Prepare version 0.8.0 release

.woodpecker/release.yml

@@ -1,3 +1,4 @@
+# Advisory validation after source publication, never a release-completion gate.
 when:
   - event: tag
 
@@ -5,29 +6,4 @@ steps:
   - name: validate-release
     image: golang:1.25.5
     commands:
-      - |
-        set -eu
-
-        version="$CI_COMMIT_TAG"
-        release_note="docs/releases/$version.md"
-
-        if ! printf '%s\n' "$version" | grep -E -x 'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)' >/dev/null; then
-          printf '%s\n' "invalid release tag: $version" >&2
-          exit 1
-        fi
-        if [ ! -s "$release_note" ]; then
-          printf '%s\n' "missing release note: $release_note" >&2
-          exit 1
-        fi
-        if ! grep -F -x "# Notarius $version" "$release_note" >/dev/null; then
-          printf '%s\n' "release note heading does not match $version" >&2
-          exit 1
-        fi
-        for heading in '## Summary' '## Compatibility' '## Upgrade' '## Changes'; do
-          if ! grep -F -x "$heading" "$release_note" >/dev/null; then
-            printf '%s\n' "release note is missing heading: $heading" >&2
-            exit 1
-          fi
-        done
-
-        ./scripts/check-release-source.sh "$version"
+      - ./scripts/check-release-source.sh "$CI_COMMIT_TAG"

AGENTS.md

@@ -1,2 +1,5 @@
 Please review `docs/development.md` for initial orientation in this repository
 and follow its task-specific reading guide.
+
+Review and follow `docs/policy/agents.md` before recommending, planning, or
+implementing any changes to this repository.

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/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/adr/0017-defer-cache-access-policy-to-deployment.md

@@ -0,0 +1,59 @@
+# ADR-0017: Defer cache access policy to the deployment
+
+**Status:** Accepted
+**Date:** 2026-09-11
+
+## Context
+
+Canonical chunk plans and invocation checkpoints are reconstructible cache
+state. They may include source-derived content, so access and retention matter,
+but Notarius cannot determine whether a deployment needs a private cache or a
+cache shared by a trusted group. The current application-enforced `0700` and
+`0600` modes prevent ordinary Unix provisioning from making that choice.
+
+[ADR-0005](0005-cache-canonical-chunk-plans-by-source.md) governs canonical
+chunk-plan reuse, while [ADR-0006](0006-separate-output-cache-and-debug-state.md)
+established the output, cache, and debug separation refined by
+[ADR-0007](0007-separate-checkpoint-recording-from-reuse.md). None of those
+decisions requires Notarius to own exact cache access modes.
+
+## Decision
+
+Deployments own cache access policy through provisioned directories, process
+umask, setgid inheritance, and default ACLs. Notarius requests conventional
+maximum creation modes for cache entries and leaves the kernel to select their
+effective permissions.
+
+Notarius continues to own confined paths, symlink rejection, validation,
+atomic publication, cleanup, and recovery. Those correctness boundaries are
+independent from deciding which accounts may access cache state.
+
+Source-derived cache contents can be sensitive. Operators own their access and
+retention policy, while API credentials remain outside cache payloads under the
+existing secret boundary. Existing cache inode modes are not migrated
+automatically.
+
+This decision does not supersede ADR-0005, ADR-0006, or ADR-0007.
+
+## Alternatives considered
+
+- Continue enforcing private `0700` directory and `0600` file modes. This
+  makes one deployment choice for every operator and prevents trusted-group
+  provisioning.
+- Add configurable Notarius permission modes or a private/group cache setting.
+  This preserves application ownership of a deployment concern and expands the
+  public configuration surface without a durable policy reason.
+- Recursively converge existing cache ownership, modes, or ACLs. This would
+  make upgrades mutate operator-managed state and cannot safely infer an
+  intended policy.
+
+## Consequences
+
+Operators can use a restrictive umask for private cache state or provision a
+setgid/default-ACL cache root for trusted-group access. They must deliberately
+recreate or administratively migrate pre-existing cache trees when their
+existing modes do not match the desired deployment policy.
+
+Notarius does not add user or group management, ACL mutation, permission
+configuration, or recursive cache migration. Output and debug permission
+contracts remain separate from this decision.

docs/adr/0018-adopt-configkit.md

@@ -0,0 +1,53 @@
+# ADR-0018: Adopt ConfigKit for configuration loading
+
+**Status:** Accepted
+**Date:** 2026-09-13
+
+## Context
+
+Notarius's former configuration boundary owned strict YAML parsing, input
+presence bookkeeping, defaults, normalization, environment overlays, and
+configuration inspection. Those general-purpose responsibilities had grown
+alongside application-specific pipeline and module semantics.
+
+ConfigKit v0.3.0 provides strict typed YAML loading, imports, configuration
+profiles, explicit CLI overrides, recorded defaults and normalization, safe
+inspection rendering, and released source-origin APIs. The application needs
+those source records to resolve filesystem paths according to the file that
+supplied a surviving declaration.
+
+## Decision
+
+Notarius will adopt the root public `gitea.maximumdirect.net/eric/ConfigKit`
+package at v0.3.0 for its version 5 configuration contract. ConfigKit owns
+source composition, strict decoding, input presence, metadata, and safe
+effective/explanation rendering. Notarius retains its typed document schema,
+field semantics, runtime conversion, catalog-dependent pipeline resolution,
+PromptKit profile handling, and physical run preparation.
+
+Version 5 is a deliberate breaking file-contract change. It uses canonical
+object module bindings and reference sources, and enables ConfigKit imports
+and configuration profiles. The selected configuration file's canonical
+directory is the base for its declared application paths; ConfigKit's released
+`Resolution.InputInfo` and `Resolution.Source` APIs provide their origins.
+
+`NOTARIUS_CONFIG` remains file discovery, while field-level operational
+environment overlays are retired. Explicit configuration-valued CLI flags are
+represented as ConfigKit overrides; invocation actions remain outside that
+boundary.
+
+## Alternatives considered
+
+- Continue extending the custom YAML loader. This preserves duplicated
+  general-purpose parsing, presence, composition, and inspection machinery.
+- Retain version 4 alongside version 5. This would leave a supported dual
+  configuration boundary and delay removal of the machinery being replaced.
+- Wrap ConfigKit in a compatibility pre-parser. This weakens ConfigKit's
+  strict contract and recreates the application-owned parser complexity.
+
+## Consequences
+
+The released application has one ConfigKit-backed loader and one version 5
+contract. Operator documentation and maintained examples move with the cutover.
+ConfigKit metadata is retained only at the configuration publication boundary;
+its types do not become module or runtime pipeline interfaces.

docs/cli.md

@@ -12,8 +12,8 @@ defined in [Operations](operations.md).
 notarius help
 notarius --version
 notarius run <pipeline-id> --input path/to/source.json [--json] [flags]
-notarius config validate [--config path/to/config.yml] [--pipeline pipeline-id] [--only lane-a,lane-b]
-notarius pipelines list [--config path/to/config.yml] [--json]
+notarius config validate [--config path/to/config.yml] [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
+notarius pipelines list [--config path/to/config.yml] [--config-overlay name] [--json]
 ~~~
 
 Running Notarius without arguments, or with **help**, **--help**, or **-h**,
@@ -42,6 +42,7 @@ pipeline ID and **--input** are required.
 | Flag | Meaning |
 | --- | --- |
 | **--config path** | Use this configuration file. When omitted, configuration discovery applies; see [Configuration](config.md). |
+| **--config-overlay name** | Select a ConfigKit configuration overlay. Empty and **default** select the composed base configuration. |
 | **--input path** | Source input file to process. Required. |
 | **--output-dir path** | Override the configured output root for this run. |
 | **--json** | Write the successful run-result receipt as JSON to standard output. |
@@ -55,8 +56,8 @@ 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 +82,62 @@ 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
@@ -125,10 +155,10 @@ OPENROUTER_API_KEY=your-api-key \
 ## config validate
 
 ~~~
-notarius config validate [--config path/to/config.yml] [--pipeline pipeline-id] [--only lane-a,lane-b]
+notarius config validate [--config path/to/config.yml] [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
 ~~~
 
-This command loads and validates a configuration. With **--pipeline**, it also
+This command loads and validates the selected composed configuration. With **--pipeline**, it also
 resolves that pipeline against the production module catalog. **--only** selects
 lanes during that resolution and requires **--pipeline**.
 
@@ -151,10 +181,32 @@ OPENROUTER_API_KEY=validation-placeholder \
 The placeholder in the second command is sufficient only for offline
 validation; it cannot run a provider-backed pipeline.
 
+## config effective
+
+~~~
+notarius config effective [--config path/to/config.yml] [--config-overlay name]
+~~~
+
+This command writes the selected composed configuration as deterministic,
+redacted YAML. Standard output contains only that YAML: it does not validate a
+pipeline, prepare modules, materialize references, or contact a provider.
+
+## config explain
+
+~~~
+notarius config explain [--config path/to/config.yml] [--config-overlay name] <json-pointer>
+~~~
+
+This command writes ConfigKit's redacted, human-readable explanation for one
+canonical JSON Pointer. Supply `""` explicitly to explain the root. A malformed
+pointer is a syntax error; a well-formed unknown path is a command failure.
+The explanation can describe a value displaced by a selected overlay. Standard
+output contains only the explanation.
+
 ## pipelines list
 
 ~~~
-notarius pipelines list [--config path/to/config.yml] [--json]
+notarius pipelines list [--config path/to/config.yml] [--config-overlay name] [--json]
 ~~~
 
 This command lists configured pipeline IDs in sorted order. By default, it
@@ -171,7 +223,11 @@ 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
+Configuration load errors include ConfigKit's field diagnostics and source
+coordinates, followed by a source-ID legend identifying the acquired files or
+source labels. This includes imported files and the selected configuration overlay.
+
+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.
 
@@ -182,6 +238,6 @@ standard error. Parse the result only after the process exits with status 0.
 | 2 | The command or flag syntax was invalid, including unknown commands, missing required arguments, invalid flag values, or invalid flag combinations. |
 
 The root help spellings are the supported help path. Invoking **--help** on
-**run**, **config validate**, or **pipelines list** is handled by the flag
+**run**, a **config** subcommand, or **pipelines list** is handled by the flag
 parser as a usage error: it writes an error to standard error and exits with
 status 2.

docs/config.md

@@ -1,7 +1,7 @@
 # Configuration
 
 This is the canonical reference for Notarius configuration. Configuration files
-are YAML and must declare version 4. They select pipelines and their modules;
+are YAML and must declare version 5. They select pipelines and their modules;
 the [CLI reference](cli.md) owns invocation syntax, and
 [Operations](operations.md) owns run-state procedures.
 
@@ -19,12 +19,16 @@ The command fails if none of these paths provides a configuration file.
 For configuration values, precedence is:
 
 1. built-in defaults;
-2. the selected YAML file;
-3. supported operational environment variables; and
-4. the CLI run overrides that apply to a command.
+2. composed YAML sources, including imports;
+3. the selected ConfigKit overlay; and
+4. explicitly supplied configuration-valued run flags.
 
-Environment variables do not provide a second configuration schema. They only
-override the fields listed below.
+`--config-overlay` selects a ConfigKit overlay from `overlays/<name>.yml`
+beneath the selected configuration root. Empty and `default` select no overlay.
+`profiles/<name>.yml` is not a fallback location, and `--config-profile` is no
+longer accepted. Configuration overlays are distinct from PromptKit LLM profiles selected by
+`llm_profile` or `--llm-profile`. The retired operational environment variables
+have no effect; `NOTARIUS_CONFIG` remains a source-discovery input.
 
 ## Maintained Examples
 
@@ -33,7 +37,8 @@ override the fields listed below.
 - [Complete D&D configuration](../examples/dnd-complete.config.yml) uses
   ordered steps, all implemented D&D lanes, generated references, state
   settings, bounded LLM concurrency, and the maintained
-  [operator profile](../examples/profiles/dnd-extraction.yml).
+  [LLM profile](../examples/llm-profiles/dnd-extraction.yml) and imports its
+  pipeline from [a fragment](../examples/pipelines/dnd-session.yml).
 
 Use these complete files as starting points rather than combining the
 illustrative fragments in this reference.
@@ -46,7 +51,7 @@ other than **version** is optional.
 
 | Field | Type | Default | Rules |
 | --- | --- | --- | --- |
-| **version** | integer | none | Required; must be 4. |
+| **version** | integer or quoted string | none | Required; must be 5. Maintained files use `version: 5`. |
 | **promptkit** | object | none | Profile source and optional local-backend configuration. |
 | **pipelines** | map | empty | Maps pipeline IDs to pipeline definitions. |
 | **concurrency** | object | see below | Global LLM and extraction limits. |
@@ -69,6 +74,22 @@ Built-in defaults are:
 
 An empty cache directory in YAML deliberately selects the corresponding
 per-user root. An explicit empty output or debug directory is invalid.
+Selecting a chunk-plan or checkpoint directory does not configure filesystem
+access policy. Cache permissions come from deployment provisioning, process
+umask, setgid inheritance, and default ACLs; see [Operations](operations.md#cache-provisioning).
+
+Every nonempty relative configured destination or reference path resolves from
+the canonical directory of the file that supplies its surviving declaration.
+This includes imported fragments and selected configuration overlays. Built-in
+relative destinations use the selected main configuration file's directory;
+explicit `--output-dir` and `--debug-dir` values use the invocation working
+directory. Absolute values remain unchanged.
+
+The root file can use ConfigKit imports to compose disjoint definitions. Imports
+and selected overlays remain under the configuration root; they do not merge or
+append a pipeline's `steps` sequence. An overlay replaces sequences as defined
+by ConfigKit, so changing an individual ordered binding requires supplying the
+whole replacement sequence.
 
 ## PromptKit Profiles
 
@@ -76,11 +97,12 @@ The optional **promptkit** object selects one source of profile definitions and
 may register one conventional local OpenAI-compatible backend:
 
 ~~~yaml
-version: 4
+version: 5
 
 promptkit:
   profile_dir: ./profiles
   # profile_file: ./profiles.yml
+  transient_retries: 2
   local_backend:
     endpoint: http://localhost:8000/v1
     concurrency_limit: 2
@@ -90,21 +112,23 @@ promptkit:
 | --- | --- | --- |
 | **profile_dir** | string | Non-empty directory containing profile files. |
 | **profile_file** | string | Non-empty profile file. |
+| **transient_retries** | integer | Non-negative additional built-in HTTP attempts for eligible transient provider failures; defaults to 2. Set 0 to disable these retries. |
 | **local_backend** | object | Optional registration for the conventional PromptKit backend ID **local**. |
 | **local_backend.endpoint** | string | Required when **local_backend** is present; absolute HTTP or HTTPS URL with a host. |
 | **local_backend.concurrency_limit** | integer | Optional non-negative limit; defaults to 0. |
 
-Set at most one of **profile_dir** and **profile_file**. Relative values use
-the process working directory, not the configuration file's directory. The
-complete example's `./examples/profiles/dnd-extraction.yml` value is therefore
-valid when Notarius is launched from the repository root; use an absolute path
-for services and containers.
+Set at most one of **profile_dir** and **profile_file**. Every nonempty relative
+configured filesystem path is resolved from the canonical directory of the
+file that supplies it. The complete example's
+`./llm-profiles/dnd-extraction.yml` value therefore resolves beside that example.
+Use absolute paths for services and containers when their deployment layout is
+not configuration-relative.
 
 An operator source is optional. For a requested ID, PromptKit checks the
 configured operator source first, then Notarius's embedded fallback profiles,
 then its own built-in catalog. A matching profile is complete: it replaces a
 lower-precedence definition rather than merging with it. The maintained
-[`dnd-extraction` operator profile](../examples/profiles/dnd-extraction.yml)
+[`dnd-extraction` LLM profile](../examples/llm-profiles/dnd-extraction.yml)
 is a secret-free deployment artifact; production, development, and local
 deployments can each provide a complete definition with that same workload ID.
 Use workload-oriented IDs for new profiles instead of model names.
@@ -147,30 +171,15 @@ requiring credential values. It rejects absent, malformed, or incompatible
 profiles before a run prepares modules. Credential availability is checked only
 when a generation is prepared.
 
-## Migrating Version 3 Configuration
+## Version 5 Migration
 
-Version 3 files are not decoded or rewritten. Change **version: 3** to
-**version: 4** and rename the top-level **scriptorium:** section to
-**promptkit:**. Version 4 decoding is strict, so a remaining **scriptorium**
-field is rejected as unknown.
-
-## Operational Environment Variables
-
-These variables are applied after YAML values:
-
-| Variable | Overrides | Rules |
-| --- | --- | --- |
-| **NOTARIUS_TOTAL_LLM_CONCURRENCY** | **concurrency.total_llm** | Integer. |
-| **NOTARIUS_STAGE_WORKERS_EXTRACT** | **concurrency.stage_workers.extract** | Integer. |
-| **NOTARIUS_OUTPUT_DIR** | **output.directory** | Non-empty path. |
-| **NOTARIUS_CACHE_CHUNK_PLANS_MODE** | **cache.chunk_plans.mode** | **auto**, **bypass**, or **refresh**. |
-| **NOTARIUS_CACHE_CHUNK_PLANS_DIR** | **cache.chunk_plans.directory** | Non-empty path. |
-| **NOTARIUS_CACHE_CHECKPOINTS_DIR** | **cache.checkpoints.directory** | Non-empty path. |
-| **NOTARIUS_DEBUG_DIR** | **debug.directory** | Non-empty path. |
-
-Integer values are trimmed then parsed as base-10 integers. Directory and
-output values reject NUL characters. **NOTARIUS_CONFIG** participates only in
-configuration discovery.
+Version 3 and version 4 files are rejected and are not rewritten at load time.
+Use `version: 5`, rename the former `scriptorium` section to `promptkit`, and
+convert every module binding to an object containing `module`. Convert every
+external reference to `{path: ...}` and move any lane-level references to the
+corresponding extraction binding. Put durable operational values in YAML or
+use the documented explicit run flags; the former field-level environment
+variables are not read.
 
 ## Concurrency, Output, Cache, And Debug
 
@@ -195,11 +204,14 @@ debug:
 **concurrency.total_llm** must be greater than zero. The only supported
 **concurrency.stage_workers** key is **extract**; its value must be from 1
 through **total_llm**. When omitted, it is recalculated from the effective
-**total_llm** after YAML and environment precedence.
+**total_llm** after imports, overlay selection, and explicit ConfigKit
+overrides.
 
 **cache.chunk_plans.mode** accepts **auto**, **bypass**, or **refresh**.
 **cache.checkpoints.enabled** is a boolean. The CLI can override the output
 directory and chunk-plan mode for one run; see [CLI reference](cli.md#run).
+Cache directory fields select only locations; Notarius has no cache permission
+field or operational environment override.
 
 ## Pipelines
 
@@ -209,14 +221,20 @@ Each **pipelines** entry has a unique, non-empty ID and the following shape:
 pipelines:
   dnd-session:
     llm_profile: dnd-extraction
-    input: seriatim
-    chunk: generic
-    output: json
+    input:
+      module: seriatim
+    chunk:
+      module: generic
+    output:
+      module: json
     artifacts:
       spells:
-        extract: dnd/spells
-        merge: appendorder
-        normalize: dnd/spells
+        extract:
+          module: dnd/spells
+        merge:
+          module: appendorder
+        normalize:
+          module: dnd/spells
 ~~~
 
 | Field | Type | Default | Rules |
@@ -286,19 +304,20 @@ A lane has these fields:
 | **extract** | module binding | none | Required. |
 | **merge** | module binding | **appendorder** | Optional. |
 | **normalize** | module binding | **noop** | Optional. |
-| **references** | map | none | Supported compatibility alias for **extract.references**. |
+| **references** | map | none | Not supported at lane scope; place references on the eligible binding. |
 | **validators** | list | none | Non-empty lane-level lists are rejected. Set validator overrides on a binding instead. |
 
-The lane-level **references** alias remains accepted. When the alias and
-**extract.references** bind the same slot, **extract.references** wins. Use
-the binding-local form in new configurations.
+Lane-level **references** are not supported. Put a lane extraction reference
+under **extract.references**.
 
 ## Module Bindings And Validators
 
-Use a module key directly when no other binding fields are needed:
+Every module and validator binding uses an object, even when it has no fields
+beyond its module key:
 
 ~~~yaml
-input: seriatim
+input:
+  module: seriatim
 ~~~
 
 Use an object for fields:
@@ -309,7 +328,8 @@ extract:
   llm_profile: dnd-extraction
   retries: 2
   references:
-    spell_catalog: ./dnd-spell-catalog.json
+    spell_catalog:
+      path: ./dnd-spell-catalog.json
 ~~~
 
 | Binding field | Type | Default | Rules |
@@ -381,18 +401,20 @@ When enabled, it publishes the selected source-unit excerpt defined by the
 
 ## References And Ordered Handoffs
 
-Reference maps bind named slots that the selected target declares. A scalar is
-an external path. Pipeline-level maps accept only external paths; step-local
-and binding-local maps may also select a normalized artifact from an earlier
-step:
+Reference maps bind named slots that the selected target declares. An external
+reference is an object with `path`. Pipeline-level maps accept only external
+paths; step-local and binding-local maps may also select a normalized artifact
+from an earlier step:
 
 ~~~yaml
 steps:
   - id: describe-session
     artifacts:
       npc-registry:
-        extract: dnd/npc-registry
-        normalize: dnd/npc-registry
+        extract:
+          module: dnd/npc-registry
+        normalize:
+          module: dnd/npc-registry
   - id: extract-events
     references:
       npc_registry:
@@ -401,8 +423,10 @@ steps:
           lane: npc-registry
     artifacts:
       spells:
-        extract: dnd/spells
-        normalize: dnd/spells
+        extract:
+          module: dnd/spells
+        normalize:
+          module: dnd/spells
 ~~~
 
 An artifact selector contains only **step** and **lane**. The producer must be
@@ -411,10 +435,16 @@ 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. A YAML reference
+path resolves from the file that supplies its surviving declaration, including
+an imported fragment or selected overlay. CLI reference paths resolve relative
+to the process working directory.
 
 ### D&D Reference Slots
 
@@ -504,11 +534,33 @@ 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:
+    - module: generic/valid_json
+    - module: extract/dnd/scene-descriptions/shape
+    - module: extract/dnd/scene-descriptions/source_refs
+    - module: generic/valid_json_schema
+    - module: 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.
@@ -541,7 +593,16 @@ go run ./cmd/notarius config validate \
 ~~~
 
 Configuration validation rejects invalid YAML, unsupported fields, invalid
-defaults or environment overrides, incompatible module keys, unknown options,
+defaults or explicit CLI overrides, incompatible module keys, unknown options,
 invalid reference bindings, missing required reference slots, invalid validator
 overrides, and incompatible generated artifact handoffs. Use
 [pipelines list](cli.md#pipelines-list) to inspect configured IDs.
+
+## Inspection
+
+Use [config effective](cli.md#config-effective) to write the selected composed
+configuration as deterministic, redacted YAML. Use
+[config explain](cli.md#config-explain) with a canonical JSON Pointer to inspect
+the effective or historical provenance for one path, including a value displaced
+by a selected overlay. Neither command resolves a pipeline, prepares modules,
+materializes references, or requires provider credentials.

docs/consumers/dnd-pipeline.md

@@ -23,14 +23,14 @@ A deployment must provide its own PromptKit profile and campaign reference
 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
-- `promptkit.profile_file` is resolved relative to the Notarius process working
-  directory.
+- reference and PromptKit profile paths in YAML are resolved relative to the
+  file that supplies their surviving declaration;
+- reference paths passed with `--reference` are resolved relative to the
+  Notarius process working directory.
 
 Do not copy the repository example's relative profile path into a deployment
-without also controlling that working directory. The complete path and profile
-rules are defined in [Configuration](../config.md).
+without preserving its configuration-relative layout. The complete path and
+profile rules are defined in [Configuration](../config.md).
 
 Preflight the deployed configuration before processing sessions and whenever
 it changes:
@@ -73,9 +73,21 @@ 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 +104,7 @@ 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.
 
@@ -145,7 +157,8 @@ 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

@@ -15,9 +15,12 @@ configured lane inventory, and downstream acceptance checklist, see
 Optionally preflight a selected configuration and pipeline before work starts:
 
 ```sh
-notarius config validate --config /path/to/notarius.yml --pipeline pipeline-id
+notarius config validate --config /path/to/notarius.yml --config-overlay production --pipeline pipeline-id
 ```
 
+Omit `--config-overlay` when the composed base configuration is the intended
+selection; it is independent from any `--llm-profile` run override.
+
 Invoke the run with explicit paths and machine-readable output. Capture
 standard output and standard error separately; do not combine them before
 processing the result.
@@ -25,19 +28,43 @@ processing the result.
 ```sh
 notarius run pipeline-id \
   --config /path/to/notarius.yml \
+  --config-overlay production \
   --input /path/to/source.json \
   --output-dir /path/to/output-root \
   --json
 ```
 
 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
@@ -79,7 +106,7 @@ 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/development.md

@@ -9,6 +9,9 @@ the [README](../README.md) for product context, [Architecture](policy/architectu
 for system boundaries, and [Internal Overview](internal/overview.md) for the
 implemented component map.
 
+Read [Agent Policy](policy/agents.md) before recommending, planning, or
+implementing changes. It owns implementation discipline and local commit rules.
+
 ## What To Read
 
 | When working on | Read | Why |
@@ -25,7 +28,7 @@ implemented component map.
 | LLM clients, prompts, schemas, profiles, or scheduling | [LLM Runtime](internal/llm.md) | It documents the transport boundary and PromptKit integration. |
 | Output, cache, resume, or debug artifacts | [Run State Internals](internal/state.md), [Operations](operations.md), and [Configuration](config.md) | These separate implementation details, operator behavior, and configuration contracts. |
 | External input formats, artifact schemas, or durable output files | [Integration Contracts](integrations/) | Integration documents define external and durable data contracts. |
-| Release preparation, tagging, publication, or verification | [Source Releases](release.md) and [Documentation Policy](policy/documentation.md) | The release procedure owns maintainer guards and immutable-tag recovery; the policy assigns release-note ownership. |
+| Release preparation, tagging, publication, or optional diagnostics | [Release Policy](policy/release.md) and [Source Releases](release.md) | Policy owns release requirements and completion; the runbook supplies the publication command and recovery procedure. |
 | Proposed or unimplemented behavior | [Roadmap](roadmap/) | Future work belongs only in roadmap documentation until implemented. |
 
 For an existing subsystem, also inspect its focused tests and the package-local

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. |
 
@@ -132,7 +135,7 @@ 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`,
@@ -142,9 +145,47 @@ 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-configkit.md

@@ -0,0 +1,13 @@
+# ConfigKit Integration
+
+Notarius uses the public root package
+[`gitea.maximumdirect.net/eric/ConfigKit`](https://gitea.maximumdirect.net/eric/ConfigKit)
+at v0.4.0 for version 5 configuration acquisition, imports, overlays, explicit
+overrides, diagnostics, and safe inspection rendering.
+
+Notarius owns the typed document, domain defaults and validation, runtime
+conversion, catalog-dependent pipeline resolution, source-aware configured
+paths, and command discovery. It uses only ConfigKit's public path, result,
+diagnostic, and resolution-source APIs. Upstream documentation owns general
+ConfigKit semantics; [Configuration](../config.md) owns Notarius's file and
+operator contract.

docs/integrations/pkg-promptkit.md

@@ -1,11 +1,11 @@
 # PromptKit Integration
 
 Notarius pins
-[`gitea.maximumdirect.net/eric/promptkit` v0.9.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0)
+[`gitea.maximumdirect.net/eric/promptkit` v0.11.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0)
 as its in-process prompt engine. The upstream
-[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/consumers/pkg-promptkit.md)
+[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/consumers/pkg-promptkit.md)
 owns the public engine API, and the upstream
-[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md)
+[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/formats.md)
 owns prompt, profile, and schema file contracts.
 
 ## Supported Boundary
@@ -27,7 +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/v0.9.0/backends.go)
+[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/backends.go)
 owns the registration and backend-capacity contract.
 
 For one completion, the adapter calls `PrepareExecution`, takes a
@@ -53,7 +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 v0.9.0's 256-code-point session limit. Session IDs are non-secret
+PromptKit'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.
 
@@ -71,7 +71,15 @@ endpoint-only profiles have no such backend limit. The adapter translates
 PromptKit capacity rejection into the provider-neutral Notarius
 `ErrLLMCapacityExceeded` contract. It may include the normalized selected
 backend ID in safe diagnostic context, without exposing PromptKit's capacity
-error type, and leaves retries to the calling pipeline stage.
+error type. PromptKit's built-in client retries eligible transient HTTP failures
+with its v0.11.0 default of two additional attempts per generation; Notarius
+keeps pipeline-stage retries outside that completion boundary.
+
+The adapter projects built-in generation failures as the application-owned
+`LLMGenerationError`, retaining a safe failure kind, HTTP status when present,
+and retry-exhaustion attempt count. Transport, malformed successful envelopes,
+provider statuses, and generic generation failures remain distinct without
+exposing PromptKit error values to Notarius callers.
 
 ## Profile Sources And Compatibility
 
@@ -86,16 +94,16 @@ 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/v0.9.0/docs/formats.md).
+[pinned format rules](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/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.
 
-PromptKit v0.9.0 accepts only the `developer`, `system`, `user`, and
+PromptKit 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
+For application-owned semantic correction, Notarius uses PromptKit'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`.
@@ -112,12 +120,28 @@ 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
+PromptKit 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,
+External catalog profile names and settings can change between dependency
+releases, including removal without aliases. Review directly selected IDs and
+inherited `base_profile` values when upgrading. An application-owned alias that
+inherits an external profile does not freeze that profile's settings.
+PromptKit's opt-in `promptkit-openrouter-standard` and
+`promptkit-openrouter-auto` names provide name continuity within its
+selected-and-tested dependency scope, not a fixed model or provider availability
+guarantee. See the upstream
+[catalog compatibility policy](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/policy/release.md#catalog-profile-compatibility).
+
+Use a complete operator-owned profile with explicit backend or endpoint, model,
+and settings when those choices must remain operator-controlled. The embedded
+`dnd-extraction` fallback already follows this model and does not inherit an
+external catalog profile. An upgrade updates Notarius's compiled-in catalog
+checkpoint identity so prior profile-source checkpoints are not reused.
+
+Notarius supports this boundary against PromptKit v0.11.0. 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,
@@ -146,3 +170,9 @@ profile/configuration contracts against the upstream documentation. Maintained
 production prompts use PromptKit's bounded structural-repair contract; their
 current declaration is one additional repair attempt. Notarius retains the
 transport-neutral boundary and does not expose PromptKit types to modules.
+
+PromptKit v0.11.0 is licensed under
+[LGPL-3.0-or-later](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/LICENSE).
+Its accompanying [GPLv3 notice](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/COPYING)
+is incorporated by that license. This dependency change does not alter
+Notarius's source license or distribution workflow.

docs/integrations/run-result.md

@@ -9,18 +9,22 @@ 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. |
@@ -34,14 +38,18 @@ means one or more otherwise accepted results advanced under validator-failure
 
 ```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,
+  "warning_group_count": 1,
+  "warning_occurrence_count": 2,
+  "diagnostic_group_count": 3,
+  "diagnostic_occurrence_count": 5,
+  "diagnostics_truncated": false,
   "validation_status": "incomplete",
   "validation_summaries": [
     {

docs/internal/cli.md

@@ -25,7 +25,8 @@ preparation, and runner mechanics after their inputs are supplied.
 ## Dispatch And Configuration Handoff
 
 The root dispatcher handles help, version reporting, configuration validation,
-pipeline listing, and a pipeline run. Version reporting resolves build
+effective-configuration inspection, explanation inspection, pipeline listing,
+and a pipeline run. Version reporting resolves build
 information through `internal/buildinfo` before production composition, so the
 diagnostic remains available without configuration or runtime collaborators.
 The public syntax, streams, exit classes, and version semantics are defined by
@@ -33,14 +34,15 @@ the [CLI reference](../cli.md). Other root commands normalize injectable
 options before dispatch so that a missing production dependency fails as a
 command error rather than reaching execution.
 
-Commands that need configuration use one shared loader. The CLI discovers the
-file, parses it through **internal/core/config**, starts from defaults, applies
-the file and supported environment overrides, and then validates it for the
-command. The configured discovery and precedence contract is in
+Commands that need configuration use one shared ConfigKit-backed loader. The
+CLI discovers the file, passes the requested overlay and only explicit
+configuration-valued run overrides, and receives a **LoadedConfig**. The
+configured discovery and precedence contract is in
 [Configuration](../config.md), while the loading and resolution mechanics are
 in [Configuration Internals](configuration.md).
 
-Configuration validation without a selected pipeline checks structural
+`config effective` and `config explain` only render the loaded ConfigKit view;
+they do not resolve pipelines or construct providers. Configuration validation without a selected pipeline checks structural
 configuration only. Validation with a selected pipeline also builds the
 effective catalog, resolves the pipeline, and verifies every explicit effective
 PromptKit profile. Selected LLM-backed input, chunk, lane, output, and validator
@@ -84,8 +86,7 @@ field.
 After parsing and validating a run invocation, the CLI performs this ordered
 handoff:
 
-1. load and validate configuration, then apply command-level operational
-   overrides;
+1. load and validate configuration with ConfigKit command overrides;
 2. create and validate a safe run identity, then allocate a debug bundle only
    when requested;
 3. build the effective catalog, resolve requested reference changes, resolve
@@ -115,6 +116,16 @@ 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 +177,9 @@ 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

@@ -8,17 +8,17 @@ either interface.
 
 ## Boundary
 
-The configuration package turns a selected YAML file and supported environment
-values into a validated, independently owned configuration. It then resolves a
+The configuration package turns a selected ConfigKit-composed YAML document
+into a validated, independently owned configuration. It then resolves a
 requested pipeline against a module catalog before the framework prepares or
 runs anything.
 
 | Boundary | Inputs | Outputs | Does not own |
 | --- | --- | --- | --- |
-| Loading | Selected file path and environment lookup | Parsed file model and a populated **Config** | Choosing the file path or reporting a command result. |
+| Loading | Selected file path, profile, overrides, and invocation directory | **LoadedConfig** with ConfigKit inspection metadata and a populated **Config** | Choosing the file path or reporting a command result. |
 | Validation | **Config** | Structural configuration errors with pipeline, lane, or binding context | Module availability, capabilities, or construction. |
 | Resolution | Valid **Config**, selected pipeline and lanes, runtime reference changes, LLM override, and module catalog | **EffectiveConfig** with a **ResolvedPipeline** | Materializing reference bytes, preparing modules, execution, or filesystem state. |
-| Summary | **Config** or **EffectiveConfig** | Detached redacted payload suitable for debug summaries | Redacting arbitrary process state or provider traffic. |
+| Inspection | **LoadedConfig** and an optional JSON Pointer | Redacted effective YAML or explanation text | Pipeline selection, module preparation, or provider work. |
 
 The CLI discovers a configuration file, invokes this package, and supplies the
 result to the framework. Configuration never reads an input file, constructs a
@@ -28,23 +28,15 @@ at their respective [CLI](cli.md), [pipeline](pipeline.md), and
 
 ## Loading And Validation
 
-The CLI loads configuration in this order:
-
-1. parse the selected YAML file strictly into the file model;
-2. start from **Default**;
-3. apply the file model; and
-4. apply the supported environment overrides.
-
-This establishes the public precedence order without giving environment input a
-second file schema. Loading and application reject malformed YAML, unsupported
-file versions, unknown fields, invalid values, and identifiers that are empty
-or collide after whitespace normalization. The file application also makes the
-effective extraction-worker default follow the effective LLM limit. A present
-PromptKit local-backend object requires and trims its endpoint, defaults its
-omitted concurrency limit to zero, and is copied so the parsed file model
-cannot alias the populated **Config**. A pipeline `llm_profile` is
-presence-aware: omission remains empty, while a present blank value is
-rejected and a non-empty file value is trimmed before it reaches **Config**.
+ConfigKit acquires the selected root, imports, optional selected overlay, and
+explicit CLI overrides in precedence order. Its document definition supplies
+typed defaults, source-aware path normalization, and structural validation
+before conversion to the independently owned runtime **Config**. Only
+`NOTARIUS_CONFIG` participates in discovery; retired field-level environment
+variables do not participate in loading. Version 5 decoding rejects scalar
+module bindings and scalar external references. ConfigKit retains redacted,
+historical inspection metadata while **LoadedConfig** keeps that implementation
+detail private behind effective and explanation writers.
 
 **Config.Validate** checks configuration-only invariants before resolution. It
 rejects incompatible profile sources, invalid state-surface values, unsupported
@@ -52,18 +44,21 @@ concurrency settings, malformed bindings and references, invalid retries, and
 invalid pipeline, step, or lane structure. PromptKit local-backend validation
 accepts only an absolute HTTP or HTTPS endpoint with a host and no user
 information, query, or fragment, and rejects a negative local concurrency
-limit. Its errors retain the closest known pipeline, lane, and binding context.
+limit. It also rejects a negative `promptkit.transient_retries` value; decoded
+omission resolves to two, while explicit zero remains a distinct opt-out. Its
+errors retain the closest known pipeline, lane, and binding context.
 It deliberately does not require modules to be registered: that requires a
 catalog and belongs to resolution.
 
 The exact user-selectable values and validation rules are defined in
-[Configuration](../config.md). Keep additions to the file model, an
-environment override, its validation, and that reference in the same change.
+[Configuration](../config.md). Keep document schema, source-aware normalization,
+runtime conversion, and that reference aligned in the same change.
 
 ## Effective Resolution
 
-**Config.Resolve** first recomputes derived concurrency defaults and validates
-the configuration. It normalizes the requested pipeline ID, copies the selected
+**Config.Resolve** derives an absent extraction-worker value in an owned copy
+and validates the configuration. A programmatically supplied worker value is
+always explicit. It normalizes the requested pipeline ID, copies the selected
 profile, and passes the non-empty command-level LLM profile override, requested
 lane selection, and reference changes to the framework resolver.
 
@@ -110,43 +105,39 @@ bytes, a debug payload, or all runtime state. The CLI records it as invocation
 provenance before execution; cache and checkpoint identity have additional
 owners in [Run State Internals](state.md).
 
-Configuration summaries must use **Redacted**, **RedactedSummaryPayload**, or
-**RedactedResolvedPipelinePayload**, never a direct configuration marshal.
-Those methods copy every binding and nested option container, replace values
-whose key is credential-shaped with **[REDACTED]**, and omit materialized
-reference content while retaining safe binding and reference provenance. The
-payload must not alias the source configuration or resolved pipeline.
-PromptKit's local endpoint and concurrency limit are preserved as non-secret
-configuration metadata in the independently owned summary; the object contains
-no credential value. This redaction is deliberately narrow: it protects
-configuration summaries and does not authorize recording arbitrary environment
-values or provider requests.
+Configuration publication uses ConfigKit's redacted effective YAML. Resolved
+pipeline publication uses **RedactedResolvedPipelinePayload**, which copies
+bindings and nested option containers, replaces credential-shaped values with
+**[REDACTED]**, and omits materialized reference content while retaining safe
+provenance. This redaction is deliberately narrow and does not authorize
+recording arbitrary environment values or provider requests.
 
 ## Invariants To Preserve
 
-- Defaults, YAML values, and environment values are applied in one direction;
-  later sources may override only their supported operational settings.
+- ConfigKit composition applies defaults, imports, an optional overlay, and
+  explicit overrides in one direction with retained origin metadata.
 - A configuration is structurally valid before it is resolved, and a resolved
   pipeline is compatible with the supplied catalog before preparation begins.
 - Whitespace-normalized identifiers are unique wherever they identify a
   pipeline, step, lane, worker, or reference slot.
-- Resolution and summary generation return detached data. Redaction must cover
-  every configured and resolved binding, including nested validator bindings.
+- Runtime conversion and resolved-pipeline publication return detached data.
+- The resolved PromptKit transient-retry budget is always passed to the shared
+  built-in client; document presence preserves an explicit zero.
 - The resolved digest changes when resolved composition changes and never
   includes itself.
 
 ## Focused Tests
 
-- **internal/core/config/file_config_contract_test.go** covers strict file
-  parsing, normalization, file application, and structural rejection.
-- **internal/core/config/env_contract_test.go** covers supported operational
-  overrides and their precedence.
+- **internal/core/config/document_test.go** and
+  **operational_document_test.go** cover ConfigKit document loading,
+  normalization, imports, overlays, and overrides.
 - **internal/core/config/validation_contract_test.go** covers configuration
   invariants and contextual failures.
 - **internal/core/config/effective_config_contract_test.go** covers defaults,
   selections, overrides, resolution context, digest changes, and ownership.
-- **internal/core/config/redaction_test.go** covers recursive credential
-  redaction, reference-content exclusion, and non-aliasing payloads.
+- **internal/core/config/loaded_config_test.go** covers safe effective and
+  explanation rendering; **redaction_test.go** covers resolved-pipeline
+  redaction and reference-content exclusion.
 
 Run **go test ./internal/core/config** after changing this boundary. Changes to
 the handoff or resolved-composition semantics also need the focused framework

docs/internal/dnd.md

@@ -29,13 +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. 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. 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
@@ -117,14 +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
@@ -134,22 +151,119 @@ relatedness validators report advisory evidence concerns. The configured order
 is documented in
 [Configuration](../config.md#production-validator-keys-and-default-chains).
 
-Every D&D rejection describes the correction in transcript-grounded domain
-terms, using contextual names, artifact fields, and source segment ranges when
-useful. The guidance must not ask the model to reproduce durable entity IDs,
-hashes, validator module keys, or reason codes. Those identifiers remain in
-ordinary validation provenance; only the actionable semantic guidance is
-eligible for the correction prompt.
+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.
@@ -158,13 +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

@@ -216,6 +216,19 @@ than output-validation failures. Apart from documented context, capacity, and
 invalid-output categories, provider error values and types do not cross the
 adapter error chain; callers receive only a credential-redacted diagnostic.
 
+`LLMGenerationError` preserves the provider-neutral failure kind (`transport`,
+`provider_status`, `malformed_response`, or generic `generation`) and any HTTP
+status. When PromptKit exhausts an eligible transient-status retry budget, it
+also records that exhaustion and the attempts in that generation. Those fields
+do not include repair generations, and zero attempts means unavailable rather
+than proof that no request was sent.
+
+Every execution that reaches PromptKit returns an optional request-local
+`execution` summary. It contains only terminal aggregate attempts, retry and
+repair counts, durations, outcome and failure classification, status, and
+presence-aware token usage. Preparation failures have no summary; later local
+decode failures retain the upstream terminal summary without changing it.
+
 When PromptKit rejects backend admission before generation, the adapter maps
 `promptkit.ErrCapacityExceeded` to
 `contracts.ErrLLMCapacityExceeded`, retaining prompt context and a redacted
@@ -247,21 +260,29 @@ redaction boundary.
 Structural repair does not replace pipeline retry behavior: a binding's
 configured retry count reruns its complete stage attempt after an operational
 or structural error, module-requested retry, or actionable semantic rejection.
-The pipeline owns attempt lifecycle, validation chains, and retry diagnostics;
-see [Pipeline Internals](pipeline.md#validation-retries-and-output) and the
+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
 
 The caller context remains the outer cancellation authority. PromptKit applies
-a positive effective generation timeout as an inner request deadline; an
-explicit zero disables only that generation deadline. The HTTP client timeout
-is a separate transport-wide cap. Notarius forwards the caller context and
-does not install another timeout wrapper around PromptKit.
+a positive effective generation timeout as an inner request deadline for each
+built-in generation attempt; an explicit zero disables only that generation
+deadline. The HTTP client timeout is a separate transport-wide cap. Notarius
+forwards the caller context and does not install another timeout wrapper around
+PromptKit.
 
-The selected PromptKit profile owns generation settings. Notarius binding
-retries remain outside the adapter and repeat the complete module operation
-and validation chain. PromptKit does not add a provider retry loop.
+The selected PromptKit profile owns generation settings. The resolved global
+`promptkit.transient_retries` configuration is passed to the shared PromptKit
+engine and controls additional built-in attempts for eligible transient HTTP
+failures in one generation. It defaults to two; explicit zero disables those
+attempts. Notarius binding retries remain outside the adapter and repeat the
+complete module operation and validation chain.
 Operator-facing behavior is summarized in
 [Operations](../operations.md#operational-limits), and the pinned upstream
 contract is identified in

docs/internal/modules.md

@@ -95,19 +95,24 @@ 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
@@ -115,6 +120,16 @@ 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
 mounts the core-owned protocol and candidate/transcript presentation assets.

docs/internal/overview.md

@@ -26,7 +26,7 @@ physical state roots.
 | --- | --- | --- |
 | Executable and command boundary | **cmd/notarius**, **internal/cli** | Process entry, command dispatch, configuration discovery, production composition, runtime collaborator setup, durable file placement, and user-facing reporting. |
 | Build information | **internal/buildinfo** | Resolves a stable linked release tag or build metadata for the diagnostic root version command. |
-| Configuration | **internal/core/config** | Defaults, strict YAML parsing, environment overrides, structural validation, effective resolution, redaction, and resolved-composition summaries. |
+| Configuration | **internal/core/config** | ConfigKit document definition, source-aware normalization, runtime conversion, structural validation, safe inspection, effective resolution, and resolved-composition redaction. |
 | Generic models | **internal/core/source**, **internal/core/artifacts**, **internal/framework/contracts** | Source documents and chunks, manifests and provenance, plus typed artifact, reference, validation, output, and structured-completion contracts. |
 | Pipeline framework | **internal/framework/pipeline** | Registries, profile and reference resolution, typed preparation, validation, retry coordination, ordered execution, handoff, and result assembly. |
 | LLM and prompt runtime | **internal/framework/llm**, **internal/framework/promptfs** | Provider-neutral structured completions, scheduling, profile recording, prompt assets, schema registration, and credential-shaped-value redaction. |

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.
 
@@ -52,7 +52,7 @@ consumers do not repeat profile inheritance or configuration resolution.
 Each LLM request receives its own copy of that resolved value. PromptKit spends
 it only for structural correction inside one completion; the runner's binding
 retry policy remains the separate outer budget for complete stage attempts.
-Configuration resolution supplies the selected profile and catalog; see
+Configuration resolution supplies the selected overlay and catalog; see
 [Configuration Internals](configuration.md).
 
 External reference materialization happens before preparation. The materializer
@@ -97,6 +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
@@ -150,34 +157,41 @@ 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, plus one fixed warning per validator whose execution
+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.
+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 semantic retries. A semantic retry is available only for a valid latest
-`single_response_v1` candidate; a deterministic or no-model rejection instead
-settles the semantic policy immediately. Structural-output errors alone use the
-structural policy, and validation failure without rejection settles the
-validator-failure policy without regenerating the producer.
+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
-warnings, nor overwrites the stored record. An incomplete cache validation
+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
@@ -224,6 +238,8 @@ 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

@@ -1,5 +1,9 @@
 # Run State Internals
 
+LLM debug response artifacts and their safe per-call references retain an
+optional terminal execution summary when one is available. This metadata is
+diagnostic only and does not change receipts, checkpoints, or aggregate usage.
+
 This document describes the implementation collaborators behind output, cache,
 and debug state. User-visible fields belong in [Configuration](../config.md),
 and physical layout, retention, recovery, reason codes, and cleanup belong in
@@ -32,11 +36,18 @@ owned by the [Accepted Chunk Map contract](../integrations/chunk-map.md).
 
 `internal/framework/chunkplan` owns source-addressed plan storage, validation,
 and atomic publication. Its store is constructed only when the selected mode is
-not `bypass`.
+not `bypass`. It preserves rooted entry checks while creating new cache roots
+and source-digest directories with requested mode `0777`, then creates an
+atomic temporary file through rooted `O_CREATE|O_EXCL` with requested mode
+`0666` before sync and rename.
 
 `internal/framework/checkpoint` owns checkpoint identity, manifests, payload
 codecs, loader, and recorder. The CLI constructs a recorder whenever checkpoint
 recording is enabled and constructs a loader only for a `--resume` invocation.
+The recorder writes manifest and payload JSON through
+`fileio.WriteJSONWithProcessPermissions`, whose confined atomic writer creates
+new directories with requested mode `0777` and temporary files with requested
+mode `0666` before publication.
 Identity incorporates explicit stable semantic fingerprints collected from
 prepared modules and validators in addition to configuration, input,
 references, runtime overrides, observed LLM profiles, and the LLM runtime's
@@ -64,12 +75,13 @@ 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
@@ -85,7 +97,10 @@ and debug-summary JSON.
 canonical operator-facing reason-code reference.
 
 `internal/core/fileio` provides confined atomic file writes used by state
-collaborators. The chunk-plan store retains its stronger entry validation.
+collaborators. Its exact-mode writer remains available to output and debug
+callers; its process-permission writer is used by checkpoints. The chunk-plan
+store retains its stronger rooted entry validation. [Operations](../operations.md#cache-provisioning)
+owns cache mode and provisioning guidance.
 
 The CLI constructs selective-recomputation policy from resolved generated
 artifact dependencies. It forces the selected step and transitive consumers,
@@ -97,7 +112,7 @@ 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

@@ -50,11 +50,45 @@ directory selects a per-user root:
 
 The field definitions and configuration examples are in [Configuration](config.md).
 On supported Unix systems, output directories and files are created with
-requested modes **0755** and **0644**. Chunk-plan, checkpoint, and debug
-directories and files use **0700** and **0600**. The operating system's umask
-may impose stricter output modes. Cache and debug roots may contain sensitive
-source-derived data, so provision them for one trusted account or service. An
-output bundle can also contain source content when its JSON output enables
+requested modes **0755** and **0644**; debug directories and files use
+**0700** and **0600**. The operating system's umask may impose stricter output
+or debug modes. Chunk-plan and checkpoint caches instead request maximum modes
+of **0777** for new directories and **0666** for new files, including atomic
+temporary files. The kernel applies the service's umask and applicable default
+ACLs when those entries are created.
+
+## Cache Provisioning
+
+Cache locations select where Notarius writes, not who can access them. A
+private service can use a restrictive umask:
+
+~~~sh
+umask 0077
+notarius run example --config /etc/notarius/config.yml --input /srv/input.txt
+~~~
+
+A trusted-group deployment can provision a setgid cache root and use a
+group-oriented umask (or an equivalent default ACL):
+
+~~~sh
+install -d -m 2770 -g notarius-operators /srv/notarius/chunk-plans /srv/notarius/checkpoints
+umask 0007
+notarius run example --config /etc/notarius/config.yml --input /srv/input.txt
+~~~
+
+Setgid and default ACLs control inherited group ownership and access policy;
+Notarius does not copy a parent directory's literal mode bits. Cache state is
+reconstructible but can contain source-derived data, so operators own its
+access and retention policy. Existing cache inodes retain their prior modes
+after an upgrade. Recreate a cache beneath correctly provisioned parents, or
+stop writers and perform a deliberate administrative migration when a
+whole-tree change is required.
+
+Cache publication remains atomic and stored records remain validated. Those
+guarantees do not add cross-process locking or transactional coordination for
+simultaneous writers to one checkpoint identity.
+
+Output bundles can also contain source content when JSON output enables
 evidence publication. Apply an appropriate umask and output-root access policy
 before enabling that option; the requested output modes alone may not be
 suitable for transcript-bearing bundles.
@@ -81,7 +115,7 @@ precedence rules are owned by [Configuration](config.md#promptkit-profiles).
 Use a profile source owned by the service account, keep it readable only by
 the intended operator, and supply provider credentials through the service
 environment—not in the Notarius configuration or profile YAML. The maintained
-[operator profile](../examples/profiles/dnd-extraction.yml) is secret-free and
+[LLM profile](../examples/llm-profiles/dnd-extraction.yml) is secret-free and
 can be copied as a format starting point. Validate a deployment without a
 provider call or credentials:
 
@@ -92,11 +126,11 @@ notarius config validate --config /etc/notarius/config.yml --pipeline dnd-sessio
 An unset optional `api_key_env` reaches the provider without authorization and
 may receive a 401 or 403 response.
 
-Profile paths are currently resolved from the process working directory, not
-from the configuration file. The complete example's
-`./examples/profiles/dnd-extraction.yml` path is valid for a repository-root
-invocation only. Use absolute paths such as
-`/etc/notarius/profiles/dnd-extraction.yml` for services and containers.
+Configured PromptKit profile paths resolve from the file that supplies their
+surviving declaration. The complete example's
+`./llm-profiles/dnd-extraction.yml` path is therefore relative to that example.
+Use absolute paths such as `/etc/notarius/profiles/dnd-extraction.yml` for
+services and containers when that is clearer for the deployment.
 
 ## Run Lifecycle
 
@@ -105,19 +139,40 @@ 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
 
+When debug recording is enabled, per-call response artifacts include the
+terminal execution summary for retry, repair, timing, and presence-aware usage
+diagnosis. Normal output, receipts, and warnings do not aggregate or publish
+this diagnostic metadata.
+
 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.
+PromptKit's structural-repair calls inside one completion, its configured
+`promptkit.transient_retries` built-in HTTP attempts, and an LLM-backed
+validator's own retry budget. The transient setting defaults to two additional
+attempts for eligible provider statuses; set it explicitly to zero when an
+operator needs one HTTP attempt per generation. 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.
+
+These budgets multiply only by their declared scopes. With two outer producer
+retries, two transient retries, and one structural repair, a single producer
+can issue up to 18 HTTP attempts: three producer attempts, each with an initial
+generation and one repair generation, each allowing three HTTP attempts. This
+is an illustration, not a total-deadline or exactly-once guarantee. Every
+built-in HTTP attempt has its own generation timeout; caller cancellation
+remains the overall authority and interrupts Retry-After waiting. The shared
+Notarius scheduler keeps its permit through those waits, so retry budgets can
+increase latency and provider cost. Observed usage can remain incomplete when
+providers omit accounting fields.
 
 After the applicable budgets are exhausted, the resolved
 [`validation_policy`](config.md#pipelines) determines the result. Structural
@@ -127,13 +182,14 @@ 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 abandoned-attempt warnings.
+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 count when an orchestrator
-requires complete validation. The durable fields and their meanings are owned
-by the [run-result receipt](integrations/run-result.md) and
+`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
@@ -271,8 +327,9 @@ Only a [debug-enabled run](cli.md#run) creates a bundle:
   trace/
 ~~~
 
-The summary contains redacted invocation and resolution information plus run,
-warning, checkpoint, chunk-plan, and terminal reporting artifacts. Attempt
+The summary contains redacted invocation and resolution information, including
+`effective-config.yml` and the separately redacted resolved pipeline, plus run,
+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
@@ -281,7 +338,8 @@ 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.
+was explicitly cleared. It records `config_overlay` only when a non-base
+configuration overlay was selected.
 
 Notarius never creates debug state without an explicit request and never
 automatically deletes a requested bundle. If allocation succeeds, the command
@@ -313,7 +371,9 @@ Provider execution settings and the generation timeout come from the selected
 PromptKit profile. The invocation-only **--reasoning-effort** and
 **--clear-reasoning-effort** controls may replace or clear that profile setting
 for all LLM-backed calls in one run without changing the profile. PromptKit
-structural output repair happens within one structured-completion call. Its
+v0.11.0 also retries eligible transient HTTP failures with its built-in default
+of two additional attempts per generation. This retry work is separate from
+structural output repair, which happens within one structured-completion call. Its
 effective `structured_output_repair_attempts` limit is resolved from the
 selected binding, then the pipeline, then the prompt declaration; see
 [module bindings](config.md#module-bindings-and-validators). This is distinct
@@ -321,19 +381,20 @@ from Notarius binding **retries**, which rerun the complete module operation
 and validation chain and do not consume or replenish the structural-repair
 limit. The maintained production prompts declare one repair attempt, paid only
 after a structural failure. One structured completion with repair budget **R**
-makes at most **R + 1** serial provider calls. If one stage attempt performs
-**C** structured completions, a binding with **retries: N** has a maximum of
-**(N + 1) * C * (R + 1)** provider calls; LLM-backed validators have their own
-corresponding invocation counts and budgets. This is an upper bound, not a
-promise that every call reaches a provider.
+contains at most **R + 1** serial generations. Each generation can now contain
+more than one provider call when PromptKit retries an eligible transient
+failure. LLM-backed validators have their own corresponding invocation and
+repair budgets. These are bounds on attempted work, not a promise that every
+call reaches a provider.
 
 Timeouts are layered. Caller cancellation is the outer authority. A positive
-effective generation timeout adds an inner request deadline, while zero
-disables only that generation deadline. The HTTP client timeout remains a
-transport-wide cap. Notarius does not add another timeout around PromptKit.
-Repairs are serial within the same caller context, so their worst-case latency
-and cost follow the provider-call bound above; provision run deadlines and
-provider budgets accordingly. Credentials remain optional unless the selected
+effective generation timeout adds an inner request deadline to each built-in
+generation attempt, while zero disables only that generation deadline. The HTTP
+client timeout remains a transport-wide cap. Notarius does not add another
+timeout around PromptKit. Retries and repairs are serial within the same caller
+context, so their worst-case latency and cost can exceed the structural-repair
+bound above; provision run deadlines and provider budgets accordingly.
+Credentials remain optional unless the selected
 PromptKit profile requires one, in which case preparation fails before a
 provider call when its configured credential is unavailable.
 The pinned upstream boundary and profile-format links are in

docs/policy/agents.md

@@ -0,0 +1,90 @@
+# Agent Policy
+
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius adaptations preserve operator choice, existing explicit contracts, and
+the deployment-permissions example below.
+
+## Purpose
+
+Agents working in this repository must read and adhere to the following house rules.
+
+## Implementation Discipline
+
+Simplify the mechanism, not the user's choices.
+
+**KEY PRINCIPLES:** Write the simplest idiomatic code that satisfies the user's
+request. Do not implicitly strengthen requirements, restrictions, or guarantees
+beyond what the user actually requested.
+
+**Keep implementation machinery proportionate.** Use straightforward,
+idiomatic code and introduce only the machinery needed to support the requested
+behavior and flexibility. Optimize for conceptual simplicity and
+maintainability, not for the fewest lines, fewest options, or narrowest product
+surface.
+
+**Do not invent restrictions.** Do not reject, prohibit, constrain, cap,
+harden, or otherwise narrow behavior unless required by the user, an explicit
+repository contract, or an unavoidable correctness constraint. Preserve
+operator and caller choice where policy has not been established. If an
+additional restriction appears advisable, present it as a decision rather than
+silently implementing it.
+
+**Do not invent guarantees.** Do not strengthen public contracts, guarantees,
+or invariants beyond those expressly requested by the user.
+
+**Use the standard library.** For example: in almost every case, the best
+approach to copying files is to use the standard library. Writing custom code to
+implement power-loss protection at the application level, or to guard against
+bad actors with privileged and concurrent access to the host's filesystem,
+is wasteful, unnecessary, and not idiomatic. Stick to the simple, standard, and
+idiomatic approach unless another path is expressly requested by the user.
+
+**An implementation detail is not repository policy.** Tests and documentation
+must not elevate incidental behavior into a durable guarantee or restriction
+unless that behavior was expressly selected as part of the contract. Existing
+explicit contracts remain authoritative; changing or weakening one requires a
+deliberate decision rather than being inferred from these principles.
+
+[ADR-0017](../adr/0017-defer-cache-access-policy-to-deployment.md) illustrates
+why Unix deployments own cache access policy through permissions, umask, setgid,
+and default ACLs. Notarius retains its separately adopted correctness boundaries.
+
+## Git Commit Policy
+
+By default, create a commit after completing each coherent unit of work, using a
+concise, plain-English message describing what changed. This policy is standing
+authorization to create those local commits; a separate user request is not
+required. An explicit user instruction to leave work uncommitted takes precedence.
+
+A unit of work is a focused, reviewable outcome, including its relevant tests and
+documentation. Do not commit after every intermediate edit or create empty commits.
+Complete the appropriate validation before committing. If work is incomplete or
+validation is blocked, report that state rather than presenting it as completed.
+
+Before committing:
+
+1. Run `git status --short` and review the relevant diff, including any staged
+   changes. Identify changes that predate the current unit of work.
+2. Exclude secrets, local environment files, generated junk, and unrelated changes.
+3. Stage only the files or hunks belonging to the completed unit. Use `git add .`
+   only when every included change belongs to that unit. Do not disturb unrelated
+   staged or unstaged work.
+4. Review `git diff --staged` and run appropriate checks for the change. Do not
+   repeat checks already passed for unchanged content merely to create a commit.
+5. Review `git log --oneline -n 10` and use
+   `git commit -m "<concise plain-English commit message>"`.
+6. Report the commit identifier and any remaining uncommitted work.
+
+If the unit cannot be safely separated from pre-existing work, leave it uncommitted
+and explain the overlap rather than silently including that work. If a commit
+fails, preserve the changes and report the failure; do not bypass hooks.
+
+Default commit authorization does not authorize pushing, publishing, amending,
+resetting, reverting, or rewriting history. Those actions require a separate user
+instruction. Do not use `git commit -a` or `git commit --no-verify` to bypass the
+staging and validation discipline above.

docs/policy/architecture.md

@@ -1,11 +1,25 @@
 # Architecture
 
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius adapts the base template to its fixed pipeline, artifact families,
+external adapters, state boundaries, and source-only distribution.
+
 This document defines the intended high-level architecture of Notarius and the
 invariants that changes must preserve. Implemented component details belong in
 [Internal Overview](../internal/overview.md) and its linked documents. The
 reasoning behind significant architectural choices belongs in
 [ADRs](../adr/).
 
+An adopted constraint can govern future changes without implying that every
+component it could govern exists. Descriptions of implemented structure must
+remain accurate; proposals and transition status belong in roadmaps and ADRs.
+Constraints reflect selected requirements, not incidental implementation choices.
+
 ## System Shape
 
 Notarius is a small, dependency-light Go application for extracting structured
@@ -38,6 +52,8 @@ Notarius is contract-first without being abstraction-heavy. Interfaces and
 extension points should protect demonstrated boundaries. New abstraction is not
 itself an architectural goal.
 
+Implementation discipline is owned by [Agent Policy](agents.md).
+
 ## Layers And Dependency Direction
 
 The application boundary is the composition root and may depend on concrete
@@ -146,8 +162,9 @@ 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.
@@ -159,17 +176,26 @@ 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,
-reject, fail, or skip when a runtime prerequisite is unavailable.
+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
@@ -191,6 +217,13 @@ 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,
 including with an explicitly empty chain. Configured validator order is
@@ -212,12 +245,14 @@ 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 rejection, 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).
+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
@@ -242,9 +277,11 @@ beneath this mandatory application-wide scheduler.
 
 ## Configuration And Provenance
 
-Configuration loading, precedence, defaults, environment overrides, redaction,
-and validation are centralized. Named pipeline definitions make structural
-composition explicit and discoverable. Operational overrides are permitted
+Configuration loading, precedence, defaults, redaction, and validation are
+centralized through the ConfigKit-backed boundary. `NOTARIUS_CONFIG` is limited
+to discovery; field-level environment overlays are not configuration input.
+Named pipeline definitions make structural composition explicit and discoverable.
+Explicit command overrides are permitted
 when they do not obscure the configured pipeline structure.
 
 Run preparation fails before stage execution when statically discoverable
@@ -258,6 +295,13 @@ model, correction, or large payload content.
 
 ## State, Output, And Safety
 
+Source material and model responses cross application validation boundaries.
+Host access policy belongs to the deployment; hypothetical privileged actors
+concurrently changing the filesystem do not justify new application defenses
+without an explicit decision. This scope statement does not remove the adopted
+correctness boundaries below or claim excluded scenarios cannot occur. Follow
+[Agent Policy](agents.md) when a new threat or guarantee needs a decision.
+
 Notarius exposes three filesystem surfaces with independent roots and
 lifecycle:
 
@@ -274,6 +318,15 @@ 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.
 
+Deployments own cache access policy. Cache writers request conventional maximum
+creation modes and let the process umask, setgid inheritance, and default ACLs
+determine effective access. Notarius owns the separate safety boundary:
+confined paths, symlink rejection, atomic publication, validation, and
+recovery. It does not chmod, chown, chgrp, or migrate cache entries to impose
+an application policy. This rule applies to chunk-plan and checkpoint cache
+only; output and debug retain their separate contracts. [Operations](../operations.md)
+owns deployment provisioning and retention guidance.
+
 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
@@ -304,8 +357,9 @@ best-effort development and compilation environment, while Windows is
 unsupported. Notarius distributes source releases only: an immutable source
 tag and its checked-in release note identify a release. The project does not
 publish executable binaries, archives, installers, container images,
-checksums, signatures, or package-manager entries. Maintainer release commands
-and tag guards belong to [Source Releases](../release.md).
+checksums, signatures, or package-manager entries. [Release Policy](release.md)
+owns publication requirements and completion; maintainer commands and guards
+belong to [Source Releases](../release.md).
 
 ## Architectural Non-Goals
 

docs/policy/documentation.md

@@ -1,5 +1,14 @@
 # Documentation Policy
 
+**Base Policy Version:** 1.3
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius retains subprocess-consumer ownership, the distinction between
+descriptive behavior and policy, and no placeholders for absent interfaces.
+
 ## Purpose
 
 This policy assigns each documentation topic to one canonical owner. Its goal is
@@ -24,12 +33,28 @@ contract, maintain it only in the owning document.
 Outside `docs/roadmap/`, documentation describes implemented behavior only.
 Partial features may be described only to their implemented boundary.
 
-ADRs are the narrow exception: an ADR may record an accepted architectural
-decision before implementation, but acceptance must not be presented as proof
+Architectural constraints may govern future changes without implying that the
+components or capabilities they would govern already exist. This does not permit
+describing a proposed system structure as implemented.
+
+ADRs may record an accepted architectural decision before implementation, but
+acceptance must not be presented as proof
 that the behavior exists. The roadmap owns implementation status and sequencing
 until the decision is implemented. Current architecture, user, operator,
 integration, and internal documentation are updated when the behavior lands.
 
+### Implementation Details And Policy
+
+Documentation must distinguish an explicit contract from an incidental
+implementation choice. Do not describe a restriction, guarantee, security
+policy, default, or limit as authoritative merely because the current code
+implements it. Documentation may accurately describe current behavior without
+making it a permanent requirement; distinguish descriptive behavior from
+normative policy. The
+[implementation-discipline rules](agents.md#implementation-discipline)
+own the criteria for adopting product restrictions and for elevating behavior
+into repository policy.
+
 ### Audience And Detail
 
 Write for the document's stated audience and include only the detail needed for
@@ -59,16 +84,18 @@ secret values.
 | --- | --- | --- | --- |
 | Product orientation and minimal end-to-end quickstart | `README.md` | What Notarius is, why it is useful, one shortest successful invocation, and links onward. | Complete command reference, configuration reference, operational procedures, implementation detail. |
 | Contributor entry point | `docs/development.md` | Task-oriented reading guide, minimal contributor orientation, baseline validation commands, and links to canonical docs. | Package inventory, architecture rules, subsystem behavior, detailed change recipes. |
-| Current application architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Application architecture policy | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Agent policy | `docs/policy/agents.md` | Implementation discipline, rules against invented restrictions or guarantees, and local commit authorization. | Application contracts, architecture, testing policy, and contributor procedures. |
 | Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
 | Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression-test policy, and criteria for adding, rewriting, or deleting tests. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
 | CLI contract | `docs/cli.md` | Commands, arguments, flags, invocation semantics, and exit codes. | End-to-end operating procedures, configuration field definitions, runtime filesystem layout, module implementation details. |
 | Configuration contract | `docs/config.md` | Discovery and precedence, file schema, fields, defaults, environment overrides, validation rules, and user-selectable module or validator keys. | Complete example files, CLI syntax, runtime state lifecycle, module implementation details. |
 | Operations | `docs/operations.md` | Runtime workflows, physical filesystem and state layout, output, cache, and debug handling, resume, cleanup, permissions, recovery, and operational limits. | CLI flag syntax, configuration field definitions, logical output schemas, implementation mechanics. |
-| Source release procedure | `docs/release.md` | Maintainer release selection, candidate validation, tagging, publication guards, verification, and immutable-tag recovery. | Product installation summary, CLI version semantics, historical release summaries, CI implementation detail. |
+| Release policy | `docs/policy/release.md` | Versioning, compatibility, integrity, immutability, proportionate gate selection, publication controls, and release completion. | Concrete commands, product installation, historical notes, and CI implementation detail. |
+| Repository-specific release procedure | `docs/release.md` | Prerequisites, candidate preparation, publication command, automated guards, optional diagnostics, and failure recovery. | General release requirements, product installation guidance, historical summaries, and current compatibility contracts. |
 | Release-note history | `docs/releases/` | One checked-in historical summary for each source release made under the procedure. The note at the immutable tag is that release's record. | Current commands, behavior, contracts, and compatibility definitions. |
 | Public HTTP contract, if introduced | `docs/api.md` | Routes, authentication, media types, request and response schemas, status codes, pagination, caching, idempotency, rate limits, and HTTP retry semantics. | Client walkthroughs, upstream or downstream integration internals, implementation detail. |
-| Consumer guidance, if a public package or API is introduced | `docs/consumers/` | Task-oriented use of the public interface, minimal client examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| Consumer guidance | `docs/consumers/` | Task-oriented subprocess use, minimal client examples, consumer responsibilities, and guidance for any future public package or API. | Wire semantics, external protocol contracts, internal implementation detail. |
 | External and durable integration contracts | `docs/integrations/` | External file formats and protocols, upstream and downstream contracts, logical output bundle paths and schemas, media types, and compatibility behavior. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
 | Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
 | Internal component behavior | Other files under `docs/internal/` | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
@@ -99,11 +126,13 @@ other contracts.
 
 ### Releases
 
-`docs/release.md` owns the source-release procedure. Release notes are
+The [release policy](release.md) owns requirements and invariants; the
+[repository procedure](../release.md) owns the commands that satisfy them.
+Release notes are
 historical summaries, not current-state contract owners: the checked-in note at
 an immutable tag records that release, while current canonical documentation
-must change with the behavior it describes. Do not use a release note to defer
-or replace current documentation updates.
+must change with the behavior it describes. Do not use a release procedure or
+note to weaken policy, redefine current contracts, or defer documentation updates.
 
 ### Contracts And Implementation
 

docs/policy/release.md

@@ -0,0 +1,204 @@
+# Release Policy
+
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius adaptations: source-only delivery, lightweight tags, branch-before-tag
+publication, no prereleases, and the justified native command check below.
+
+## Purpose
+
+This policy defines a proportionate workflow for validating and publishing an
+intended source revision under a fresh version. Releases should require little
+routine intervention, avoid repeated work, and preserve published contents.
+
+The repository-specific commands belong in `docs/release.md`. Historical release
+summaries belong in `docs/releases/`. The procedure must agree with the adopted
+policy and link to it rather than introduce additional release requirements.
+
+## Release Model And Completion
+
+A source release identifies one exact committed source revision through an
+immutable version tag and a version-matched, checked-in release note. The
+primary release branch is `main` and the authoritative remote is `origin`.
+Tags are lightweight. Publish the candidate to `main` before pushing its tag.
+
+A successful push of the intended tag, with its source commit available on the
+authoritative primary branch, completes the release. Notarius uses
+branch-before-tag publication; it does not require an atomic push.
+
+Release commands must finish at successful publication. They must not wait for
+or poll CI, require successful CI status, inspect downloadable artifacts, or
+check consumer installation or Go module availability as completion gates.
+A failed or ambiguous push requires resolving the authoritative remote state
+before declaring success or retrying; successful pushes do not require a
+separate remote read-back gate.
+
+Notarius publishes source only: no binaries, archives, images, packages,
+checksums, signatures, or Gitea release objects. Tag CI may provide advisory
+validation but publishes no deliverables and is not a completion gate.
+
+## Versioning And Compatibility
+
+Stable releases use Semantic Versioning in the form `vMAJOR.MINOR.PATCH`, with
+no leading zeroes. Notarius does not support prereleases.
+
+The selected version must be unused locally and on the authoritative remote.
+Do not require queries to every configured remote or publication service.
+A version known to have already been published must not be reused, even if the
+normal guards do not inspect that channel. Adding a distribution channel
+requires a deliberate update to the policy and procedure where needed.
+
+For releases at or after `v1.0.0`:
+
+- a major release may make intentional incompatible changes;
+- a minor release may add backward-compatible behavior; and
+- a patch release may make backward-compatible fixes.
+
+Before `v1.0.0`, a minor release may change a documented public, operational,
+configuration, data, integration, or compatibility contract when the release
+note explains the impact and required consumer action. A patch release must
+not intentionally break those contracts within its minor line.
+
+Additional compatibility promises and deliberate exceptions belong in the
+adopted policy or canonical contract documentation. Release notes explain
+changes; they do not redefine current compatibility contracts.
+
+## Release Documentation And Preparation
+
+Every new release must include `docs/releases/<version>.md` in the tagged
+source tree. Use the complete version tag as the filename and begin with a
+title containing the product name and that exact version.
+
+Include a short summary and any relevant compatibility impact or upgrade
+instructions. Keep the note proportionate to the change. Fixed sections and
+explicit no-action boilerplate are not required. Existing historical notes
+need not be rewritten to adopt this format.
+
+Keep exact current contracts in their canonical documents and update affected
+documentation with the behavior change. Link to those documents from the note
+where useful. A release note must not substitute for current documentation.
+Do not create retrospective notes for the pre-procedure v0.1.0, v0.2.0, or
+v0.3.0 tags.
+
+Version selection, release-note content, and retirement of completed roadmaps
+are preparation responsibilities, not separate human approval gates. Preserve
+planned and deferred work and repair affected links when retiring roadmaps.
+Ordinary change review covers unintended files, credentials, and private data;
+do not repeat it as a release-specific sign-off. Ignored local files are not
+prohibited merely because they exist: they are not included in the tagged source.
+
+## Candidate Validation
+
+Provide one publication command that validates the final committed candidate,
+checks publication state, and publishes the intended refs. A validation-only
+mode or command may be provided for diagnosis, but must not be a required step
+before publication repeats the same checks.
+
+Run the complete deterministic, offline default test suite once for the final
+candidate, following the [testing policy](testing.md). For repositories that
+publish builds, verify the required build targets. Select those targets for
+the application's actual support commitments; native testing on each platform
+and access to particular hardware are not default release prerequisites.
+
+Additional gates must address a concrete risk that the existing checks do not
+adequately cover. Do not automatically require linting, race testing, module
+hygiene audits, documentation audits, credential scans, example validation, or
+expanded platform matrices solely because they are possible. Retain such
+checks when their distinct protection justifies their cost. A one-time migration
+check must not silently become a permanent release requirement.
+
+Release tests must not require real credentials, mutable application services,
+paid APIs, provider probes, or inference calls. The procedure may describe dependency
+preparation separately from offline validation.
+
+Avoid duplicate validation across release phases. Successful checks for the
+same final candidate need not be repeated solely to satisfy another phase;
+any reuse must clearly apply to that exact candidate. Persistent validation
+caches are not required. If the candidate changes, validate the new candidate.
+Cheap publication-state checks may be repeated immediately before pushing
+because local or remote state may have changed during validation.
+
+Automate mechanical guards. Human review gates are strongly disfavored: add one
+only through an explicit project-specific policy decision explaining the
+concrete risk and why ordinary change review or automation is insufficient.
+Do not add routine confirmations for checks the command can perform itself.
+
+### Notarius Gate Selection
+
+Run the default offline Go suite once and build the native `cmd/notarius` with
+release version injection, checking its `--version` output. The command check
+protects the actual executable entry point and linker-supplied version, which
+package tests alone do not verify. Build products are temporary, not published.
+Use `GOWORK=off` so a sibling checkout cannot substitute for declared dependencies.
+
+Race tests, standalone vet, formatting/whitespace checks, module tidy audits,
+standalone example validation, and Linux/Darwin cross-builds remain useful
+change-specific development checks, not mandatory release gates. Default tests
+already cover maintained examples. Source-only publication does not require a
+binary distribution matrix. No new platform or hardware prerequisites are added.
+
+## Publication Guards
+
+Automate the checks needed to publish the intended source under a fresh version:
+
+- the version has the adopted format and is unused locally and on the
+  authoritative remote;
+- the worktree and index are clean on the primary release branch;
+- the source being published is the exact candidate that passed validation,
+  including its matching release note;
+- the branch push advances authoritative `main` to the candidate without
+  discarding remote history, before the tag is published; and
+- the tag identifies the validated candidate and only the intended refs are
+  published.
+
+Fetch the authoritative branch and tag state as needed for these checks.
+Do not force branch or tag updates to bypass a failed guard. Publish explicit
+refs rather than all local tags. Do not require CI success, publication-service
+availability checks, or a separate manual confirmation for each guard.
+
+## Immutability And Failure Recovery
+
+Published version tags and published contents are immutable. Do not move,
+overwrite, recreate, or delete a published tag, or replace a published artifact,
+checksum, signature, provenance statement, or release note under that version.
+The checked-in note at the immutable tag is that release's historical record.
+
+If validation fails before publication, correct and commit the candidate and
+rerun the publication command. If a push fails or its outcome is ambiguous,
+inspect the authoritative remote before retrying. Remove an unpublished local
+tag only after confirming it was never published or used by a release channel.
+An unavailable remote leaves publication status unknown; do not overwrite a tag
+to resolve uncertainty.
+
+Transient network, authentication, CI, or consumer-availability failures do not
+by themselves require a new version. Advisory CI or optional installation checks
+may be retried against the same immutable tag. Notarius has no asynchronous
+artifact delivery to recover.
+
+A defect requiring changes to published source or contents requires a corrected
+commit, a new semantically appropriate version, a matching release note, and
+the release procedure again. Leave the original release intact. A channel's
+yank, deprecation, or advisory mechanism may warn consumers away from a defective
+release without replacing its contents or reusing its version.
+
+## Repository Release Procedure
+
+Maintain `docs/release.md` as a short, executable maintainer runbook. It must
+identify:
+
+1. prerequisites, primary branch, and authoritative remote;
+2. version and release-note preparation;
+3. one publication command with its selected local checks and automated guards;
+4. declared deliverables, required build targets, and automation responsibilities;
+5. the successful-push completion point and separately reported delivery status;
+   and
+6. failure handling and ambiguous-push recovery.
+
+Use copyable commands where practical. Keep optional diagnostics and consumer
+verification clearly optional. The procedure must not turn preparation work,
+ordinary change review, or post-publication investigation into additional
+release gates.

docs/policy/testing.md

@@ -1,5 +1,14 @@
 # Testing Policy
 
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Adopted without substantive deviations; Notarius's established contracts remain
+subject to [Agent Policy](agents.md).
+
 ## Purpose
 
 Our tests exist to make **incorrect changes expensive and correct changes cheap**.
@@ -8,29 +17,16 @@ We do not optimize for test count, line coverage, exhaustive isolation, or the f
 
 ## Every test has a cost
 
-Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
-
-A test must be:
-
-- written and reviewed;
-- understood by future maintainers and coding agents;
-- executed in local and CI workflows;
-- diagnosed when it fails;
-- updated when legitimate behavior changes;
-- maintained as fixtures, APIs, and dependencies evolve; and
-- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
-
-Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+Every test has an immediate and continuing cost: writing, review, execution,
+diagnosis, fixture maintenance, and updates when legitimate behavior changes.
+Tests can also constrain refactoring, duplicate policy, and add failure noise.
 
 A test is warranted only when the confidence it provides justifies these costs.
-
-Apply this cost-benefit analysis at two levels:
+Evaluate that tradeoff at two levels:
 
 1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
 2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
 
-The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
-
 Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
 
 The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
@@ -55,7 +51,7 @@ Control clocks, randomness, environment variables, and other process-global or m
 
 ## What deserves tests
 
-Prioritize tests for:
+To the extent the following features are present in the application, prioritize tests for:
 
 1. Public and package-level contracts.
 2. Domain rules and important invariants.
@@ -67,7 +63,18 @@ Prioritize tests for:
 
 A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
 
-For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+Choose tests for requested behavior and established contracts. Tests must not
+introduce new restrictions, guarantees, or threat scenarios merely because they
+could be asserted. Follow the [agent policy](agents.md) when deciding whether a
+behavior belongs to the contract; current implementation behavior alone is not
+sufficient authority.
+
+For established behavior involving **data integrity, destructive operations,
+compatibility, security, concurrency, idempotency, or recovery**, favor durable
+tests for realistic failures whose consequences warrant protection. Identify
+the failure and the existing contract it would violate. These categories call
+for attention, not an automatic test inventory or permission to strengthen the
+application's guarantees. Account for protection already provided at other layers.
 
 Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
 
@@ -85,11 +92,7 @@ Do not force all behavior through oversized end-to-end tests. Do not test every
 
 ## Test behavior, not implementation
 
-A test should protect a decision, contract, or invariant—not memorialize the current implementation.
-
-Before adding or retaining a test, ask:
-
-> What realistic defect would this test catch?
+A test should protect a decision, contract, or invariant. A test must not memorialize the current implementation.
 
 A test is suspect when its main purpose is to detect that someone:
 
@@ -102,8 +105,6 @@ A test is suspect when its main purpose is to detect that someone:
 
 Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
 
-A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
-
 ## Expected effects of different changes
 
 Use the following expectations when evaluating test failures and test maintenance:
@@ -121,10 +122,11 @@ A test failing is not the same as a test needing to be edited. Many tests may co
 
 Configurable thresholds and defaults must not be duplicated throughout the test suite.
 
-For example, do not encode an internal concurrency limit indirectly:
+For this example, assume the selected contract rejects a request to start more
+processes than a configured limit. Do not encode its production default indirectly:
 
 ```go
-// Production policy:
+// Production default:
 const maxConcurrency = 4
 
 // Brittle test:
@@ -142,25 +144,30 @@ require.NoError(t, runner.Start(limit))
 require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
 ```
 
-The test should prove:
+For this selected contract, the test should prove:
 
 - the configured limit is accepted; and
 - one beyond the configured limit is rejected.
 
+Rejection is not a general requirement for concurrency control. If the selected
+contract queues excess work or limits active workers while accepting larger
+batches, test that behavior instead. Do not add rejection or a configurable
+limit solely to reproduce this example.
+
 The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
 
 Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
 
-For concurrency limits, test both kinds of behavior when relevant:
+For established concurrency limits, distinguish these behaviors where the contract requires them:
 
-1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+1. **Configuration handling:** values are accepted or rejected according to the selected contract.
 2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
 
 Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
 
 ## Avoid semantic duplication across layers
 
-Each behavior should have a clear test owner.
+Each behavior should have a clear test owner. Where these layers exist:
 
 - Parser tests own parsing cases.
 - Validator tests own validation rules.
@@ -172,8 +179,6 @@ Each behavior should have a clear test owner.
 
 Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
 
-Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
-
 ## Use test doubles deliberately
 
 Choose the least elaborate test double that provides the required control or observation.
@@ -197,15 +202,16 @@ Do not use mocks merely to isolate every object or reproduce the implementation'
 
 ## Go-specific guidance
 
-Use:
+Select techniques according to the behavior and risk being tested. The following
+are options, not a checklist that every package or application must satisfy:
 
 - table-driven tests for meaningful behavioral categories and boundaries;
 - `t.TempDir()` for real filesystem behavior;
 - `httptest.Server` for realistic HTTP interactions;
-- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- fuzz tests when exploring broad input spaces, such as parsers or path handling, adds useful protection beyond representative cases;
 - golden files only when the complete output is intentionally stable;
 - integration tests where correctness depends on component interaction; and
-- a small number of representative end-to-end tests.
+- a small number of representative end-to-end tests when assembled workflows expose meaningful risks not adequately covered at narrower boundaries.
 
 Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
 
@@ -223,8 +229,6 @@ Use it to find untested critical branches and unexpectedly weak packages. Do not
 
 Pure domain logic will often warrant higher coverage than CLI wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
 
-Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
-
 ## Regression tests
 
 A bug fix should normally include a regression test that fails before the fix and passes afterward.
@@ -253,44 +257,26 @@ Strong candidates include tests that:
 
 Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
 
-Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
-
 ## Reviewing a proposed test
 
 Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
 
-1. What realistic defect would it catch?
-2. How likely is that defect?
-3. How consequential would it be?
-4. Is the behavior already protected elsewhere?
-5. At which layer should this behavior be owned?
-6. Does the test assert a durable contract or an incidental implementation detail?
-7. Could the implementation be refactored without changing the behavior and without editing this test?
-8. What should cause this test to fail?
-9. What legitimate changes should not cause this test to fail?
-10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
-11. Is there a smaller or more direct test that protects the same risk?
-
-Do not add the test when its expected lifetime cost exceeds its expected protective value.
-
-When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+1. Which requested behavior or established contract does it protect, and what realistic defect would violate it?
+2. How likely and consequential is that defect, and is it already protected elsewhere?
+3. What is the narrowest stable boundary that can test it clearly?
+4. Would a correct internal refactor or a change to an incidental default require test edits?
+5. Does its confidence justify its maintenance, execution, and diagnostic cost, or would a smaller test suffice?
 
 ## Definition of sufficient
 
 A test suite is sufficient when:
 
-- important contracts and invariants are protected;
-- meaningful boundaries and failure modes are exercised;
-- realistic and consequential regressions are credibly protected against silent recurrence;
-- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
-- important external boundaries have realistic integration coverage;
-- representative complete workflows are tested;
+- important requested behavior and established contracts are protected against realistic, consequential failures;
+- boundaries and component interactions have coverage where they introduce meaningful risks not protected elsewhere;
 - failures provide useful signal rather than redundant noise;
 - legitimate internal changes usually do not require test edits; and
 - additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
 
-Sufficiency is a risk judgment, not a coverage percentage or test count. Reassess it as the application, its users, and the consequences of failure evolve.
-
-The governing rule is:
-
-> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.
+Sufficiency is a risk judgment, not a coverage percentage, test count, or required
+mix of testing techniques. Reassess it as the application, its users, and the
+consequences of failure evolve.

docs/release.md

@@ -1,164 +1,84 @@
 # Source Releases
 
-This procedure is for maintainers publishing Notarius source releases. A
-release is an immutable lightweight `vMAJOR.MINOR.PATCH` tag on `main` together
-with its checked-in `docs/releases/<tag>.md` note. Tag CI validates that source
-candidate after publication; it does not publish or repair a release.
+This is the executable runbook for [Release Policy](policy/release.md).
+Notarius publishes an immutable lightweight tag and its checked-in release note,
+not binaries or Gitea release objects. Linux is supported; macOS is best-effort
+for development, and Windows is unsupported. `main` on `origin` is authoritative.
 
-Notarius publishes no binaries, archives, checksums, signatures, containers,
-package-manager entries, or Gitea release objects. Windows is not supported.
-Do not create retrospective notes for the pre-procedure `v0.1.0`, `v0.2.0`, or
-`v0.3.0` tags.
+## Prepare
 
-## Select And Describe The Release
-
-Choose an unused stable semantic version in the form `vMAJOR.MINOR.PATCH`.
-Prereleases are not supported. Before `v1.0.0`, a minor release may change a
-documented CLI, configuration, durable artifact, integration, or operating
-contract when its note explains the impact and required operator action. A
-patch release must not intentionally break those documented contracts within
-its minor line.
-
-Create the version-matched note as part of the candidate. Every new note uses
-this structure, with concise, truthful content in each section:
-
-```markdown
-# Notarius vMAJOR.MINOR.PATCH
-
-This release ...
-
-## Summary
-
-## Compatibility
-
-## Upgrade
-
-## Changes
-```
-
-The note is a historical summary. Link to current canonical documentation for
-exact behavior, and update that documentation in the candidate rather than
-using the note as a substitute.
-
-## Prepare The Candidate
-
-Set the selected release version and disable Go workspace use for every
-candidate command:
+Use Git, a POSIX shell, and Go 1.25.5 or newer. Configure ordinary Git credentials
+for `origin` and private-module access without putting secrets in the repository.
+Prepare dependencies before offline validation where necessary:
 
 ```sh
-RELEASE_VERSION=vMAJOR.MINOR.PATCH
-export RELEASE_VERSION GOWORK=off
+GOWORK=off GOPRIVATE=gitea.maximumdirect.net go mod download
 ```
 
-Run the shared source-candidate checks from the repository. They cover module
-hygiene, tests, race tests, vet, builds, formatting, whitespace, maintained
-configuration validation, and the Linux and Darwin command-build matrix:
+Choose an unused stable version under the policy's compatibility rules. Add
+`docs/releases/<version>.md`, starting with `# Notarius <version>`, with a concise
+summary and relevant compatibility/upgrade information. No fixed section list is
+required. Update current documentation with behavior changes; retire only
+completed roadmaps and preserve planned/deferred work. Commit the candidate on
+`main` after ordinary change review. There is no separate release sign-off.
+
+## Publish
+
+Run one command from the repository with the actual selected version:
 
 ```sh
-./scripts/check-release-source.sh "$RELEASE_VERSION"
+./scripts/release.sh vMAJOR.MINOR.PATCH
 ```
 
-Before committing, manually follow every changed local Markdown link and
-review the candidate for unintended files, generated output, credentials, or
-other unrelated changes. Commit the release note and all affected current
-documentation, then run the shared checker against that exact candidate. Push
-the candidate commit to `main` only after it succeeds. Record the exact commit
-only after that push:
+The command checks version availability on `origin`, branch and worktree state,
+and the matching note. It runs the complete offline default test suite once and
+a native CLI build/version check with `GOWORK=off`. The
+[gate rationale](policy/release.md#notarius-gate-selection) explains the selected
+checks. It then rechecks candidate and publication state, pushes the exact
+candidate to `main` without force, creates a lightweight tag, and pushes only
+that tag. It neither rewrites history nor publishes other local tags.
+
+A successful tag push completes the source release. The command does not wait
+for CI or verify consumer installation/module availability. Tag CI is advisory
+validation only; there are no asynchronous deliverables to await.
+
+## Failure And Recovery
+
+Before publication, correct failed checks, commit any changes, and rerun the
+command. If the candidate changes during validation, validation must run again
+on the new candidate. A failed branch push publishes no tag.
+
+On a failed or ambiguous push, the command queries the affected remote ref. If
+a tag push reports failure but the intended lightweight tag is confirmed on the
+remote, publication succeeded. Otherwise it stops and reports failure or unknown
+status, leaving any local tag intact for inspection. Do not blindly rerun, force
+push, or delete a possibly published tag.
+
+Inspect authoritative state before retrying:
 
 ```sh
-RELEASE_COMMIT=$(git rev-parse 'HEAD^{commit}')
-export RELEASE_COMMIT
+git ls-remote origin refs/heads/main refs/tags/vMAJOR.MINOR.PATCH
+git rev-parse vMAJOR.MINOR.PATCH
 ```
 
-For private-module installation, configure standard `GOPRIVATE` matching this
-module and ordinary Git authentication for the hosting service before running
-the verification below. The exact authentication mechanism belongs to the
-maintainer environment; never record credentials or environment dumps in a
-release note, command history, or repository file.
+If the intended tag exists remotely, leave it immutable. If it is absent and
+you have confirmed that the local tag was never published or used by a release
+channel, remove only that unpublished local tag, then rerun the command. A
+failed remote query is not proof of absence. Network/authentication failures
+alone do not require a new version. A defect requiring changes to published
+source requires a new semantically appropriate version and matching note.
 
-## Guard And Publish The Tag
+## Optional Diagnostics
 
-Fetch current remote references, then run this guard without editing the
-candidate. It requires `main`, a clean worktree and index, disabled workspace
-use, a stable release version, the recorded and pushed commit, a matching note,
-and unused local and remote tags:
+`./scripts/check-release-source.sh <version>` runs validation without publication.
+It is useful for diagnosis, but is not a required preliminary step that must
+repeat the publication command's checks.
 
-```sh
-git fetch origin main --tags
+Race tests, vet, module hygiene, formatting, example checks, and cross-builds may
+be run when relevant to a change; they are not additional release gates. The
+development guide and testing policy govern change-specific verification.
 
-if ! printf '%s\n' "$RELEASE_VERSION" |
-  grep -E -x 'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)' >/dev/null
-then
-  printf '%s\n' "invalid release version: $RELEASE_VERSION" >&2
-  exit 1
-fi
-test "$GOWORK" = off
-test "$(git branch --show-current)" = main
-test -z "$(git status --porcelain)"
-test "$RELEASE_COMMIT" = "$(git rev-parse 'HEAD^{commit}')"
-test "$RELEASE_COMMIT" = "$(git rev-parse 'origin/main^{commit}')"
-test -s "docs/releases/$RELEASE_VERSION.md"
-grep -F -x "# Notarius $RELEASE_VERSION" "docs/releases/$RELEASE_VERSION.md"
-for heading in '## Summary' '## Compatibility' '## Upgrade' '## Changes'; do
-  grep -F -x "$heading" "docs/releases/$RELEASE_VERSION.md"
-done
-if git rev-parse -q --verify "refs/tags/$RELEASE_VERSION" >/dev/null; then
-  printf '%s\n' "local tag already exists: $RELEASE_VERSION" >&2
-  exit 1
-fi
-if git ls-remote --exit-code --tags origin "refs/tags/$RELEASE_VERSION" >/dev/null 2>&1; then
-  printf '%s\n' "remote tag already exists: $RELEASE_VERSION" >&2
-  exit 1
-fi
-```
-
-Create an explicitly lightweight tag against the guarded commit, verify its
-target, and push only that tag ref:
-
-```sh
-git -c tag.gpgSign=false tag "$RELEASE_VERSION" "$RELEASE_COMMIT"
-test "$(git cat-file -t "$RELEASE_VERSION")" = commit
-test "$(git rev-parse "$RELEASE_VERSION^{commit}")" = "$RELEASE_COMMIT"
-git push origin "refs/tags/$RELEASE_VERSION:refs/tags/$RELEASE_VERSION"
-```
-
-Never use `git push --tags`, move a published tag, or delete a published tag.
-
-## Verify The Published Release
-
-Confirm that the remote tag still points at the guarded commit and that the
-note is available from the tagged tree:
-
-```sh
-REMOTE_TAG_COMMIT=$(git ls-remote origin "refs/tags/$RELEASE_VERSION" | awk '{print $1}')
-test "$REMOTE_TAG_COMMIT" = "$RELEASE_COMMIT"
-git show "$RELEASE_VERSION:docs/releases/$RELEASE_VERSION.md" >/dev/null
-```
-
-Verify a fresh source installation and its diagnostic version. The temporary
-directory confines the installed command to this check:
-
-```sh
-release_verification_dir=$(mktemp -d)
-trap 'rm -rf "$release_verification_dir"' 0 HUP INT TERM
-mkdir -p "$release_verification_dir/bin"
-GOWORK=off GOBIN="$release_verification_dir/bin" go install \
-  "gitea.maximumdirect.net/eric/notarius/cmd/notarius@$RELEASE_VERSION"
-test "$("$release_verification_dir/bin/notarius" --version)" = "notarius $RELEASE_VERSION"
-```
-
-An exact fresh checkout and `GOWORK=off go build ./cmd/notarius` is an
-equivalent source verification when local installation policy requires it.
-`notarius --version` is diagnostic only; downstream compatibility remains
-defined by the published receipt and artifact contracts.
-
-## Failure And Correction Policy
-
-If candidate validation fails before publication, fix the candidate on `main`,
-rerun the shared checker, and repeat the guards. An unpublished local tag may
-be deleted after inspection.
-
-If the remote tag or tag CI reveals a defect, leave the published tag intact.
-Fix the defect on `main`, choose a new patch version, write a new matching
-note, and repeat this procedure. Do not weaken tag immutability or add release
-assets as a workaround.
+A fresh source install or exact-tag checkout/build may be used to investigate
+consumer availability after release. Such checks are optional and do not delay
+or reverse release completion. `notarius --version` is diagnostic; subprocess
+compatibility is defined by the receipt and artifact contracts.

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/releases/v0.7.0.md

@@ -0,0 +1,88 @@
+# Notarius v0.7.0
+
+This release simplifies configuration management with ConfigKit and makes
+cache access permissions a deployment responsibility.
+
+## Summary
+
+Notarius adopts ConfigKit v0.3.0 for strict configuration loading, imports,
+configuration profiles, provenance, and redacted inspection. Operators can
+compose whole pipeline definitions from separate files and select development
+overlays without duplicating the base configuration.
+
+Configured relative filesystem paths now resolve against their declaring
+files, including imported definitions and profile replacements. The effective
+configuration and its explanation reflect the paths used at runtime.
+
+## Compatibility
+
+- Configuration version 5 replaces version 4; there is no compatibility loader.
+  Numeric `version: 5` is the canonical spelling; quoted `"5"` also works.
+- Module and validator bindings require objects with `module:`; external
+  references require objects with `path:`. Lane-level reference aliases are
+  removed; use binding-local references instead.
+- Strict ConfigKit YAML parsing rejects aliases, merge keys, and unsupported
+  scalar forms. See the [configuration reference](../config.md) for the
+  complete contract and migration guidance.
+- The seven operational environment overrides are removed:
+  `NOTARIUS_TOTAL_LLM_CONCURRENCY`, `NOTARIUS_STAGE_WORKERS_EXTRACT`,
+  `NOTARIUS_OUTPUT_DIR`, `NOTARIUS_CACHE_CHUNK_PLANS_MODE`,
+  `NOTARIUS_CACHE_CHUNK_PLANS_DIR`, `NOTARIUS_CACHE_CHECKPOINTS_DIR`, and
+  `NOTARIUS_DEBUG_DIR`. `NOTARIUS_CONFIG` still selects the configuration
+  file; PromptKit credential environment variables remain supported.
+- Configured relative reference, PromptKit profile, output, cache, and debug
+  paths use the canonical declaring file's directory. Built-in relative
+  defaults use the main configuration directory. CLI-relative paths still use
+  the invocation directory; absolute paths are unchanged.
+- Requested debug bundles now contain `effective-config.yml` instead of
+  `effective-config.json`. Inspection output is redacted and is not promised
+  to be reloadable configuration.
+- Chunk-plan and checkpoint cache writers respect deployment permissions
+  through ordinary creation modes, umask, setgid inheritance, and default ACLs;
+  they no longer impose private permissions. Existing entries are not chmodded
+  or migrated. Output and debug access contracts are unchanged.
+- Durable extraction artifact schemas and subprocess receipt formats are
+  unchanged by this release.
+
+## Upgrade
+
+1. Migrate deployed configurations to version 5 using the
+   [configuration migration guide](../config.md#version-5-migration). Expand
+   shorthand bindings and references, and move retired environment settings
+   into YAML or supported CLI overrides.
+2. Review relative paths, particularly output/cache/debug directories and
+   PromptKit profile sources. Use absolute paths when preserving an existing
+   deployment location is important.
+3. Review deployment access settings for cache directories. Provision umask,
+   groups, and ACLs as appropriate; Notarius does not choose the installation's
+   access policy. See [Operations](../operations.md).
+4. Use `config validate --config <path> --pipeline <id>` for each pipeline,
+   including selected configuration profiles. Inspect resolved settings with
+   `config effective` and their provenance with `config explain`; see the
+   [CLI reference](../cli.md).
+5. Update any debug-bundle tooling that consumes the former effective JSON
+   file. Subprocess callers may optionally select an overlay through
+   `--config-profile`; this is separate from PromptKit's `--llm-profile`.
+
+The maintained [minimal](../../examples/dnd-minimal.config.yml) and
+[complete](../../examples/dnd-complete.config.yml) D&D examples demonstrate
+self-contained configuration and composed pipelines with a development profile.
+Build from the source tag using the [source-release procedure](../release.md).
+
+## Changes
+
+- Replaced custom YAML parsing and environment-overlay machinery with a
+  ConfigKit-backed typed document boundary and detached runtime conversion.
+- Added imports, configuration profiles, source-aware normalization, safe
+  effective/explanation commands, and source-file legends for load errors.
+- Preserved explicit zeroes and empty validator overrides at supported stages,
+  rejected invalid reference branches before conversion, and retained
+  whitespace normalization for worker identifiers.
+- Corrected normalization recording for ordered steps and validator sequences,
+  with regression coverage for runtime values and provenance.
+- Accepted ConfigKit's integer representation at module-option boundaries
+  through checked shared conversion.
+- Updated examples, configuration and subprocess guidance, and debug rendering.
+- Reworked cache creation to respect deployment-selected permissions while
+  retaining confined paths, symlink rejection, atomic publication, validation,
+  and recovery.

docs/releases/v0.7.1.md

@@ -0,0 +1,44 @@
+# Notarius v0.7.1
+
+This release updates the prompt engine and its provider catalogs while retaining
+Notarius's application-owned D&D defaults and subprocess contracts.
+
+## Summary
+
+- Upgrade PromptKit from v0.9.0 to v0.10.0 and adopt its selected provider catalogs.
+- Refresh the profile-source checkpoint identity for the updated catalog.
+- Document profile ownership and upstream catalog compatibility.
+
+## Compatibility
+
+Notarius's CLI, configuration schema, durable artifact formats, and subprocess
+receipt are unchanged. The `dnd-extraction` fallback retains its explicit model
+and settings; the upgrade does not switch it to an upstream default profile.
+
+Upstream catalog selections require review: `deepseek-3-2`, `deepseek-4-flash`,
+and `deepseek-4-pro` are no longer present. Operators selecting or inheriting
+those profiles must supply a complete operator-owned definition or select an
+available profile. `deepseek-flash-latest` and `deepseek-pro-latest` are not
+equivalent aliases and need not preserve previous settings. See the
+[PromptKit integration contract](../integrations/pkg-promptkit.md#profile-sources-and-compatibility).
+
+Existing checkpoints with the previous profile-source identity are not reused.
+Source-keyed chunk plans and existing output artifacts are unaffected.
+
+## Upgrade
+
+Review selected LLM profile IDs and inherited bases before rebuilding from this
+tag. No Notarius configuration migration is required for the maintained examples
+or application-owned D&D defaults. For controlled model settings, use a complete
+[operator profile](../config.md#promptkit-profiles), rather than inheriting a
+volatile catalog entry. Follow the [source release procedure](../release.md)
+for source installation and verification.
+
+## Changes
+
+- PromptKit v0.10.0 preserves the public engine API and adds two opt-in maintained
+  profile names; Notarius continues using its existing adapter and defaults.
+- The dependency graph selects OpenRouter catalog v2.20260922.1 and Rakestrawhome
+  catalog v2.20260922.0.
+- Profile checkpoint fingerprinting and current integration documentation now
+  identify the updated engine.

docs/releases/v0.8.0.md

@@ -0,0 +1,31 @@
+# Notarius v0.8.0
+
+This release upgrades ConfigKit and PromptKit and makes provider retry behavior
+explicit, inspectable, and available in debug diagnostics.
+
+## Compatibility
+
+Configuration selection now uses `--config-overlay` and
+`overlays/<name>.yml`. The retired `--config-profile` flag and
+`profiles/<name>.yml` lookup are no longer accepted. PromptKit LLM profiles are
+unchanged.
+
+`promptkit.transient_retries` defaults to two additional eligible transient
+HTTP attempts per generation. Set it to `0` to disable built-in transient
+retries. Existing pipeline and validator retry settings remain separate.
+
+## Upgrade
+
+Move selected configuration files from `profiles/` to `overlays/` and replace
+every `--config-profile` invocation with `--config-overlay`. Review retry and
+timeout budgets before deployment, especially where providers charge per
+attempt. See [Configuration](../config.md), [Operations](../operations.md), and
+the [PromptKit integration contract](../integrations/pkg-promptkit.md).
+
+## Changes
+
+- Upgrade ConfigKit to v0.4.0 and PromptKit to v0.11.0.
+- Add application-owned generation failure and request-local execution-summary
+  metadata, including retry exhaustion details.
+- Record execution summaries in existing debug response artifacts and safe
+  per-call references.

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/future.md

@@ -10,77 +10,13 @@ not as committed release dates.
 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. The remaining near-term work applies those completed foundations
-to domain review and operator-facing diagnostics.
+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.
 
-### 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.
-
-### 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

@@ -0,0 +1,403 @@
+# ConfigKit And PromptKit Upgrade Implementation Plan
+
+## Purpose And Scope
+
+Implement [ConfigKit And PromptKit Upgrades](upgrade-kit.md). That document
+owns the desired behavior and operator migration; this document owns execution
+order. Each numbered stage is one bounded implementation prompt for a
+gpt-5.6-terra coding agent. Complete and validate stages in order.
+
+The inspected baseline pins ConfigKit v0.3.0 and PromptKit v0.10.0. Adopt
+exactly ConfigKit v0.4.0 and PromptKit v0.11.0, not whatever release happens to
+be newest during implementation. The release guides and public APIs are
+available in the sibling checkouts `../ConfigKit` and `../promptkit`; consult
+the target tags or downloaded modules if those checkouts have advanced.
+Build and test Notarius with `GOWORK=off`, without local dependency replacements.
+
+Read `docs/development.md` and all five policies in `docs/policy/` before
+implementation. In particular, simplify machinery without narrowing operator
+choices or elevating incidental details into requirements. Preserve existing
+contracts unless this roadmap expressly changes them. Use graph discovery for
+code, then inspect the relevant source and tests; locations below identify the
+reviewed baseline, not a requirement to preserve private file structure.
+
+## Decisions And Boundaries
+
+- Configuration selection becomes `--config-overlay` and `overlays/<name>.yml`,
+  with no legacy alias or fallback. Keep empty, omitted, and `default` selection
+  behavior. Do not rename PromptKit profiles or pipeline/runtime profile types.
+- Keep configuration version 5, source-file-relative declarations, CLI override
+  precedence, strict loading, inspection provenance, and redaction.
+- Add only the global `promptkit.transient_retries` setting: nonnegative integer,
+  default **2 additional attempts**, explicit zero disables. No arbitrary cap,
+  per-binding setting, new CLI retry flag, or new total-deadline setting.
+- PromptKit owns transient HTTP retries and structural repairs; Notarius keeps
+  stage/validator retry budgets and disposition unchanged. The existing shared
+  scheduler retains its completion-scoped permit through upstream waiting.
+- Add provider-neutral failure metadata and request-local execution summaries.
+  Do not change domain artifacts, result receipts, prompt content, validators,
+  acceptance policy, or introduce a new telemetry service/artifact.
+- Successful retry recovery is ordinary success, not a warning. Debug remains
+  opt-in. Existing token fields keep their meaning; observed usage is separate.
+- No release, tag, push, source-license change, or binary packaging is part of
+  these stages. Record upgrade implications in current integration/operator
+  documentation; the eventual release note is a separate release task.
+
+For each stage, inspect the starting worktree, implement code with its focused
+tests and current-behavior documentation, run the listed relevant checks, and
+make one coherent local commit under agent policy. Report checks and any gaps.
+Do not repeat passed checks for unchanged content merely to make a commit.
+
+## Stage 1 — Cut Over Configuration Selection To Overlays
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Adopt ConfigKit v0.4.0 and finish its narrow operator-facing rename in
+one coherent change. This is one prompt: it is a naming migration, not a new
+configuration engine.
+
+**Inspect:** `internal/core/config/load.go`, `loaded_config.go`, configuration
+document/path/operational tests; `internal/cli/run.go`,
+`configkit_commands.go`, CLI configuration and example contract tests;
+`examples/profiles/development.yml`; upstream v0.4.0 migration guide.
+
+**Work:**
+
+1. Upgrade ConfigKit to v0.4.0 in `go.mod`/`go.sum`. Change `LoadOptions.Profile`
+   to `Overlay` and pass it into `configkit.Options[Document].Overlay`. Rename
+   only helpers, fields, locals, fixtures, and descriptions representing the
+   selected configuration layer.
+2. Replace `--config-profile` with `--config-overlay` for `run`, `config validate`,
+   `config effective`, `config explain`, and `pipelines list`. Update usage,
+   flag parsing, argument classification, explicit-empty handling, and error
+   paths together. Retired flags use normal unknown-flag behavior.
+3. Move `examples/profiles/development.yml` to
+   `examples/overlays/development.yml`; update maintained-example discovery and
+   selected-layer fixtures. Check file-relative paths after moving. Leave
+   `examples/llm-profiles/`, PromptKit profile sources, and ordinary explicit
+   imports alone unless an import actually points at the moved example.
+4. Consume upstream overlay provenance and diagnostic names directly. Do not
+   rebuild composition, implement directory fallback, or translate metadata
+   back to profile vocabulary.
+5. Update `docs/config.md`, `docs/cli.md`, `docs/integrations/pkg-configkit.md`,
+   `docs/internal/configuration.md`, relevant portions of `docs/internal/cli.md`,
+   and `docs/consumers/subprocess.md`. Review README, operations, and other
+   consumer examples for affected invocations. Explain the clean migration and
+   distinction from LLM profiles. Preserve accepted ADR-0018 and release history.
+
+**Acceptance:** All five commands select overlays consistently. Representative
+existing tests cover selected/base/empty/default behavior, overlay and explicit
+override precedence, declaration-relative paths, redacted effective/explain
+output, and source-aware failure labels. Include a selected legacy-only file
+case proving there is no silent fallback. Retain ordinary-import behavior and
+LLM profile coverage; do not duplicate every upstream parser case.
+
+**Validate:** `GOWORK=off go test ./internal/core/config ./internal/cli`,
+`GOWORK=off go mod tidy -diff`, `git diff --check`. Review occurrences of
+`config-profile` and `examples/profiles` in current docs/code semantically;
+historical docs and explicit legacy-rejection tests are not defects.
+
+## Stage 2 — Upgrade The PromptKit Runtime Baseline
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Adopt the compatible upstream dependency and document its changed
+default before adding Notarius controls. One small prompt.
+
+**Inspect:** `internal/framework/llm/promptkit_client.go`, asset/profile identity
+helpers and tests, `go.mod`, and upstream v0.11.0 release/consumer documentation.
+
+**Work:**
+
+1. Pin PromptKit v0.11.0 and tidy. Keep the Go version and provider catalog
+   selections unless the actual dependency graph requires a change; the target
+   release does not require either. Do not alter embedded D&D model settings.
+2. Update `promptKitBuiltinProfileCatalogID` to the v0.11.0 identity in accordance
+   with the existing checkpoint fingerprint convention. No other cache policy
+   change or blanket invalidation is part of this work.
+3. Update the pinned integration links in `docs/integrations/pkg-promptkit.md`,
+   including the relocated upstream `docs/policy/release.md`. Record the
+   LGPL-3.0-or-later dependency license and reference its license/notices; do not
+   relicense Notarius or invent a new distribution workflow.
+4. Remove the obsolete no-provider-retry-loop statement in `docs/internal/llm.md`.
+   Describe the now-active upstream default and per-attempt timeouts in the
+   appropriate integration/operations sections. Do not yet describe the setting
+   or observations from later stages as implemented.
+
+**Acceptance:** The prepared-execution adapter builds and its existing tests
+pass without prompt/catalog changes. Current documentation acknowledges the
+dependency's default retry behavior and license accurately.
+
+**Validate:** `GOWORK=off go test ./internal/framework/llm ./internal/cli`,
+`GOWORK=off go mod tidy -diff`, `git diff --check`.
+
+## Stage 3 — Wire The Explicit Transient Retry Setting
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Make retry policy inspectable and configurable through the one shared
+engine. One prompt covering a single vertical configuration slice.
+
+**Inspect:** `internal/core/config/{document,config,load,operational_document,
+loaded_config,validation}.go`, effective-config tests;
+`internal/cli/catalog.go:buildProductionLLMClient`;
+`internal/framework/llm/promptkit_client.go:NewPromptKitClient` and HTTP tests.
+
+**Work:**
+
+1. Add `TransientRetries *int` to `PromptKitDocument` and a resolved integer to
+   `PromptKitConfig`. Supply 2 through both the document default path and the
+   runtime `Default()` path. Preserve explicit zero; never use zero as omission.
+   Include it in document-to-runtime conversion and effective configuration.
+2. Reject negative values through both existing document and direct runtime
+   validation entry points. Follow current source-aware diagnostics; do not
+   impose a new maximum. Keep ConfigKit's existing decoding behavior.
+3. Add `TransientRetries *int` to `PromptKitClientConfig`: nil follows upstream
+   default, non-nil includes explicit zero. Production construction always
+   supplies the resolved configuration value. Pass it into `promptkit.Config`,
+   not per-request options. All calls through the injected shared client thus
+   share the setting without module changes.
+4. Add the setting to `examples/dnd-complete.config.yml` and `docs/config.md`;
+   keep the minimal example minimal. Update configuration/LLM internal guides
+   for the new plumbing and operations for the zero-value opt-out.
+
+**Acceptance/tests:** Demonstrate omitted default, explicit zero, a positive
+override, negative rejection, and effective/explain inspection. Use an existing
+CLI factory test to prove the resolved value reaches construction. Exercise
+built-in generation with a local HTTP server: eligible failure then success,
+disabled retries, exhausted configured count, and a noneligible 400 response.
+An injected `LLMClient` stub does not test built-in HTTP retry wiring. Avoid
+testing the upstream jitter formula or every equivalent HTTP status.
+
+**Validate:** `GOWORK=off go test ./internal/core/config ./internal/cli ./internal/framework/llm`
+and `git diff --check`.
+
+## Stage 4 — Preserve Typed Failure And Exhaustion Metadata
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Distinguish provider failures without coupling callers to PromptKit.
+One prompt; defer execution observations to the next stage.
+
+**Inspect:** `internal/framework/contracts/errors.go`, adapter error conversion
+in `promptkit_client.go`, redaction helpers, and existing pipeline handling of
+`ErrLLMGeneration`, `ErrLLMCapacityExceeded`, and invalid structured output.
+
+**Work:**
+
+1. Extend the existing application-owned `LLMGenerationError` rather than
+   creating a parallel provider-error hierarchy. Retain `StatusCode()` and
+   `errors.Is(err, ErrLLMGeneration)`. Add application-owned failure-kind,
+   exhaustion, and exhausted-generation attempt metadata/accessors; use uint64
+   for upstream attempt counts. Preserve the existing constructor's callers
+   with sensible unspecified/generation metadata, adding a metadata constructor
+   or private adapter helper as needed.
+2. Define kinds for `transport`, `provider_status`, `malformed_response`, and
+   generic `generation`. Exhaustion is an independent boolean and attempt count,
+   not a replacement kind. Zero attempts means unavailable in this error
+   projection, not proof no request happened. The count comes from
+   `RetryExhaustedError.Attempts()` and covers the exhausted generation, not all
+   repair generations combined.
+3. Use `errors.Is`/`errors.As` on upstream sentinels/types. Keep cancellation and
+   deadline identity first, capacity handling intact, and HTTP status available
+   through an exhaustion wrapper. Do not unwrap application errors to upstream
+   provider objects or classify by prose. Preserve sanitized diagnostic text.
+4. A malformed provider envelope is a generation failure; schema-invalid model
+   content stays `ErrInvalidStructuredOutput`. Do not change outer retry or
+   validator disposition logic. Document classification at the integration and
+   internal LLM boundaries.
+
+**Acceptance/tests:** Adapter tests cover transport, malformed response,
+noneligible status, exhausted eligible status with count/status preserved, and
+cancellation/deadline/capacity regression behavior. Check typed fields and
+identities, not full messages. Existing credential-redaction cases remain valid
+for nested exhaustion diagnostics; use fake secrets only.
+
+**Validate:** `GOWORK=off go test ./internal/framework/contracts ./internal/framework/llm ./internal/framework/pipeline`
+and `git diff --check`.
+
+## Stage 5 — Project Request-Local Execution Observations
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Return a lightweight terminal summary even when generation fails.
+One prompt focused on contracts and adapter mapping, without persistence changes.
+
+**Inspect:** `internal/framework/contracts/contracts.go`, `CompleteStructured`,
+`responseFromResult`, prepared execution tests, upstream `observations.go` and
+prepared-execution observer documentation.
+
+**Work:**
+
+1. Add an optional `Execution *LLMExecutionSummary` field to
+   `StructuredCompletionResponse` (`execution,omitempty`). Own these DTOs in
+   Notarius contracts; do not alias or serialize upstream event types.
+2. The summary contains uint64 `attempts`, `transient_retries`, `repair_calls`;
+   integer millisecond `capacity_wait_ms`, `generation_ms`, `retry_delay_ms`,
+   `duration_ms`; `outcome`, `failure_kind`, `status_code`, `retries_exhausted`;
+   and `usage`. Usage contains prompt/completion/total/cached/cache-write counts,
+   each with `value`, `known`, `complete` fields. Do not omit these booleans or
+   known-zero values. The summary needs no prompts, response text, endpoints,
+   credentials, raw errors, or new copies of run/profile identifiers.
+3. Map terminal outcome/failure values explicitly to application-owned strings.
+   Preserve the distinctions exposed by v0.11.0 (including canceled, deadline,
+   capacity, invalid output, configuration/load/render failures); unknown future
+   failures map to `other`. Reuse relevant application failure vocabulary from
+   Stage 4, without forcing every execution error into a generation error.
+4. Attach a local observer to the `RunRequest` before `PrepareExecution` freezes
+   it. Retain the `ExecutionFinished` projection only: upstream supplies aggregate
+   counts, durations, and usage. Do not sum attempt events and terminal totals,
+   retain an event log, or reconstruct a second accounting system. Synchronous
+   request-local collection needs no global state or asynchronous queue.
+5. Attach that summary to every subsequent return from `RunPrepared`, including
+   context/capacity/provider failure and post-result decode/validation failure.
+   Preparation failure has no execution event and returns no summary. Preserve
+   upstream execution outcome as such: a later Notarius decode error remains a
+   separate returned error, not a fabricated upstream terminal event.
+6. Keep existing successful-result token and repair fields unchanged. Observed
+   `repair_calls` represents execution activity and is not a renamed legacy
+   `RepairAttempts`. No observation field is a billable-cost guarantee.
+
+**Acceptance/tests:** Success, failed execution, cancellation, and no-observation
+preparation failure retain the right presence. Test known zero versus unknown
+and partial aggregate usage with controlled provider responses. One concurrent
+completion case proves no summary cross-contamination. Check relationships and
+nonnegative durations, not exact timing. Document the DTO and mapping in
+`docs/internal/llm.md`; public dependency semantics stay in pkg-promptkit.
+
+**Validate:** `GOWORK=off go test ./internal/framework/contracts ./internal/framework/llm`
+and `GOWORK=off go test -race ./internal/framework/llm`.
+
+## Stage 6 — Carry Summaries Through Existing Debug Paths
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Make observations useful to operators without a new output channel.
+One prompt, limited to debug propagation and terminal allowlists.
+
+**Inspect:** `internal/framework/pipeline/debug.go`, its response-content/fallback
+helpers, terminal recording tests, and `debugLLMScope.references`. Trace
+`debugLLMCallReference` through the existing timed stage/terminal envelopes;
+updating only the detailed response file is insufficient. The CLI's separate
+run-summary files do not need a new run-wide accounting mechanism.
+
+**Work:**
+
+1. Project the optional execution summary into existing debug response artifacts
+   and per-call references used by stage/terminal records. Explicitly preserve it
+   when `Debug.Response` or `RunResult` is absent and through content splitting
+   and fallback construction. Keep copies detached if existing clone paths
+   could otherwise share mutable pointers.
+2. Use explicit typed fields as the allowlist for this metadata-only terminal
+   projection. Do not carry raw upstream events, newly copied provider diagnostics, or model
+   content into safe terminal summaries. Preserve established redaction and
+   detailed-trace content boundaries.
+3. Leave warnings, normal console output, receipts, domain artifacts, checkpoint
+   payloads, and no-debug behavior unchanged. No new telemetry file or aggregation
+   of observed token totals into legacy totals.
+4. Update `docs/internal/state.md`/relevant debug internals and `docs/operations.md`
+   to identify where this diagnostic information is available. Link to the
+   canonical explanation rather than repeating every field in consumer docs.
+
+**Acceptance/tests:** Through the real debug wrapper and existing recorder seam,
+verify success and failed/canceled completions retain summaries in response and
+safe terminal records. Verify content splitting/fallback does not discard them,
+and debug disabled writes nothing. Reuse existing redaction tests, adding only
+the new projection assertions. No new warnings result from recovered retries.
+
+**Validate:** `GOWORK=off go test ./internal/framework/pipeline ./internal/cli`,
+`GOWORK=off go test -race ./internal/framework/pipeline`, `git diff --check`.
+
+## Stage 7 — Verify Retry Composition And Finish Operator Guidance
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Validate the cross-boundary behavior that individual stages cannot
+prove, then reconcile current documentation. One prompt; no generalized retry
+test framework or runtime redesign.
+
+**Inspect:** Existing adapter HTTP fixtures, scheduler cancellation tests,
+pipeline stage retry/debug tests, CLI production tests; current config, LLM,
+operations, and consumer documentation.
+
+**Work and acceptance:**
+
+- Add one representative local-HTTP scenario combining a recoverable transient
+  status with structurally invalid generated content followed by a successful
+  repair. Prove transient budgets apply independently per generation and
+  terminal attempt/repair/usage accounting is not doubled. Use small explicit
+  budgets rather than executing the illustrative 18-request worst case.
+- Exercise cancellation during provider-directed retry waiting without a long
+  real sleep: signal that the request was received, cancel a context while a
+  long Retry-After is pending, and wait with a bounded test timeout. Preserve
+  cancellation identity and the terminal observation. Do not build a second
+  backoff implementation or test exact random delays.
+- Review shared scheduler tests for completion-scoped permit ownership and
+  cancellation/release. Add a focused integration case only where existing
+  coverage does not protect that boundary across the new retry path. Keep
+  upstream backend capacity algorithms upstream.
+- Confirm a recovered generation passes through ordinary success without
+  producing process warnings. Preserve existing outer stage retries and semantic
+  feedback. Do not alter limits to hide budget multiplication.
+- Finish operations guidance: distinguish transient, structural, and stage/
+  validator budgets; give the 18-attempt single-producer illustration with its
+  assumptions; explain per-attempt timeout versus overall caller cancellation,
+  Retry-After waits, cost/latency implications, incomplete usage accounting, and
+  the outer scheduler holding its permit during waits. No new total deadline or
+  exactly-once guarantee is implied.
+- Review all canonical docs changed earlier for consistency. Keep exact fields
+  and defaults in config, flags in CLI, dependency semantics in integrations,
+  mechanism in internals, and operating implications in operations. Keep the
+  migration note about overlays distinct from unchanged LLM profiles. Retain
+  historical ADR/release content; no new ADR is needed for these upgrades.
+
+**Validate:** Run the focused adapter/pipeline/CLI tests and race tests covering
+modified concurrency paths. Review tests for unnecessary upstream duplication
+or exact-prose/change-detector assertions; consolidate rather than multiply them.
+
+## Stage 8 — Final Integration Verification And Handoff
+
+**Status:** Completed 2026-09-25.
+
+**Goal:** Verify the complete upgraded application and documentation against the
+feature roadmap. One final verification prompt; fix only defects within scope.
+
+**Work:** Review all feature acceptance items and staged diffs. Confirm no
+upstream types escaped the LLM boundary, no references to the retired config
+flag remain in current examples/help, no unintended LLM profile rename occurred,
+and no source-only/license/release scope expanded. Check edited relative links.
+Leave the roadmap available for the user's completion review; do not publish a
+release or delete planning documents automatically.
+
+**Validation commands:**
+
+```sh
+GOWORK=off go test ./...
+GOWORK=off go vet ./...
+GOWORK=off go build ./...
+GOWORK=off go mod tidy -diff
+GOWORK=off go test -race ./internal/framework/llm ./internal/framework/pipeline
+GOWORK=off go run ./cmd/notarius config validate --config examples/dnd-minimal.config.yml
+GOWORK=off go run ./cmd/notarius config validate --config examples/dnd-complete.config.yml
+GOWORK=off go run ./cmd/notarius config validate --config examples/dnd-complete.config.yml --config-overlay development
+git diff --check
+```
+
+Reuse already-passed checks for the same unchanged candidate where appropriate.
+The example/CLI tests own maintained examples; the commands above are a one-time
+migration smoke check, not new permanent release gates. No live model calls or
+real API keys are required. Report environment/dependency-access failures
+separately from implementation failures; do not claim blocked checks passed.
+
+**Completion criteria:** Both exact dependency pins are active; overlay migration,
+retry setting, typed errors, observations, debug propagation, tests, examples,
+and current documentation satisfy the feature roadmap. Report commits and
+checks, plus any residual limitation. Future release preparation should explain
+the breaking overlay cutover, retry default/opt-out, and upstream license change
+in a minor-release note, but is not part of this implementation.
+
+## Open Questions
+
+None. The feature roadmap supplies the policy choices needed for implementation.
+The DTO naming and stage boundaries above settle remaining implementation choices
+without adding new product restrictions or compatibility promises.

docs/roadmap/remediation.md

@@ -0,0 +1,123 @@
+# Kit Upgrade Review Remediation
+
+## Resolution
+
+Completed 2026-09-26. No production changes were needed.
+
+- `internal/framework/llm/promptkit_observation_test.go` now covers retries in
+  both initial and repair generations, aggregate versus legacy accounting,
+  unknown/known-zero/partial usage, retry-wait cancellation, exhausted execution,
+  preparation failure, and overlapping request isolation.
+- `internal/framework/pipeline/debug_execution_test.go` exercises the real
+  scheduled client and debug wrapper together. It verifies returned and persisted
+  summaries for provider errors and cancellation without a body, split debug and
+  fallback content, terminal references, and explicit usage-presence JSON fields.
+- Temporary regression probes confirmed that discarding failed responses in the
+  scheduler and collapsing usage presence into known zero cause the new tests
+  to fail. Those probes were reverted.
+- Focused LLM/pipeline tests, their race-enabled suites, and
+  `GOWORK=off go test ./...` passed. Existing no-debug and redaction coverage was
+  retained. Runtime budgets, scheduler policy, and output contracts are unchanged.
+
+The original finding below is retained as review context.
+
+## Review Scope
+
+Reviewed implementation commits `31b48d2` through `7ac9d94` against
+[the feature roadmap](upgrade-kit.md) and [the implementation plan](implementation.md).
+The review covered configuration selection/defaults, production client wiring,
+error translation, execution observation, scheduler/debug propagation, current
+documentation, and tests. No production model calls were made.
+
+The implemented architecture is consistent with the intended design: ConfigKit
+v0.4.0 owns composition and overlay provenance behind the configuration boundary;
+PromptKit v0.11.0 owns transient retries and structural repairs behind the LLM
+adapter. Notarius retains stage retry/disposition policy, completion-scoped
+scheduling, application-owned error/observation types, and opt-in debug output.
+No new runtime defect or unrequested restriction was established in this review.
+
+## Finding 1 — Complete The Missing Retry And Observation Verification
+
+**Priority:** Medium. This is a verification gap, not a demonstrated runtime
+failure. Stages 5–7 are marked complete, but their consequential cross-boundary
+acceptance tests were not implemented.
+
+### Evidence And Risk
+
+- `internal/framework/llm/promptkit_client_test.go:58` adds only a successful,
+  single-attempt observation assertion with positive prompt-token usage. It
+  does not exercise absent versus partial/known-zero usage, failed/canceled
+  execution summaries, preparation failure, or concurrent request isolation.
+- `TestPromptKitClientAppliesConfiguredBuiltInTransientRetries` covers basic
+  status recovery, explicit zero, exhaustion, and a noneligible status. It
+  discards the response, so does not protect terminal accounting. There is no
+  combined transient-retry/structural-repair scenario or cancellation while
+  waiting on Retry-After.
+- `internal/framework/llm/scheduled_client.go:40` now preserves a response when
+  the inner call fails. No corresponding test protects that behavior. Returning
+  an empty response again would silently remove precisely the failed-call
+  observations this feature is intended to retain.
+- `internal/framework/pipeline/debug.go:267` and `:277` add execution summaries
+  to response artifacts and call references, plus fallback/content-splitting
+  propagation. Neither the debug nor terminal-debug tests were extended to
+  exercise these fields. Passing existing tests does not prove that failed-call
+  metadata reaches persisted records.
+
+These are Notarius-owned integration risks, not a request to re-test PromptKit's
+backoff algorithm. Losing failed-call observations or misrepresenting incomplete
+usage would be difficult to detect in ordinary successful runs. A focused set
+of behavioral tests is justified under the testing policy.
+
+### Required Remediation
+
+Use existing test assets, local HTTP fixtures, scheduler wrappers, and debug
+recorders. Combine related assertions; no fixed test count or new test framework
+is required.
+
+1. Add an adapter scenario with a transient failure, a structurally invalid
+   completion, and a successful repair. Exercise retries in both generations
+   with small explicit budgets, and assert terminal attempts/retries/repair
+   counts without double-counting. Use controlled usage fields to distinguish
+   known zero, unknown, and partial aggregate usage. Preserve legacy usage
+   semantics independently.
+2. Cover failed and canceled completions, including cancellation during a
+   Retry-After wait. Verify caller cancellation identity, bounded completion,
+   and retained execution metadata. Use signaling and a generous bounded test
+   timeout, not exact sleep durations or jitter assertions. Cover a preparation
+   failure with no execution summary. Add a small concurrent-call check for
+   request-local summary isolation, which race tests alone do not prove.
+3. Exercise an error response through the real scheduled client and existing
+   debug wrapper/recorder seams. Assert that the summary reaches response
+   artifacts and terminal call references even without a provider response
+   body. Cover the content-splitting/fallback path with a representative response.
+   Reuse existing no-debug and redaction tests rather than duplicating their
+   whole inventories. Keep new summary assertions metadata-only.
+
+Retain existing scheduler policy, stage/validator budgets, cancellation
+disposition, warning semantics, and debug opt-in behavior. Do not introduce a
+total deadline, new retry caps, telemetry artifacts, or live API requirements
+to make these tests easier. Fix production code only if these checks reveal a
+violation of the already selected behavior.
+
+**Acceptance:** The added tests fail if failed responses lose their execution
+summary at an adapter/wrapper boundary, usage presence is collapsed into zero,
+or retry/repair totals are double-counted. They pass with the intended behavior
+and provide evidence for the previously unverified stage acceptance criteria.
+
+**Validation:** Run focused LLM/pipeline tests, race-enabled tests for those
+packages, and `GOWORK=off go test ./...`. No production credentials are needed.
+
+## Checks Completed During Review
+
+- `GOWORK=off go test ./...`
+- `GOWORK=off go vet ./...`
+- `GOWORK=off go build ./...`
+- `GOWORK=off go mod tidy -diff`
+- `GOWORK=off go test -race ./internal/framework/llm ./internal/framework/pipeline`
+- CLI configuration validation for the minimal and complete examples, including
+  the complete example with `--config-overlay development`.
+- `git diff --check`
+
+All passed. Existing green tests do not substitute for the missing behavioral
+coverage above. Temporary diagnostic test material used during review was
+removed; no application code was changed.

docs/roadmap/upgrade-kit.md

@@ -0,0 +1,337 @@
+# ConfigKit And PromptKit Upgrades
+
+## Purpose
+
+Adopt ConfigKit v0.4.0 and PromptKit v0.11.0 in one coordinated work set:
+
+- Consistently call ConfigKit's selected configuration layer an **overlay**,
+  distinct from PromptKit LLM profiles.
+- Adopt configurable transient HTTP retries, provider-neutral failure
+  classifications, and execution observations from PromptKit.
+
+Preserve application/framework boundaries and keep the two upgrades as distinct
+implementation workstreams. Neither requires a redesign of pipeline execution.
+
+This feature roadmap owns the target state and scope. The ordered work and
+stage-specific verification belong to the [implementation plan](implementation.md).
+Neither document authorizes release publication.
+
+## Current Baseline
+
+The reviewed baseline is Notarius v0.7.1 (`53d9284`):
+
+- `go.mod` pins ConfigKit v0.3.0 and PromptKit v0.10.0.
+- `internal/core/config/load.go` passes `LoadOptions.Profile` into
+  `configkit.Options[Document].Profile`.
+- `--config-profile` selects `profiles/<name>.yml` beneath the configuration
+  root for `run`, `config validate`, `config effective`, `config explain`, and
+  `pipelines list`.
+- The maintained selected-layer example is
+  `examples/profiles/development.yml`; configuration and CLI tests create
+  equivalent directories and exercise selection and inspection.
+- ConfigKit owns composition and inspection metadata. Notarius owns its
+  version 5 document, defaults, validation, source-relative path resolution,
+  runtime conversion, and CLI discovery.
+
+A dependency-only trial against v0.4.0 fails to compile at `load.go:33` because
+`configkit.Options.Profile` no longer exists. The review found no direct
+Notarius references to `LayerProfile`, `CodeProfile*`, or `LoadSecretsDir`.
+
+An isolated dependency-only trial against PromptKit v0.11.0 passes the complete
+Notarius test suite and `go vet ./...`. That proves baseline compatibility, not
+adoption of the new controls or observations. The current adapter does not
+configure transient retries or attach an execution observer. It classifies
+generation errors broadly and records successful-result usage and structural
+repair counts. The outer scheduler holds a permit for the whole completion.
+
+## Upstream Changes
+
+### ConfigKit v0.4.0
+
+The [ConfigKit v0.4.0 release guide](https://gitea.maximumdirect.net/eric/ConfigKit/src/tag/v0.4.0/docs/releases/v0.4.0.md)
+defines four breaking naming changes:
+
+| Surface | Previous | Target |
+| --- | --- | --- |
+| Selection API | `Options.Profile` | `Options.Overlay` |
+| Selected file | `profiles/<name>.yml` | `overlays/<name>.yml` |
+| Provenance layer | `LayerProfile` / `profile` | `LayerOverlay` / `overlay` |
+| Diagnostic identifiers | `CodeProfile*` / `profile.*` | `CodeOverlay*` / `overlay.*` |
+
+There are no legacy aliases or directory fallback. Explicit imports from a
+directory named `profiles` remain valid ordinary imports. Merge rules,
+precedence, lifecycle, source-origin APIs, minimum Go version, and YAML parser
+dependency are unchanged.
+
+The intervening v0.3.1 release makes the opt-in `LoadSecretsDir` API ignore
+dot-prefixed entries. Notarius does not use this API; adding secrets-directory
+bootstrap is outside this work.
+
+### PromptKit v0.11.0
+
+The [PromptKit v0.11.0 release guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/releases/v0.11.0.md)
+introduces additive runtime APIs and one changed default:
+
+- Built-in generation defaults to two additional transient HTTP attempts for
+  statuses 429, 500, 502, 503, and 504. Engine and request controls can override
+  the count, including zero. Injected `LLMClient` implementations are not
+  automatically replayed; a custom HTTP client on the built-in path still is.
+- Backoff is cancellable and honors valid provider `Retry-After` timing.
+  Generation and HTTP timeouts apply per attempt; the caller context bounds
+  the whole operation. Transport failures, malformed responses, other HTTP
+  statuses, and invalid generated content are not transient-retry candidates.
+- `RunRequest.Observer` reports lifecycle metadata, attempts, waits, outcomes,
+  and presence-aware usage, including work before a failed execution. It works
+  with prepared execution; preparation itself emits no execution events.
+- `ErrTransport`, `ErrMalformedResponse`, `ErrRetriesExhausted`, and
+  `RetryExhaustedError` support typed classification without parsing error text.
+- The license changes to LGPL-3.0-or-later. Catalog selections and the minimum
+  Go version are unchanged from v0.10.0.
+
+## Target End State
+
+### Dependency And Ownership
+
+Pin ConfigKit v0.4.0 and PromptKit v0.11.0 and tidy the dependency graph.
+Continue using only ConfigKit's
+public root API behind `internal/core/config`. No compatibility parser,
+application-owned composition engine, or new configuration abstraction is
+needed. PromptKit remains behind `internal/framework/llm`; its types must not
+leak into domain modules or pipeline contracts. Retain the existing prepared
+execution path, profile choices, and D&D model settings.
+
+### Consistent Overlay Selection
+
+- Replace the configuration-selection CLI flag with `--config-overlay` on all
+  commands that currently accept `--config-profile`.
+- Rename application-owned fields, variables, helpers, help text, and test
+  descriptions that specifically represent this selection to overlay terms.
+- Pass the selection through `configkit.Options.Overlay`.
+- Select files only from `overlays/<name>.yml`, following ConfigKit's contract.
+- Preserve empty and `default` selection as the composed base configuration
+  without a selected overlay.
+- Use a clean cutover: no old CLI alias, directory fallback, automatic file
+  moves, or custom migration machinery. Normal unknown-flag handling is
+  sufficient for the retired flag; explain migration in documentation.
+
+This rename is deliberately narrow. Preserve `llm_profile`, `--llm-profile`,
+`promptkit.profile_dir`, `promptkit.profile_file`, `base_profile`, embedded
+`dnd-extraction` profiles, and `examples/llm-profiles/`. Other application
+concepts named `Profile`, including pipeline/runtime types, must not be renamed
+merely because they share the word.
+
+### Preserved Behavior
+
+Retain configuration version 5, with the additive retry setting described below.
+Preserve existing defaults other than the explicitly adopted transient retry policy,
+imports, overlay merge behavior (including sequence replacement), explicit CLI
+override precedence, strict decoding, domain validation, and secret redaction.
+Do not add new restrictions or relax existing explicit contracts as part of the
+rename.
+
+Relative paths continue to resolve from the file supplying the surviving
+declaration, including overlay files. Invocation-relative CLI paths and
+main-file-relative built-in defaults retain their current behavior.
+
+Effective configuration, explanations, and configuration-load diagnostics use
+the new upstream overlay provenance and diagnostic vocabulary. Preserve source
+file labels and actionable error context; do not translate the new metadata
+back into legacy profile terminology.
+
+Pipeline stage order, semantic acceptance, LLM selection, domain artifact
+schemas, and subprocess result receipts remain unchanged. No blanket checkpoint
+invalidation is added merely for the ConfigKit rename. Update the existing
+PromptKit-version identity used by checkpoint fingerprinting with its dependency
+pin, following the current adapter convention; no cache redesign is needed.
+
+### Transient Retry Policy
+
+Expose `promptkit.transient_retries` as a nonnegative integer setting with a
+Notarius default of **2 additional attempts**. Explicit zero disables transient
+retries. Apply the resolved setting to the shared PromptKit engine so modules,
+normalizers, validators, and semantic correction calls receive the same policy.
+Include the effective value in configuration inspection. Do not impose an
+arbitrary application maximum or add per-binding controls in this work set.
+
+Maintain three separate budgets and owners:
+
+| Budget | Owner | Repeated work |
+| --- | --- | --- |
+| Transient retries | PromptKit | The same generation request after eligible HTTP failures |
+| Structural repairs | PromptKit | A corrective generation for invalid structured output |
+| Producer/validator retries | Notarius | The configured stage or validator operation, with semantic feedback where applicable |
+
+Do not silently reduce, repurpose, or remove existing stage/validator budgets.
+Each structural repair generation receives its own transient budget. For a
+single-completion producer, two additional stage attempts, one structural
+repair, and two transient retries permit up to 18 HTTP attempts, excluding any
+separate validator calls. Document this multiplication and the possibility of
+additional cost; retries do not imply exactly-once provider execution.
+
+Preserve caller cancellation throughout generation and waiting. Do not describe
+a profile's generation timeout as a total completion deadline: provider
+`Retry-After` can extend waits beyond local backoff, and each HTTP attempt has
+its own timeout. A new total-deadline configuration surface is outside this
+upgrade; accurately document the current controls and limitations.
+
+Retain Notarius's completion-scoped scheduler permit during upstream backoff.
+PromptKit releases its active backend permit during that wait, but this does
+not release the outer Notarius permit. This remains bounded and correct; do
+not add an attempt-level scheduler redesign.
+
+### Provider-Neutral Failure Diagnostics
+
+Translate the new typed errors into application-owned classifications for
+transport failures and malformed provider responses. Represent transient retry
+exhaustion separately from its underlying failure/status, retaining the actual
+attempt count where available. Preserve existing HTTP-status reporting,
+capacity classification, cancellation, deadline handling, and redaction.
+
+Do not turn a malformed provider envelope into a semantic rejection or confuse
+it with a structurally invalid model response. Do not expose PromptKit error
+types across the adapter boundary or parse prose to recover classifications.
+Existing stage disposition policies continue to decide what happens after a
+completion fails; this work does not redesign those policies.
+
+### Execution Observations
+
+Attach a request-local observer through the existing prepared-execution path.
+Translate selected metadata into Notarius-owned completion/debug summaries:
+
+- dispatched attempt, transient retry, and structural repair-call counts;
+- capacity-wait, generation, retry-delay, and total execution durations;
+- terminal outcome, failure classification, HTTP status, and exhaustion state;
+- observed token counts with explicit known/complete indicators.
+
+Retain available summaries on failed completions, including cancellation, rather
+than discarding them with the absent `RunResult`. Preparation failures have no
+execution observation; do not fabricate one. Keep existing successful-result
+usage fields and their meanings unchanged. Observation totals are separate
+presence-aware metadata, not a second set of tokens to add to existing usage.
+Unknown counts are not known zero; partial totals are not a billing ledger.
+
+Use lightweight per-request accumulation with no unbounded event history or
+asynchronous queue. Callbacks are synchronous: avoid blocking I/O or acquisition
+of scheduler resources inside them, and isolate concurrent requests. Persist
+the projection through the existing opt-in debug path, including safe terminal
+summaries where applicable. Do not serialize upstream event structs directly;
+their JSON layout is not a public contract. Allowlist metadata and preserve
+existing identifier/content handling and redaction boundaries.
+
+Successful transient recovery is ordinary success, not a warning. This work
+does not add routine console messages, change the warning policy, introduce a
+new durable telemetry artifact, or add complete cost accounting.
+
+## Required Changes
+
+### Code And Examples
+
+- Update `go.mod` and `go.sum` and the configuration loader/options boundary.
+- Update CLI selection plumbing in `internal/cli/run.go` and
+  `configkit_commands.go`, including usage, flag parsing, argument classification,
+  explicit empty-value handling, and error paths.
+- Move the maintained example to `examples/overlays/development.yml` and update
+  example discovery and related fixtures. Inspect relative paths when moving
+  files; do not assume all occurrences of `profiles` are configuration overlays.
+- Update affected configuration and CLI tests, particularly document paths,
+  operational defaults, safe inspection, command selection, diagnostic source
+  labels, run contracts, and maintained-example coverage.
+- Add the retry setting through the typed configuration document, defaults,
+  validation, runtime conversion, and CLI client-construction boundary. Update
+  the complete example to demonstrate it without adding unnecessary settings
+  to the minimal example.
+- Update the PromptKit adapter's engine construction, request observation,
+  error conversion, and checkpoint version identity. Extend transport-neutral
+  response/debug contracts only with the application-owned metadata needed.
+- Carry observations through the pipeline debug wrapper and safe terminal
+  reporting without double-counting usage or losing failed-call summaries.
+
+### Documentation
+
+Update current documentation with the implementation, keeping canonical
+ownership under [Documentation Policy](../policy/documentation.md):
+
+- `docs/cli.md`: new flag on every affected command and matching help behavior.
+- `docs/config.md`: overlay terminology, directory layout, precedence, and the
+  explicit distinction from LLM profiles; transient retry setting and default.
+- `docs/integrations/pkg-configkit.md`: pinned release and supported upstream
+  naming/inspection contract.
+- `docs/internal/configuration.md` and relevant CLI internals: new selection
+  terminology and unchanged boundaries.
+- `docs/consumers/subprocess.md` and any other current callers/examples: updated
+  invocation syntax. Review operations and README references where applicable.
+- `docs/integrations/pkg-promptkit.md`: updated pin, typed errors, observer
+  support, and retry contract. Refresh pinned upstream links, including its
+  relocated release-policy document.
+- `docs/internal/llm.md` and relevant pipeline/debug internals: observation
+  projection, failure handling, scheduler interaction, and separate retry
+  budgets. Remove the obsolete claim that PromptKit has no provider retry loop.
+- `docs/operations.md`: latency/cost implications, timeout limits, opt-in debug
+  evidence, and how to disable transient retries.
+
+Preserve historical release notes and the accepted text of ADR-0018. Its
+ConfigKit ownership decision remains valid; a terminology migration alone does
+not require a new ADR or supersession. Current documentation must not portray
+the old dependency pin as the current supported version.
+
+Record PromptKit's LGPL-3.0-or-later change in the integration documentation and
+eventual release note. Review its notices and distribution implications within
+Notarius's source-only release model. Do not silently relicense Notarius, copy
+upstream release policy into this repository, or introduce binary packaging.
+Any future binary distribution needs its own license-compliance review.
+
+## Operator Migration And Release Impact
+
+This work belongs in the next **minor** Notarius release, not a patch, because
+both the documented selected-file layout and CLI syntax change.
+
+Operators move selected configuration files from `profiles/` to `overlays/`
+and replace `--config-profile` with `--config-overlay` in scripts, service
+definitions, and subprocess callers. Preserve unrelated PromptKit profile
+directories and explicit imports. Review relative paths and explicit imports
+that refer to moved files. No YAML body version migration is required.
+
+The eventual release note must explain this cutover, the newly enabled transient
+retries and zero-value opt-out, and the dependency license change. PromptKit's
+catalogs are unchanged from v0.10.0; this upgrade adds no further profile-name
+migration or change to the D&D model defaults. Release preparation,
+tagging, and publication remain a separately requested task under
+[Source Releases](../release.md).
+
+## Acceptance And Verification
+
+- Notarius builds against ConfigKit v0.4.0 and PromptKit v0.11.0 without
+  compatibility shims.
+- All five affected commands use the new flag; non-default overlay selection,
+  omitted selection, empty selection, and `default` retain their intended
+  behavior. The old flag is not accepted.
+- Selected overlays load from the new directory. A legacy-only selected file
+  does not silently load; ordinary explicit imports remain supported.
+- Representative tests demonstrate overlay precedence, CLI overrides,
+  declaration-relative paths, redacted inspection, and source-aware failures.
+- Maintained examples validate, including the development overlay. LLM profile
+  loading and defaults remain unaffected.
+- Tests prove the default transient retry policy, explicit zero, configured
+  counts, and negative-value rejection through the relevant configuration and
+  adapter boundaries. Use local HTTP test servers for eligible/noneligible
+  statuses, exhaustion, cancellation, and interaction with structural repair;
+  an injected LLM fake alone does not exercise built-in HTTP retries.
+- Provider-neutral errors preserve underlying status and cancellation identity
+  without leaking credentials or upstream types.
+- Observations are available for success and failure, preserve unknown versus
+  known-zero usage, avoid aggregate double-counting, and stay isolated under
+  concurrent completions. Successful retry recovery emits no warning.
+- Representative scheduler/repair/retry checks preserve bounded execution and
+  cancellation; use existing coverage where it already proves the contract.
+- Current documentation, help text, examples, and tests agree. Searches for old
+  vocabulary are reviewed semantically, excluding intentional historical and
+  LLM-profile references rather than enforcing a blanket text ban.
+- Run focused `go test ./internal/core/config ./internal/cli ./internal/framework/llm ./internal/framework/pipeline`, followed by
+  `go test ./...`, `go vet ./...`, `go build ./...`, `go mod tidy -diff`, and
+  `git diff --check`, with `GOWORK=off`. Run race-enabled tests for the modified
+  adapter, scheduler, and observation integration. Validate both maintained configurations
+  and the complete configuration with `--config-overlay development`.
+
+Use existing behavioral tests where possible, following
+[Testing Policy](../policy/testing.md). No live model calls or production API
+credentials are needed to validate this migration.

examples/dnd-complete.config.yml

@@ -1,6 +1,10 @@
-version: 4
+version: 5
+configkit:
+  imports:
+    - path: pipelines/dnd-session.yml
 promptkit:
-  profile_file: ./examples/profiles/dnd-extraction.yml
+  profile_file: ./llm-profiles/dnd-extraction.yml
+  transient_retries: 2
 concurrency:
   total_llm: 2
   stage_workers:
@@ -16,145 +20,3 @@ cache:
     directory: ./notarius-cache/checkpoints
 debug:
   directory: ./notarius-debug
-pipelines:
-  dnd-session:
-    llm_profile: dnd-extraction
-    input: seriatim
-    # Stable campaign context is shared by every module that accepts these slots.
-    references:
-      party: ./dnd-party.txt
-      glossary: ./dnd-glossary.txt
-    chunk:
-      module: dnd/scenes
-      retries: 2
-    output:
-      module: json
-      options:
-        include_chunk_map: true
-        evidence_context:
-          enabled: true
-          window_units: 3
-          lanes:
-            - item-occurrences
-            - item-registry
-            - location-registry
-            - location-occurrences
-            - npc-registry
-            - spells
-            - combat-turns
-            - npc-occurrences
-            - enemy-events
-    steps:
-      # Establish session-wide reference artifacts before their consumers.
-      - id: describe-session
-        artifacts:
-          item-registry:
-            extract:
-              module: dnd/item-registry
-              retries: 2
-            merge: appendorder
-            normalize: dnd/item-registry
-          npc-registry:
-            extract:
-              module: dnd/npc-registry
-              retries: 2
-            merge: appendorder
-            normalize:
-              module: dnd/npc-registry
-              retries: 2
-          location-registry:
-            extract:
-              module: dnd/location-registry
-              retries: 2
-            merge: appendorder
-            normalize:
-              module: dnd/location-registry
-              retries: 2
-          scene-descriptions:
-            extract:
-              module: dnd/scene-descriptions
-              retries: 2
-            merge: appendorder
-            normalize: dnd/scene-descriptions
-      - id: extract-events
-        # Accepted registry artifacts and scene-description eligibility artifacts
-        # are supplied in memory to their compatible consumers in this step.
-        references:
-          location_registry:
-            artifact:
-              step: describe-session
-              lane: location-registry
-          npc_registry:
-            artifact:
-              step: describe-session
-              lane: npc-registry
-          scene_descriptions:
-            artifact:
-              step: describe-session
-              lane: scene-descriptions
-          item_registry:
-            artifact:
-              step: describe-session
-              lane: item-registry
-        artifacts:
-          item-occurrences:
-            extract:
-              module: dnd/item-occurrences
-              retries: 2
-            merge: appendorder
-            normalize: dnd/item-occurrences
-          spells:
-            extract:
-              module: dnd/spells
-              retries: 2
-              references:
-                spell_catalog: ./dnd-spell-catalog.json
-            merge: appendorder
-            # Stage-local file references are intentionally bound at each stage.
-            normalize:
-              module: dnd/spells
-              references:
-                spell_catalog: ./dnd-spell-catalog.json
-          combat-turns:
-            extract:
-              module: dnd/combat-turns
-              retries: 2
-            merge: appendorder
-            normalize: dnd/combat-turns
-          npc-occurrences:
-            extract:
-              module: dnd/npc-occurrences
-              retries: 2
-            merge: appendorder
-            normalize: dnd/npc-occurrences
-          location-occurrences:
-            extract:
-              module: dnd/location-occurrences
-              retries: 2
-            merge: appendorder
-            normalize: dnd/location-occurrences
-      - id: track-enemies
-        references:
-          npc_registry:
-            artifact:
-              step: describe-session
-              lane: npc-registry
-          scene_descriptions:
-            artifact:
-              step: describe-session
-              lane: scene-descriptions
-          combat_turns:
-            artifact:
-              step: extract-events
-              lane: combat-turns
-          npc_occurrences:
-            artifact:
-              step: extract-events
-              lane: npc-occurrences
-        artifacts:
-          enemy-events:
-            extract:
-              module: dnd/enemy-events
-              retries: 2
-            merge: appendorder
-            normalize: dnd/enemy-events

examples/dnd-minimal.config.yml

@@ -1,9 +1,12 @@
-version: 4
+version: 5
 pipelines:
   dnd-session:
     llm_profile: dnd-extraction
-    input: seriatim
+    input:
+      module: seriatim
     artifacts:
       spells:
-        extract: dnd/spells
-        normalize: dnd/spells
+        extract:
+          module: dnd/spells
+        normalize:
+          module: dnd/spells

examples/overlays/development.yml

@@ -0,0 +1,8 @@
+# This selects operational settings only; choose a different LLM profile with
+# the pipeline's llm_profile field or the run command's --llm-profile flag.
+concurrency:
+  total_llm: 1
+  stage_workers:
+    extract: 1
+output:
+  directory: ./notarius-output-development

examples/pipelines/dnd-session.yml

@@ -0,0 +1,164 @@
+pipelines:
+  dnd-session:
+    llm_profile: dnd-extraction
+    input:
+      module: seriatim
+    # Stable campaign context is shared by every module that accepts these slots.
+    references:
+      party:
+        path: ../dnd-party.txt
+      glossary:
+        path: ../dnd-glossary.txt
+    chunk:
+      module: dnd/scenes
+      retries: 2
+    output:
+      module: json
+      options:
+        include_chunk_map: true
+        evidence_context:
+          enabled: true
+          window_units: 3
+          lanes:
+            - item-occurrences
+            - item-registry
+            - location-registry
+            - location-occurrences
+            - npc-registry
+            - spells
+            - combat-turns
+            - npc-occurrences
+            - enemy-events
+    steps:
+      # Establish session-wide reference artifacts before their consumers.
+      - id: describe-session
+        artifacts:
+          item-registry:
+            extract:
+              module: dnd/item-registry
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/item-registry
+          npc-registry:
+            extract:
+              module: dnd/npc-registry
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/npc-registry
+              retries: 2
+          location-registry:
+            extract:
+              module: dnd/location-registry
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/location-registry
+              retries: 2
+          scene-descriptions:
+            extract:
+              module: dnd/scene-descriptions
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/scene-descriptions
+      - id: extract-events
+        # Accepted registry artifacts and scene-description eligibility artifacts
+        # are supplied in memory to their compatible consumers in this step.
+        references:
+          location_registry:
+            artifact:
+              step: describe-session
+              lane: location-registry
+          npc_registry:
+            artifact:
+              step: describe-session
+              lane: npc-registry
+          scene_descriptions:
+            artifact:
+              step: describe-session
+              lane: scene-descriptions
+          item_registry:
+            artifact:
+              step: describe-session
+              lane: item-registry
+        artifacts:
+          item-occurrences:
+            extract:
+              module: dnd/item-occurrences
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/item-occurrences
+          spells:
+            extract:
+              module: dnd/spells
+              retries: 2
+              references:
+                spell_catalog:
+                  path: ../dnd-spell-catalog.json
+            merge:
+              module: appendorder
+            # Stage-local file references are intentionally bound at each stage.
+            normalize:
+              module: dnd/spells
+              references:
+                spell_catalog:
+                  path: ../dnd-spell-catalog.json
+          combat-turns:
+            extract:
+              module: dnd/combat-turns
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/combat-turns
+          npc-occurrences:
+            extract:
+              module: dnd/npc-occurrences
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/npc-occurrences
+          location-occurrences:
+            extract:
+              module: dnd/location-occurrences
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/location-occurrences
+      - id: track-enemies
+        references:
+          npc_registry:
+            artifact:
+              step: describe-session
+              lane: npc-registry
+          scene_descriptions:
+            artifact:
+              step: describe-session
+              lane: scene-descriptions
+          combat_turns:
+            artifact:
+              step: extract-events
+              lane: combat-turns
+          npc_occurrences:
+            artifact:
+              step: extract-events
+              lane: npc-occurrences
+        artifacts:
+          enemy-events:
+            extract:
+              module: dnd/enemy-events
+              retries: 2
+            merge:
+              module: appendorder
+            normalize:
+              module: dnd/enemy-events

go.mod

@@ -3,14 +3,16 @@ module gitea.maximumdirect.net/eric/notarius
 go 1.25.5
 
 require (
-	gitea.maximumdirect.net/eric/promptkit v0.9.0
+	gitea.maximumdirect.net/eric/ConfigKit v0.4.0
+	gitea.maximumdirect.net/eric/promptkit v0.11.0
 	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2
-	gopkg.in/yaml.v3 v3.0.1
+	gopkg.in/yaml.v3 v3.0.1 // indirect
 )
 
 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
+	gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1 // indirect
+	gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0 // indirect
+	go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect
 )

go.sum

@@ -1,13 +1,17 @@
-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=
+gitea.maximumdirect.net/eric/ConfigKit v0.4.0 h1:fo+vKff3zTGeWqJIA38NtT84b6OUP//9ZmBH3lFuKgk=
+gitea.maximumdirect.net/eric/ConfigKit v0.4.0/go.mod h1:tYWpsrbLyu05r8+1MapWp5fN6ZrX9I8hU2xEPedYmYk=
+gitea.maximumdirect.net/eric/promptkit v0.11.0 h1:x7JguOUiInPVXK0GEitjyoRFqqvcuSTjYUvqJ+whtik=
+gitea.maximumdirect.net/eric/promptkit v0.11.0/go.mod h1:0zEBVlJJC+ldDdXazAgybYBXNSe74vwFx5dyOuSos/w=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1 h1:S2/Q/JDE+m8Ca3pDhAjVbfYrdlEhs9blj0STgNopHo0=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1/go.mod h1:a3VBbh9oaGXD6ruxhtwHj+AWW6DY/SSyby1QtMmYSUo=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0 h1:Y0EXc3eh12eKW2NrOviVqdk3oBo6UUjopt1YAa1B3Oc=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0/go.mod h1:mtgD1E0SGm0e/I8Ffhz+AYX7GN7RKUWJXehNFIKHs4Y=
 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=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.2/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU=
+go.yaml.in/yaml/v4 v4.0.0-rc.6 h1:1h7H1ohdUh93/FyE4YaDa1Zh64K6VVbjF4K6WUxMtH4=
+go.yaml.in/yaml/v4 v4.0.0-rc.6/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0=
 golang.org/x/text v0.40.0 h1:Ub2Z6/xjgF1WrYQz2nuITOEegKFtiIy+rieRJ5lHZKs=
 golang.org/x/text v0.40.0/go.mod h1:hpnzDAfGV753zIKo+wk3u1bVKCGPbrnF7+7LBF/UHVY=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=

internal/cli/assembled_spell_pipeline_contract_test.go

@@ -71,28 +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 {
@@ -155,14 +166,14 @@ 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{})
@@ -190,12 +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) != 2 || output.Warnings[0].ReasonCode != spellnormalize.ReasonCodeSpellNameUnresolved || output.Warnings[0].Scope != "spell_casts[0]" || output.Warnings[1].ReasonCode != "spell_not_near_source" {
-		t.Fatalf("warnings = %#v, want complete terminal normalize validation warnings", 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 {
@@ -219,14 +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)
 	}
 }
 
@@ -429,6 +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/cache_contract_test.go

@@ -23,7 +23,7 @@ func TestRunChunkPlanModePrecedenceAndValidation(t *testing.T) {
 	}{
 		{name: "default", wantStores: 1},
 		{name: "file", fileMode: "bypass"},
-		{name: "environment", envMode: "bypass"},
+		{name: "retired environment has no effect", envMode: "bypass", wantStores: 1},
 		{name: "cli", envMode: "refresh", cliMode: "bypass"},
 	}
 	for _, tt := range tests {
@@ -80,25 +80,14 @@ func TestRunChunkPlanModePrecedenceAndValidation(t *testing.T) {
 	})
 
 	for _, tt := range []struct {
-		name       string
-		fileConfig bool
+		name string
 	}{
-		{name: "invalid environment mode"},
-		{name: "invalid file mode", fileConfig: true},
+		{name: "invalid file mode"},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
 			roots := newStateTestRoots(t)
 			opts := newStateTestHarness().options()
-			if tt.fileConfig {
-				replaceStateTestConfigLine(t, roots.config, "    mode: auto\n", "    mode: invalid\n")
-			} else {
-				opts.LookupEnv = func(name string) (string, bool) {
-					if name == "NOTARIUS_CACHE_CHUNK_PLANS_MODE" {
-						return "invalid", true
-					}
-					return "", false
-				}
-			}
+			replaceStateTestConfigLine(t, roots.config, "    mode: auto\n", "    mode: invalid\n")
 			var stdout, stderr bytes.Buffer
 			code := RunWithOptions([]string{"run", "sample", "--config", roots.config, "--input", roots.input}, &stdout, &stderr, opts)
 			if code != 1 || stdout.Len() != 0 || stderr.Len() == 0 {
@@ -207,16 +196,20 @@ func TestRunAutoReusesPlanWhenRunInputsChange(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	configText := replaceRequiredOnce(t, string(data), "    chunk: test/chunk\n", `    chunk:
+	configText := replaceRequiredOnce(t, string(data), "    chunk:\n      module: test/chunk\n", `    chunk:
       module: test/chunk
       options:
         strategy: first
 `)
-	configText = replaceRequiredOnce(t, configText, "    output: test/output\n", `      other:
-        extract: test/extract
-        merge: test/merge
-        normalize: test/normalize
-    output: test/output
+	configText = replaceRequiredOnce(t, configText, "    output:\n      module: test/output\n", `      other:
+        extract:
+          module: test/extract
+        merge:
+          module: test/merge
+        normalize:
+          module: test/normalize
+    output:
+      module: test/output
 `)
 	if err := os.WriteFile(roots.config, []byte(configText), 0o600); err != nil {
 		t.Fatal(err)

internal/cli/catalog.go

@@ -157,13 +157,15 @@ func buildProductionLLMClient(ctx context.Context, cfg config.Config, profileID
 		return nil, nil, fmt.Errorf("production asset registry must not be nil")
 	}
 	recorder := llm.NewLLMProfileRecorder()
+	transientRetries := cfg.PromptKit.TransientRetries
 	client, err := llm.NewPromptKitClient(llm.PromptKitClientConfig{
-		ProfileDir:      cfg.PromptKit.ProfileDir,
-		ProfileFile:     cfg.PromptKit.ProfileFile,
-		LocalBackend:    mapPromptKitLocalBackend(cfg.PromptKit.LocalBackend),
-		Assets:          assets,
-		Recorder:        recorder,
-		ReasoningEffort: overrides.ReasoningEffort,
+		ProfileDir:       cfg.PromptKit.ProfileDir,
+		ProfileFile:      cfg.PromptKit.ProfileFile,
+		TransientRetries: &transientRetries,
+		LocalBackend:     mapPromptKitLocalBackend(cfg.PromptKit.LocalBackend),
+		Assets:           assets,
+		Recorder:         recorder,
+		ReasoningEffort:  overrides.ReasoningEffort,
 	})
 	if err != nil {
 		return nil, nil, fmt.Errorf("create PromptKit-backed LLM client: %w", err)

internal/cli/command_contract_test.go

@@ -140,7 +140,7 @@ func TestConfigLoadingFailuresReturnOneWithPathContext(t *testing.T) {
 	stdout.Reset()
 	stderr.Reset()
 	code = RunWithOptions([]string{"config", "validate", "--config", malformed}, &stdout, &stderr, commandContractOptions(t))
-	if code != 1 || !strings.Contains(stderr.String(), malformed) || !strings.Contains(stderr.String(), "parse config file") || stdout.Len() != 0 {
+	if code != 1 || !strings.Contains(stderr.String(), "configuration contains invalid YAML") || stdout.Len() != 0 {
 		t.Fatalf("malformed config: code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 	}
 }
@@ -185,14 +185,16 @@ func TestConfigValidateResolvesPipelineAndChecksSelection(t *testing.T) {
 }
 
 func TestConfigValidatePipelineDefaultProfileIsOffline(t *testing.T) {
-	configPath := writeCommandConfigContent(t, `version: 4
+	configPath := writeCommandConfigContent(t, `version: 5
 pipelines:
   demo:
     llm_profile: dnd-extraction
-    input: seriatim
+    input:
+      module: seriatim
     artifacts:
       spells:
-        extract: dnd/spells
+        extract:
+          module: dnd/spells
 `)
 	var stdout, stderr bytes.Buffer
 	code := RunWithOptions([]string{"config", "validate", "--config", configPath, "--pipeline", "demo"}, &stdout, &stderr, Options{})
@@ -261,19 +263,21 @@ func commandContractOptionsWithLookup(t *testing.T, lookup func(string) (string,
 
 func writeCommandConfig(t *testing.T, firstID, secondID string) string {
 	t.Helper()
-	content := fmt.Sprintf("version: 4\npipelines:\n  %q:\n    input: seriatim\n  %q:\n    input: seriatim\n", firstID, secondID)
+	content := fmt.Sprintf("version: 5\npipelines:\n  %q:\n    input:\n      module: seriatim\n  %q:\n    input:\n      module: seriatim\n", firstID, secondID)
 	return writeCommandConfigContent(t, content)
 }
 
 func writeResolvableCommandConfig(t *testing.T) string {
 	t.Helper()
-	return writeCommandConfigContent(t, `version: 4
+	return writeCommandConfigContent(t, `version: 5
 pipelines:
   demo:
-    input: seriatim
+    input:
+      module: seriatim
     artifacts:
       spells:
-        extract: dnd/spells
+        extract:
+          module: dnd/spells
 `)
 }
 

internal/cli/configkit_commands.go

@@ -0,0 +1,157 @@
+package cli
+
+import (
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"strings"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+)
+
+func runConfigEffective(args []string, stdout, stderr io.Writer, opts Options) int {
+	fs := flag.NewFlagSet("config effective", flag.ContinueOnError)
+	fs.SetOutput(io.Discard)
+	configPath := fs.String("config", "", "config file path")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
+	if err := fs.Parse(args); err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 2
+	}
+	if fs.NArg() != 0 {
+		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
+		return 2
+	}
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
+	if err != nil {
+		writeConfigLoadError(stderr, err)
+		return 1
+	}
+	if err := loaded.WriteEffective(stdout); err != nil {
+		fmt.Fprintf(stderr, "notarius: write effective configuration: %v\n", err)
+		return 1
+	}
+	return 0
+}
+
+func runConfigExplain(args []string, stdout, stderr io.Writer, opts Options) int {
+	fs := flag.NewFlagSet("config explain", flag.ContinueOnError)
+	fs.SetOutput(io.Discard)
+	configPath := fs.String("config", "", "config file path")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
+	if err := fs.Parse(args); err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 2
+	}
+	if fs.NArg() != 1 {
+		fmt.Fprintln(stderr, "notarius: config explain requires exactly one JSON Pointer")
+		return 2
+	}
+	pointer := fs.Arg(0)
+	parsed, err := configkit.ParsePath(pointer)
+	if err != nil || parsed.String() != pointer {
+		fmt.Fprintf(stderr, "notarius: %q is not a canonical JSON Pointer\n", pointer)
+		return 2
+	}
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
+	if err != nil {
+		writeConfigLoadError(stderr, err)
+		return 1
+	}
+	if err := loaded.WriteExplanation(stdout, pointer); err != nil {
+		fmt.Fprintf(stderr, "notarius: write configuration explanation: %v\n", err)
+		return 1
+	}
+	return 0
+}
+
+func runConfigValidate(args []string, stdout, stderr io.Writer, opts Options) int {
+	fs := flag.NewFlagSet("config validate", flag.ContinueOnError)
+	fs.SetOutput(io.Discard)
+	configPath := fs.String("config", "", "config file path")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
+	pipelineID := fs.String("pipeline", "", "pipeline ID")
+	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
+	if err := fs.Parse(args); err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 2
+	}
+	if fs.NArg() != 0 {
+		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
+		return 2
+	}
+	if strings.TrimSpace(*onlyRaw) != "" && strings.TrimSpace(*pipelineID) == "" {
+		fmt.Fprintln(stderr, "notarius: --only requires --pipeline")
+		return 2
+	}
+	only, err := parseOnly(*onlyRaw)
+	if err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 2
+	}
+	loaded, path, err := loadConfigKit(*configPath, *overlay, nil, opts)
+	if err != nil {
+		writeConfigLoadError(stderr, err)
+		return 1
+	}
+	if strings.TrimSpace(*pipelineID) == "" {
+		fmt.Fprintf(stdout, "config %q is valid\n", path)
+		return 0
+	}
+	catalog, err := effectiveCatalog(opts)
+	if err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 1
+	}
+	effective, err := loaded.Config.Resolve(config.ResolveInput{PipelineID: *pipelineID, Only: only, Catalog: catalog})
+	if err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 1
+	}
+	if err := validateExplicitPromptKitProfiles(context.Background(), effective.Config, effectiveLLMProfileIDs(effective.ResolvedPipeline), opts.promptKitAssets); err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 1
+	}
+	fmt.Fprintf(stdout, "config %q is valid for pipeline %q\n", path, strings.TrimSpace(*pipelineID))
+	return 0
+}
+
+func runPipelinesList(args []string, stdout, stderr io.Writer, opts Options) int {
+	fs := flag.NewFlagSet("pipelines list", flag.ContinueOnError)
+	fs.SetOutput(io.Discard)
+	configPath := fs.String("config", "", "config file path")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
+	jsonOutput := fs.Bool("json", false, "write JSON output")
+	if err := fs.Parse(args); err != nil {
+		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		return 2
+	}
+	if fs.NArg() != 0 {
+		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
+		return 2
+	}
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
+	if err != nil {
+		writeConfigLoadError(stderr, err)
+		return 1
+	}
+	ids := sortedPipelineIDs(loaded.Config)
+	if *jsonOutput {
+		payload, err := json.Marshal(struct {
+			Pipelines []string `json:"pipelines"`
+		}{Pipelines: ids})
+		if err != nil {
+			fmt.Fprintf(stderr, "notarius: marshal pipeline list: %v\n", err)
+			return 1
+		}
+		fmt.Fprintf(stdout, "%s\n", payload)
+		return 0
+	}
+	for _, id := range ids {
+		fmt.Fprintln(stdout, id)
+	}
+	return 0
+}

internal/cli/configkit_commands_test.go

@@ -0,0 +1,159 @@
+package cli
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
+	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+func TestConfigKitReadOnlyCommandsLoadImportsAndOverlays(t *testing.T) {
+	t.Parallel()
+	root := t.TempDir()
+	path := filepath.Join(root, "notarius.yml")
+	writeConfigKitFile(t, filepath.Join(root, "fragment.yml"), "pipelines:\n  imported:\n    input: {module: input}\n")
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "development.yml"), "pipelines:\n  overlaid:\n    input: {module: input}\n")
+	writeConfigKitFile(t, path, "version: 5\nconfigkit:\n  imports:\n    - path: fragment.yml\npipelines:\n  base:\n    input: {module: input}\n")
+	var stdout, stderr bytes.Buffer
+	if code := RunWithOptions([]string{"pipelines", "list", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || stdout.String() != "base\nimported\noverlaid\n" {
+		t.Fatalf("list = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+	stdout.Reset()
+	stderr.Reset()
+	if code := RunWithOptions([]string{"config", "validate", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || !strings.Contains(stdout.String(), "is valid") {
+		t.Fatalf("validate = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+	for _, selection := range [][]string{nil, {"--config-overlay", ""}, {"--config-overlay", "default"}} {
+		stdout.Reset()
+		stderr.Reset()
+		args := append([]string{"pipelines", "list", "--config", path}, selection...)
+		if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != 0 || stdout.String() != "base\nimported\n" || stderr.Len() != 0 {
+			t.Fatalf("base list %q = %d, %q, %q", selection, code, stdout.String(), stderr.String())
+		}
+	}
+}
+
+func TestRetiredConfigProfileFlagUsesNormalUnknownFlagBehavior(t *testing.T) {
+	t.Parallel()
+
+	for _, args := range [][]string{
+		{"run", "sample", "--config-profile=development"},
+		{"config", "validate", "--config-profile=development"},
+		{"config", "effective", "--config-profile=development"},
+		{"config", "explain", "--config-profile=development", ""},
+		{"pipelines", "list", "--config-profile=development"},
+	} {
+		t.Run(strings.Join(args[:2], " "), func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != 2 || stdout.Len() != 0 || !strings.Contains(stderr.String(), "not defined") {
+				t.Fatalf("args=%q code=%d stdout=%q stderr=%q", args, code, stdout.String(), stderr.String())
+			}
+		})
+	}
+}
+
+func TestConfigKitReadOnlyCommandsRejectVersionFour(t *testing.T) {
+	t.Parallel()
+	path := filepath.Join(t.TempDir(), "notarius.yml")
+	writeConfigKitFile(t, path, "version: 4\n")
+	var stdout, stderr bytes.Buffer
+	if code := runPipelinesList([]string{"--config", path}, &stdout, &stderr, commandContractOptions(t)); code != 1 || !strings.Contains(stderr.String(), "version") {
+		t.Fatalf("list = %d, %q", code, stderr.String())
+	}
+}
+
+func TestConfigurationInspectionCommandsRenderSafeConfigKitViews(t *testing.T) {
+	t.Parallel()
+	root := t.TempDir()
+	path := filepath.Join(root, "notarius.yml")
+	writeConfigKitFile(t, path, `version: 5
+concurrency:
+  total_llm: 4
+pipelines:
+  "0":
+    input: {module: input}
+`)
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "development.yml"), "concurrency:\n  total_llm: 3\n")
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "retired.yml"), "pipelines: null\n")
+
+	providerCalled := false
+	opts := commandContractOptions(t)
+	opts.LLMClientFactory = func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		providerCalled = true
+		return nil, nil, errors.New("LLM client must not be constructed")
+	}
+
+	var stdout, stderr bytes.Buffer
+	if code := RunWithOptions([]string{"config", "effective", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), "total_llm: 3") || stderr.Len() != 0 {
+		t.Fatalf("effective = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+	if providerCalled {
+		t.Fatal("config effective constructed an LLM client")
+	}
+
+	stdout.Reset()
+	stderr.Reset()
+	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-overlay", "development", "/concurrency/total_llm"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/concurrency/total_llm"`) || stderr.Len() != 0 {
+		t.Fatalf("explain effective = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+
+	stdout.Reset()
+	stderr.Reset()
+	if code := RunWithOptions([]string{"config", "explain", "--config", path, ""}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: ""`) || stderr.Len() != 0 {
+		t.Fatalf("explain root = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+
+	stdout.Reset()
+	stderr.Reset()
+	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-overlay", "retired", "/pipelines/0"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/pipelines/0"`) || stderr.Len() != 0 {
+		t.Fatalf("explain removed mapping key = %d, %q, %q", code, stdout.String(), stderr.String())
+	}
+}
+
+func TestConfigExplainRejectsInvalidAndUnknownPointers(t *testing.T) {
+	t.Parallel()
+	path := writeCommandConfig(t, "session", "other")
+	for _, test := range []struct {
+		name     string
+		pointer  string
+		wantCode int
+	}{
+		{name: "missing", wantCode: 2},
+		{name: "malformed escape", pointer: "/pipelines~2session", wantCode: 2},
+		{name: "unknown", pointer: "/unknown", wantCode: 1},
+		{name: "invalid sequence index", pointer: "/pipelines/session/steps/+0", wantCode: 1},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			args := []string{"config", "explain", "--config", path}
+			if test.name != "missing" {
+				args = append(args, test.pointer)
+			}
+			if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != test.wantCode || stdout.Len() != 0 || stderr.Len() == 0 {
+				t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+			}
+		})
+	}
+}
+
+func TestConfigInspectionCommandsReportWriterFailures(t *testing.T) {
+	t.Parallel()
+	path := writeCommandConfig(t, "session", "other")
+	var stderr bytes.Buffer
+	if code := RunWithOptions([]string{"config", "effective", "--config", path}, failingInspectionWriter{}, &stderr, commandContractOptions(t)); code != 1 || !strings.Contains(stderr.String(), "write effective configuration") {
+		t.Fatalf("effective writer failure = %d, %q", code, stderr.String())
+	}
+}
+
+type failingInspectionWriter struct{}
+
+func (failingInspectionWriter) Write([]byte) (int, error) {
+	return 0, io.ErrClosedPipe
+}

internal/cli/configkit_load.go

@@ -0,0 +1,49 @@
+package cli
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+)
+
+func loadConfigKit(configPath, overlay string, overrides []configkit.Override, opts Options) (*config.LoadedConfig, string, error) {
+	path, err := discoverConfigPath(configPath, opts)
+	if err != nil {
+		return nil, "", err
+	}
+	workingDirectory, err := os.Getwd()
+	if err != nil {
+		return nil, "", fmt.Errorf("determine invocation directory: %w", err)
+	}
+	loaded, err := config.Load(path, config.LoadOptions{Overlay: overlay, Overrides: overrides, InvocationDirectory: workingDirectory})
+	if err != nil {
+		return nil, "", err
+	}
+	return loaded, path, nil
+}
+
+func writeConfigLoadError(output io.Writer, err error) {
+	var loadError *configkit.LoadError
+	if errors.As(err, &loadError) {
+		if writeErr := configkit.WriteDiagnostics(output, loadError.Diagnostics()); writeErr == nil {
+			// ConfigKit's locations contain source IDs. Keep its renderer and
+			// supply the acquired-source legend so imported/overlay errors can
+			// be traced to a file. Quoting also escapes terminal controls.
+			for _, source := range loadError.Sources() {
+				label := source.Path
+				if label == "" {
+					label = source.Label
+				}
+				if _, err := fmt.Fprintf(output, "source %q: %q\n", source.ID, label); err != nil {
+					return
+				}
+			}
+			return
+		}
+	}
+	fmt.Fprintf(output, "notarius: %v\n", err)
+}

internal/cli/configkit_load_test.go

@@ -0,0 +1,85 @@
+package cli
+
+import (
+	"bytes"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"testing"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+)
+
+func TestConfigLoadDiagnosticsIdentifyComposedSources(t *testing.T) {
+	t.Parallel()
+	for _, overlay := range []bool{false, true} {
+		t.Run(strconv.FormatBool(overlay), func(t *testing.T) {
+			root := t.TempDir()
+			main := filepath.Join(root, "config.yml")
+			fragment := filepath.Join(root, "pipeline\n\x1b.yml")
+			args := []string{"config", "validate", "--config", main}
+			if overlay {
+				fragment = filepath.Join(root, "overlays", "development.yml")
+				writeConfigKitFile(t, main, "version: 5\n")
+				args = append(args, "--config-overlay", "development")
+			} else {
+				writeConfigKitFile(t, main, "version: 5\nconfigkit:\n  imports:\n    - path: "+strconv.Quote(filepath.Base(fragment))+"\n")
+			}
+			writeConfigKitFile(t, fragment, "concurrency: {total_llm: invalid-private-value}\n")
+			var stdout, stderr bytes.Buffer
+			if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != 1 {
+				t.Fatalf("status = %d; %s", code, stderr.String())
+			}
+			got := stderr.String()
+			if stdout.Len() != 0 || !strings.Contains(got, "decode.type") ||
+				!strings.Contains(got, "source-2:1:") ||
+				!strings.Contains(got, strconv.Quote(fragment)) ||
+				strings.Contains(got, "invalid-private-value") ||
+				strings.ContainsRune(got, '\x1b') {
+				t.Fatalf("stdout = %q; stderr = %q", stdout.String(), got)
+			}
+		})
+	}
+}
+
+func TestLoadConfigKitUsesDiscoveryOverlayAndOverrides(t *testing.T) {
+	t.Parallel()
+	root := t.TempDir()
+	path := filepath.Join(root, "notarius.yml")
+	writeConfigKitFile(t, path, "version: 5\npipelines:\n  base:\n    input: {module: input}\n")
+	overlays := filepath.Join(root, "overlays")
+	writeConfigKitFile(t, filepath.Join(overlays, "development.yml"), "pipelines:\n  overlaid:\n    input: {module: input}\n")
+	loaded, gotPath, err := loadConfigKit(path, "development", []configkit.Override{{Path: configkit.Keys("output", "directory"), Value: "generated", Flag: "--output"}}, commandContractOptions(t))
+	if err != nil {
+		t.Fatalf("loadConfigKit() error = %v", err)
+	}
+	if gotPath != path || loaded.Config.Pipelines["overlaid"].Input.Module != "input" || loaded.Config.Output.Directory == "generated" {
+		t.Fatalf("loaded = %#v, path=%q", loaded.Config, gotPath)
+	}
+}
+
+func TestWriteConfigLoadErrorUsesSafeDiagnostics(t *testing.T) {
+	t.Parallel()
+	path := filepath.Join(t.TempDir(), "invalid.yml")
+	writeConfigKitFile(t, path, "version: 4\n")
+	_, _, err := loadConfigKit(path, "", nil, commandContractOptions(t))
+	if err == nil {
+		t.Fatal("loadConfigKit() error = nil")
+	}
+	var output bytes.Buffer
+	writeConfigLoadError(&output, err)
+	if !strings.Contains(output.String(), "version") || strings.Contains(output.String(), "notarius: configkit") {
+		t.Fatalf("diagnostics = %q", output.String())
+	}
+}
+
+func writeConfigKitFile(t *testing.T, path, contents string) {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(path), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(contents), 0o600); err != nil {
+		t.Fatal(err)
+	}
+}

internal/cli/dnd_enemy_events_contract_test.go

@@ -264,13 +264,29 @@ func completeExampleConfigWithTemporaryCache(t *testing.T) string {
 	updated := strings.Replace(string(content), "directory: ./notarius-cache/chunk-plans", fmt.Sprintf("directory: %q", filepath.Join(cacheRoot, "chunk-plans")), 1)
 	updated = strings.Replace(updated, "directory: ./notarius-cache/checkpoints", fmt.Sprintf("directory: %q", filepath.Join(cacheRoot, "checkpoints")), 1)
 	for relative, absolute := range map[string]string{
-		"./dnd-party.txt":          repositoryPath("examples", "dnd-party.txt"),
-		"./dnd-glossary.txt":       repositoryPath("examples", "dnd-glossary.txt"),
-		"./dnd-spell-catalog.json": repositoryPath("examples", "dnd-spell-catalog.json"),
+		"./llm-profiles/dnd-extraction.yml": repositoryPath("examples", "llm-profiles", "dnd-extraction.yml"),
 	} {
 		updated = strings.ReplaceAll(updated, relative, fmt.Sprintf("%q", absolute))
 	}
-	path := filepath.Join(t.TempDir(), "dnd-complete.config.yml")
+	root := t.TempDir()
+	fragment, err := os.ReadFile(repositoryPath("examples", "pipelines", "dnd-session.yml"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	for relative, absolute := range map[string]string{
+		"../dnd-party.txt":          repositoryPath("examples", "dnd-party.txt"),
+		"../dnd-glossary.txt":       repositoryPath("examples", "dnd-glossary.txt"),
+		"../dnd-spell-catalog.json": repositoryPath("examples", "dnd-spell-catalog.json"),
+	} {
+		fragment = []byte(strings.ReplaceAll(string(fragment), relative, fmt.Sprintf("%q", absolute)))
+	}
+	if err := os.MkdirAll(filepath.Join(root, "pipelines"), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "pipelines", "dnd-session.yml"), fragment, 0o600); err != nil {
+		t.Fatal(err)
+	}
+	path := filepath.Join(root, "dnd-complete.config.yml")
 	if err := os.WriteFile(path, []byte(updated), 0o600); err != nil {
 		t.Fatal(err)
 	}

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

@@ -136,19 +136,26 @@ func TestMaintainedConfigurationExampleSet(t *testing.T) {
 		t.Fatalf("maintained configuration examples = %q, want only the minimal and complete D&D examples", got)
 	}
 
-	profileEntries, err := os.ReadDir(repositoryPath("examples", "profiles"))
+	overlayEntries, err := os.ReadDir(repositoryPath("examples", "overlays"))
 	if err != nil {
 		t.Fatal(err)
 	}
 	names = names[:0]
-	for _, entry := range profileEntries {
+	for _, entry := range overlayEntries {
 		if !entry.IsDir() && strings.HasSuffix(entry.Name(), ".yml") {
 			names = append(names, entry.Name())
 		}
 	}
 	sort.Strings(names)
-	if got := strings.Join(names, ","); got != "dnd-extraction.yml" {
-		t.Fatalf("maintained operator profiles = %q, want dnd-extraction.yml", got)
+	if got := strings.Join(names, ","); got != "development.yml" {
+		t.Fatalf("maintained configuration overlays = %q, want development.yml", got)
+	}
+	llmProfiles, err := os.ReadDir(repositoryPath("examples", "llm-profiles"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(llmProfiles) != 1 || llmProfiles[0].Name() != "dnd-extraction.yml" {
+		t.Fatalf("maintained LLM profiles = %#v, want dnd-extraction.yml", llmProfiles)
 	}
 }
 
@@ -239,10 +246,10 @@ 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/npc_registry_contract_test.go

@@ -23,7 +23,7 @@ func TestOversizedNPCRegistryFailsBeforeRuntimeAndCheckpointConstruction(t *test
 		t.Fatal(err)
 	}
 	checkpointRoot := filepath.Join(t.TempDir(), "checkpoints")
-	content := fmt.Sprintf(`version: 4
+	content := fmt.Sprintf(`version: 5
 cache:
   chunk_plans:
     mode: bypass
@@ -32,14 +32,17 @@ cache:
     directory: %q
 pipelines:
   dnd-session:
-    input: seriatim
+    input:
+      module: seriatim
     artifacts:
       spells:
         extract:
           module: dnd/spells
           references:
-            npc_registry: %q
-        normalize: dnd/spells
+            npc_registry:
+              path: %q
+        normalize:
+          module: dnd/spells
 `, checkpointRoot, npcPath)
 	configPath := filepath.Join(t.TempDir(), "config.yml")
 	if err := os.WriteFile(configPath, []byte(content), 0o600); err != nil {

internal/cli/production_contract_test.go

@@ -19,8 +19,6 @@ import (
 	"testing/fstest"
 	"time"
 
-	"gopkg.in/yaml.v3"
-
 	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/chunkmap"
@@ -333,61 +331,8 @@ func TestProductionPromptAssetsPrepareWithoutProviderCredentials(t *testing.T) {
 	if err != nil {
 		t.Fatalf("production prompt engine: %v", err)
 	}
-	promptFS, err := components.assets.PromptFS()
-	if err != nil {
-		t.Fatalf("production prompt assets: %v", err)
-	}
-	type manifest struct {
-		ID      string `yaml:"id"`
-		Version string `yaml:"version"`
-		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 {
-		if walkErr != nil {
-			return walkErr
-		}
-		if entry.IsDir() || filepath.Base(path) != "prompt.yaml" {
-			return nil
-		}
-		data, err := fs.ReadFile(promptFS, path)
-		if err != nil {
-			return err
-		}
-		var prompt manifest
-		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(`{}`)
-		}
-		prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
-			PromptID: prompt.ID, PromptVersion: prompt.Version, ProfileID: "assembled-prompt-test", Inputs: inputs,
-		})
-		if err != nil {
-			return fmt.Errorf("prepare production prompt %q: %w", prompt.ID, err)
-		}
-		if prepared.OutputContract.RepairAttempts != 1 {
-			return fmt.Errorf("prompt %q repair attempts = %d, want 1", prompt.ID, prepared.OutputContract.RepairAttempts)
-		}
-		preparedPrompts++
-		return nil
-	}); err != nil {
-		t.Fatal(err)
-	}
-	if preparedPrompts == 0 {
-		t.Fatal("prepared no production prompts")
+	if engine == nil {
+		t.Fatal("production PromptKit engine is nil")
 	}
 }
 
@@ -612,11 +557,16 @@ func TestNormalizeOptionsSharesProductionProfileAssetsWithDefaultRuntime(t *test
 func TestProductionLLMClientFactoryUsesConfiguredLocalBackend(t *testing.T) {
 	var providerCalls atomic.Int32
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		providerCalls.Add(1)
+		call := providerCalls.Add(1)
 		if r.URL.Path != "/v1/chat/completions" {
 			t.Errorf("provider path = %q, want /v1/chat/completions", r.URL.Path)
 		}
 		w.Header().Set("Content-Type", "application/json")
+		if call == 1 {
+			w.WriteHeader(http.StatusServiceUnavailable)
+			_, _ = w.Write([]byte(`{"error":{"message":"transient"}}`))
+			return
+		}
 		_, _ = w.Write([]byte(`{
 			"choices": [{"message": {"role": "assistant", "content": "{\"ok\":true}"}}],
 			"usage": {"prompt_tokens": 3, "completion_tokens": 4, "total_tokens": 7}
@@ -651,6 +601,7 @@ output:
 
 	cfg := config.Default()
 	cfg.PromptKit.ProfileFile = profilePath
+	cfg.PromptKit.TransientRetries = 1
 	cfg.PromptKit.LocalBackend = &config.PromptKitLocalBackendConfig{
 		Endpoint:         server.URL + "/v1",
 		ConcurrencyLimit: 2,
@@ -679,8 +630,8 @@ output:
 	if err != nil {
 		t.Fatalf("CompleteStructured() error = %v, want nil", err)
 	}
-	if providerCalls.Load() != 1 {
-		t.Fatalf("provider calls = %d, want 1", providerCalls.Load())
+	if providerCalls.Load() != 2 {
+		t.Fatalf("provider calls = %d, want 2", providerCalls.Load())
 	}
 	provider, ok := client.(contracts.LLMProfileManifestProvider)
 	if !ok {
@@ -737,13 +688,13 @@ func TestProductionConfigValidationCoversModuleAndVariantFailures(t *testing.T)
 	}{
 		{
 			name:      "unknown module",
-			content:   replaceRequiredOnce(t, base, "    input: seriatim\n", "    input: missing/input\n"),
+			content:   replaceRequiredOnce(t, base, "    input:\n      module: seriatim\n", "    input:\n      module: missing/input\n"),
 			options:   productionCLIOptions(t),
 			fragments: []string{"pipeline \"dnd-session\"", "input", "missing/input"},
 		},
 		{
 			name:      "unknown validator",
-			content:   replaceRequiredOnce(t, base, "        extract: dnd/spells\n", "        extract:\n          module: dnd/spells\n          validators:\n            - module: missing/validator\n"),
+			content:   replaceRequiredOnce(t, base, "        extract:\n          module: dnd/spells\n", "        extract:\n          module: dnd/spells\n          validators:\n            - module: missing/validator\n"),
 			options:   productionCLIOptions(t),
 			fragments: []string{"validator", "missing/validator"},
 		},
@@ -755,7 +706,7 @@ func TestProductionConfigValidationCoversModuleAndVariantFailures(t *testing.T)
 		},
 		{
 			name:      "deterministic validator with profile",
-			content:   replaceRequiredOnce(t, base, "        extract: dnd/spells\n", "        extract:\n          module: dnd/spells\n          validators:\n            - module: generic/valid_json\n              llm_profile: forbidden-profile\n"),
+			content:   replaceRequiredOnce(t, base, "        extract:\n          module: dnd/spells\n", "        extract:\n          module: dnd/spells\n          validators:\n            - module: generic/valid_json\n              llm_profile: forbidden-profile\n"),
 			options:   productionCLIOptions(t),
 			fragments: []string{"deterministic validator", "llm_profile"},
 		},
@@ -779,7 +730,7 @@ func TestProductionConfigValidationCoversModuleAndVariantFailures(t *testing.T)
 
 func TestProductionNormalizeValidatorOverrideRemainsAuthoritative(t *testing.T) {
 	base := string(readRepositoryFile(t, "examples", "dnd-minimal.config.yml"))
-	content := replaceRequiredOnce(t, base, "        normalize: dnd/spells\n", "        normalize:\n          module: dnd/spells\n          validators:\n            - module: generic/always_accept\n            - module: generic/valid_json\n")
+	content := replaceRequiredOnce(t, base, "        normalize:\n          module: dnd/spells\n", "        normalize:\n          module: dnd/spells\n          validators:\n            - module: generic/always_accept\n            - module: generic/valid_json\n")
 	path := writeProductionContractConfig(t, content)
 	components := productionTestComponents(t)
 	effective, err := loadMaintainedExample(t, path).Resolve(resolveInputForMaintainedExample(components, "dnd-session"))
@@ -852,10 +803,10 @@ 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())
@@ -879,7 +830,7 @@ func maintainedExampleFiles(t *testing.T) []maintainedExample {
 
 func productionSpellCatalogContractConfig(t *testing.T) string {
 	t.Helper()
-	return fmt.Sprintf(`version: 4
+	return fmt.Sprintf(`version: 5
 cache:
   chunk_plans:
     mode: bypass
@@ -888,21 +839,26 @@ cache:
     directory: ""
 pipelines:
   dnd-session:
-    input: seriatim
+    input:
+      module: seriatim
     references:
-      party: %q
-      glossary: %q
+      party:
+        path: %q
+      glossary:
+        path: %q
     artifacts:
       spells:
         extract:
           module: dnd/spells
           retries: 2
           references:
-            spell_catalog: %q
+            spell_catalog:
+              path: %q
         normalize:
           module: dnd/spells
           references:
-            spell_catalog: %q
+            spell_catalog:
+              path: %q
 `, repositoryPath("examples", "dnd-party.txt"), repositoryPath("examples", "dnd-glossary.txt"), repositoryPath("examples", "dnd-spell-catalog.json"), repositoryPath("examples", "dnd-spell-catalog.json"))
 }
 
@@ -913,14 +869,11 @@ func writeProductionSpellCatalogContractConfig(t *testing.T) string {
 
 func loadMaintainedExample(t *testing.T, path string) config.Config {
 	t.Helper()
-	fileConfig, err := config.LoadFileConfig(path)
+	loaded, err := config.Load(path, config.LoadOptions{})
 	if err != nil {
 		t.Fatalf("load maintained config %q: %v", path, err)
 	}
-	cfg := config.Default()
-	if err := cfg.ApplyFileConfig(fileConfig); err != nil {
-		t.Fatalf("apply maintained config %q: %v", path, err)
-	}
+	cfg := loaded.RuntimeConfig()
 	if err := cfg.Validate(); err != nil {
 		t.Fatalf("validate maintained config %q: %v", path, err)
 	}
@@ -977,7 +930,7 @@ func productionRunOptions(t *testing.T, fake *productionFakeLLMClient) Options {
 }
 
 func productionRunConfig(outputRoot, chunkModule string) string {
-	return fmt.Sprintf(`version: 4
+	return fmt.Sprintf(`version: 5
 output:
   directory: %q
 cache:
@@ -988,8 +941,10 @@ debug:
   directory: %q
 pipelines:
   dnd-session:
-    input: seriatim
-    chunk: %s
+    input:
+      module: seriatim
+    chunk:
+      module: %s
     output:
       module: json
       options:
@@ -998,8 +953,10 @@ pipelines:
       - id: identify-items
         artifacts:
           item-registry:
-            extract: dnd/item-registry
-            normalize: dnd/item-registry
+            extract:
+              module: dnd/item-registry
+            normalize:
+              module: dnd/item-registry
       - id: extract-occurrences
         references:
           item_registry:
@@ -1008,9 +965,11 @@ pipelines:
               lane: item-registry
         artifacts:
           spells:
-            extract: dnd/spells
+            extract:
+              module: dnd/spells
           item-occurrences:
-            extract: dnd/item-occurrences
+            extract:
+              module: dnd/item-occurrences
 `, outputRoot, filepath.Join(filepath.Dir(outputRoot), "debug"), chunkModule)
 }
 

internal/cli/recompute_execution_contract_test.go

@@ -194,7 +194,7 @@ func (recomputeTestOutput) Encode(_ context.Context, req contracts.OutputRequest
 func newRecomputeTestRoots(t *testing.T) stateTestRoots {
 	t.Helper()
 	roots := newStateTestRoots(t)
-	config := fmt.Sprintf(`version: 4
+	config := fmt.Sprintf(`version: 5
 output:
   directory: %q
 cache:
@@ -208,19 +208,27 @@ debug:
   directory: %q
 pipelines:
   sample:
-    input: test/input
-    chunk: test/chunk
+    input:
+      module: test/input
+    chunk:
+      module: test/chunk
     steps:
       - id: first
         artifacts:
           producer:
-            extract: test/extract/producer
-            merge: test/merge
-            normalize: test/normalize
+            extract:
+              module: test/extract/producer
+            merge:
+              module: test/merge
+            normalize:
+              module: test/normalize
           unrelated:
-            extract: test/extract/unrelated
-            merge: test/merge
-            normalize: test/normalize
+            extract:
+              module: test/extract/unrelated
+            merge:
+              module: test/merge
+            normalize:
+              module: test/normalize
       - id: second
         references:
           upstream:
@@ -229,9 +237,12 @@ pipelines:
               lane: producer
         artifacts:
           middle:
-            extract: test/extract/middle
-            merge: test/merge
-            normalize: test/normalize
+            extract:
+              module: test/extract/middle
+            merge:
+              module: test/merge
+            normalize:
+              module: test/normalize
       - id: third
         references:
           upstream:
@@ -240,10 +251,14 @@ pipelines:
               lane: middle
         artifacts:
           dependent:
-            extract: test/extract/dependent
-            merge: test/merge
-            normalize: test/normalize
-    output: test/recompute-output
+            extract:
+              module: test/extract/dependent
+            merge:
+              module: test/merge
+            normalize:
+              module: test/normalize
+    output:
+      module: test/recompute-output
 `, roots.output, roots.plans, roots.checkpoints, roots.debug)
 	if err := os.WriteFile(roots.config, []byte(config), 0o600); err != nil {
 		t.Fatal(err)

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) {
@@ -217,7 +289,7 @@ func TestReferenceMaterializationSeparatesCLIAndConfigPathOrigins(t *testing.T)
 	workingDir := t.TempDir()
 	cfg := referenceContractConfig()
 	configPath := filepath.Join(configDir, "config.yml")
-	if err := os.WriteFile(configPath, []byte("version: 4\n"), 0o600); err != nil {
+	if err := os.WriteFile(configPath, []byte("version: 5\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
 	if err := os.WriteFile(filepath.Join(configDir, "required.txt"), []byte("config reference"), 0o600); err != nil {
@@ -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

@@ -4,7 +4,6 @@ import (
 	"context"
 	"crypto/sha256"
 	"encoding/hex"
-	"encoding/json"
 	"errors"
 	"flag"
 	"fmt"
@@ -16,6 +15,7 @@ import (
 	"strings"
 	"time"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/notarius/internal/buildinfo"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
@@ -34,8 +34,10 @@ const usage = `Usage:
   notarius help
   notarius --version
   notarius run <pipeline-id> --input path/to/source.json [--json] [flags]
-  notarius config validate --config path/to/config.yml [--pipeline pipeline-id] [--only lane-a,lane-b]
-  notarius pipelines list --config path/to/config.yml [--json]
+  notarius config validate --config path/to/config.yml [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
+  notarius config effective --config path/to/config.yml [--config-overlay name]
+  notarius config explain --config path/to/config.yml [--config-overlay name] <json-pointer>
+  notarius pipelines list --config path/to/config.yml [--config-overlay name] [--json]
 `
 
 type Options struct {
@@ -158,6 +160,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	fs := flag.NewFlagSet("run", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
+	configOverlay := fs.String("config-overlay", "", "configuration overlay")
 	inputPath := fs.String("input", "", "source input file path")
 	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
 	outputDir := fs.String("output-dir", "", "output directory")
@@ -177,7 +180,7 @@ 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 {
@@ -265,20 +268,12 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		return 2
 	}
 
-	cfg, loadedConfigPath, err := loadConfig(*configPath, opts)
+	loaded, loadedConfigPath, err := loadConfigKit(*configPath, *configOverlay, runConfigurationOverrides(chunkCache, *outputDir, *debugDir), opts)
 	if err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
+		writeConfigLoadError(stderr, err)
 		return 1
 	}
-	if chunkCache.set {
-		cfg.Cache.ChunkPlans.Mode = chunkCache.value
-	}
-	if dir := strings.TrimSpace(*outputDir); dir != "" {
-		cfg.Output.Directory = dir
-	}
-	if dir := strings.TrimSpace(*debugDir); dir != "" {
-		cfg.Debug.Directory = dir
-	}
+	cfg := loaded.RuntimeConfig()
 	if err := cfg.Validate(); err != nil {
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
 		return 1
@@ -336,6 +331,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		InputPath:               strings.TrimSpace(*inputPath),
 		ConfigPath:              loadedConfigPath,
 		ConfigSource:            configSource(*configPath),
+		ConfigOverlay:           effectiveConfigOverlay(*configOverlay),
 		OnlyLanes:               append([]string(nil), only...),
 		ChunkCacheOverride:      chunkCache.explicitValue(),
 		ReasoningEffortOverride: runtimeOverrides.ReasoningEffort,
@@ -375,7 +371,7 @@ 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,
 	})
@@ -394,7 +390,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	if err := writeSummary(summary, func() error { return summary.WriteInvocation(invocation) }); err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("write debug invocation metadata: %w", err))
 	}
-	if err := writeSummary(summary, func() error { return summary.WriteRedactedEffectiveConfig(effective) }); err != nil {
+	if err := writeSummary(summary, func() error { return summary.WriteEffectiveConfig(loaded) }); err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("write debug effective config: %w", err))
 	}
 	if err := writeSummary(summary, func() error { return summary.WriteResolvedPipeline(effective) }); err != nil {
@@ -462,7 +458,7 @@ 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 +467,10 @@ 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 +478,21 @@ 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 +518,25 @@ 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 +556,7 @@ 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)
@@ -859,13 +877,39 @@ func reorderRunArgs(args []string) []string {
 
 func runFlagTakesValue(arg string) bool {
 	switch arg {
-	case "--config", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--reasoning-effort", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
+	case "--config", "--config-overlay", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--reasoning-effort", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
 		return true
 	default:
 		return false
 	}
 }
 
+func runConfigurationOverrides(chunkCache chunkCacheFlag, outputDir, debugDir string) []configkit.Override {
+	overrides := make([]configkit.Override, 0, 3)
+	if chunkCache.set {
+		overrides = append(overrides, configkit.Override{
+			Path:  configkit.Keys("cache", "chunk_plans", "mode"),
+			Value: string(chunkCache.value),
+			Flag:  "--chunk_cache",
+		})
+	}
+	if outputDir = strings.TrimSpace(outputDir); outputDir != "" {
+		overrides = append(overrides, configkit.Override{
+			Path:  configkit.Keys("output", "directory"),
+			Value: outputDir,
+			Flag:  "--output-dir",
+		})
+	}
+	if debugDir = strings.TrimSpace(debugDir); debugDir != "" {
+		overrides = append(overrides, configkit.Override{
+			Path:  configkit.Keys("debug", "directory"),
+			Value: debugDir,
+			Flag:  "--debug-dir",
+		})
+	}
+	return overrides
+}
+
 type chunkCacheFlag struct {
 	value pipeline.ChunkCacheMode
 	set   bool
@@ -919,7 +963,7 @@ func chunkPlanStoreForRun(cfg config.ChunkPlanCacheConfig, opts Options) (pipeli
 
 func validateRunFlagValues(args []string) error {
 	for i, arg := range args {
-		if arg != "--session-id" && arg != "--reasoning-effort" {
+		if arg != "--config-overlay" && arg != "--session-id" && arg != "--reasoning-effort" {
 			continue
 		}
 		if i+1 >= len(args) || strings.HasPrefix(args[i+1], "-") {
@@ -929,6 +973,13 @@ func validateRunFlagValues(args []string) error {
 	return nil
 }
 
+func effectiveConfigOverlay(overlay string) string {
+	if overlay == "" || overlay == "default" {
+		return ""
+	}
+	return overlay
+}
+
 func flagWasProvided(args []string, name string) bool {
 	for _, arg := range args {
 		if arg == name || strings.HasPrefix(arg, name+"=") {
@@ -1004,6 +1055,10 @@ func runConfig(args []string, stdout, stderr io.Writer, opts Options) int {
 	switch args[0] {
 	case "validate":
 		return runConfigValidate(args[1:], stdout, stderr, opts)
+	case "effective":
+		return runConfigEffective(args[1:], stdout, stderr, opts)
+	case "explain":
+		return runConfigExplain(args[1:], stdout, stderr, opts)
 	default:
 		fmt.Fprintf(stderr, "notarius: unknown config subcommand %q\n", args[0])
 		writeUsage(stderr)
@@ -1011,67 +1066,6 @@ func runConfig(args []string, stdout, stderr io.Writer, opts Options) int {
 	}
 }
 
-func runConfigValidate(args []string, stdout, stderr io.Writer, opts Options) int {
-	fs := flag.NewFlagSet("config validate", flag.ContinueOnError)
-	fs.SetOutput(io.Discard)
-	configPath := fs.String("config", "", "config file path")
-	pipelineID := fs.String("pipeline", "", "pipeline ID")
-	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
-	if err := fs.Parse(args); err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 2
-	}
-	if fs.NArg() != 0 {
-		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
-		return 2
-	}
-	if strings.TrimSpace(*onlyRaw) != "" && strings.TrimSpace(*pipelineID) == "" {
-		fmt.Fprintln(stderr, "notarius: --only requires --pipeline")
-		return 2
-	}
-	only, err := parseOnly(*onlyRaw)
-	if err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 2
-	}
-
-	cfg, path, err := loadConfig(*configPath, opts)
-	if err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 1
-	}
-
-	if strings.TrimSpace(*pipelineID) != "" {
-		catalog, err := effectiveCatalog(opts)
-		if err != nil {
-			fmt.Fprintf(stderr, "notarius: %v\n", err)
-			return 1
-		}
-		effective, err := cfg.Resolve(config.ResolveInput{
-			PipelineID: *pipelineID,
-			Only:       only,
-			Catalog:    catalog,
-		})
-		if err != nil {
-			fmt.Fprintf(stderr, "notarius: %v\n", err)
-			return 1
-		}
-		if err := validateExplicitPromptKitProfiles(context.Background(), effective.Config, effectiveLLMProfileIDs(effective.ResolvedPipeline), opts.promptKitAssets); err != nil {
-			fmt.Fprintf(stderr, "notarius: %v\n", err)
-			return 1
-		}
-		fmt.Fprintf(stdout, "config %q is valid for pipeline %q\n", path, strings.TrimSpace(*pipelineID))
-		return 0
-	}
-
-	if err := cfg.Validate(); err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 1
-	}
-	fmt.Fprintf(stdout, "config %q is valid\n", path)
-	return 0
-}
-
 func runPipelines(args []string, stdout, stderr io.Writer, opts Options) int {
 	if len(args) == 0 {
 		fmt.Fprintln(stderr, "notarius: pipelines requires a subcommand")
@@ -1088,70 +1082,6 @@ func runPipelines(args []string, stdout, stderr io.Writer, opts Options) int {
 	}
 }
 
-func runPipelinesList(args []string, stdout, stderr io.Writer, opts Options) int {
-	fs := flag.NewFlagSet("pipelines list", flag.ContinueOnError)
-	fs.SetOutput(io.Discard)
-	configPath := fs.String("config", "", "config file path")
-	jsonOutput := fs.Bool("json", false, "write JSON output")
-	if err := fs.Parse(args); err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 2
-	}
-	if fs.NArg() != 0 {
-		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
-		return 2
-	}
-
-	cfg, _, err := loadConfig(*configPath, opts)
-	if err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 1
-	}
-	if err := cfg.Validate(); err != nil {
-		fmt.Fprintf(stderr, "notarius: %v\n", err)
-		return 1
-	}
-
-	ids := sortedPipelineIDs(cfg)
-	if *jsonOutput {
-		payload := struct {
-			Pipelines []string `json:"pipelines"`
-		}{Pipelines: ids}
-		encoded, err := json.Marshal(payload)
-		if err != nil {
-			fmt.Fprintf(stderr, "notarius: marshal pipeline list: %v\n", err)
-			return 1
-		}
-		fmt.Fprintf(stdout, "%s\n", encoded)
-		return 0
-	}
-
-	for _, id := range ids {
-		fmt.Fprintln(stdout, id)
-	}
-	return 0
-}
-
-func loadConfig(configPath string, opts Options) (config.Config, string, error) {
-	path, err := discoverConfigPath(configPath, opts)
-	if err != nil {
-		return config.Config{}, "", err
-	}
-
-	fileCfg, err := config.LoadFileConfig(path)
-	if err != nil {
-		return config.Config{}, "", err
-	}
-	cfg := config.Default()
-	if err := cfg.ApplyFileConfigWithLookup(fileCfg, opts.LookupEnv); err != nil {
-		return config.Config{}, "", err
-	}
-	if err := cfg.ApplyEnvOverridesWithLookup(opts.LookupEnv); err != nil {
-		return config.Config{}, "", err
-	}
-	return cfg, path, nil
-}
-
 func discoverConfigPath(configPath string, opts Options) (string, error) {
 	if path := strings.TrimSpace(configPath); path != "" {
 		if err := requireConfigFile(path); err != nil {
@@ -1265,11 +1195,21 @@ 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 +1218,7 @@ 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 +1268,14 @@ 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 +1283,9 @@ 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 +1306,153 @@ 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 +1669,77 @@ 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 +1747,19 @@ 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

@@ -12,6 +12,7 @@ import (
 
 	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+	"gitea.maximumdirect.net/eric/notarius/internal/core/debugbundle"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/checkpoint"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
@@ -37,6 +38,9 @@ func TestRunControlsRejectSyntaxWithoutAllocatingState(t *testing.T) {
 		{name: "blank debug directory", args: func(roots stateTestRoots) []string {
 			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--debug-dir", ""}
 		}},
+		{name: "missing configuration overlay", args: func(roots stateTestRoots) []string {
+			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--config-overlay"}
+		}},
 		{name: "debug directory without debug", args: func(roots stateTestRoots) []string {
 			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--debug-dir", filepath.Join(filepath.Dir(roots.debug), "requested-debug")}
 		}},
@@ -79,7 +83,7 @@ func TestRecomputeStepCLIContract(t *testing.T) {
 		{
 			name: "explicit step",
 			configure: func(t *testing.T, roots stateTestRoots) {
-				replaceStateTestConfigLine(t, roots.config, "    artifacts:\n      items:\n        extract: test/extract\n        merge: test/merge\n        normalize: test/normalize\n", "    steps:\n      - id: chosen\n        artifacts:\n          items:\n            extract: test/extract\n            merge: test/merge\n            normalize: test/normalize\n")
+				replaceStateTestConfigLine(t, roots.config, "    artifacts:\n      items:\n        extract:\n          module: test/extract\n        merge:\n          module: test/merge\n        normalize:\n          module: test/normalize\n", "    steps:\n      - id: chosen\n        artifacts:\n          items:\n            extract:\n              module: test/extract\n            merge:\n              module: test/merge\n            normalize:\n              module: test/normalize\n")
 			},
 			flags:      []string{"--resume", "--recompute-step", "chosen"},
 			wantCode:   0,
@@ -161,7 +165,7 @@ func TestRunOnlyExecutesSelectedLanes(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	data = []byte(replaceRequiredOnce(t, string(data), "    output: test/output\n", "      other:\n        extract: test/extract\n    output: test/output\n"))
+	data = []byte(replaceRequiredOnce(t, string(data), "    output:\n      module: test/output\n", "      other:\n        extract:\n          module: test/extract\n    output:\n      module: test/output\n"))
 	if err := os.WriteFile(roots.config, data, 0o600); err != nil {
 		t.Fatal(err)
 	}
@@ -180,7 +184,7 @@ func TestRunOnlyExecutesSelectedLanes(t *testing.T) {
 }
 
 func TestRunStateRootsHonorEnvironmentFlagsAndDefaults(t *testing.T) {
-	t.Run("environment roots", func(t *testing.T) {
+	t.Run("retired environment roots have no effect", func(t *testing.T) {
 		roots := newStateTestRoots(t)
 		environmentOutput := filepath.Join(t.TempDir(), "environment-output")
 		environmentDebug := filepath.Join(t.TempDir(), "environment-debug")
@@ -193,10 +197,10 @@ func TestRunStateRootsHonorEnvironmentFlagsAndDefaults(t *testing.T) {
 		if result.code != 0 {
 			t.Fatalf("code=%d stdout=%q stderr=%q", result.code, result.stdout, result.stderr)
 		}
-		assertFile(t, filepath.Join(environmentOutput, filepath.Base(onlyChildDir(t, environmentOutput)), "result.json"))
-		onlyChildDir(t, environmentDebug)
-		assertAbsent(t, roots.output)
-		assertAbsent(t, roots.debug)
+		assertFile(t, filepath.Join(roots.output, filepath.Base(onlyChildDir(t, roots.output)), "result.json"))
+		onlyChildDir(t, roots.debug)
+		assertAbsent(t, environmentOutput)
+		assertAbsent(t, environmentDebug)
 	})
 
 	t.Run("command flags override environment", func(t *testing.T) {
@@ -227,23 +231,57 @@ func TestRunStateRootsHonorEnvironmentFlagsAndDefaults(t *testing.T) {
 			t.Fatal(err)
 		}
 		text := string(data)
-		text = replaceRequiredOnce(t, text, fmt.Sprintf("  directory: %q\n", roots.output), "")
-		text = replaceRequiredOnce(t, text, fmt.Sprintf("  directory: %q\n", roots.debug), "")
+		text = replaceRequiredOnce(t, text, fmt.Sprintf("output:\n  directory: %q\n", roots.output), "")
+		text = replaceRequiredOnce(t, text, fmt.Sprintf("debug:\n  directory: %q\n", roots.debug), "")
 		if err := os.WriteFile(roots.config, []byte(text), 0o600); err != nil {
 			t.Fatal(err)
 		}
-		workDir := t.TempDir()
-		t.Chdir(workDir)
 		opts := newStateTestHarness().options()
 		result := runWithStateRoots(t, roots, opts, nil)
 		if result.code != 0 {
 			t.Fatalf("code=%d stdout=%q stderr=%q", result.code, result.stdout, result.stderr)
 		}
-		assertFile(t, filepath.Join(workDir, "notarius-output", filepath.Base(onlyChildDir(t, filepath.Join(workDir, "notarius-output"))), "result.json"))
-		onlyChildDir(t, filepath.Join(workDir, "notarius-debug"))
+		configDirectory := filepath.Dir(roots.config)
+		assertFile(t, filepath.Join(configDirectory, "notarius-output", filepath.Base(onlyChildDir(t, filepath.Join(configDirectory, "notarius-output"))), "result.json"))
+		onlyChildDir(t, filepath.Join(configDirectory, "notarius-debug"))
 	})
 }
 
+func TestRunAppliesConfigurationOverlayAndRecordsSelection(t *testing.T) {
+	roots := newStateTestRoots(t)
+	overlayDirectory := filepath.Join(filepath.Dir(roots.config), "overlays")
+	overlayOutput := filepath.Join(overlayDirectory, "overlay-output")
+	overlayDebug := filepath.Join(overlayDirectory, "overlay-debug")
+	writeConfigKitFile(t, filepath.Join(overlayDirectory, "development.yml"), "output:\n  directory: ./overlay-output\ndebug:\n  directory: ./overlay-debug\n")
+
+	var stdout, stderr bytes.Buffer
+	code := RunWithOptions([]string{
+		"run", "sample", "--config", roots.config, "--config-overlay", "development",
+		"--input", roots.input, "--chunk_cache", "bypass", "--debug",
+	}, &stdout, &stderr, newStateTestHarness().options())
+	if code != 0 || stderr.Len() != 0 {
+		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+	}
+	assertFile(t, filepath.Join(overlayOutput, filepath.Base(onlyChildDir(t, overlayOutput)), "result.json"))
+	bundle := onlyChildDir(t, overlayDebug)
+	var invocation debugbundle.Invocation
+	readStateTestSummaryJSON(t, bundle, "invocation.json", &invocation)
+	if invocation.ConfigOverlay != "development" {
+		t.Fatalf("config overlay = %q, want development", invocation.ConfigOverlay)
+	}
+}
+
+func TestEffectiveConfigOverlayOmitsBaseSelection(t *testing.T) {
+	for _, overlay := range []string{"", "default"} {
+		if got := effectiveConfigOverlay(overlay); got != "" {
+			t.Fatalf("effectiveConfigOverlay(%q) = %q, want empty", overlay, got)
+		}
+	}
+	if got := effectiveConfigOverlay("development"); got != "development" {
+		t.Fatalf("effectiveConfigOverlay(development) = %q", got)
+	}
+}
+
 func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 	t.Run("one effective profile reaches the factory and modules", func(t *testing.T) {
 		roots := newStateTestRoots(t)
@@ -567,7 +605,7 @@ func TestRunFactoryAndPreparationFailuresAreProcessFailures(t *testing.T) {
 		if err != nil {
 			t.Fatal(err)
 		}
-		data = []byte(replaceRequiredOnce(t, string(data), "extract: test/extract", "extract: test/failing-extract"))
+		data = []byte(replaceRequiredOnce(t, string(data), "module: test/extract", "module: test/failing-extract"))
 		if err := os.WriteFile(roots.config, data, 0o600); err != nil {
 			t.Fatal(err)
 		}
@@ -590,10 +628,11 @@ 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 +640,12 @@ 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

@@ -8,26 +8,39 @@ import (
 	"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"`
-	ValidationSummaries   []artifacts.ValidationSummary `json:"validation_summaries,omitempty"`
-	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")
 	}
@@ -53,15 +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,
-		ValidationSummaries:   cloneValidationSummaries(output.Manifest.ValidationSummaries),
+		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) != "" {

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          validation_policy:\n            semantic_rejection: reject_output\n"))
+	configBytes = []byte(replaceRequiredOnce(t, string(configBytes), "        normalize:\n          module: test/normalize\n", "        normalize:\n          module: test/normalize\n          validators:\n            - module: 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 +52,8 @@ 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)
@@ -182,8 +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",
 		},
 	}
 
@@ -92,8 +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) != 1 || output.Warnings[0].ReasonCode != "spell_not_near_source" {
-					t.Fatalf("warnings = %#v, want complete terminal validation warnings", 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
 			}
@@ -108,8 +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

@@ -57,7 +57,6 @@ func TestRunStateSurfaceMatrix(t *testing.T) {
 						}
 					}
 					assertAnyFile(t, roots.checkpoints)
-					assertRestrictedTree(t, roots.checkpoints)
 					if debug {
 						bundle := onlyChildDir(t, roots.debug)
 						assertFile(t, filepath.Join(bundle, "summary", "invocation.json"))
@@ -259,7 +258,7 @@ func TestRunRedactsSensitiveModuleOptionsFromConfigAndPipelineSummaries(t *testi
 	if err != nil {
 		t.Fatal(err)
 	}
-	configText := replaceRequiredOnce(t, string(data), "    input: test/input\n", `    input:
+	configText := replaceRequiredOnce(t, string(data), "    input:\n      module: test/input\n", `    input:
       module: test/input
       options:
         api_key: CONFIG_SUMMARY_SECRET_SENTINEL
@@ -277,7 +276,7 @@ func TestRunRedactsSensitiveModuleOptionsFromConfigAndPipelineSummaries(t *testi
 		t.Fatalf("code=%d stderr=%q", result.code, result.stderr)
 	}
 	summaryRoot := filepath.Join(onlyChildDir(t, roots.debug), "summary")
-	for _, name := range []string{"effective-config.json", "resolved-pipeline.json"} {
+	for _, name := range []string{"effective-config.yml", "resolved-pipeline.json"} {
 		contents, err := os.ReadFile(filepath.Join(summaryRoot, name))
 		if err != nil {
 			t.Fatal(err)
@@ -349,7 +348,7 @@ 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 +430,7 @@ 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 +444,7 @@ 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 +453,7 @@ 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 +462,10 @@ 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)
@@ -688,7 +687,7 @@ func newStateTestRoots(t *testing.T) stateTestRoots {
 		t.Fatal(err)
 	}
 	roots.config = filepath.Join(base, "config.yml")
-	config := fmt.Sprintf("version: 4\noutput:\n  directory: %q\ncache:\n  chunk_plans:\n    directory: %q\n    mode: auto\n  checkpoints:\n    enabled: true\n    directory: %q\ndebug:\n  directory: %q\npipelines:\n  sample:\n    input: test/input\n    chunk: test/chunk\n    artifacts:\n      items:\n        extract: test/extract\n        merge: test/merge\n        normalize: test/normalize\n    output: test/output\n", roots.output, roots.plans, roots.checkpoints, roots.debug)
+	config := fmt.Sprintf("version: 5\noutput:\n  directory: %q\ncache:\n  chunk_plans:\n    directory: %q\n    mode: auto\n  checkpoints:\n    enabled: true\n    directory: %q\ndebug:\n  directory: %q\npipelines:\n  sample:\n    input:\n      module: test/input\n    chunk:\n      module: test/chunk\n    artifacts:\n      items:\n        extract:\n          module: test/extract\n        merge:\n          module: test/merge\n        normalize:\n          module: test/normalize\n    output:\n      module: test/output\n", roots.output, roots.plans, roots.checkpoints, roots.debug)
 	if err := os.WriteFile(roots.config, []byte(config), 0o600); err != nil {
 		t.Fatal(err)
 	}
@@ -863,11 +862,12 @@ 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 +892,7 @@ 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 +931,7 @@ 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 +999,28 @@ 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/cli/state_permissions_unix_test.go

@@ -0,0 +1,53 @@
+//go:build linux || darwin
+
+package cli
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"syscall"
+	"testing"
+)
+
+func TestRunCreatesReusableGroupAccessibleCacheState(t *testing.T) {
+	previousUmask := syscall.Umask(0o007)
+	defer syscall.Umask(previousUmask)
+
+	roots := newStateTestRoots(t)
+	harness := newStateTestHarness()
+	result := runStateTest(t, roots, harness.options(), false, false, "auto")
+	if result.code != 0 {
+		t.Fatalf("initial run code=%d stderr=%q", result.code, result.stderr)
+	}
+	assertGroupAccessibleCacheTree(t, roots.plans)
+	assertGroupAccessibleCacheTree(t, roots.checkpoints)
+
+	extractCalls := harness.extractCalls
+	result = runStateTest(t, roots, harness.options(), false, true, "auto")
+	if result.code != 0 {
+		t.Fatalf("resume run code=%d stderr=%q", result.code, result.stderr)
+	}
+	if harness.extractCalls != extractCalls {
+		t.Fatalf("extract calls after reuse = %d, want %d", harness.extractCalls, extractCalls)
+	}
+}
+
+func assertGroupAccessibleCacheTree(t *testing.T, root string) {
+	t.Helper()
+	if err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+		want := os.FileMode(0o660)
+		if info.IsDir() {
+			want = 0o770
+		}
+		if got := info.Mode().Perm(); got != want {
+			return fmt.Errorf("%s mode = %#o, want %#o", path, got, want)
+		}
+		return nil
+	}); err != nil {
+		t.Fatal(err)
+	}
+}

internal/core/config/config.go

@@ -4,8 +4,6 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
-const SupportedFileConfigVersion = 4
-
 type Config struct {
 	PromptKit   PromptKitConfig                     `json:"promptkit,omitempty"`
 	Pipelines   map[string]pipeline.PipelineProfile `json:"pipelines"`
@@ -16,9 +14,10 @@ type Config struct {
 }
 
 type PromptKitConfig struct {
-	ProfileDir   string                       `json:"profile_dir,omitempty"`
-	ProfileFile  string                       `json:"profile_file,omitempty"`
-	LocalBackend *PromptKitLocalBackendConfig `json:"local_backend,omitempty"`
+	ProfileDir       string                       `json:"profile_dir,omitempty"`
+	ProfileFile      string                       `json:"profile_file,omitempty"`
+	TransientRetries int                          `json:"transient_retries"`
+	LocalBackend     *PromptKitLocalBackendConfig `json:"local_backend,omitempty"`
 }
 
 type PromptKitLocalBackendConfig struct {
@@ -29,12 +28,12 @@ type PromptKitLocalBackendConfig struct {
 type ConcurrencyConfig struct {
 	TotalLLM     int            `json:"total_llm"`
 	StageWorkers map[string]int `json:"stage_workers"`
-
-	extractWorkersConfigured bool
-	defaultedExtractWorkers  int
 }
 
-const defaultLLMConcurrency = 16
+const (
+	defaultLLMConcurrency   = 16
+	defaultTransientRetries = 2
+)
 
 type OutputConfig struct {
 	Directory string `json:"directory"`
@@ -60,11 +59,11 @@ type DebugConfig struct {
 
 func Default() Config {
 	return Config{
+		PromptKit: PromptKitConfig{TransientRetries: defaultTransientRetries},
 		Pipelines: map[string]pipeline.PipelineProfile{},
 		Concurrency: ConcurrencyConfig{
-			TotalLLM:                defaultLLMConcurrency,
-			StageWorkers:            map[string]int{"extract": defaultLLMConcurrency},
-			defaultedExtractWorkers: defaultLLMConcurrency,
+			TotalLLM:     defaultLLMConcurrency,
+			StageWorkers: map[string]int{"extract": defaultLLMConcurrency},
 		},
 		Output: OutputConfig{Directory: "./notarius-output"},
 		Cache:  CacheConfig{ChunkPlans: ChunkPlanCacheConfig{Mode: pipeline.ChunkCacheAuto}},
@@ -97,21 +96,15 @@ func cloneIntMap(in map[string]int) map[string]int {
 	return out
 }
 
-func (c *ConcurrencyConfig) recomputeStageWorkerDefaults() {
-	if c == nil {
-		return
-	}
-	if c.StageWorkers == nil {
-		c.StageWorkers = make(map[string]int)
-	}
-	if !c.extractWorkersConfigured {
-		if value, ok := c.StageWorkers["extract"]; ok && (c.defaultedExtractWorkers == 0 || value != c.defaultedExtractWorkers) {
-			c.extractWorkersConfigured = true
-			return
+func (c Config) withDerivedStageWorkers() Config {
+	c = cloneConfig(c)
+	if _, exists := c.Concurrency.StageWorkers["extract"]; !exists {
+		if c.Concurrency.StageWorkers == nil {
+			c.Concurrency.StageWorkers = make(map[string]int)
 		}
-		c.StageWorkers["extract"] = c.TotalLLM
-		c.defaultedExtractWorkers = c.TotalLLM
+		c.Concurrency.StageWorkers["extract"] = c.Concurrency.TotalLLM
 	}
+	return c
 }
 
 func clonePipelineProfile(in pipeline.PipelineProfile) pipeline.PipelineProfile {

internal/core/config/document.go

@@ -0,0 +1,123 @@
+package config
+
+import "gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
+
+// Document is the version 5 configuration shape decoded by ConfigKit. It is
+// deliberately separate from Config so YAML presence and composition concerns
+// do not reach runtime pipeline types.
+type Document struct {
+	Version     string                      `yaml:"version"`
+	PromptKit   *PromptKitDocument          `yaml:"promptkit,omitempty"`
+	Pipelines   map[string]PipelineDocument `yaml:"pipelines,omitempty"`
+	Concurrency *ConcurrencyDocument        `yaml:"concurrency,omitempty"`
+	Output      *OutputDocument             `yaml:"output,omitempty"`
+	Cache       *CacheDocument              `yaml:"cache,omitempty"`
+	Debug       *DebugDocument              `yaml:"debug,omitempty"`
+}
+
+type PromptKitDocument struct {
+	ProfileDir       *string                        `yaml:"profile_dir,omitempty"`
+	ProfileFile      *string                        `yaml:"profile_file,omitempty"`
+	TransientRetries *int                           `yaml:"transient_retries,omitempty"`
+	LocalBackend     *PromptKitLocalBackendDocument `yaml:"local_backend,omitempty"`
+}
+
+type PromptKitLocalBackendDocument struct {
+	Endpoint         *string `yaml:"endpoint,omitempty"`
+	ConcurrencyLimit *int    `yaml:"concurrency_limit,omitempty"`
+}
+
+type PipelineDocument struct {
+	LLMProfile                     *string                         `yaml:"llm_profile,omitempty"`
+	StructuredOutputRepairAttempts *int                            `yaml:"structured_output_repair_attempts,omitempty"`
+	ValidationPolicy               *ValidationPolicyDocument       `yaml:"validation_policy,omitempty"`
+	Input                          BindingDocument                 `yaml:"input"`
+	Chunk                          *BindingDocument                `yaml:"chunk,omitempty"`
+	Artifacts                      map[string]ArtifactLaneDocument `yaml:"artifacts,omitempty"`
+	Steps                          []PipelineStepDocument          `yaml:"steps,omitempty"`
+	Output                         *BindingDocument                `yaml:"output,omitempty"`
+	References                     map[string]ReferenceDocument    `yaml:"references,omitempty"`
+}
+
+type PipelineStepDocument struct {
+	ID         string                          `yaml:"id"`
+	Artifacts  map[string]ArtifactLaneDocument `yaml:"artifacts"`
+	References map[string]ReferenceDocument    `yaml:"references,omitempty"`
+}
+
+type ArtifactLaneDocument struct {
+	Extract    BindingDocument              `yaml:"extract"`
+	Merge      *BindingDocument             `yaml:"merge,omitempty"`
+	Normalize  *BindingDocument             `yaml:"normalize,omitempty"`
+	Validators []ValidatorDocument          `yaml:"validators,omitempty"`
+	References map[string]ReferenceDocument `yaml:"references,omitempty"`
+}
+
+// BindingDocument represents a producer binding. ValidatorDocument is a
+// deliberately finite subset so nested validator-only fields are rejected by
+// ConfigKit's strict decoder.
+type BindingDocument struct {
+	Module                         string                       `yaml:"module"`
+	LLMProfile                     *string                      `yaml:"llm_profile,omitempty"`
+	StructuredOutputRepairAttempts *int                         `yaml:"structured_output_repair_attempts,omitempty"`
+	ValidationPolicy               *ValidationPolicyDocument    `yaml:"validation_policy,omitempty"`
+	Retries                        int                          `yaml:"retries,omitempty"`
+	Options                        map[string]any               `yaml:"options,omitempty"`
+	References                     map[string]ReferenceDocument `yaml:"references,omitempty"`
+	Validators                     []ValidatorDocument          `yaml:"validators,omitempty"`
+}
+
+type ValidatorDocument struct {
+	Module                         string         `yaml:"module"`
+	LLMProfile                     *string        `yaml:"llm_profile,omitempty"`
+	StructuredOutputRepairAttempts *int           `yaml:"structured_output_repair_attempts,omitempty"`
+	Retries                        int            `yaml:"retries,omitempty"`
+	Options                        map[string]any `yaml:"options,omitempty"`
+}
+
+// ReferenceDocument is the explicit version 5 union. Later document
+// validation establishes which locations permit generated references and
+// requires exactly one populated branch.
+type ReferenceDocument struct {
+	Path     *string                    `yaml:"path,omitempty"`
+	Artifact *ArtifactReferenceDocument `yaml:"artifact,omitempty"`
+}
+
+type ArtifactReferenceDocument struct {
+	Step string `yaml:"step"`
+	Lane string `yaml:"lane"`
+}
+
+type ValidationPolicyDocument struct {
+	ProducerStructuralFailure *pipeline.ProducerStructuralFailureAction `yaml:"producer_structural_failure,omitempty"`
+	SemanticRejection         *pipeline.SemanticRejectionAction         `yaml:"semantic_rejection,omitempty"`
+	ValidatorFailure          *pipeline.ValidatorFailureAction          `yaml:"validator_failure,omitempty"`
+}
+
+type ConcurrencyDocument struct {
+	TotalLLM     *int           `yaml:"total_llm,omitempty"`
+	StageWorkers map[string]int `yaml:"stage_workers,omitempty"`
+}
+
+type OutputDocument struct {
+	Directory *string `yaml:"directory,omitempty"`
+}
+
+type CacheDocument struct {
+	ChunkPlans  *ChunkPlanCacheDocument  `yaml:"chunk_plans,omitempty"`
+	Checkpoints *CheckpointCacheDocument `yaml:"checkpoints,omitempty"`
+}
+
+type ChunkPlanCacheDocument struct {
+	Directory *string `yaml:"directory,omitempty"`
+	Mode      *string `yaml:"mode,omitempty"`
+}
+
+type CheckpointCacheDocument struct {
+	Enabled   *bool   `yaml:"enabled,omitempty"`
+	Directory *string `yaml:"directory,omitempty"`
+}
+
+type DebugDocument struct {
+	Directory *string `yaml:"directory,omitempty"`
+}

internal/core/config/document_normalizer.go

@@ -0,0 +1,26 @@
+package config
+
+import configkit "gitea.maximumdirect.net/eric/ConfigKit"
+
+// documentNormalizer records leaf provenance and, while visiting a sequence,
+// covers its atomic mutation boundary. ConfigKit audits sequences as whole
+// values, so a record for only a descendant is insufficient. The outermost
+// sequence covers nested sequences as well; unchanged visits record nothing.
+type documentNormalizer struct {
+	*configkit.Resolution
+	sequence *configkit.Path
+}
+
+func (n documentNormalizer) inSequence(path configkit.Path) documentNormalizer {
+	if n.sequence == nil {
+		n.sequence = &path
+	}
+	return n
+}
+
+func (n documentNormalizer) Record(path configkit.Path, kind configkit.ChangeKind, reason string, dependencies ...configkit.Path) {
+	n.Resolution.Record(path, kind, reason, dependencies...)
+	if n.sequence != nil {
+		n.Resolution.Record(*n.sequence, kind, reason, append([]configkit.Path{path}, dependencies...)...)
+	}
+}

internal/core/config/document_paths.go

@@ -0,0 +1,167 @@
+package config
+
+import (
+	"path/filepath"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+)
+
+func normalizeDocumentPaths(document *Document, context *configkit.Resolution, state *documentLoadState) {
+	resolution := documentNormalizer{Resolution: context}
+	if document == nil {
+		return
+	}
+	initializeDocumentMainDirectory(resolution, state)
+	if document.PromptKit != nil {
+		normalizeDocumentPath(resolution, state, configkit.Keys("promptkit", "profile_dir"), &document.PromptKit.ProfileDir, false, "resolved PromptKit profile directory")
+		normalizeDocumentPath(resolution, state, configkit.Keys("promptkit", "profile_file"), &document.PromptKit.ProfileFile, false, "resolved PromptKit profile file")
+	}
+	if document.Output != nil {
+		normalizeDocumentPath(resolution, state, configkit.Keys("output", "directory"), &document.Output.Directory, false, "resolved output directory")
+	}
+	if document.Cache != nil {
+		if document.Cache.ChunkPlans != nil {
+			normalizeDocumentPath(resolution, state, configkit.Keys("cache", "chunk_plans", "directory"), &document.Cache.ChunkPlans.Directory, true, "resolved chunk-plan cache directory")
+		}
+		if document.Cache.Checkpoints != nil {
+			normalizeDocumentPath(resolution, state, configkit.Keys("cache", "checkpoints", "directory"), &document.Cache.Checkpoints.Directory, true, "resolved checkpoint cache directory")
+		}
+	}
+	if document.Debug != nil {
+		normalizeDocumentPath(resolution, state, configkit.Keys("debug", "directory"), &document.Debug.Directory, false, "resolved debug directory")
+	}
+	for _, id := range sortedPipelineDocumentKeys(document.Pipelines) {
+		profile := document.Pipelines[id]
+		path := configkit.Keys("pipelines").Key(id)
+		normalizeDocumentReferencePaths(profile.References, path.Key("references"), resolution, state)
+		normalizeDocumentBindingReferencePaths(&profile.Input, path.Key("input"), resolution, state)
+		if profile.Chunk != nil {
+			normalizeDocumentBindingReferencePaths(profile.Chunk, path.Key("chunk"), resolution, state)
+		}
+		if profile.Output != nil {
+			normalizeDocumentBindingReferencePaths(profile.Output, path.Key("output"), resolution, state)
+		}
+		for laneID, lane := range profile.Artifacts {
+			normalizeDocumentLaneReferencePaths(&lane, path.Key("artifacts").Key(laneID), resolution, state)
+			profile.Artifacts[laneID] = lane
+		}
+		stepResolution := resolution.inSequence(path.Key("steps"))
+		for index := range profile.Steps {
+			step := &profile.Steps[index]
+			stepPath := path.Key("steps").Index(index)
+			normalizeDocumentReferencePaths(step.References, stepPath.Key("references"), stepResolution, state)
+			for laneID, lane := range step.Artifacts {
+				normalizeDocumentLaneReferencePaths(&lane, stepPath.Key("artifacts").Key(laneID), stepResolution, state)
+				step.Artifacts[laneID] = lane
+			}
+		}
+		document.Pipelines[id] = profile
+	}
+}
+
+func initializeDocumentMainDirectory(resolution documentNormalizer, state *documentLoadState) {
+	if state == nil || state.mainDirectory != "" {
+		return
+	}
+	info, ok := resolution.InputInfo(configkit.Keys("version"))
+	if ok && len(info.Origin.Locations) == 1 {
+		if source, sourceOK := resolution.Source(info.Origin.Locations[0].SourceID); sourceOK && source.Path != "" {
+			state.mainDirectory = filepath.Dir(source.Path)
+			return
+		}
+	}
+	// Byte and reader sources are test-oriented private entry points. They have
+	// no declaring file, so preserve their established invocation-directory
+	// semantics without extending that fallback to file-backed production loads.
+	state.mainDirectory = cleanInvocationDirectory(".")
+}
+
+func normalizeDocumentBindingReferencePaths(binding *BindingDocument, path configkit.Path, resolution documentNormalizer, state *documentLoadState) {
+	if binding != nil {
+		normalizeDocumentReferencePaths(binding.References, path.Key("references"), resolution, state)
+	}
+}
+
+func normalizeDocumentLaneReferencePaths(lane *ArtifactLaneDocument, path configkit.Path, resolution documentNormalizer, state *documentLoadState) {
+	if lane == nil {
+		return
+	}
+	normalizeDocumentBindingReferencePaths(&lane.Extract, path.Key("extract"), resolution, state)
+	normalizeDocumentBindingReferencePaths(lane.Merge, path.Key("merge"), resolution, state)
+	normalizeDocumentBindingReferencePaths(lane.Normalize, path.Key("normalize"), resolution, state)
+}
+
+func normalizeDocumentReferencePaths(references map[string]ReferenceDocument, path configkit.Path, resolution documentNormalizer, state *documentLoadState) {
+	for _, name := range sortedReferenceDocumentKeys(references) {
+		source := references[name]
+		sourcePath := path.Key(name).Key("path")
+		if source.Path != nil && *source.Path != "" && !filepath.IsAbs(*source.Path) {
+			base, ok := documentPathBase(resolution, state, sourcePath)
+			if !ok {
+				resolution.Error(sourcePath, "relative path has no unambiguous source base")
+			} else {
+				value := filepath.Clean(filepath.Join(base, *source.Path))
+				if value != *source.Path {
+					resolution.Record(sourcePath, configkit.ChangeNormalization, "resolved external reference path")
+					source.Path = documentValue(value)
+				}
+			}
+			references[name] = source
+		}
+	}
+}
+
+func normalizeDocumentPath(resolution documentNormalizer, state *documentLoadState, path configkit.Path, target **string, optional bool, reason string) {
+	if target == nil || *target == nil || **target == "" || filepath.IsAbs(**target) {
+		return
+	}
+	base, ok := documentPathBase(resolution, state, path)
+	if !ok {
+		resolution.Error(path, "relative path has no unambiguous source base")
+		return
+	}
+	value := filepath.Clean(filepath.Join(base, **target))
+	if optional && **target == "" {
+		return
+	}
+	if *target != nil && **target != value {
+		configkit.Normalize(resolution.Resolution, path, target, documentValue(value), reason)
+	}
+}
+
+func documentPathBase(resolution documentNormalizer, state *documentLoadState, path configkit.Path) (string, bool) {
+	original := path
+	if state != nil {
+		original = state.originalPath(path)
+	}
+	info, ok := resolution.InputInfo(original)
+	if !ok {
+		if state != nil && state.mainDirectory != "" {
+			return state.mainDirectory, true
+		}
+		return "", false
+	}
+	switch info.Origin.Layer {
+	case configkit.LayerCLI:
+		if state != nil && state.invocationDirectory != "" {
+			return state.invocationDirectory, true
+		}
+	case configkit.LayerDefault:
+		if state != nil && state.mainDirectory != "" {
+			return state.mainDirectory, true
+		}
+	default:
+		if len(info.Origin.Locations) != 1 {
+			return "", false
+		}
+		source, ok := resolution.Source(info.Origin.Locations[0].SourceID)
+		if !ok || source.Path == "" {
+			if state != nil && state.mainDirectory != "" {
+				return state.mainDirectory, true
+			}
+			return "", false
+		}
+		return filepath.Dir(source.Path), true
+	}
+	return "", false
+}

internal/core/config/document_paths_test.go

@@ -0,0 +1,144 @@
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+)
+
+func TestDocumentPathsUseDeclaringFilesAcrossImports(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	main := filepath.Join(root, "notarius.yml")
+	fragmentDirectory := filepath.Join(root, "fragments")
+	if err := os.MkdirAll(fragmentDirectory, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	fragment := filepath.Join(fragmentDirectory, "pipelines.yml")
+	if err := os.WriteFile(fragment, []byte(`pipelines:
+  imported:
+    input: {module: input}
+    references:
+      source: {path: ./from-fragment.txt}
+`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(main, []byte(`version: 5
+configkit:
+  imports:
+    - path: fragments/pipelines.yml
+output: {directory: ./output}
+debug: {directory: ./debug}
+pipelines:
+  main:
+    input: {module: input}
+    references:
+      source: {path: ./from-main.txt}
+`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	loaded, err := loadConfig(main, LoadOptions{})
+	if err != nil {
+		t.Fatalf("loadConfig() error = %v", err)
+	}
+	if got := loaded.Config.Pipelines["main"].References["source"].Path; got != filepath.Join(root, "from-main.txt") {
+		t.Fatalf("main reference = %q", got)
+	}
+	if got := loaded.Config.Pipelines["imported"].References["source"].Path; got != filepath.Join(fragmentDirectory, "from-fragment.txt") {
+		t.Fatalf("fragment reference = %q", got)
+	}
+	if loaded.Config.Output.Directory != filepath.Join(root, "output") || loaded.Config.Debug.Directory != filepath.Join(root, "debug") {
+		t.Fatalf("destinations = %#v %#v", loaded.Config.Output, loaded.Config.Debug)
+	}
+}
+
+func TestDocumentPathsUseInvocationDirectoryForOverrides(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	configPath := filepath.Join(root, "notarius.yml")
+	if err := os.WriteFile(configPath, []byte("version: 5\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	invocation := t.TempDir()
+	loaded, err := loadConfig(configPath, LoadOptions{
+		InvocationDirectory: invocation,
+		Overrides:           []configkit.Override{{Path: configkit.Keys("output", "directory"), Value: "generated", Flag: "--output"}},
+	})
+	if err != nil {
+		t.Fatalf("loadConfig() error = %v", err)
+	}
+	if got := loaded.Config.Output.Directory; got != filepath.Join(invocation, "generated") {
+		t.Fatalf("override directory = %q", got)
+	}
+}
+
+func TestDocumentPathsUseOverlayDeclarationDirectory(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	overlays := filepath.Join(root, "overlays")
+	if err := os.MkdirAll(overlays, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	main := filepath.Join(root, "notarius.yml")
+	if err := os.WriteFile(main, []byte("version: 5\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(overlays, "development.yml"), []byte("output: {directory: ./overlay-output}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	loaded, err := loadConfig(main, LoadOptions{Overlay: "development"})
+	if err != nil {
+		t.Fatalf("loadConfig() error = %v", err)
+	}
+	if got := loaded.Config.Output.Directory; got != filepath.Join(overlays, "overlay-output") {
+		t.Fatalf("overlay directory = %q", got)
+	}
+}
+
+func TestDocumentLoadDoesNotFallBackToLegacyProfileDirectory(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	main := filepath.Join(root, "notarius.yml")
+	if err := os.WriteFile(main, []byte("version: 5\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	legacy := filepath.Join(root, "profiles", "development.yml")
+	if err := os.MkdirAll(filepath.Dir(legacy), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(legacy, []byte("output: {directory: ./legacy-output}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	if _, err := loadConfig(main, LoadOptions{Overlay: "development"}); err == nil {
+		t.Fatal("loadConfig() error = nil, want missing overlay")
+	}
+}
+
+func TestDocumentPathsKeepEmptyCacheDirectoriesAndAbsolutePaths(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	configPath := filepath.Join(root, "notarius.yml")
+	abs := filepath.Join(t.TempDir(), "outside.txt")
+	if err := os.WriteFile(configPath, []byte("version: 5\ncache:\n  chunk_plans: {directory: ''}\n  checkpoints: {directory: ''}\npipelines:\n  main:\n    input: {module: input}\n    references:\n      outside: {path: "+abs+"}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	loaded, err := loadConfig(configPath, LoadOptions{})
+	if err != nil {
+		t.Fatalf("loadConfig() error = %v", err)
+	}
+	if loaded.Config.Cache.ChunkPlans.Directory != "" || loaded.Config.Cache.Checkpoints.Directory != "" {
+		t.Fatalf("cache directories = %#v", loaded.Config.Cache)
+	}
+	if got := loaded.Config.Pipelines["main"].References["outside"].Path; got != abs {
+		t.Fatalf("absolute reference = %q", got)
+	}
+}

internal/core/config/document_test.go

@@ -0,0 +1,154 @@
+package config
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+)
+
+func TestDocumentLoadAcceptsVersion5AndCanonicalShapes(t *testing.T) {
+	t.Parallel()
+
+	for _, source := range []string{
+		`version: 5
+pipelines:
+  session:
+    input:
+      module: seriatim
+    references:
+      party:
+        path: ./party.txt
+    steps:
+      - id: describe
+        artifacts:
+          spells:
+            extract:
+              module: dnd/spells
+              validators:
+                - module: generic/valid_json
+                  options:
+                    strict: true
+      - id: use-artifact
+        references:
+          spells:
+            artifact:
+              step: describe
+              lane: spells
+        artifacts:
+          locations:
+            extract:
+              module: dnd/locations
+`,
+		`version: "5"
+pipelines:
+  session:
+    input:
+      module: seriatim
+`,
+	} {
+		loaded, err := loadDocumentSource(configkit.Bytes("document.yml", []byte(source)), LoadOptions{})
+		if err != nil {
+			t.Fatalf("loadDocumentSource() error = %v", err)
+		}
+		profile, ok := loaded.Config.Pipelines["session"]
+		if !ok || profile.Input.Module != "seriatim" {
+			t.Fatalf("pipeline = %#v", profile)
+		}
+	}
+}
+
+func TestDocumentLoadRejectsUnsupportedVersionsBeforeDocumentDecoding(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name   string
+		source string
+		code   configkit.Code
+	}{
+		{name: "missing", source: "unknown: value\n", code: configkit.CodeVersionError},
+		{name: "null", source: "version: null\nunknown: value\n", code: configkit.CodeVersionError},
+		{name: "version four", source: "version: 4\nunknown: value\n", code: configkit.CodeVersionError},
+		{name: "unsupported", source: "version: 6\nunknown: value\n", code: configkit.CodeVersionError},
+		{name: "noninteger", source: "version: 5.0\n", code: configkit.CodeVersionType},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			_, err := loadDocumentSource(configkit.Bytes("document.yml", []byte(test.source)), LoadOptions{})
+			var loadError *configkit.LoadError
+			if !errors.As(err, &loadError) {
+				t.Fatalf("error = %v, want ConfigKit load error", err)
+			}
+			diagnostics := loadError.Diagnostics()
+			if len(diagnostics) != 1 || diagnostics[0].Code != test.code || diagnostics[0].Path.String() != "/version" {
+				t.Fatalf("diagnostics = %#v", diagnostics)
+			}
+			if strings.Contains(diagnostics[0].Message, "unknown") {
+				t.Fatalf("document decoding ran before version validation: %#v", diagnostics)
+			}
+		})
+	}
+}
+
+func TestDocumentLoadRejectsShorthandAndUnknownFields(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name   string
+		source string
+		code   configkit.Code
+		path   string
+	}{
+		{
+			name:   "binding shorthand",
+			source: "version: 5\npipelines:\n  session:\n    input: seriatim\n",
+			code:   configkit.CodeDecodeType,
+			path:   "/pipelines/session/input",
+		},
+		{
+			name:   "validator nested producer field",
+			source: "version: 5\npipelines:\n  session:\n    input:\n      module: seriatim\n      validators:\n        - module: generic/valid_json\n          validators: []\n",
+			code:   configkit.CodeDecodeUnknownField,
+			path:   "/pipelines/session/input/validators/0/validators",
+		},
+		{
+			name:   "unknown field",
+			source: "version: 5\npipelines:\n  session:\n    input:\n      module: seriatim\n      unknown: true\n",
+			code:   configkit.CodeDecodeUnknownField,
+			path:   "/pipelines/session/input/unknown",
+		},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			_, err := loadDocumentSource(configkit.Bytes("document.yml", []byte(test.source)), LoadOptions{})
+			var loadError *configkit.LoadError
+			if !errors.As(err, &loadError) {
+				t.Fatalf("error = %v, want ConfigKit load error", err)
+			}
+			diagnostics := loadError.Diagnostics()
+			if len(diagnostics) != 1 || diagnostics[0].Code != test.code || diagnostics[0].Path.String() != test.path {
+				t.Fatalf("diagnostics = %#v", diagnostics)
+			}
+		})
+	}
+}
+
+func TestDocumentDefinitionReturnsIndependentDefaults(t *testing.T) {
+	t.Parallel()
+
+	first, err := loadDocumentSource(configkit.Bytes("first.yml", []byte("version: 5\n")), LoadOptions{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	first.Config.Pipelines["changed"] = PipelineDocument{}
+	first.Config.Concurrency.StageWorkers["extract"] = 3
+
+	second, err := loadDocumentSource(configkit.Bytes("second.yml", []byte("version: 5\n")), LoadOptions{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(second.Config.Pipelines) != 0 || second.Config.Concurrency.StageWorkers["extract"] != defaultLLMConcurrency || len(second.Config.Concurrency.StageWorkers) != 1 {
+		t.Fatalf("defaults aliased across loads: %#v", second.Config)
+	}
+}

internal/core/config/effective_config.go

@@ -26,7 +26,7 @@ type EffectiveConfig struct {
 }
 
 func (c Config) Resolve(input ResolveInput) (EffectiveConfig, error) {
-	c.Concurrency.recomputeStageWorkerDefaults()
+	c = c.withDerivedStageWorkers()
 	if err := c.Validate(); err != nil {
 		return EffectiveConfig{}, err
 	}

internal/core/config/env.go

@@ -1,103 +0,0 @@
-package config
-
-import (
-	"fmt"
-	"os"
-	"strconv"
-	"strings"
-
-	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
-)
-
-func LoadFromEnv() (Config, error) {
-	cfg := Default()
-	if err := cfg.applyEnvOverridesWithLookup(os.LookupEnv); err != nil {
-		return Config{}, err
-	}
-	return cfg, nil
-}
-
-func (c *Config) ApplyEnvOverrides() error {
-	return c.applyEnvOverridesWithLookup(os.LookupEnv)
-}
-
-func (c *Config) ApplyEnvOverridesWithLookup(lookup func(string) (string, bool)) error {
-	return c.applyEnvOverridesWithLookup(lookup)
-}
-
-func (c *Config) applyEnvOverridesWithLookup(lookup func(string) (string, bool)) error {
-	if c == nil {
-		return fmt.Errorf("config must not be nil")
-	}
-	if raw, ok := lookup("NOTARIUS_TOTAL_LLM_CONCURRENCY"); ok {
-		value, err := parseIntEnv("NOTARIUS_TOTAL_LLM_CONCURRENCY", raw)
-		if err != nil {
-			return err
-		}
-		c.Concurrency.TotalLLM = value
-	}
-	if raw, ok := lookup("NOTARIUS_STAGE_WORKERS_EXTRACT"); ok {
-		value, err := parseIntEnv("NOTARIUS_STAGE_WORKERS_EXTRACT", raw)
-		if err != nil {
-			return err
-		}
-		if c.Concurrency.StageWorkers == nil {
-			c.Concurrency.StageWorkers = make(map[string]int)
-		}
-		c.Concurrency.StageWorkers["extract"] = value
-		c.Concurrency.extractWorkersConfigured = true
-	}
-	c.Concurrency.recomputeStageWorkerDefaults()
-	if raw, ok := lookup("NOTARIUS_OUTPUT_DIR"); ok {
-		c.Output.Directory = strings.TrimSpace(raw)
-		if c.Output.Directory == "" {
-			return fmt.Errorf("NOTARIUS_OUTPUT_DIR: must not be empty")
-		}
-		if strings.ContainsRune(c.Output.Directory, '\x00') {
-			return fmt.Errorf("NOTARIUS_OUTPUT_DIR: must not contain NUL")
-		}
-	}
-	if raw, ok := lookup("NOTARIUS_CACHE_CHUNK_PLANS_MODE"); ok {
-		mode, err := pipeline.ParseChunkCacheMode(raw)
-		if err != nil {
-			return fmt.Errorf("NOTARIUS_CACHE_CHUNK_PLANS_MODE: %w", err)
-		}
-		c.Cache.ChunkPlans.Mode = mode
-	}
-	if raw, ok := lookup("NOTARIUS_CACHE_CHUNK_PLANS_DIR"); ok {
-		c.Cache.ChunkPlans.Directory = cleanOptionalPath(raw)
-		if c.Cache.ChunkPlans.Directory == "" {
-			return fmt.Errorf("NOTARIUS_CACHE_CHUNK_PLANS_DIR: must not be empty")
-		}
-		if strings.ContainsRune(c.Cache.ChunkPlans.Directory, '\x00') {
-			return fmt.Errorf("NOTARIUS_CACHE_CHUNK_PLANS_DIR: must not contain NUL")
-		}
-	}
-	if raw, ok := lookup("NOTARIUS_CACHE_CHECKPOINTS_DIR"); ok {
-		c.Cache.Checkpoints.Directory = cleanOptionalPath(raw)
-		if c.Cache.Checkpoints.Directory == "" {
-			return fmt.Errorf("NOTARIUS_CACHE_CHECKPOINTS_DIR: must not be empty")
-		}
-		if strings.ContainsRune(c.Cache.Checkpoints.Directory, '\x00') {
-			return fmt.Errorf("NOTARIUS_CACHE_CHECKPOINTS_DIR: must not contain NUL")
-		}
-	}
-	if raw, ok := lookup("NOTARIUS_DEBUG_DIR"); ok {
-		c.Debug.Directory = strings.TrimSpace(raw)
-		if c.Debug.Directory == "" {
-			return fmt.Errorf("NOTARIUS_DEBUG_DIR: must not be empty")
-		}
-		if strings.ContainsRune(c.Debug.Directory, '\x00') {
-			return fmt.Errorf("NOTARIUS_DEBUG_DIR: must not contain NUL")
-		}
-	}
-	return nil
-}
-
-func parseIntEnv(name string, raw string) (int, error) {
-	value, err := strconv.Atoi(strings.TrimSpace(raw))
-	if err != nil {
-		return 0, fmt.Errorf("%s: must be an integer", name)
-	}
-	return value, nil
-}

internal/core/config/env_contract_test.go

@@ -1,239 +0,0 @@
-package config
-
-import (
-	"errors"
-	"reflect"
-	"strings"
-	"testing"
-
-	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
-)
-
-func TestPrecedenceFileValuesOverrideBuiltInDefaults(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 4
-concurrency:
-  total_llm: 4
-  stage_workers:
-    extract: 2
-output:
-  directory: ./file-output
-cache:
-  chunk_plans:
-    directory: ./file-plans
-    mode: refresh
-  checkpoints:
-    directory: ./file-checkpoints
-debug:
-  directory: ./file-debug
-`)
-	if cfg.Concurrency.TotalLLM != 4 || cfg.Concurrency.StageWorkers["extract"] != 2 ||
-		cfg.Output.Directory != "./file-output" || cfg.Cache.ChunkPlans.Directory != "file-plans" ||
-		cfg.Cache.ChunkPlans.Mode != pipeline.ChunkCacheRefresh || cfg.Cache.Checkpoints.Directory != "file-checkpoints" ||
-		cfg.Debug.Directory != "./file-debug" {
-		t.Fatalf("file values did not override defaults: %#v", cfg)
-	}
-}
-
-func TestPrecedenceOperationalEnvironmentOverridesFileValues(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 4
-concurrency:
-  total_llm: 2
-  stage_workers:
-    extract: 1
-output:
-  directory: ./file-output
-cache:
-  chunk_plans:
-    directory: ./file-plans
-    mode: refresh
-  checkpoints:
-    directory: ./file-checkpoints
-debug:
-  directory: ./file-debug
-`)
-	env := map[string]string{
-		"NOTARIUS_TOTAL_LLM_CONCURRENCY":  "8",
-		"NOTARIUS_STAGE_WORKERS_EXTRACT":  "6",
-		"NOTARIUS_OUTPUT_DIR":             "/env/output",
-		"NOTARIUS_CACHE_CHUNK_PLANS_MODE": "bypass",
-		"NOTARIUS_CACHE_CHUNK_PLANS_DIR":  "/env/plans",
-		"NOTARIUS_CACHE_CHECKPOINTS_DIR":  "/env/checkpoints",
-		"NOTARIUS_DEBUG_DIR":              "/env/debug",
-	}
-	if err := cfg.ApplyEnvOverridesWithLookup(lookupValues(env)); err != nil {
-		t.Fatal(err)
-	}
-	if cfg.Concurrency.TotalLLM != 8 || cfg.Concurrency.StageWorkers["extract"] != 6 ||
-		cfg.Output.Directory != "/env/output" || cfg.Cache.ChunkPlans.Directory != "/env/plans" ||
-		cfg.Cache.ChunkPlans.Mode != pipeline.ChunkCacheBypass || cfg.Cache.Checkpoints.Directory != "/env/checkpoints" ||
-		cfg.Debug.Directory != "/env/debug" {
-		t.Fatalf("environment values did not override file values: %#v", cfg)
-	}
-}
-
-func TestPrecedenceExtractWorkersFollowEffectiveConcurrencyUnlessExplicit(t *testing.T) {
-	tests := []struct {
-		name       string
-		file       string
-		env        map[string]string
-		wantTotal  int
-		wantWorker int
-	}{
-		{
-			name:       "default follows environment total",
-			file:       "version: 4\n",
-			env:        map[string]string{"NOTARIUS_TOTAL_LLM_CONCURRENCY": "5"},
-			wantTotal:  5,
-			wantWorker: 5,
-		},
-		{
-			name:       "file worker is retained",
-			file:       "version: 4\nconcurrency:\n  total_llm: 3\n  stage_workers:\n    extract: 2\n",
-			env:        map[string]string{"NOTARIUS_TOTAL_LLM_CONCURRENCY": "6"},
-			wantTotal:  6,
-			wantWorker: 2,
-		},
-		{
-			name: "environment worker is retained",
-			file: "version: 4\nconcurrency:\n  total_llm: 2\n",
-			env: map[string]string{
-				"NOTARIUS_TOTAL_LLM_CONCURRENCY": "6",
-				"NOTARIUS_STAGE_WORKERS_EXTRACT": "4",
-			},
-			wantTotal:  6,
-			wantWorker: 4,
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			cfg := applyFileConfig(t, tt.file)
-			if err := cfg.ApplyEnvOverridesWithLookup(lookupValues(tt.env)); err != nil {
-				t.Fatal(err)
-			}
-			if cfg.Concurrency.TotalLLM != tt.wantTotal || cfg.Concurrency.StageWorkers["extract"] != tt.wantWorker {
-				t.Fatalf("concurrency = %#v, want total %d and extract %d", cfg.Concurrency, tt.wantTotal, tt.wantWorker)
-			}
-		})
-	}
-}
-
-func TestPrecedenceEmptyFileCacheDirectoriesDeferPerUserResolution(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 4
-cache:
-  chunk_plans:
-    directory: ""
-  checkpoints:
-    directory: ""
-`)
-	if err := cfg.Validate(); err != nil {
-		t.Fatalf("empty file cache directories should be valid: %v", err)
-	}
-	if cfg.Cache.ChunkPlans.Directory != "" || cfg.Cache.Checkpoints.Directory != "" {
-		t.Fatalf("empty cache directories were not preserved for deferred resolution: %#v", cfg.Cache)
-	}
-	resolver := func() (string, error) { return "/user/cache", nil }
-	chunkPlans, err := DefaultChunkPlanRoot(resolver)
-	if err != nil {
-		t.Fatal(err)
-	}
-	checkpoints, err := DefaultCheckpointRoot(resolver)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if chunkPlans != "/user/cache/notarius/chunk-plans" || checkpoints != "/user/cache/notarius/checkpoints" {
-		t.Fatalf("deferred cache roots = %q, %q", chunkPlans, checkpoints)
-	}
-}
-
-func TestDefaultCacheRootsRejectInvalidUserCacheResolvers(t *testing.T) {
-	tests := []struct {
-		name     string
-		resolver func() (string, error)
-		want     string
-	}{
-		{name: "nil resolver", want: "must not be nil"},
-		{
-			name: "resolver failure",
-			resolver: func() (string, error) {
-				return "", errors.New("cache home unavailable")
-			},
-			want: "resolve user cache directory",
-		},
-		{name: "empty directory", resolver: func() (string, error) { return "  ", nil }, want: "must not be empty"},
-	}
-	families := []struct {
-		name string
-		root func(func() (string, error)) (string, error)
-	}{
-		{name: "chunk plans", root: DefaultChunkPlanRoot},
-		{name: "checkpoints", root: DefaultCheckpointRoot},
-	}
-
-	for _, family := range families {
-		for _, tt := range tests {
-			t.Run(family.name+"/"+tt.name, func(t *testing.T) {
-				_, err := family.root(tt.resolver)
-				if err == nil || !strings.Contains(err.Error(), tt.want) {
-					t.Fatalf("error = %v, want substring %q", err, tt.want)
-				}
-			})
-		}
-	}
-}
-
-func TestEnvEmptyDirectoryOverridesAreErrors(t *testing.T) {
-	tests := []string{
-		"NOTARIUS_OUTPUT_DIR",
-		"NOTARIUS_CACHE_CHUNK_PLANS_DIR",
-		"NOTARIUS_CACHE_CHECKPOINTS_DIR",
-		"NOTARIUS_DEBUG_DIR",
-	}
-	for _, name := range tests {
-		t.Run(name, func(t *testing.T) {
-			cfg := Default()
-			err := cfg.ApplyEnvOverridesWithLookup(lookupValues(map[string]string{name: " \t"}))
-			if err == nil || !strings.Contains(err.Error(), name) {
-				t.Fatalf("error = %v, want responsible environment variable", err)
-			}
-		})
-	}
-}
-
-func TestEnvInvalidIntegersAndChunkCacheModesReportTheirNames(t *testing.T) {
-	tests := map[string]string{
-		"NOTARIUS_TOTAL_LLM_CONCURRENCY":  "not-an-integer",
-		"NOTARIUS_STAGE_WORKERS_EXTRACT":  "not-an-integer",
-		"NOTARIUS_CACHE_CHUNK_PLANS_MODE": "not-a-cache-mode",
-	}
-	for name, value := range tests {
-		t.Run(name, func(t *testing.T) {
-			cfg := Default()
-			err := cfg.ApplyEnvOverridesWithLookup(lookupValues(map[string]string{name: value}))
-			if err == nil || !strings.Contains(err.Error(), name) {
-				t.Fatalf("error = %v, want responsible environment variable", err)
-			}
-		})
-	}
-}
-
-func TestEnvRemovedProviderVariablesAreIgnored(t *testing.T) {
-	before := Default()
-	cfg := Default()
-	removed := map[string]string{
-		"NOTARIUS_LLM_DEFAULT_ENDPOINT": "ignored-provider-setting",
-		"NOTARIUS_LLM_DEFAULT_MODEL":    "ignored-provider-setting",
-	}
-	if err := cfg.ApplyEnvOverridesWithLookup(lookupValues(removed)); err != nil {
-		t.Fatal(err)
-	}
-	if !reflect.DeepEqual(cfg, before) {
-		t.Fatalf("removed provider variables changed configuration: %#v", cfg)
-	}
-}
-
-func lookupValues(values map[string]string) func(string) (string, bool) {
-	return func(name string) (string, bool) {
-		value, ok := values[name]
-		return value, ok
-	}
-}