+	if err := collector.Add(invalid); err == nil {
+		t.Fatal("Add() error = nil, want invalid diagnostic error")
+	}
+
+	limited := NewCollector()
+	for index := 0; index < contracts.MaxProducerDiagnosticGroups; index++ {
+		diagnostic := advisory("scope", "message")
+		diagnostic.ReasonCode = "reason-" + string(rune('a'+index))
+		if err := limited.Add(diagnostic); err != nil {
+			t.Fatalf("Add(%d) error = %v", index, err)
+		}
+	}
+	if err := limited.Add(advisory("overflow", "overflow")); err == nil {
+		t.Fatal("Add() error = nil, want local group limit error")
+	}
+}
+
+func TestCollectorReturnsIndependentSnapshots(t *testing.T) {
+	collector := NewCollector()
+	if err := collector.Add(advisory("scope", "message")); err != nil {
+		t.Fatalf("Add() error = %v", err)
+	}
+	first := collector.Diagnostics()
+	first[0].Samples[0].Message = "changed"
+	second := collector.Diagnostics()
+	if second[0].Samples[0].Message != "message" {
+		t.Fatalf("collector snapshot changed = %#v", second)
+	}
+}
+
+func advisory(scope string, message string) contracts.ProducerDiagnostic {
+	return contracts.ProducerDiagnostic{
+		Disposition:     contracts.DiagnosticDispositionAdvisory,
+		Category:        contracts.DiagnosticCategoryDataQuality,
+		ReasonCode:      "source_unrelated",
+		OccurrenceCount: 1,
+		Samples:         []contracts.DiagnosticSample{{Scope: scope, Message: message}},
+	}
+}
+
+func equalStrings(left []string, right []string) bool {
+	if len(left) != len(right) {
+		return false
+	}
+	for index := range left {
+		if left[index] != right[index] {
+			return false
+		}
+	}
+	return true
+}

internal/framework/llm/promptkit_client.go

@@ -119,6 +119,10 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 	if req.StructuredOutputRepairAttempts != nil && (*req.StructuredOutputRepairAttempts < 0 || *req.StructuredOutputRepairAttempts > 3) {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("structured output repair attempts must be between zero and three")
 	}
+	appendedMessages, err := promptKitCorrectionMessages(req.Correction)
+	if err != nil {
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("structured completion correction: %w", err)
+	}
 	promptID := strings.TrimSpace(req.PromptID)
 	if promptID == "" {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("structured completion prompt_id must not be empty")
@@ -133,13 +137,14 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 	}
 
 	runReq := promptkit.RunRequest{
-		PromptID:      promptID,
-		PromptVersion: strings.TrimSpace(req.PromptVersion),
-		ProfileID:     strings.TrimSpace(req.ProfileID),
-		SessionID:     sessionID,
-		Inputs:        promptKitInputs(req.Inputs),
-		Vars:          promptKitVars(req, sessionID),
-		Execution:     execution,
+		PromptID:         promptID,
+		PromptVersion:    strings.TrimSpace(req.PromptVersion),
+		ProfileID:        strings.TrimSpace(req.ProfileID),
+		SessionID:        sessionID,
+		Inputs:           promptKitInputs(req.Inputs),
+		Vars:             promptKitVars(req, sessionID),
+		Execution:        execution,
+		AppendedMessages: appendedMessages,
 	}
 	if req.StructuredOutputRepairAttempts != nil {
 		inspection, err := c.engine.InspectPrompt(ctx, promptID, strings.TrimSpace(req.PromptVersion))
@@ -213,6 +218,20 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 	return response, nil
 }
 
+func promptKitCorrectionMessages(correction *contracts.SemanticCorrection) ([]promptkit.RenderedMessage, error) {
+	owned, err := contracts.CloneSemanticCorrection(correction)
+	if err != nil {
+		return nil, err
+	}
+	if owned == nil {
+		return nil, nil
+	}
+	return []promptkit.RenderedMessage{
+		{Role: promptkit.RoleAssistant, Content: string(owned.AssistantResponse)},
+		{Role: promptkit.RoleUser, Content: owned.UserGuidance},
+	}, nil
+}
+
 func promptKitDebugGenerationError(prepared *promptkit.PreparedRun, generationErr *promptkit.GenerationError) *contracts.LLMDebugResponse {
 	if generationErr == nil {
 		return nil

internal/framework/llm/promptkit_client_test.go

@@ -1,6 +1,7 @@
 package llm
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
@@ -116,6 +117,74 @@ func TestPromptKitClientMapsPromptRequestAndUnmarshalsOutput(t *testing.T) {
 	}
 }
 
+func TestPromptKitClientAppendsSemanticCorrectionAfterRenderedPrompt(t *testing.T) {
+	fake := &fakePromptKitLLM{content: `{"ok":true}`}
+	client := newTestPromptKitClient(t, fake)
+	request := contracts.StructuredCompletionRequest{
+		PromptID:  "adapter.direct-session",
+		ProfileID: "explicit-profile",
+		SessionID: "correction-session",
+		Inputs: contracts.LLMInputSet{
+			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
+		},
+		Vars: map[string]any{"custom": "value"},
+	}
+
+	var ordinary map[string]any
+	if _, err := client.CompleteStructured(context.Background(), request, &ordinary); err != nil {
+		t.Fatalf("ordinary CompleteStructured() error = %v", err)
+	}
+	ordinaryMessages := append([]promptkit.RenderedMessage(nil), fake.lastRequest().Prompt.Messages...)
+	if len(ordinaryMessages) != 1 {
+		t.Fatalf("ordinary rendered messages = %#v, want only the declared prompt message", ordinaryMessages)
+	}
+
+	correction, err := contracts.NewSemanticCorrection([]byte(`{"previous":"response"}`), "Return the corrected JSON object.")
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
+	request.Correction = correction
+	var corrected map[string]any
+	if _, err := client.CompleteStructured(context.Background(), request, &corrected); err != nil {
+		t.Fatalf("corrected CompleteStructured() error = %v", err)
+	}
+	correctedMessages := fake.lastRequest().Prompt.Messages
+	if len(correctedMessages) != len(ordinaryMessages)+2 {
+		t.Fatalf("corrected message count = %d, want %d", len(correctedMessages), len(ordinaryMessages)+2)
+	}
+	if !reflect.DeepEqual(correctedMessages[:len(ordinaryMessages)], ordinaryMessages) {
+		t.Fatalf("ordinary rendered prefix changed: got %#v, want %#v", correctedMessages[:len(ordinaryMessages)], ordinaryMessages)
+	}
+	if got, want := correctedMessages[len(ordinaryMessages)], (promptkit.RenderedMessage{Role: promptkit.RoleAssistant, Content: `{"previous":"response"}`}); got != want {
+		t.Fatalf("assistant correction message = %#v, want %#v", got, want)
+	}
+	if got, want := correctedMessages[len(ordinaryMessages)+1], (promptkit.RenderedMessage{Role: promptkit.RoleUser, Content: "Return the corrected JSON object."}); got != want {
+		t.Fatalf("user correction message = %#v, want %#v", got, want)
+	}
+}
+
+func TestPromptKitClientRejectsInvalidCorrectionsBeforePromptPreparation(t *testing.T) {
+	const sensitiveResponse = "assistant-response-must-not-appear-in-errors"
+	for _, correction := range []*contracts.SemanticCorrection{
+		{AssistantResponse: []byte(sensitiveResponse), UserGuidance: " \t"},
+		{AssistantResponse: bytes.Repeat([]byte(sensitiveResponse), contracts.MaxAssistantResponseBytes/len(sensitiveResponse)+1), UserGuidance: "Use a smaller response."},
+	} {
+		fake := &fakePromptKitLLM{content: `{"ok":true}`}
+		client := newTestPromptKitClient(t, fake)
+		var out map[string]any
+		_, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{Correction: correction}, &out)
+		if err == nil || !strings.Contains(err.Error(), "correction") {
+			t.Fatalf("CompleteStructured() error = %v, want correction validation failure", err)
+		}
+		if strings.Contains(err.Error(), sensitiveResponse) {
+			t.Fatalf("correction validation error leaked response content: %v", err)
+		}
+		if got := atomic.LoadInt32(&fake.calls); got != 0 {
+			t.Fatalf("provider calls = %d, want no provider call after invalid correction", got)
+		}
+	}
+}
+
 func TestPromptKitClientUsesOnePreparedSnapshotForDebugAndGeneration(t *testing.T) {
 	const initialPrompt = `id: snapshot.test
 version: "v1"
@@ -960,6 +1029,10 @@ func TestPromptKitClientRepairsStructuredOutputAndReportsCumulativeUsage(t *test
 		{Content: `{"ok":true}`, Usage: promptkit.TokenUsage{PromptTokens: 7, CompletionTokens: 11, TotalTokens: 18}},
 	}}
 	client := newTestPromptKitClient(t, fake)
+	correction, err := contracts.NewSemanticCorrection([]byte(`{"bad":true}`), "Return the required ok field.")
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
 
 	var out struct {
 		OK bool `json:"ok"`
@@ -968,6 +1041,7 @@ func TestPromptKitClientRepairsStructuredOutputAndReportsCumulativeUsage(t *test
 		PromptID:                       "adapter.test",
 		StructuredOutputRepairAttempts: &attempts,
 		SessionID:                      "repair-test",
+		Correction:                     correction,
 		Inputs: contracts.LLMInputSet{
 			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
 		},
@@ -978,6 +1052,17 @@ func TestPromptKitClientRepairsStructuredOutputAndReportsCumulativeUsage(t *test
 	if got := atomic.LoadInt32(&fake.calls); got != 2 {
 		t.Fatalf("provider calls = %d, want initial generation and one repair", got)
 	}
+	requests := fake.requestsSnapshot()
+	if len(requests) != 2 || len(requests[0].Prompt.Messages) < 3 {
+		t.Fatalf("repair requests = %#v, want correction messages on the initial prepared request", requests)
+	}
+	messages := requests[0].Prompt.Messages
+	if got, want := messages[len(messages)-2], (promptkit.RenderedMessage{Role: promptkit.RoleAssistant, Content: `{"bad":true}`}); got != want {
+		t.Fatalf("repair assistant correction = %#v, want %#v", got, want)
+	}
+	if got, want := messages[len(messages)-1], (promptkit.RenderedMessage{Role: promptkit.RoleUser, Content: "Return the required ok field."}); got != want {
+		t.Fatalf("repair user correction = %#v, want %#v", got, want)
+	}
 	if response.RepairAttempts != 1 || response.PromptTokens != 10 || response.CompletionTokens != 16 || response.TotalTokens != 26 {
 		t.Fatalf("response repair and usage = %#v, want one repair and PromptKit cumulative usage", response)
 	}

internal/framework/llm/promptkit_profile_fingerprint.go

@@ -17,7 +17,7 @@ const (
 	promptKitLocalBackendMarker          = "notarius:promptkit-local-backend:v1"
 	// The built-in profile catalog is compiled into this pinned PromptKit
 	// release. Update this identity when the dependency is upgraded.
-	promptKitBuiltinProfileCatalogID = "promptkit:v0.8.0:builtin-profiles"
+	promptKitBuiltinProfileCatalogID = "promptkit:v0.9.0:builtin-profiles"
 )
 
 func promptKitProfileFingerprint(profileDir, profileFile, fallbackProfileDigest string) (CheckpointFingerprint, error) {

internal/framework/pipeline/checkpoint.go

@@ -28,14 +28,14 @@ type CheckpointRecorder interface {
 	SourceSucceeded(moduleKey string, doc *source.SourceDocument) error
 	SourceFailed(moduleKey string, err error) error
 	ExtractRunning(laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	ExtractSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput, warnings []contracts.Warning) error
+	ExtractSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput) error
 	ExtractFailed(laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 	MergeRunning(laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	MergeSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error
+	MergeSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact) error
 	MergeRejected(laneID string, moduleKey string, dependencies []CheckpointFingerprint, rejected contracts.RejectedOutput) error
 	MergeFailed(laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 	NormalizeRunning(laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	NormalizeSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error
+	NormalizeSucceeded(laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact) error
 	NormalizeRejected(laneID string, moduleKey string, dependencies []CheckpointFingerprint, rejected contracts.RejectedOutput) error
 	NormalizeFailed(laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 }
@@ -45,14 +45,14 @@ type CheckpointRecorder interface {
 // remain available for callers that do not have step context.
 type StepCheckpointRecorder interface {
 	ExtractRunningForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	ExtractSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput, warnings []contracts.Warning) error
+	ExtractSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput) error
 	ExtractFailedForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 	MergeRunningForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	MergeSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error
+	MergeSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact) error
 	MergeRejectedForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, rejected contracts.RejectedOutput) error
 	MergeFailedForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 	NormalizeRunningForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint) error
-	NormalizeSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error
+	NormalizeSucceededForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, output CheckpointArtifact) error
 	NormalizeRejectedForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, rejected contracts.RejectedOutput) error
 	NormalizeFailedForStep(stepID, laneID string, moduleKey string, dependencies []CheckpointFingerprint, err error) error
 }
@@ -89,6 +89,7 @@ const (
 	CheckpointReasonReused                      CheckpointReasonCode = "checkpoint_reused"
 	CheckpointReasonAcceptedArtifactReused      CheckpointReasonCode = "accepted_artifact_reused"
 	CheckpointReasonRecomputeStep               CheckpointReasonCode = "recompute_step"
+	CheckpointReasonValidationIncompleteLineage CheckpointReasonCode = "validation_incomplete_lineage"
 )
 
 type CheckpointDecision struct {
@@ -160,6 +161,8 @@ func checkpointDecisionDetail(reasonCode CheckpointReasonCode) string {
 		return "accepted normalized artifact is reusable"
 	case CheckpointReasonRecomputeStep:
 		return "selected step requires execution"
+	case CheckpointReasonValidationIncompleteLineage:
+		return "checkpoint reuse is disabled by validation-incomplete input lineage"
 	default:
 		return "checkpoint decision"
 	}
@@ -276,21 +279,27 @@ type CheckpointArtifact struct {
 	ChunkRef     source.SourceRef
 	Artifact     contracts.SerializedArtifact
 	SchemaDigest string
+	Diagnostics  []CheckpointDiagnostic
+}
+
+// CheckpointDiagnostic stores a producer-local diagnostic alongside a
+// reusable artifact. The runner supplies the current run's origin when it
+// promotes this value into a diagnostic group.
+type CheckpointDiagnostic struct {
+	Diagnostic   contracts.ProducerDiagnostic `json:"diagnostic"`
+	ValidatorKey string                       `json:"validator_key,omitempty"`
 }
 
 type ExtractCheckpoint struct {
 	Outputs  []CheckpointArtifact
 	Rejected []contracts.RejectedOutput
-	Warnings []contracts.Warning
 }
 
 type MergeCheckpoint struct {
-	Output   CheckpointArtifact
-	Warnings []contracts.Warning
+	Output CheckpointArtifact
 }
 type NormalizeCheckpoint struct {
-	Output   CheckpointArtifact
-	Warnings []contracts.Warning
+	Output CheckpointArtifact
 }
 
 type CheckpointLoader interface {
@@ -323,14 +332,14 @@ func (noopCheckpointRecorder) SourceFailed(string, error) error
 func (noopCheckpointRecorder) ExtractRunning(string, string, []CheckpointFingerprint) error {
 	return nil
 }
-func (noopCheckpointRecorder) ExtractSucceeded(string, string, []CheckpointFingerprint, []CheckpointArtifact, []contracts.RejectedOutput, []contracts.Warning) error {
+func (noopCheckpointRecorder) ExtractSucceeded(string, string, []CheckpointFingerprint, []CheckpointArtifact, []contracts.RejectedOutput) error {
 	return nil
 }
 func (noopCheckpointRecorder) ExtractFailed(string, string, []CheckpointFingerprint, error) error {
 	return nil
 }
 func (noopCheckpointRecorder) MergeRunning(string, string, []CheckpointFingerprint) error { return nil }
-func (noopCheckpointRecorder) MergeSucceeded(string, string, []CheckpointFingerprint, CheckpointArtifact, []contracts.Warning) error {
+func (noopCheckpointRecorder) MergeSucceeded(string, string, []CheckpointFingerprint, CheckpointArtifact) error {
 	return nil
 }
 func (noopCheckpointRecorder) MergeRejected(string, string, []CheckpointFingerprint, contracts.RejectedOutput) error {
@@ -342,7 +351,7 @@ func (noopCheckpointRecorder) MergeFailed(string, string, []CheckpointFingerprin
 func (noopCheckpointRecorder) NormalizeRunning(string, string, []CheckpointFingerprint) error {
 	return nil
 }
-func (noopCheckpointRecorder) NormalizeSucceeded(string, string, []CheckpointFingerprint, CheckpointArtifact, []contracts.Warning) error {
+func (noopCheckpointRecorder) NormalizeSucceeded(string, string, []CheckpointFingerprint, CheckpointArtifact) error {
 	return nil
 }
 func (noopCheckpointRecorder) NormalizeRejected(string, string, []CheckpointFingerprint, contracts.RejectedOutput) error {
@@ -375,11 +384,11 @@ func checkpointExtractRunning(recorder CheckpointRecorder, stepID, laneID, modul
 	}
 	return recorder.ExtractRunning(laneID, moduleKey, deps)
 }
-func checkpointExtractSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput, warnings []contracts.Warning) error {
+func checkpointExtractSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, outputs []CheckpointArtifact, rejected []contracts.RejectedOutput) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {
-		return stepAware.ExtractSucceededForStep(stepID, laneID, moduleKey, deps, outputs, rejected, warnings)
+		return stepAware.ExtractSucceededForStep(stepID, laneID, moduleKey, deps, outputs, rejected)
 	}
-	return recorder.ExtractSucceeded(laneID, moduleKey, deps, outputs, rejected, warnings)
+	return recorder.ExtractSucceeded(laneID, moduleKey, deps, outputs, rejected)
 }
 func checkpointExtractFailed(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, err error) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {
@@ -393,11 +402,11 @@ func checkpointMergeRunning(recorder CheckpointRecorder, stepID, laneID, moduleK
 	}
 	return recorder.MergeRunning(laneID, moduleKey, deps)
 }
-func checkpointMergeSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error {
+func checkpointMergeSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, output CheckpointArtifact) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {
-		return stepAware.MergeSucceededForStep(stepID, laneID, moduleKey, deps, output, warnings)
+		return stepAware.MergeSucceededForStep(stepID, laneID, moduleKey, deps, output)
 	}
-	return recorder.MergeSucceeded(laneID, moduleKey, deps, output, warnings)
+	return recorder.MergeSucceeded(laneID, moduleKey, deps, output)
 }
 func checkpointMergeRejected(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, rejected contracts.RejectedOutput) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {
@@ -417,11 +426,11 @@ func checkpointNormalizeRunning(recorder CheckpointRecorder, stepID, laneID, mod
 	}
 	return recorder.NormalizeRunning(laneID, moduleKey, deps)
 }
-func checkpointNormalizeSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, output CheckpointArtifact, warnings []contracts.Warning) error {
+func checkpointNormalizeSucceeded(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, output CheckpointArtifact) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {
-		return stepAware.NormalizeSucceededForStep(stepID, laneID, moduleKey, deps, output, warnings)
+		return stepAware.NormalizeSucceededForStep(stepID, laneID, moduleKey, deps, output)
 	}
-	return recorder.NormalizeSucceeded(laneID, moduleKey, deps, output, warnings)
+	return recorder.NormalizeSucceeded(laneID, moduleKey, deps, output)
 }
 func checkpointNormalizeRejected(recorder CheckpointRecorder, stepID, laneID, moduleKey string, deps []CheckpointFingerprint, rejected contracts.RejectedOutput) error {
 	if stepAware, ok := recorder.(StepCheckpointRecorder); ok {

internal/framework/pipeline/chunk_plan_store.go

@@ -8,7 +8,7 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 )
 
-const ChunkPlanSchemaVersion = "notarius.chunk-plan.v2"
+const ChunkPlanSchemaVersion = "notarius.chunk-plan.v3"
 
 type ChunkPlanProducer struct {
 	InputModule string                          `json:"input_module"`
@@ -19,13 +19,13 @@ type ChunkPlanProducer struct {
 }
 
 type ChunkPlanRecord struct {
-	SchemaVersion string              `json:"schema_version"`
-	SourceDigest  string              `json:"source_digest"`
-	PlanDigest    string              `json:"plan_digest"`
-	Plan          source.ChunkPlan    `json:"plan"`
-	Producer      ChunkPlanProducer   `json:"producer"`
-	Warnings      []contracts.Warning `json:"warnings,omitempty"`
-	CreatedAt     time.Time           `json:"created_at"`
+	SchemaVersion string                         `json:"schema_version"`
+	SourceDigest  string                         `json:"source_digest"`
+	PlanDigest    string                         `json:"plan_digest"`
+	Plan          source.ChunkPlan               `json:"plan"`
+	Producer      ChunkPlanProducer              `json:"producer"`
+	Diagnostics   []contracts.ProducerDiagnostic `json:"diagnostics,omitempty"`
+	CreatedAt     time.Time                      `json:"created_at"`
 }
 
 type ChunkPlanStore interface {

internal/framework/pipeline/debug.go

@@ -42,6 +42,7 @@ type debugTimedEnvelope struct {
 	LaneID      string                  `json:"lane_id,omitempty"`
 	ModuleKey   string                  `json:"module_key,omitempty"`
 	Attempt     int                     `json:"attempt,omitempty"`
+	AttemptKind string                  `json:"attempt_kind,omitempty"`
 	StartedAt   time.Time               `json:"started_at"`
 	CompletedAt time.Time               `json:"completed_at"`
 	DurationMS  int64                   `json:"duration_ms"`
@@ -51,11 +52,10 @@ type debugTimedEnvelope struct {
 }
 
 type debugBinaryEnvelope struct {
-	ContentBase64 string              `json:"content_base64,omitempty"`
-	ContentDigest string              `json:"content_digest,omitempty"`
-	MediaType     string              `json:"media_type,omitempty"`
-	Metadata      map[string]any      `json:"metadata,omitempty"`
-	Warnings      []contracts.Warning `json:"warnings,omitempty"`
+	ContentBase64 string         `json:"content_base64,omitempty"`
+	ContentDigest string         `json:"content_digest,omitempty"`
+	MediaType     string         `json:"media_type,omitempty"`
+	Metadata      map[string]any `json:"metadata,omitempty"`
 }
 
 type debugSourceInput struct {
@@ -110,24 +110,7 @@ type debugSerializedOutput struct {
 	Content       debugBinaryEnvelope      `json:"content"`
 }
 
-type debugLLMInputMaterial struct {
-	Name      string `json:"name"`
-	MediaType string `json:"media_type,omitempty"`
-	Content   string `json:"content_base64,omitempty"`
-	Digest    string `json:"digest,omitempty"`
-	OriginURI string `json:"origin_uri,omitempty"`
-	SizeBytes int64  `json:"size_bytes,omitempty"`
-}
-
-type debugStructuredCompletionRequest struct {
-	StageName     string                           `json:"stage_name"`
-	PromptID      string                           `json:"prompt_id,omitempty"`
-	PromptVersion string                           `json:"prompt_version,omitempty"`
-	ProfileID     string                           `json:"profile_id,omitempty"`
-	SessionID     string                           `json:"session_id,omitempty"`
-	Inputs        map[string]debugLLMInputMaterial `json:"inputs,omitempty"`
-	Vars          map[string]any                   `json:"vars,omitempty"`
-}
+type debugStructuredCompletionRequest = contracts.DebugStructuredCompletionRequest
 
 type debugStructuredCompletionResponse struct {
 	Content          string `json:"content,omitempty"`
@@ -160,9 +143,37 @@ type debugLLMCallReference struct {
 	PromptID            string `json:"prompt_id,omitempty"`
 	ProfileID           string `json:"profile_id,omitempty"`
 	Model               string `json:"model,omitempty"`
+	PromptTokens        int    `json:"prompt_tokens,omitempty"`
+	CompletionTokens    int    `json:"completion_tokens,omitempty"`
+	TotalTokens         int    `json:"total_tokens,omitempty"`
+	RepairAttempts      int    `json:"repair_attempts,omitempty"`
 	Error               bool   `json:"error,omitempty"`
 }
 
+type debugValidationOutcome struct {
+	ValidatorName string `json:"validator_name"`
+	Outcome       string `json:"outcome"`
+	AttemptCount  int    `json:"attempt_count"`
+	ReasonCode    string `json:"reason_code,omitempty"`
+}
+
+type debugProducerAttempt struct {
+	Number     int                      `json:"number"`
+	Kind       string                   `json:"kind"`
+	Outcome    string                   `json:"outcome"`
+	Validation []debugValidationOutcome `json:"validation,omitempty"`
+}
+
+type debugProducerTerminal struct {
+	ProducerAttemptCount int                         `json:"producer_attempt_count"`
+	Attempts             []debugProducerAttempt      `json:"attempts,omitempty"`
+	AggregateReasonCodes []string                    `json:"aggregate_reason_codes,omitempty"`
+	ValidationComplete   bool                        `json:"validation_complete"`
+	EffectivePolicy      ValidationPolicy            `json:"effective_policy"`
+	TerminalAction       string                      `json:"terminal_action"`
+	Summary              artifacts.ValidationSummary `json:"validation_summary"`
+}
+
 type debugValidationCall struct {
 	ValidatorName string                     `json:"validator_name"`
 	Request       any                        `json:"request"`
@@ -267,6 +278,10 @@ func (client *debugLLMClient) CompleteStructured(ctx context.Context, req contra
 		PromptID:            req.PromptID,
 		ProfileID:           debugFirstNonEmptyString(response.ProfileID, req.ProfileID),
 		Model:               debugFirstNonEmptyString(response.Model, debugResponseModel(response)),
+		PromptTokens:        response.PromptTokens,
+		CompletionTokens:    response.CompletionTokens,
+		TotalTokens:         response.TotalTokens,
+		RepairAttempts:      response.RepairAttempts,
 		Error:               err != nil,
 	}
 	if scope := debugLLMScopeFromContext(ctx); scope != nil {
@@ -423,14 +438,57 @@ func (r attemptTerminalRecorder) record(payload any, terminalErr error) error {
 	return terminalErr
 }
 
-func debugContentEnvelope(content []byte, mediaType string, metadata map[string]any, warnings []contracts.Warning) debugBinaryEnvelope {
+func debugValidationOutcomes(report validationReport) []debugValidationOutcome {
+	if len(report.records) == 0 {
+		return nil
+	}
+	outcomes := make([]debugValidationOutcome, 0, len(report.records))
+	for _, record := range report.records {
+		outcomes = append(outcomes, debugValidationOutcome{
+			ValidatorName: record.validatorName,
+			Outcome:       string(record.outcome),
+			AttemptCount:  record.attemptCount,
+			ReasonCode:    record.reasonCode,
+		})
+	}
+	return outcomes
+}
+
+func writeProducerTerminalDebug(recorder DebugRecorder, name string, terminal producerAttemptTerminal, policy ValidationPolicy, summary artifacts.ValidationSummary) error {
+	attempts := make([]debugProducerAttempt, 0, len(terminal.Provenance))
+	for _, item := range terminal.Provenance {
+		attempts = append(attempts, debugProducerAttempt{
+			Number:     item.Number,
+			Kind:       string(item.Kind),
+			Outcome:    string(item.Outcome),
+			Validation: debugValidationOutcomes(item.Validation),
+		})
+	}
+	return writeDebugTimed(recorder, name, debugTimedEnvelope{
+		Stage:     summary.Stage,
+		StepID:    summary.StepID,
+		LaneID:    summary.LaneID,
+		ModuleKey: summary.ModuleKey,
+		StartedAt: time.Now().UTC(),
+		Payload: debugProducerTerminal{
+			ProducerAttemptCount: summary.ProducerAttemptCount,
+			Attempts:             attempts,
+			AggregateReasonCodes: append([]string(nil), summary.ReasonCodes...),
+			ValidationComplete:   summary.Status == "complete",
+			EffectivePolicy:      policy,
+			TerminalAction:       string(terminal.Action),
+			Summary:              artifacts.CloneValidationSummary(summary),
+		},
+	})
+}
+
+func debugContentEnvelope(content []byte, mediaType string, metadata map[string]any, _ any) debugBinaryEnvelope {
 	content = redactSecretBytes(content)
 	return debugBinaryEnvelope{
 		ContentBase64: base64.StdEncoding.EncodeToString(content),
 		ContentDigest: debugContentDigest(content),
 		MediaType:     mediaType,
 		Metadata:      redactSensitiveMap(metadata),
-		Warnings:      cloneWarnings(warnings),
 	}
 }
 
@@ -567,29 +625,7 @@ func debugOutputFiles(files []contracts.OutputFile) []debugOutputFile {
 }
 
 func debugCompletionRequest(req contracts.StructuredCompletionRequest) debugStructuredCompletionRequest {
-	inputs := make(map[string]debugLLMInputMaterial, len(req.Inputs))
-	for key, material := range req.Inputs {
-		inputs[key] = debugLLMInputMaterial{
-			Name:      material.Name,
-			MediaType: material.MediaType,
-			Content:   base64.StdEncoding.EncodeToString(redactSecretBytes(material.Content)),
-			Digest:    material.Digest,
-			OriginURI: material.OriginURI,
-			SizeBytes: material.SizeBytes,
-		}
-	}
-	if len(inputs) == 0 {
-		inputs = nil
-	}
-	return debugStructuredCompletionRequest{
-		StageName:     req.StageName,
-		PromptID:      req.PromptID,
-		PromptVersion: req.PromptVersion,
-		ProfileID:     req.ProfileID,
-		SessionID:     req.SessionID,
-		Inputs:        inputs,
-		Vars:          redactSensitiveMap(req.Vars),
-	}
+	return req.DebugSummary()
 }
 
 func debugCompletionResponse(response contracts.StructuredCompletionResponse) debugStructuredCompletionResponse {
@@ -691,21 +727,11 @@ func debugResponseModel(response contracts.StructuredCompletionResponse) string
 
 func debugValidationResultEnvelope(result contracts.ValidationResult) contracts.ValidationResult {
 	result.Message = string(redactSecretBytes([]byte(result.Message)))
+	result.CorrectionGuidance = ""
 	result.DiagnosticArtifactPath = string(redactSecretBytes([]byte(result.DiagnosticArtifactPath)))
-	for i := range result.Warnings {
-		result.Warnings[i].Message = string(redactSecretBytes([]byte(result.Warnings[i].Message)))
-	}
 	return result
 }
 
-func debugWarningEnvelopes(warnings []contracts.Warning) []contracts.Warning {
-	out := cloneWarnings(warnings)
-	for i := range out {
-		out[i].Message = string(redactSecretBytes([]byte(out[i].Message)))
-	}
-	return out
-}
-
 func debugRejectedOutputEnvelope(rejected contracts.RejectedOutput) contracts.RejectedOutput {
 	rejected.Message = string(redactSecretBytes([]byte(rejected.Message)))
 	rejected.DiagnosticArtifactPath = string(redactSecretBytes([]byte(rejected.DiagnosticArtifactPath)))

internal/framework/pipeline/debug_test.go

@@ -47,6 +47,32 @@ func TestDebugLLMPathsKeepDotIdentitiesDistinct(t *testing.T) {
 	}
 }
 
+func TestDebugCompletionRequestOmitsCorrectionContent(t *testing.T) {
+	const assistantResponse = `{"secret":"assistant response"}`
+	const userGuidance = "secret user guidance"
+	correction, err := contracts.NewSemanticCorrection([]byte(assistantResponse), userGuidance)
+	if err != nil {
+		t.Fatalf("NewSemanticCorrection() error = %v", err)
+	}
+	summary := debugCompletionRequest(contracts.StructuredCompletionRequest{
+		Inputs:     contracts.LLMInputSet{"source": contracts.NewLLMInputMaterial("source", "application/json", []byte(`{"source":true}`), "", "")},
+		Vars:       map[string]any{"custom": "value"},
+		Correction: correction,
+	})
+	encoded, err := json.Marshal(summary)
+	if err != nil {
+		t.Fatalf("marshal completion summary: %v", err)
+	}
+	for _, secret := range []string{assistantResponse, userGuidance} {
+		if strings.Contains(string(encoded), secret) {
+			t.Fatalf("debug completion summary leaked %q: %s", secret, encoded)
+		}
+	}
+	if summary.InputCount != 1 || summary.VariableCount != 1 || summary.Correction == nil {
+		t.Fatalf("debug completion summary = %#v, want counts and correction metadata", summary)
+	}
+}
+
 func TestDebugSourceDocumentPreservesUnitReferences(t *testing.T) {
 	doc := validSourceDocument()
 	envelope := debugSourceDocumentEnvelope(doc)