Make PromptKit profile handling safer and more consistent

.gitignore

@@ -2,6 +2,7 @@
 notarius
 notarius-output
 workspace/
+.codebase-memory/
 
 # ---> Go
 # If you prefer the allow list template instead of the deny list, see community template:
@@ -73,4 +74,3 @@ Icon
 Network Trash Folder
 Temporary Items
 .apdisk
-

README.md

@@ -36,6 +36,8 @@ demonstrates all implemented D&D lanes and the supporting campaign references.
   handling.
 - [Integration contracts](docs/integrations/) — Seriatim input and published
   artifact formats.
+- [Subprocess consumer guide](docs/consumers/subprocess.md) — invoke Notarius
+  from an orchestrator and consume a published result.
 - [Internal overview](docs/internal/overview.md) — implemented component map
   for maintainers.
 - [Developer guide](docs/development.md) — contributor orientation and

docs/adr/0010-workload-oriented-llm-profile-defaults.md

@@ -0,0 +1,49 @@
+# ADR-0010: Use workload-oriented LLM profile defaults
+
+**Status:** Accepted
+**Date:** 2026-08-03
+
+## Context
+
+LLM-backed D&D operations share an execution-policy choice, but repeating a
+provider or model-named profile on every module binding ties pipeline structure
+to a deployment decision. Different environments may require different model,
+backend, timeout, or reasoning settings while retaining the same workload.
+
+Notarius also needs a usable default for maintained D&D prompts without making
+an operator profile mandatory. That default must remain owned by the D&D
+family, while generic LLM infrastructure stays unaware of domain-specific
+policy.
+
+## Decision
+
+Pipelines may name one workload-oriented default profile, inherited only by
+selected LLM-backed bindings and validators. Binding-level profile IDs remain
+intentional exceptions, and the run-wide CLI profile override has highest
+precedence.
+
+The D&D family owns an embedded fallback profile named `dnd-extraction`.
+Operators may provide a complete profile with the same ID through a PromptKit
+filesystem source. PromptKit selects the higher-precedence matching definition;
+Notarius does not merge profile documents. Production, development, and local
+deployments can therefore use different execution policy behind one unchanged
+pipeline ID.
+
+## Alternatives considered
+
+- Repeat a model-named profile on every binding. This makes routine deployment
+  policy changes noisy and obscures the shared workload intent.
+- Require every deployment to install a profile file. This adds configuration
+  friction and leaves maintained D&D prompts without an application-owned
+  fallback.
+- Put D&D profile policy in generic LLM infrastructure. This breaks domain
+  ownership and makes generic code depend on one workload.
+
+## Consequences
+
+Pipeline configuration expresses workload intent rather than a specific
+provider or model. Operators can replace the complete execution policy without
+editing bindings, while binding-level and run-wide exceptions remain available.
+Profile changes affect resolved pipeline and checkpoint identity, so they may
+intentionally cause work to be recomputed. The D&D fallback becomes a
+maintained application execution-policy asset.

docs/cli.md

@@ -10,7 +10,7 @@ defined in [Operations](operations.md).
 
 ~~~
 notarius help
-notarius run <pipeline-id> --input path/to/source.json [flags]
+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]
 ~~~
@@ -21,7 +21,7 @@ writes the command summary to standard output and exits with status 0.
 ## run
 
 ~~~
-notarius run <pipeline-id> --input path/to/source.json [flags]
+notarius run <pipeline-id> --input path/to/source.json [--json] [flags]
 ~~~
 
 The **run** command executes the named pipeline for one input file. The
@@ -32,22 +32,30 @@ pipeline ID and **--input** are required.
 | **--config path** | Use this configuration file. When omitted, configuration discovery applies; see [Configuration](config.md). |
 | **--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. |
 | **--chunk_cache auto\|bypass\|refresh** | Override chunk-plan cache handling for this run. |
 | **--resume** | Reuse compatible recorded checkpoints when checkpoint recording is enabled. |
 | **--recompute-step step-id** | With **--resume**, recompute the selected ordered step and its dependent lanes. It cannot be combined with **--only**. |
 | **--debug** | Retain a debug bundle for this run. |
 | **--debug-dir path** | Override the debug-bundle root. Requires **--debug**. |
 | **--only lane-a,lane-b** | Run only the selected comma-separated artifact lanes when that selection is valid for the configured pipeline. |
-| **--llm-profile id** | Override effective LLM-capable module bindings with one configured profile. |
+| **--llm-profile id** | Highest-precedence configured profile for selected LLM-backed bindings and validators; it replaces binding and [pipeline](config.md#pipelines) defaults. |
 | **--session-id id** | Supply a non-empty prompt session identifier to 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. |
 
 **--chunk_cache** accepts only **auto**, **bypass**, or **refresh**.
 **--debug-dir**, **--output-dir**, **--session-id**, and
-**--recompute-step** reject explicit empty values. **--recompute-step**
-requires **--resume**; checkpoint requirements and reuse behavior are
-documented in [Operations](operations.md).
+**--reasoning-effort**, and **--recompute-step** reject explicit empty values.
+**--reasoning-effort** and **--clear-reasoning-effort** are mutually exclusive.
+When neither is present, reasoning effort comes from the selected PromptKit
+profile. These controls apply to the shared run client, including retries and
+LLM-backed validators, and do not modify configuration or profile files.
+Persistent reasoning settings remain a PromptKit profile concern.
+**--recompute-step** requires **--resume**; checkpoint requirements and reuse
+behavior are documented in [Operations](operations.md).
 
 ### Reference selectors
 
@@ -70,11 +78,18 @@ names, requiredness, and configured bindings are part of the
 
 ### Run output
 
-On success, 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
-envelope is defined by the [JSON output contract](integrations/json-output.md).
+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
+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
+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
+leave incomplete standard-output bytes that are not a result document.
 
 Example:
 
@@ -134,6 +149,10 @@ 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
+machine-readable success result only. Syntax and runtime diagnostics remain on
+standard error. Parse the result only after the process exits with status 0.
+
 | Status | Meaning |
 | --- | --- |
 | 0 | The command completed successfully, including root help. |

docs/config.md

@@ -1,7 +1,7 @@
 # Configuration
 
 This is the canonical reference for Notarius configuration. Configuration files
-are YAML and must declare version 3. They select pipelines and their modules;
+are YAML and must declare version 4. They select pipelines and their modules;
 the [CLI reference](cli.md) owns invocation syntax, and
 [Operations](operations.md) owns run-state procedures.
 
@@ -32,7 +32,8 @@ override the fields listed below.
   single-lane Seriatim-to-spell pipeline.
 - [Complete D&D configuration](../examples/dnd-complete.config.yml) uses
   ordered steps, all implemented D&D lanes, generated references, state
-  settings, and bounded LLM concurrency.
+  settings, bounded LLM concurrency, and the maintained
+  [operator profile](../examples/profiles/dnd-extraction.yml).
 
 Use these complete files as starting points rather than combining the
 illustrative fragments in this reference.
@@ -45,8 +46,8 @@ other than **version** is optional.
 
 | Field | Type | Default | Rules |
 | --- | --- | --- | --- |
-| **version** | integer | none | Required; must be 3. |
-| **scriptorium** | object | none | Profile source configuration. |
+| **version** | integer | none | Required; must be 4. |
+| **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. |
 | **output** | object | see below | Published output settings. |
@@ -69,20 +70,85 @@ 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.
 
-## Scriptorium Profiles
+## PromptKit Profiles
 
-The optional **scriptorium** object selects one source of profile definitions:
+The optional **promptkit** object selects one source of profile definitions and
+may register one conventional local OpenAI-compatible backend:
+
+~~~yaml
+version: 4
+
+promptkit:
+  profile_dir: ./profiles
+  # profile_file: ./profiles.yml
+  local_backend:
+    endpoint: http://localhost:8000/v1
+    concurrency_limit: 2
+~~~
 
 | Field | Type | Rules |
 | --- | --- | --- |
 | **profile_dir** | string | Non-empty directory containing profile files. |
 | **profile_file** | string | Non-empty profile file. |
+| **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 these fields. Profile IDs used by a binding must be available
-from the selected Scriptorium profile source when the pipeline is resolved.
-Keep credentials out of this file: configure a profile to read its credential
-from an environment variable, then set that environment variable only in the
-run environment.
+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.
+
+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)
+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.
+[Operations](operations.md#promptkit-profile-deployment) owns the deployment
+workflow and credential-handling guidance.
+
+When **local_backend** is present, its endpoint is trimmed and must use HTTP or
+HTTPS case-insensitively, be absolute, and have a non-empty host. URL paths are
+allowed. User information, queries, and fragments are rejected. A zero
+**concurrency_limit** leaves the local backend unrestricted inside PromptKit;
+a positive value limits simultaneous local generations. The application-wide
+**concurrency.total_llm** limit still applies in both cases. Neither local
+backend field has an environment override. Omitting **local_backend** registers
+nothing and preserves existing built-in and endpoint-only profile behavior.
+
+A file-backed PromptKit profile selects the registration by its case-sensitive
+backend ID:
+
+~~~yaml
+id: local-summary
+backend: local
+model: example-model
+~~~
+
+Keep credentials out of the local-backend object. A PromptKit profile may name
+its credential environment variable through `api_key_env`; set that variable
+only in the run environment. PromptKit owns the
+[pinned profile-file format](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/formats.md).
+The [PromptKit upstream boundary](integrations/pkg-promptkit.md) identifies the
+supported package API, and [Operations](operations.md#operational-limits)
+describes the effective concurrency layers.
+
+`notarius config validate --pipeline <id>` resolves the selected pipeline and
+inspects every explicit effective profile without contacting a provider or
+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 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
 
@@ -138,6 +204,7 @@ Each **pipelines** entry has a unique, non-empty ID and the following shape:
 ~~~yaml
 pipelines:
   dnd-session:
+    llm_profile: dnd-extraction
     input: seriatim
     chunk: generic
     output: json
@@ -150,6 +217,7 @@ pipelines:
 
 | Field | Type | Default | Rules |
 | --- | --- | --- | --- |
+| **llm_profile** | string | none | Optional non-empty default PromptKit profile ID for selected LLM-backed bindings and validators. An explicitly present blank value is invalid. |
 | **input** | module binding | none | Required. |
 | **chunk** | module binding | **generic** | Optional. |
 | **output** | module binding | **json** | Optional. |
@@ -163,6 +231,12 @@ needs a unique non-empty **id**, an **artifacts** map, and may have
 **references**. A lane ID must not appear more than once in a pipeline,
 including across explicit steps.
 
+For each selected LLM-backed binding or validator, profile selection occurs
+after module, validator, and `--only` lane selection. It uses the
+run-level **--llm-profile** value first, then the binding's **llm_profile**,
+then the pipeline's **llm_profile**, and finally the PromptKit default.
+Deterministic bindings do not receive these defaults or run overrides.
+
 A lane has these fields:
 
 | Field | Type | Default | Rules |
@@ -190,7 +264,7 @@ Use an object for fields:
 ~~~yaml
 extract:
   module: dnd/spells
-  llm_profile: gemini-2-flash
+  llm_profile: dnd-extraction
   retries: 2
   references:
     spell_catalog: ./dnd-spell-catalog.json
@@ -199,7 +273,7 @@ extract:
 | Binding field | Type | Default | Rules |
 | --- | --- | --- | --- |
 | **module** | string | none | Required for an object binding. Must be a registered compatible key. |
-| **llm_profile** | string | none | Optional non-empty Scriptorium profile ID. |
+| **llm_profile** | string | none | Optional non-empty PromptKit profile ID for an LLM-backed binding. It overrides the pipeline default unless the run supplies **--llm-profile**. |
 | **retries** | integer | 0 | Non-negative additional attempts for chunk, extract, merge, and normalize bindings. |
 | **options** | object | none | Must satisfy the selected module. |
 | **references** | map | none | Valid only on chunk, extract, merge, and normalize bindings. |
@@ -209,21 +283,44 @@ Omitting **validators** uses the registered chain. **validators: []** selects
 an empty chain; a non-empty list replaces the chain in the listed order.
 Validator bindings accept only **module**, **llm_profile**, and **options**.
 They reject **references**, **retries**, and nested **validators**. Deterministic
-validators reject an explicit **llm_profile**.
+validators reject an explicit **llm_profile**. Deterministic module bindings
+also reject an explicit **llm_profile**.
 
-The **json** output module accepts one option:
+The **json** output module accepts optional **include_chunk_map** and
+**evidence_context** settings:
 
 ~~~yaml
 output:
   module: json
   options:
     include_chunk_map: true
+    evidence_context:
+      enabled: true
+      window_units: 3
+      lanes:
+        - npcs
+        - spells
 ~~~
 
 **include_chunk_map** is a boolean and defaults to false. It adds the accepted
 chunk map when one exists; its wire format is defined in the
 [chunk-map contract](integrations/chunk-map.md).
 
+Omitting **evidence_context** disables evidence publication. When present, it
+is an object with these strict fields:
+
+| Field | Type | Rules |
+| --- | --- | --- |
+| **enabled** | boolean | Required. `false` permits no other evidence fields. |
+| **lanes** | array of strings | Required and non-empty when enabled. Each value is trimmed and must be unique; every value must name a configured pipeline lane. |
+| **window_units** | non-negative integer | Optional when enabled; defaults to 3. Zero retains only directly cited units. |
+
+Unknown outer or nested option fields are rejected, as are incompatible YAML
+types. The allowlist remains valid when a run uses lane filtering: a configured
+lane that is not active for that invocation simply contributes no evidence.
+Evidence publication is opt-in because it can persist source text and metadata.
+Its payload contract is [Published Evidence Context](integrations/evidence-context.md).
+
 ## References And Ordered Handoffs
 
 Reference maps bind named slots that the selected target declares. A scalar is

docs/consumers/subprocess.md

@@ -0,0 +1,74 @@
+# Using Notarius As A Subprocess
+
+Use this workflow when an orchestrator runs Notarius and consumes its published
+artifacts. The [CLI reference](../cli.md) owns invocation syntax and exit
+statuses, while the [run-result receipt](../integrations/run-result.md) and
+[Published JSON Output contract](../integrations/json-output.md) own the
+durable result formats.
+
+## Run And Check The Process
+
+Optionally preflight a selected configuration and pipeline before work starts:
+
+```sh
+notarius config validate --config /path/to/notarius.yml --pipeline pipeline-id
+```
+
+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.
+
+```sh
+notarius run pipeline-id \
+  --config /path/to/notarius.yml \
+  --input /path/to/source.json \
+  --output-dir /path/to/output-root \
+  --json
+```
+
+Use absolute paths for supplied input, configuration, output-root, and
+reference files. When a stable prompt session identifier or references are
+needed, pass the supported CLI flags. Supply credentials through Notarius's
+documented configuration and environment mechanisms, never as command-line
+arguments or generated secret-bearing configuration.
+
+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
+diagnosis and ignore all standard-output bytes: a failed receipt write may have
+left a partial document.
+
+## Discover Required Artifacts
+
+Decode the successful receipt and accept the schema versions supported by the
+caller. Use its `output_directory` as the bundle root. For the production JSON
+output, resolve `index_file` under that root with a confinement check and reject
+an absolute path or a result that escapes the root.
+
+Read the resulting `index.json` and locate each artifact by `lane_id`, not by a
+guessed filename. Before decoding a selected payload, verify its descriptor's
+media type and schema identity against the relevant published artifact
+contract. The JSON bundle contract links to the available lane contracts.
+
+If `index.json` has an `evidence_context` descriptor, treat it as a
+pipeline-wide artifact rather than a lane entry. Verify its six descriptor
+fields before decoding the linked file according to the [Published Evidence
+Context contract](../integrations/evidence-context.md). Use each
+`evidence_refs` entry as the citation to source material. Its surrounding
+context range and included units explain the citation, but do not widen or
+replace the cited source reference.
+
+A zero exit status may still report rejected outputs, warnings, or absent
+lanes. The caller decides which lane IDs are required for its own work and
+which are optional; it should make that decision explicitly rather than infer
+failure from the receipt counts alone.
+
+## 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
+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
+provenance records. An evidence-context artifact contains source-unit text and
+metadata, and selected lanes can cover most of an input; preserve and share it
+only when that source content is authorized for the recipient.

docs/development.md

@@ -18,10 +18,11 @@ implemented component map.
 | Any documentation addition or revision | [Documentation Policy](policy/documentation.md) | It defines canonical homes, audiences, current-behavior rules, and maintenance requirements. |
 | Adding, changing, reviewing, or deleting tests | [Testing Policy](policy/testing.md) | It defines risk-based sufficiency, durable test boundaries, test-double guidance, and criteria for retaining tests. |
 | CLI composition or command behavior | [CLI Internals](internal/cli.md) and [CLI Reference](cli.md) | The internal guide owns composition and command flow; the reference owns public syntax. |
+| Building a subprocess caller or changing its result protocol | [Subprocess Consumer Guide](consumers/subprocess.md), [Run Result Receipt](integrations/run-result.md), and [CLI Internals](internal/cli.md) | These separate caller workflow, durable receipt contract, and CLI implementation behavior. |
 | Configuration loading, resolution, or user-visible configuration behavior | [Configuration Internals](internal/configuration.md) and [Configuration](config.md) | The internal guide owns loading and resolution mechanics; the reference owns the configuration contract. |
 | Pipeline resolution or execution | [Pipeline Internals](internal/pipeline.md) | It documents profiles, references, validation, retries, checkpoints, and runner behavior. |
 | Production modules or validators | [Module Internals](internal/modules.md), [D&D Module Internals](internal/dnd.md), and [D&D integration contracts](integrations/) | The generic guide owns extension mechanics, the D&D guide owns shared family conventions, and the contracts own durable output shapes. |
-| LLM clients, prompts, schemas, profiles, or scheduling | [LLM Runtime](internal/llm.md) | It documents the transport boundary and Scriptorium integration. |
+| 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. |
 | Proposed or unimplemented behavior | [Roadmap](roadmap/) | Future work belongs only in roadmap documentation until implemented. |

docs/integrations/dnd-combat-turn-artifacts.md

@@ -15,8 +15,8 @@ complete initiative tracker, combat summary, or state model.
 | Media type | `application/json` |
 
 `v1` is a strict JSON object with required `combat_turns`; the array may be
-empty. Turn and source-reference objects reject unknown fields. A future
-incompatible shape requires a new schema version.
+empty. Turn and source-reference objects reject unknown fields. An incompatible
+shape change requires a new schema version.
 
 ## Wire shape
 

docs/integrations/dnd-item-event-artifacts.md

@@ -15,8 +15,8 @@ changes; it does not maintain an inventory, balance, or ledger.
 | Media type | `application/json` |
 
 `v1` is a strict JSON object with required `events`; the array may be empty.
-Event and source-reference objects reject unknown fields. A future incompatible
-shape requires a new schema version.
+Event and source-reference objects reject unknown fields. An incompatible
+shape change requires a new schema version.
 
 ## Wire shape
 

docs/integrations/dnd-npc-artifacts.md

@@ -16,8 +16,8 @@ character sheet or a relationship summary.
 | Identity policy | `dnd.npcs.identity.v1` |
 
 `v1` accepts one strict JSON object with required `npcs`; the array may be
-empty. NPC and source-reference objects reject unknown fields. A future
-incompatible artifact shape or identity policy uses a new version or policy.
+empty. NPC and source-reference objects reject unknown fields. An incompatible
+artifact shape or identity-policy change uses a new version or policy.
 
 ## Wire shape and identity
 

docs/integrations/dnd-npc-interaction-artifacts.md

@@ -16,8 +16,8 @@ or summarize the session.
 | Media type | `application/json` |
 
 `v1` is a strict JSON object with required `interactions`; the array may be
-empty. Interaction and source-reference objects reject unknown fields. A future
-incompatible shape requires a new schema version.
+empty. Interaction and source-reference objects reject unknown fields. An
+incompatible shape change requires a new schema version.
 
 ## Wire shape
 

docs/integrations/dnd-scene-description-artifacts.md

@@ -15,7 +15,7 @@ source-grounded title and summary.
 | Media type | `application/json` |
 
 `v1` is a strict JSON object with required non-empty `scenes`. Scene and
-source-reference objects reject unknown fields. A future incompatible shape
+source-reference objects reject unknown fields. An incompatible shape change
 requires a new schema version.
 
 ## Wire shape

docs/integrations/dnd-spell-artifacts.md

@@ -16,7 +16,7 @@ spellbook, a rules lookup result, or a record of hypothetical casts.
 
 `v1` is a single strict JSON object. It requires `spell_casts`; the array may
 be empty. Each spell-cast object and source-reference object rejects unknown
-fields. A future incompatible shape requires a new schema version.
+fields. An incompatible shape change requires a new schema version.
 
 ## Wire shape
 

docs/integrations/evidence-context.md

@@ -0,0 +1,116 @@
+# Published Evidence Context
+
+This contract defines the optional `source/evidence-context` artifact emitted
+by the production JSON output. Its configuration is owned by
+[Configuration](../config.md#module-bindings-and-validators); its logical-file
+discovery is owned by [Published JSON Output](json-output.md).
+
+## Identity And Discovery
+
+When enabled, the JSON bundle contains `evidence-context.json` and an
+`index.json` `evidence_context` descriptor with the same six fields as other
+pipeline-wide artifact descriptors.
+
+| Property | Value |
+| --- | --- |
+| Artifact kind | `source/evidence-context` |
+| Media type | `application/json` |
+| Schema ID | `notarius.source.evidence_context` |
+| Schema name | `notarius_source_evidence_context_v1` |
+| Schema version | `v1` |
+| Logical file | `evidence-context.json` |
+
+Consumers must discover the file from the descriptor, verify all six descriptor
+fields, and decode only a supported schema version. The descriptor is optional:
+its absence means evidence publication was not enabled for that bundle.
+
+## Payload
+
+The v1 payload is a JSON object with required `source_id`, `source_digest`,
+`window_units`, `selected_lanes`, and `contexts` fields. `selected_lanes` and
+`contexts` are always arrays; an enabled configuration with no accepted direct
+evidence publishes `contexts: []`.
+
+```json
+{
+  "source_id": "session-alpha",
+  "source_digest": "sha256:0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
+  "window_units": 1,
+  "selected_lanes": ["npcs", "spells"],
+  "contexts": [
+    {
+      "context_ref": {
+        "source_id": "session-alpha",
+        "start_unit_id": 10,
+        "end_unit_id": 20
+      },
+      "evidence_refs": [
+        {
+          "lane_id": "spells",
+          "source_ref": {
+            "source_id": "session-alpha",
+            "start_unit_id": 10,
+            "end_unit_id": 10
+          }
+        }
+      ],
+      "units": [
+        {
+          "id": 10,
+          "kind": "transcript_segment",
+          "text": "Aria casts Cure Wounds.",
+          "ref": {
+            "source_id": "session-alpha",
+            "start_unit_id": 10,
+            "end_unit_id": 10
+          }
+        },
+        {
+          "id": 20,
+          "kind": "transcript_segment",
+          "text": "The party regroups.",
+          "ref": {
+            "source_id": "session-alpha",
+            "start_unit_id": 20,
+            "end_unit_id": 20
+          }
+        }
+      ]
+    }
+  ]
+}
+```
+
+Each context requires a `context_ref` object and `evidence_refs` and `units`
+arrays. `context_ref` identifies the first and last included unit. Each
+evidence entry contains a selected `lane_id` and an original `source_ref`. A
+unit uses the existing source-unit shape: required `id`, `kind`, `text`, and
+self `ref`, plus optional JSON-object `metadata`. Fixed payload objects reject
+unknown fields; unit metadata may contain application-defined JSON values.
+
+## Citations And Context
+
+`evidence_refs` are the authoritative citations. They identify the direct
+references emitted by accepted normalized artifacts. `context_ref` and the
+units collection include those cited units plus nearby source units selected by
+the configured window. They are explanatory context, not widened citations.
+
+Only accepted outputs from the configured lane allowlist contribute. Rejected,
+failed, absent, and lane-filtered outputs do not contribute. The artifact never
+contains raw input bytes, prompts, model responses, auxiliary reference
+content, credentials, or filesystem paths.
+
+## Ordering And Compatibility
+
+The selected lane allowlist is lexical. Contexts and units are in source
+document position order, not numeric unit-ID order. Direct evidence entries
+are deterministically ordered by lane and source reference. Overlapping or
+contiguous windows merge, and each source unit appears at most once in the
+resulting contexts.
+
+The artifact is additive to the JSON bundle and is not a lane payload,
+normalized-output count, checkpoint, or generated reference. Consumers that
+do not need it must tolerate the absent optional descriptor. Consumers that do
+use it should preserve the artifact and its schema identity with the run
+provenance, and should treat its source text and metadata as sensitive durable
+content.

docs/integrations/json-output.md

@@ -3,14 +3,18 @@
 This document defines the logical JSON bundle emitted by the production JSON
 output encoder. The bundle’s physical destination, atomic publication, and
 retention are operational concerns; see [Operations](../operations.md#output-bundles).
-Output configuration, including chunk-map export, belongs in
+Output configuration, including chunk-map and evidence-context publication, belongs in
 [Configuration](../config.md#module-bindings-and-validators).
 
 ## 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 chunk-map
-files when their corresponding artifacts are available:
+encoder always emits the first four 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
+[run-result receipt](run-result.md), then resolves `index.json` beneath that
+root for the logical discovery described here.
 
 | Path | Purpose |
 | --- | --- |
@@ -20,6 +24,7 @@ files when their corresponding artifacts are available:
 | `warnings.json` | Accepted-output and run warnings. |
 | `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 source-context artifact, when evidence publication is enabled. |
 
 JSON files are pretty-printed with a trailing newline. Lane payloads are
 accepted only when their media type is `application/json`.
@@ -45,13 +50,15 @@ normalized lanes has this valid minimal index:
 | `rejected_file` | Yes | Always `rejected.json`. |
 | `warnings_file` | Yes | Always `warnings.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. |
 
 Each lane descriptor has required `lane_id` and `file`. It may also include
 `media_type`, `module_key`, `schema_id`, `schema_name`, and `schema_version`
-when supplied by the normalized artifact. A `chunk_map` descriptor contains
-`artifact_kind`, `file`, `media_type`, `schema_id`, `schema_name`, and
-`schema_version`; its payload is defined by the
-[Accepted Chunk Map contract](chunk-map.md).
+when supplied by the normalized artifact. Each pipeline-wide artifact
+descriptor (`chunk_map` or `evidence_context`) contains `artifact_kind`,
+`file`, `media_type`, `schema_id`, `schema_name`, and `schema_version`. Their
+payloads are defined by the [Accepted Chunk Map contract](chunk-map.md) and
+[Published Evidence Context](evidence-context.md), respectively.
 
 The lane path is derived from its lane ID. Characters outside letters, digits,
 periods, underscores, and hyphens become underscores; `..` sequences are
@@ -91,6 +98,21 @@ summarize results without embedding lane payload bytes. A chunk-plan summary is
 provenance for the plan used by this run; cache records, debug artifacts, and
 other operational state are not published as bundle files.
 
+Each `llm_profiles` entry identifies effective, non-secret LLM execution
+provenance:
+
+| Field | Required | Meaning |
+| --- | --- | --- |
+| `id` | Yes | Selected PromptKit profile identifier. |
+| `provider` | No | Notarius adapter provider identifier. |
+| `model` | No | Effective provider model identifier. |
+| `backend_id` | No | Effective PromptKit backend registration identifier. Endpoint-only profiles omit it. |
+| `reasoning_effort` | No | Effective opaque provider reasoning setting. An empty or explicitly cleared setting is omitted. |
+
+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
 
 `rejected.json` is always an object with a `rejected` array. Each entry has

docs/integrations/pkg-promptkit.md

@@ -0,0 +1,106 @@
+# PromptKit Integration
+
+Notarius pins
+[`gitea.maximumdirect.net/eric/promptkit` v0.5.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0)
+as its in-process prompt engine. The upstream
+[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/consumers/pkg-promptkit.md)
+owns the public engine API, and the upstream
+[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/formats.md)
+owns prompt, profile, and schema file contracts.
+
+## Supported Boundary
+
+Notarius relies on the root `promptkit` package to:
+
+- construct an `Engine` with filesystem-backed prompt, schema, and optional
+  operator and application-fallback profile sources;
+- prepare one frozen execution from a `RunRequest` with named inline artifacts,
+  variables, a direct session ID, prompt identity, and profile selection, then
+  record credential-redacted details and run that exact execution;
+- return rendered debug material, validated structured output, selected
+  profile, backend, effective model metadata, and token usage;
+- register the optional conventional `local` backend through `BackendLocal`,
+  `LocalBackend`, and `WithBackend`;
+- distinguish structured-output validation failure from execution failure; and
+- identify a missing explicit profile through `ErrProfileNotFound` and backend
+  admission exhaustion through `ErrCapacityExceeded`.
+
+The pinned
+[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/backends.go)
+owns the registration and backend-capacity contract.
+
+For one completion, the adapter calls `PrepareExecution`, takes a
+caller-owned `Details` snapshot, and calls `RunPrepared` for that same opaque
+prepared execution. It defers `Discard` for every unexecuted handle. Explicit
+profile preflight uses `Engine.InspectProfile`; it does not prepare a synthetic
+prompt. PromptKit's prepared handle, inspection result, and capacity-error
+types stay inside the Notarius LLM adapter.
+
+When a PromptKit profile and runtime override leave `temperature`, `max_tokens`,
+or `top_p` unset, Notarius leaves that control unset as well. Compatible
+providers therefore apply their own defaults; an operator that requires a
+specific sampling value must select it explicitly in the profile or runtime
+override.
+
+Notarius does not use PromptKit's optional `ArtifactReader`. It materializes
+source and reference content itself and supplies owned inline artifacts at the
+adapter boundary. It also retains responsibility for pipeline retries,
+scheduling, debug persistence, redaction, profile provenance, and conversion
+from private model responses into durable domain artifacts.
+
+Notarius sends its trimmed run 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. Session IDs are stable, non-secret correlation identifiers and
+may be exposed to providers and provider observability.
+
+Notarius records PromptKit's selected backend ID and effective reasoning
+setting as optional run-manifest provenance. Endpoint-only profiles have no
+backend ID. Debug prompt material also retains the selected backend ID and
+PromptKit's stable lower-case `effective_model_params` JSON, which may include
+`backend_id`. Notarius production configuration exposes one optional
+conventional `local` registration. It does not expose a general user-defined
+PromptKit backend registry. Endpoint-only profiles remain supported unchanged.
+
+Notarius retains its application-wide scheduled client around the PromptKit
+adapter. PromptKit may apply a narrower limit for the selected backend;
+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.
+
+## Profile Sources And Compatibility
+
+Notarius gives PromptKit the configured operator profile source, registered
+application fallback profile assets, and optional backend registration through
+the same construction path for inspection and execution. PromptKit owns the
+resulting source precedence and strict profile parsing: a matching operator
+profile is a complete replacement for a fallback or built-in profile, while an
+invalid matching document fails instead of falling through. The operator
+configuration and deployment workflow are defined in
+[Configuration](../config.md#promptkit-profiles) and
+[Operations](../operations.md#promptkit-profile-deployment).
+
+Notarius supports this boundary against PromptKit v0.5.0. 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,
+profile-source construction, and this compatibility statement against the
+pinned upstream documentation.
+
+## Notarius Ownership
+
+[LLM Runtime Internals](../internal/llm.md) describes how Notarius mounts
+module assets, maps its transport-neutral completion contract, prepares and
+executes requests, validates output, records provenance, captures debug
+material, redacts errors, and preserves timeout ownership.
+[D&D Module Internals](../internal/dnd.md) owns the embedded
+`dnd-extraction` fallback profile and the maintained D&D prompt defaults.
+[Configuration](../config.md#promptkit-profiles) defines how a Notarius
+configuration selects one PromptKit profile source and optionally registers
+the conventional local backend.
+
+PromptKit API or format changes outside this boundary are not implicitly
+supported. Updating the pinned version requires reviewing the adapter and
+profile/configuration contracts against the upstream documentation.

docs/integrations/run-result.md

@@ -0,0 +1,68 @@
+# Run Result Receipt
+
+`notarius run --json` writes this receipt to standard output when a run
+completes successfully. It lets a subprocess caller discover the physical root
+of the published output bundle without parsing interactive command output.
+Command syntax, streams, and exit statuses are defined in the
+[CLI reference](../cli.md); logical files within the bundle are defined in the
+[Published JSON Output contract](json-output.md).
+
+## Schema
+
+The current schema version is `notarius.run-result.v1`.
+
+| Field | Required | Meaning |
+| --- | --- | --- |
+| `schema_version` | Yes | Exactly `notarius.run-result.v1`. |
+| `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. |
+| `validation_status` | Yes | The final run manifest validation status. |
+| `debug_directory` | No | Absolute path to the run-specific debug bundle when requested debug capture completed. |
+
+For the production `json` output module, `index_file` is present only when the
+completed run returned exactly one logical output file named `index.json`.
+For another output module, its absence does not indicate a failed run.
+
+```json
+{
+  "schema_version": "notarius.run-result.v1",
+  "run_id": "run-1770000000000000000-0123456789abcdef0123456789abcdef",
+  "pipeline_id": "dnd-session",
+  "output_directory": "/work/results/run-1770000000000000000-0123456789abcdef0123456789abcdef",
+  "index_file": "index.json",
+  "normalized_output_count": 6,
+  "rejected_output_count": 2,
+  "warning_count": 1,
+  "validation_status": "rejected"
+}
+```
+
+## Paths And Bundle Discovery
+
+`output_directory` and `debug_directory`, when present, are lexical absolute
+paths. They identify the paths used by Notarius and do not resolve symlinks.
+`output_directory` is the run-specific bundle, not the configured output root.
+
+The receipt is a summary and discovery document. It does not contain lane
+descriptors, payloads, manifest data, rejections, warnings, or file contents.
+For the production JSON output, resolve `index_file` beneath
+`output_directory`, reject path escapes, and use the
+[Published JSON Output contract](json-output.md) to discover logical files and
+lane payloads.
+
+## Delivery And Compatibility
+
+Notarius writes the receipt only after the output bundle has been published and
+any requested debug terminal reporting has completed. Standard output is not
+transactional: a result-write failure returns a nonzero status and can leave
+partial bytes. Consumers must ignore standard output unless the process exits
+with status 0.
+
+Future versions may add optional fields to this schema. Consumers must tolerate
+unknown fields. An incompatible field or semantic change requires a new
+`schema_version` value.

docs/internal/cli.md

@@ -38,9 +38,14 @@ in [Configuration Internals](configuration.md).
 
 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 explicitly selected
-Scriptorium profiles. Pipeline listing validates configuration before returning
-normalized, sorted identifiers.
+effective catalog, resolves the pipeline, and verifies every explicit effective
+PromptKit profile. Selected LLM-backed input, chunk, lane, output, and validator
+profiles are inspected
+against the configured PromptKit source and backend registrations without
+loading a prompt or performing generation, so an unknown or invalid profile
+fails before pipeline preparation. Credential availability remains an
+execution-time concern. Pipeline listing validates configuration before
+returning normalized, sorted identifiers.
 
 ## Production Composition
 
@@ -51,12 +56,24 @@ catalog used for resolution and the concrete constructors used for preparation.
 Tests may provide a catalog or registries instead; production code must not
 silently merge an injected partial catalog with production registrations.
 
-The production LLM factory builds the Scriptorium-backed client from resolved
-configuration, creates one scheduler from the effective global LLM limit, and
-wraps the client before it reaches modules. Registration and LLM construction
-errors are returned before a pipeline is prepared. Concrete module keys and
-validator chains are public configuration choices and remain documented in
-[Configuration](../config.md).
+The production LLM factory builds one PromptKit-backed client from the resolved
+**promptkit.profile_dir** or **promptkit.profile_file** source, attaches the
+profile-provenance recorder, creates one scheduler from the effective global
+LLM limit, and wraps the client before it reaches modules. Registration and LLM
+construction errors are returned before a pipeline is prepared. Configuration
+field definitions remain in [Configuration](../config.md#promptkit-profiles);
+the D&D registrar's fallback profile assets and the adapter mechanics remain in
+[LLM Runtime](llm.md).
+
+The factory also accepts `LLMRuntimeOverrides`, whose reasoning pointer
+preserves inherit, replace, and clear states across the composition boundary.
+Run orchestration constructs this value from the mutually exclusive
+`--reasoning-effort` and `--clear-reasoning-effort` controls. Absence preserves
+a nil pointer, replacement is trimmed, and clear uses a non-nil empty string.
+The same override reaches the one shared production client, checkpoint
+identity, and debug invocation metadata. Persistent reasoning configuration
+remains owned by PromptKit profiles; Notarius configuration has no reasoning
+field.
 
 ## Run Orchestration
 
@@ -68,7 +85,8 @@ handoff:
 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
-   the effective pipeline, and verify explicit Scriptorium profiles;
+   the effective pipeline, and inspect its explicit effective PromptKit
+   profiles;
 4. materialize external or generated references and record redacted invocation
    and resolution provenance when debug capture is enabled;
 5. construct registries, the scheduled LLM client, prepared modules, and the
@@ -84,6 +102,18 @@ final command result. Detailed state lifecycle, resume handling, and physical
 path confinement are maintained in [Run State Internals](state.md) and
 [Operations](../operations.md).
 
+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
+output publication and requested debug terminalization succeed. A receipt-write
+failure exits with runtime status 1 and may leave partial standard-output bytes,
+but the already-published output bundle remains complete and requested debug
+reporting remains successfully terminalized. The CLI reports a bounded
+command-owned error and does not repeat terminal reporting. The receipt remains
+a CLI reporting concern rather than a framework or output-module responsibility;
+its public contract is the
+[run-result receipt](../integrations/run-result.md).
+
 ## Failure Mapping And Terminal Reporting
 
 Argument, flag, and invocation-combination failures are reported to standard

docs/internal/configuration.md

@@ -39,14 +39,22 @@ 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.
+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**.
 
 **Config.Validate** checks configuration-only invariants before resolution. It
 rejects incompatible profile sources, invalid state-surface values, unsupported
 concurrency settings, malformed bindings and references, invalid retries, and
-invalid pipeline, step, or lane structure. 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.
+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.
+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
@@ -56,13 +64,15 @@ environment override, its validation, and that reference in the same change.
 
 **Config.Resolve** first recomputes derived concurrency defaults and validates
 the configuration. It normalizes the requested pipeline ID, copies the selected
-profile, applies a non-empty command-level LLM profile override to the
-LLM-capable stage bindings, and calls the framework resolver with the requested
-lane selection and reference changes.
+profile, and passes the non-empty command-level LLM profile override, requested
+lane selection, and reference changes to the framework resolver.
 
-The command-level override does not replace an explicitly selected validator
-profile. Validator bindings remain part of the resolved validator chain and
-are resolved under their own declared configuration.
+After module and validator selection, the resolver applies the effective
+profile policy to LLM-backed bindings only: command override, binding profile,
+pipeline profile, then the prompt default. Deterministic bindings remain
+profile-free, and no second inheritance decision occurs during execution. The
+public field definitions and precedence are owned by
+[Configuration](../config.md#pipelines).
 
 The framework resolver supplies defaults, selects lanes, resolves validator
 chains, checks registered module and artifact compatibility, validates module
@@ -71,7 +81,8 @@ options, and returns the fixed ordered pipeline shape. The resulting
 changes, a clone of the input configuration, and the resolved pipeline.
 Callers may therefore retain or modify their input slices and maps without
 changing the resolved result, and later consumers cannot mutate the original
-configuration through the effective value.
+configuration through the effective value. This ownership includes the nested
+PromptKit local-backend value.
 
 Resolution failures stop before module construction and source parsing. They
 include an error path for an unconfigured pipeline, missing module, missing
@@ -83,7 +94,8 @@ runtime error class described in the [CLI reference](../cli.md#output-streams-an
 
 The framework assigns the resolved pipeline a deterministic SHA-256 digest
 after defaults, lane selection, module bindings, reference bindings, validator
-chains, and artifact schema identity have been resolved. The digest excludes
+chains, effective LLM profiles, and artifact schema identity have been
+resolved. The digest excludes
 its own stored value. It identifies resolved composition rather than raw YAML
 bytes, a debug payload, or all runtime state. The CLI records it as invocation
 provenance before execution; cache and checkpoint identity have additional
@@ -94,9 +106,12 @@ Configuration summaries must use **Redacted**, **RedactedSummaryPayload**, or
 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. This
-redaction is deliberately narrow: it protects configuration summaries and does
-not authorize recording arbitrary environment values or provider requests.
+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.
 
 ## Invariants To Preserve
 

docs/internal/dnd.md

@@ -22,10 +22,15 @@ does not repeat their JSON shapes or schemas.
 ## Family Composition
 
 The D&D registrar registers the family’s artifact codecs, extractors, typed
-append-order mergers, normalizers, validators, prompt assets, and default
-validator chains. Each extractor and normalizer has a stable module spec,
-strict option decoding, and a typed builder. Configuration remains the
-canonical owner of the exact keys and validator order.
+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 normalization are
+registered as `llm_backed`; the remaining current D&D mergers and normalizers
+are `deterministic`. The metadata is available to catalog inspection and
+resolved-pipeline debug data and determines which selected bindings inherit the
+pipeline profile. Configuration remains the canonical owner of the exact keys,
+profile precedence, and validator order.
 
 Private structured-LLM response schemas are deliberately minimal. They reject
 invalid JSON structure, missing required fields, incompatible types, and
@@ -39,9 +44,40 @@ D&D extractors assemble prompts from an ordered manifest of shared and
 module-owned assets. Reuse the shared D&D system, evidence, identity,
 reference, and transcript assets instead of copying their text into individual
 modules. A manifest’s declared sequence, including cache-control placement, is
-part of the prompt behavior, and the chunk transcript is the final message.
-Preserve that order when changing an extractor or its assets so prompt-cache
-behavior remains stable.
+part of the prompt behavior.
+
+Every maintained D&D LLM prompt selects `dnd-extraction` as its default
+profile. The D&D registrar embeds that fallback profile with the maintained
+OpenRouter model, timeout, and service-tier policy. An operator may provide a
+complete profile with the same ID through the configured PromptKit source; that
+definition replaces the fallback rather than merging with it. The fallback
+leaves reasoning and optional sampling controls unspecified. Deployment profile
+selection and the maintained operator example are documented in
+[Configuration](../config.md#promptkit-profiles).
+
+All extraction prompts share this four-message rendered prefix: the system
+message without cache control, the identity message without cache control, the
+campaign-reference message with ephemeral cache control, and the chunk
+transcript message with ephemeral cache control. This gives equivalent
+extraction requests the same reusable prefix through their source material.
+
+Extraction-evidence policy, generated NPC registries, spell catalogs, module
+tasks, and instructions follow the transcript because they are not universal
+across all extraction lanes. The final instructions message carries ephemeral
+cache control; evidence, registry, catalog, and task messages do not. Preserve
+this division when changing an extractor or its assets so prompt-cache behavior
+remains stable.
+
+The other D&D LLM prompts intentionally follow different patterns. Scene
+chunking has no sibling extraction lane with which to share its full transcript,
+so it renders campaign references before its task and instructions, then places
+the cacheable full transcript last. NPC normalization keeps its task and
+cacheable instructions before the candidate collection, followed by the
+cacheable transcript windows: candidates must be available before their
+supporting evidence is evaluated, and those windows are not a cross-lane
+prefix. Mounted assets and their declared message order determine the prompt
+fingerprint, so intentional prompt edits continue to invalidate stale
+checkpoints.
 
 All extractors use the shared prompt-input preparation rules. The current chunk
 is copied into transcript material; player, party, glossary, and compatible

docs/internal/llm.md

@@ -1,12 +1,12 @@
 # LLM Runtime Internals
 
 `internal/framework/llm` is Notarius’s provider-independent structured
-completion boundary. It adapts framework requests to Scriptorium, bounds
+completion boundary. It adapts framework requests to PromptKit, bounds
 provider calls, assembles registered prompt and schema assets, records selected
 profiles, and redacts provider errors. The architectural boundary is defined in
 [Architecture](../policy/architecture.md#llm-boundary); profile sources,
 credentials, and concurrency settings belong in
-[Configuration](../config.md#scriptorium-profiles) and
+[Configuration](../config.md#promptkit-profiles) and
 [Configuration](../config.md#concurrency-output-cache-and-debug).
 
 ## Structured Completion Boundary
@@ -24,22 +24,78 @@ adapter does not own source evidence, artifact conversion, normalization, or
 durable schemas. Those responsibilities remain with the module and its
 [integration contract](../integrations/).
 
-`ScriptoriumClient` validates the request target and prompt identity, maps each
-named material to a Scriptorium inline artifact while preserving its origin URI,
-forwards session and profile selection, then prepares and runs the prompt. It
-returns Scriptorium’s validated raw bytes rather than re-encoding the decoded
-target. An empty optional material is represented as one space so its named
-input is retained by Scriptorium.
+`PromptKitClient` validates the request target and prompt identity, maps each
+named material to a PromptKit inline artifact while preserving its origin URI,
+maps the trimmed request session to PromptKit's direct per-run session field,
+retains the same value as the `session_id` prompt variable for maintained
+prompt compatibility, and forwards profile selection. It then creates one
+frozen prepared execution, captures its caller-owned credential-redacted
+details for debug material, and executes that exact snapshot through
+PromptKit's prepared-execution boundary. The direct field
+is authoritative for provider session behavior. A session ID is a stable,
+non-secret correlation identifier and may be exposed to providers and provider
+observability. The adapter returns PromptKit’s validated raw bytes rather than
+re-encoding the decoded target. An empty optional material is represented as
+one space so its named input is retained by PromptKit.
 
-An empty request profile lets the prompt select its configured default. The CLI
-prepares every explicitly selected binding profile before a run begins, so a
-missing explicit profile fails before stage execution. Calls record the profile
-actually selected by Scriptorium; the recorder deduplicates non-secret profile
-identity, provider, and model values for manifest use.
+Client construction may also receive a run-wide reasoning-effort override from
+the CLI factory boundary. The adapter copies the caller-owned pointer and
+creates a fresh PromptKit execution override for each request: a nil pointer
+inherits the selected profile, a non-empty value replaces it, and an empty
+value clears inherited reasoning. The CLI's mutually exclusive
+`--reasoning-effort` and `--clear-reasoning-effort` controls select those
+states. With neither flag, profile behavior remains unchanged. Because
+production constructs one shared client, the selected state applies uniformly
+to module calls, retries, and LLM-backed validators for the whole run.
+
+An empty request profile lets the prompt select its configured default. Before a
+run begins, the CLI asks the adapter to inspect every explicit profile on the
+resolved selected LLM-backed bindings and validators, including inherited
+pipeline profiles. Inspection resolves the profile and its selected backend and
+target without loading a prompt, reading credentials, admitting capacity, or
+contacting a provider, so a missing or invalid explicit profile fails before
+stage execution while a valid `api_key_env` may remain unset. Calls record the
+profile actually selected by PromptKit. The recorder trims and deduplicates
+non-secret profile identity, provider, model, selected backend ID, and
+effective reasoning values for manifest use. Entries that differ in backend or
+reasoning remain distinct and deterministically ordered. Endpoint-only profiles
+retain an empty backend ID, which the published JSON omits. Successful
+completion responses and recorded profile manifests identify the adapter
+provider as `promptkit`.
+
+The CLI's profile-inspection engine and the production adapter use the same
+profile-source construction to apply the configured profile directory or file,
+the optional registered fallback profile assets, and the optional conventional
+`local` backend. Preflight therefore resolves the same profile sources and
+backend membership as runtime without performing generation. Fallback assets
+are mounted only when at least one source is registered. The production D&D
+registrar contributes its `dnd-extraction` fallback, and the maintained D&D
+prompts select that logical ID by default. PromptKit owns source precedence and
+profile parsing: an operator-provided matching profile takes precedence over a
+fallback profile without Notarius merging either document.
+When the registration is absent, a profile selecting `backend: local` fails
+inspection instead of falling back to a built-in or endpoint-only target.
+
+Before execution, the adapter also contributes a non-secret checkpoint
+fingerprint for the effective PromptKit profile source. It combines the
+identity of PromptKit's compiled-in profile catalog with a deterministic digest
+of every YAML profile in the configured profile directory, or of the configured
+profile file, and a deterministic digest of the flattened fallback profile
+assets. The fingerprint contains neither profile content nor source paths. It
+covers inherited pipeline profiles, explicit binding profiles, and
+prompt-selected defaults, so changing a model or other profile setting cannot
+reuse checkpoints created under the
+prior profile source. This cache identity is independent of durable
+profile provenance: run manifests continue to list only profiles actually
+observed during LLM calls. When the local backend is registered, a second
+fingerprint hashes its trimmed endpoint behind a stable marker. Changing that
+semantic execution target invalidates checkpoint reuse. The raw endpoint is not
+stored in checkpoint identity, and the local concurrency limit is excluded
+because it changes scheduling rather than execution semantics.
 
 ## Shared Provider-Call Limit
 
-Production construction creates one Scriptorium client and wraps it in one
+Production construction creates one PromptKit client and wraps it in one
 scheduled client. The scheduler has a fixed, positive permit limit, serves
 queued calls in FIFO order, and removes a queued call when its context is
 cancelled. A granted permit is released exactly once on every completion path.
@@ -51,12 +107,23 @@ worker counts cannot exceed the configured LLM limit. The configuration field
 and its effective default are owned by
 [Configuration](../config.md#concurrency-output-cache-and-debug).
 
+PromptKit applies a second, independent admission limit when the selected
+profile names a limited backend. It sits beneath the Notarius scheduled client,
+so it may narrow but cannot expand the application-wide limit. Built-in
+OpenRouter profiles select PromptKit's reserved backend and its upstream
+capacity policy. A positive configured local-backend limit bounds active local
+generations inside PromptKit; zero leaves that backend unlimited there.
+Endpoint-only profiles do not select a PromptKit backend and remain limited
+only by the Notarius scheduler.
+
 ## Prompt And Schema Assets
 
-An `AssetRegistry` collects prompt and schema filesystems from production module
-families. It flattens registered roots into the Scriptorium filesystems and
-rejects invalid roots, unreadable assets, duplicate paths, and missing prompt
-or schema files during preparation. The framework’s `promptfs` helper combines
+An `AssetRegistry` collects prompt, schema, and optional fallback-profile
+filesystems from production module families. It flattens registered roots into
+the corresponding PromptKit filesystems and rejects invalid roots, unreadable
+assets, duplicate paths, and missing prompt or schema files during preparation.
+Fallback assets receive a safe content digest for checkpoint identity; raw
+paths and bytes are never included. The framework’s `promptfs` helper combines
 module-owned prompt files with reusable domain fragments without making the
 framework depend on D&D content.
 
@@ -76,30 +143,43 @@ contract. Durable formats and compatibility rules remain in the
 ## Prompt Maintenance And Backend Caching
 
 Prompt message order and shared asset bytes are runtime behavior. Backend cache
-reuse depends on the same preceding messages and content, not merely equivalent
-meaning. Keep reusable shared assets byte-identical and keep stable material
-before the inputs that vary per request wherever a prompt’s declared sequence
-supports caching. Preserve the existing manifest order and cache-control hints
-when editing a prompt.
+reuse depends on identical preceding roles, rendered bytes, and cache-control
+metadata—not merely equivalent meaning. Keep reusable shared assets
+byte-identical and preserve each prompt’s declared ordering and cache controls
+when editing it.
 
-D&D extraction manifests place the changing chunk transcript at the end of the
-prompt after their reusable context. Scene chunking and NPC normalization use
-their own declared message sequences because their inputs and work differ. The
-family-specific asset and ordering rules belong in [D&D Module Internals](dnd.md).
-Do not add tests that enforce a fixed message-prefix length; prompt-asset tests
-should instead verify the meaningful asset sequence, inputs, and cache controls
-of the prompt being changed.
+For sibling prompts that can reuse the same source material, order universal
+shared context first, request source material next, and module-specific
+suffixes last. Put a cache boundary at a reusable prefix that is useful to the
+backend. Redundant intermediate cache boundaries do not extend that reusable
+prefix and add no value.
+
+Prompt-family owners may choose a different sequence when their inputs and
+reuse pattern differ. The D&D family’s extraction, scene-chunking, and NPC
+normalization policies are maintained in [D&D Module Internals](dnd.md#prompt-construction).
+Do not add tests that enforce prompt prose; prompt tests should verify the
+meaningful input placement and cache controls of the prompt being changed.
 
 ## Validation, Repair, And Retries
 
-Scriptorium performs prompt rendering, provider execution, and the prompt’s
+PromptKit performs prompt rendering, provider execution, and the prompt’s
 structured-output validation. The adapter reports an empty result, validation
 failure, empty structured body, or decode failure as
 `ErrInvalidStructuredOutput`, while retaining the returned raw bytes and debug
 material when they exist. Provider failures remain operational errors rather
 than output-validation failures.
 
-Prompt-declared repair is executed within Scriptorium’s structured-output flow.
+When PromptKit rejects backend admission before generation, the adapter maps
+`promptkit.ErrCapacityExceeded` to
+`contracts.ErrLLMCapacityExceeded`, retaining prompt context and a redacted
+upstream diagnostic without exposing the PromptKit sentinel or capacity-error
+type as a framework contract. When supplied, the normalized selected backend
+ID appears only in that safe application-owned diagnostic context. A canceled
+caller context takes precedence. The adapter does not retry capacity failures;
+the pipeline's existing binding attempt policy sees the operational error and
+decides whether to rerun the complete operation.
+
+Prompt-declared repair is executed within PromptKit’s structured-output flow.
 The current production D&D prompt manifests set repair attempts to zero. That
 setting does not replace pipeline retry behavior: a binding’s configured retry
 count reruns its stage attempt after an error or rejection, and an exhausted
@@ -108,22 +188,42 @@ 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.
+
+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.
+Operator-facing behavior is summarized in
+[Operations](../operations.md#operational-limits), and the pinned upstream
+contract is identified in
+[PromptKit Integration](../integrations/pkg-promptkit.md).
+
 ## Observability And Redaction
 
 When debug recording is enabled, the pipeline decorates the shared client. The
 wrapper records prepared prompt and response material, timing, selected profile
-and model, and call identifiers in the run’s debug bundle, including material
-available from a failed structured completion. For a successful completion, a
-debug-write failure is surfaced; when the completion already failed, its call
-error remains the result. Debug-bundle location, retention, and handling are
-operational concerns documented in [Operations](../operations.md#debug-bundles).
+and backend, effective model parameters, and call identifiers in the run’s
+debug bundle, including material available from a failed structured completion.
+Effective parameters use PromptKit's stable lower-case JSON field names and may
+include `backend_id`. For a successful completion, a debug-write failure is
+surfaced; when the completion already failed, its call error remains the
+result. Debug-bundle location, retention, and handling are operational concerns
+documented in [Operations](../operations.md#debug-bundles).
 
-Run manifests receive selected profile summaries and component identities, not
-prompt, schema, source, reference, or response content. Provider error text is
-wrapped with prompt context and bearer credentials are redacted before it
-crosses the runtime boundary. Known-secret redaction is available to other
-runtime collaborators; it does not make prompt or response contents safe for
-general logging.
+Run manifests receive selected profile summaries, including optional effective
+backend and reasoning provenance, and component identities—not prompt, schema,
+source, reference, or response content. The published field semantics belong
+to the [JSON output contract](../integrations/json-output.md#manifestjson).
+Provider error text is wrapped with prompt context and bearer credentials are
+redacted before it crosses the runtime boundary. Known-secret redaction is
+available to other runtime collaborators; it does not make prompt or response
+contents safe for general logging.
 
 ## Failure Boundaries
 
@@ -131,6 +231,8 @@ general logging.
   sources, invalid asset registration, or a non-positive scheduler limit.
 - Preparation failures, unavailable explicit profiles, provider failures, and
   context cancellation propagate to the calling stage with context.
+- Backend admission exhaustion is a provider-neutral operational error and is
+  not classified as invalid structured output or validator rejection.
 - Malformed or schema-invalid provider output is classified separately as
   invalid structured output so the module or pipeline can apply its own retry
   and rejection policy.

docs/internal/modules.md

@@ -12,9 +12,15 @@ exceptions. See [D&D Module Internals](dnd.md) rather than adding them here.
 
 A module is a typed implementation registered for one pipeline stage. Its
 `ModuleSpec` is the public-to-the-framework declaration of its stable key,
-stage, required and provided capabilities, artifact kind, and accepted
-reference slots. The framework uses that declaration to resolve a configured
-binding before it builds the implementation.
+stage, execution class, required and provided capabilities, artifact kind, and
+accepted reference slots. The execution class states whether a module is
+`deterministic` or `llm_backed`; registries retain it for catalog inspection and
+resolved-pipeline debug data without constructing the module. The framework
+uses the declaration to resolve a configured binding before it builds the
+implementation. After selection, the resolver applies profile inheritance only
+to bindings whose declared execution class is `llm_backed` and rejects a
+binding-specific profile on a deterministic module. The user-facing precedence
+contract belongs in [Configuration](../config.md#pipelines).
 
 Implementations that accept options must provide both an option validator and
 a builder. The validator is used while resolving configuration; the builder
@@ -30,6 +36,14 @@ they need, register each leaf implementation, and add any family-owned assets
 or default validator chains. They return contextual errors so production
 composition fails at startup rather than at the first run.
 
+An artifact family can register an optional typed evidence projector alongside
+its codec. The projector returns defensive copies of the artifact's direct
+generic source references and must use the codec's exact Go type. It does not
+interpret surrounding context or publish files; the pipeline validates the
+capability during preparation and the output boundary owns publication. See
+the [Published Evidence Context contract](../integrations/evidence-context.md)
+for the durable result.
+
 ## Production Composition
 
 Production composition is intentionally split by family:
@@ -39,7 +53,7 @@ Production composition is intentionally split by family:
 - The Seriatim registrar provides the transcript input adapter. Its external
   input behavior is defined by the [Seriatim contract](../integrations/seriatim.md).
 - The D&D registrar provides its codecs, extractors, mergers, normalizers,
-  validators, prompt assets, and default chains. Its behavioral conventions
+  validators, prompt assets, fallback profile asset, and default chains. Its behavioral conventions
   are documented in [D&D Module Internals](dnd.md).
 
 The CLI owns the composition that invokes these registrars. A module package
@@ -51,10 +65,10 @@ packages depend on production extensions.
 1. Choose the pipeline stage and the typed artifact boundary. Put external
    input or durable artifact formats in the relevant integration contract,
    not in this guide or in a private LLM response type.
-2. Define a stable `ModuleSpec` with the exact capabilities and reference
-   slots needed for the operation. Model a producer/consumer handoff as an
-   artifact-compatible slot; configuration then chooses an external file or a
-   generated binding.
+2. Define a stable `ModuleSpec` with an explicit execution class, the exact
+   capabilities, and reference slots needed for the operation. Model a
+   producer/consumer handoff as an artifact-compatible slot; configuration
+   then chooses an external file or a generated binding.
 3. Implement strict option decoding, construction, and the typed stage
    interface. Preserve caller ownership: do not retain mutable request data
    and return defensive copies where an implementation exposes stored data.

docs/internal/pipeline.md

@@ -34,6 +34,11 @@ requested lanes where that is supported, resolves validator chains, checks
 module capabilities and typed artifact compatibility, validates options, and
 assigns a deterministic resolved-composition digest. The resolved pipeline
 contains bindings and declared reference targets, not external reference bytes.
+After selection, the resolver applies command, binding, and pipeline profile
+precedence to LLM-backed bindings and validators only; prompt defaults remain
+an empty resolved binding profile. Deterministic bindings remain profile-free.
+These effective values are part of the digest, so execution and checkpoint
+consumers do not repeat profile inheritance.
 Configuration resolution supplies the selected profile and catalog; see
 [Configuration Internals](configuration.md).
 
@@ -52,6 +57,13 @@ checkpoint fingerprints. Missing registrations, incompatible typed entries,
 nil implementations, and constructor failures are reported before source
 parsing or any stage operation begins.
 
+An output encoder can opt into source-evidence publication through its output
+policy. Preparation keeps the configured lane allowlist and active lanes
+separate, then verifies an exact typed evidence projector and registered codec
+for each active lane. The resulting private plan is immutable; lanes excluded
+by invocation filtering remain configured but do not acquire a projector for
+that run.
+
 ## Typed Lanes And References
 
 Each resolved lane has one artifact kind, codec, and exact Go type. The
@@ -110,11 +122,16 @@ directives consume this same budget and validate any final safe fallback through
 the normalizer chain.
 
 After terminal lane work, the runner assembles manifest provenance, normalized
-artifacts, rejections, warnings, and an optional accepted chunk map. The output
-encoder returns logical files; it does not choose a physical directory. The CLI
-publishes those files only after the runner returns without a framework error.
-Logical file names and schemas are defined by the
-[output integration contracts](../integrations/).
+artifacts, rejections, warnings, 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
+also used after normalized-checkpoint reuse, so no second typed output channel
+is retained. The runner passes the resulting owned artifact to the output
+encoder, which returns logical files and does not choose a physical directory.
+The CLI publishes those files only after the runner returns without a framework
+error. Logical file names and schemas are defined by the [output integration
+contracts](../integrations/).
 
 ## Checkpoint And Debug Hooks
 

docs/internal/state.md

@@ -39,7 +39,9 @@ codecs, loader, and recorder. The CLI constructs a recorder whenever checkpoint
 recording is enabled and constructs a loader only for a `--resume` invocation.
 Identity incorporates explicit stable semantic fingerprints collected from
 prepared modules and validators in addition to configuration, input,
-references, runtime overrides, and LLM profiles.
+references, runtime overrides, observed LLM profiles, and the LLM runtime's
+non-secret effective profile-source identity. A profile source change therefore
+causes a cold miss even when the configured profile ID remains unchanged.
 The serialized
 `workspace_schema_version` identifiers are frozen wire-compatibility fields;
 they do not describe a current public state surface.

docs/operations.md

@@ -36,7 +36,47 @@ 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.
+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
+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.
+
+## PromptKit Profile Deployment
+
+Profile deployment has four distinct layers:
+
+| Layer | Owner | Operational role |
+| --- | --- | --- |
+| Prompts and schemas | Notarius module families | Embedded request and structured-output definitions. They are not deployment profile files. |
+| Fallback profiles | Notarius module families | Embedded application defaults, including D&D's `dnd-extraction` profile. |
+| Built-in profiles | PromptKit | Upstream catalog entries available when no higher-precedence source defines an ID. |
+| Operator profiles | Deployment filesystem | Complete environment-specific definitions selected by `promptkit.profile_file` or `promptkit.profile_dir`. |
+
+The maintained D&D pipeline uses the workload ID `dnd-extraction`. The
+embedded fallback makes that ID usable without an operator file. Production,
+development, and local deployments can each install a different complete
+definition for the same ID, retaining the pipeline while choosing their own
+model, backend, timeout, or reasoning policy. An operator definition wins over
+the fallback; it is not merged with it. The configuration field and full
+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
+can be copied as a format starting point. Validate a deployment without a
+provider call or credentials:
+
+~~~sh
+notarius config validate --config /etc/notarius/config.yml --pipeline dnd-session
+~~~
+
+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.
 
 ## Run Lifecycle
 
@@ -66,7 +106,13 @@ run directory remains for inspection and is never removed automatically.
 Treat an output bundle as durable user data. Do not use cache-cleanup policy to
 remove it. An optional accepted chunk map is also durable output and can carry
 source- or model-derived annotations; its content and compatibility contract
-are defined in [Accepted Chunk Map](integrations/chunk-map.md).
+are defined in [Accepted Chunk Map](integrations/chunk-map.md). An optional
+[evidence context](integrations/evidence-context.md) contains source-unit text
+and metadata. It is not a cache or debug artifact: retain it with the output
+bundle only for as long as consumers need it, and apply source-content access
+controls to the entire bundle. Selected lanes may collectively cite most of a
+transcript, so a broad allowlist can make the evidence artifact nearly as
+sensitive and large as the source itself.
 
 ## Chunk-Plan Cache
 
@@ -107,9 +153,15 @@ compatible recorded work. A resume request fails when checkpoint recording is
 disabled. Without **--resume**, a recording-enabled run executes normally and
 does not load checkpoint state. Compatibility includes the resolved pipeline,
 input, selected lanes, runtime overrides, reference provenance, LLM-profile
-provenance, and prepared-component fingerprints. A changed identity produces a
-cold miss; Notarius does not migrate, rewrite, or delete older checkpoint
-directories.
+provenance, the effective PromptKit profile-source fingerprint, and
+prepared-component fingerprints. When a local PromptKit backend is configured,
+compatibility also includes a non-secret fingerprint of its endpoint. Changing
+profile content or the local endpoint causes a cold miss; changing only the
+local concurrency limit does not. A changed identity produces a cold miss;
+Notarius does not migrate, rewrite, or delete older checkpoint directories.
+Reasoning-effort inheritance, replacement, and explicit clearing are distinct
+runtime identities, so checkpoints created under one state are not reused by
+either of the others.
 
 Checkpoint state is confined below an identity-specific path:
 
@@ -169,7 +221,9 @@ warning, checkpoint, chunk-plan, and terminal reporting artifacts. The trace
 contains allowlisted application diagnostic records and can include source or
 derived application data. Neither surface is a cache input. Do not treat a
 debug bundle as safe to share merely because its configuration summary is
-redacted.
+redacted. Invocation metadata omits reasoning effort when it is inherited,
+records the replacement value when one is supplied, and records an empty value
+when inherited reasoning was explicitly cleared.
 
 Notarius never creates debug state without an explicit request and never
 automatically deletes a requested bundle. If allocation succeeds, the command
@@ -197,9 +251,40 @@ or automatic cleanup command.
 
 ## Operational Limits
 
-Provider retries and timeouts are supplied by the selected Scriptorium profile.
-Module retry settings and concurrency limits are configuration contracts; see
-[module bindings](config.md#module-bindings-and-validators) and
+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
+v0.5.0 does not add a provider retry loop. Notarius binding retries rerun the
+complete module operation and validation chain as defined by
+[module bindings](config.md#module-bindings-and-validators).
+
+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.
+The pinned upstream boundary and profile-format links are in
+[PromptKit Integration](integrations/pkg-promptkit.md).
+
+Concurrency has two independent layers. Notarius **total_llm** is the
+application-wide provider-call limit shared by all backends, modules, retries,
+and validators. PromptKit may impose a narrower admission limit for the
+selected backend. The effective active-generation bound is the intersection of
+both limits and can therefore be lower than **total_llm**. Built-in OpenRouter
+profiles use PromptKit's upstream backend limit; endpoint-only profiles have no
+PromptKit backend limit and remain bounded by Notarius. For the configured
+local backend, a zero **concurrency_limit** leaves only the Notarius scheduler
+as a call limit. A positive value makes the effective active local-generation
+bound the smaller of **total_llm** and that local limit.
+
+For a positive local limit, PromptKit owns its default waiting capacity and
+admission behavior. When a PromptKit backend has admitted all active and queued
+work, a new call fails as capacity exhaustion before generation. The adapter
+maps that failure to Notarius's existing provider-neutral capacity error and
+does not retry it. The calling stage's configured retry policy applies
+normally, and the run fails if those attempts are exhausted. Caller
+cancellation remains authoritative. Configuration contracts are documented
+under [PromptKit profiles](config.md#promptkit-profiles) and
 [concurrency](config.md#concurrency-output-cache-and-debug). Extract-worker
 limits and actual provider-call limits are independent. Notarius writes local
 filesystem state only; remote storage, archival, and retention automation are

docs/policy/architecture.md

@@ -79,6 +79,14 @@ Pipeline resolution requires a compatible codec and matching kind-specific
 variants before a typed lane can be accepted. Framework-owned erasure remains
 private and must report type incompatibility as an error rather than a panic.
 
+An artifact kind may additionally provide a typed evidence projection that
+copies its direct generic source references. Preparation proves that projection
+matches the artifact codec's exact Go type before retaining it for an output
+policy. The runner reconstructs evidence only from accepted serialized
+normalized artifacts, and the output boundary owns any resulting publication.
+Generic framework code never infers evidence by inspecting domain JSON or
+depends on domain artifact types.
+
 Auxiliary references provide context or disambiguation. They are not source
 evidence and must not be converted into source references.
 
@@ -169,7 +177,8 @@ individual modules.
 The application-wide LLM scheduler bounds actual provider calls independently
 of framework worker limits. Every LLM-backed module, retry, and validator uses
 the single injected scheduled client, including work performed by overlapping
-lanes.
+lanes. Provider runtime adapters may enforce a narrower backend-specific limit
+beneath this mandatory application-wide scheduler.
 
 ## Configuration And Provenance
 

docs/roadmap/future.md

@@ -7,6 +7,37 @@ not as committed release dates.
 
 ## Near-Term D&D Pipeline
 
+### Combat Enemy Ledger
+
+- Add a D&D artifact that identifies enemies faced during combat and supports
+  an end-of-session encounter ledger.
+- Track each enemy's observed state using a small controlled vocabulary such as
+  `active`, `killed`, `fled`, `captured`, or `incapacitated`, while preserving
+  an explicit unresolved state when the transcript does not establish an
+  outcome.
+- Preserve the evidence for enemy participation and state changes rather than
+  inferring a terminal outcome from combat ending or an enemy disappearing
+  from the conversation.
+- Define how repeated mentions, groups of unnamed enemies, summoned or allied
+  creatures, and the same enemy appearing in multiple combats affect identity
+  and ledger entries.
+- Evaluate whether the ledger should be extracted directly, derived from
+  combat-turn artifacts, or use a sequential pipeline that consumes combat
+  turns and the normalized NPC registry as grounding references.
+
+### Location Extraction
+
+- Add a D&D artifact for locations visited by the party or otherwise mentioned
+  in the transcript.
+- Distinguish observed visits from references, plans, recalled places, and
+  uncertain or inferred locations so a mention alone is not reported as a
+  visit.
+- Preserve transcript evidence for each visit or mention and reconcile aliases,
+  nested places, and repeated appearances without collapsing distinct
+  locations that share a generic name.
+- Define how the location artifact should ground later narrative reports and
+  whether future event artifacts should retain canonical location identities.
+
 ### Evaluate Spell Extraction And Normalization
 
 - Evaluate ordinary extraction retries and the completed normalization path
@@ -16,15 +47,77 @@ not as committed release dates.
   validator, and normalizer development. Treat model-quality review as an
   iterative human evaluation aid, not a deterministic correctness gate.
 
-### Use D&D Scene Chunking
+### Evaluate The Shared D&D Scene Plan
 
-- Use ordered pipeline steps whenever a later artifact needs an accepted
-  earlier artifact as context. Keep independent lanes in the same step and do
-  not introduce a general DAG or concurrent cross-lane reconciliation model.
 - Reassess whether one shared scene plan provides enough context for NPC,
   spell, combat, interaction, and scene-description lanes after real-world use.
   Add more complex chunking only in response to demonstrated failures.
 
+## Cross-Cutting LLM Runtime
+
+### Deterministic Prompt Session Identity
+
+- Replace the source-document-ID default for prompt sessions with one
+  predictable, procedurally generated session ID for the complete
+  source-processing workload.
+- Preserve an explicit non-empty `--session-id` as the highest-precedence
+  override. Otherwise, derive the default only from the effective input module
+  identity and the exact raw input bytes.
+- Use a versioned, bounded representation such as
+  `notarius:v1:<sha256(input-module + NUL + raw-input)>`. The exact encoding
+  must fit PromptKit's session length contract and must not embed source
+  content.
+- Keep the derived session stable across runs, pipelines, selected lanes,
+  ordered steps, retries, resume, recomputation, LLM profiles, reasoning
+  overrides, and output, debug, or cache settings.
+- Do not include file-backed references, generated references, reference
+  contents, or the composition of a reference bundle in session derivation.
+  References may change between prompt calls within one pipeline without
+  changing routing affinity.
+- Resolve the authoritative session before checkpoint construction and use the
+  same value for checkpoint runtime identity, every prompt-facing module,
+  PromptKit's direct session field, the compatibility `session_id` prompt
+  variable, run-manifest metadata, and debug metadata.
+- Keep routing identity separate from cache and checkpoint content identity.
+  Exact prompt prefixes, reference contents, model settings, and other
+  generation-affecting inputs must continue to participate in their existing
+  hashes and checkpoint fingerprints even though they do not change the
+  session.
+- Treat the generated value as a provider-visible, stable pseudonymous
+  correlation identifier. Do not introduce an installation-specific HMAC or
+  secret unless a concrete multi-tenant or privacy requirement justifies
+  sacrificing deterministic identity across installations.
+
+### Raise The Default Application-Wide LLM Limit
+
+- Raise the default `concurrency.total_llm` value from 1 to 16 so ordinary
+  single-backend runs can use PromptKit's expected OpenRouter capacity and
+  lower-capacity local backends without an unnecessarily narrower Notarius
+  limit.
+- Keep the Notarius application-wide scheduler mandatory and require
+  `total_llm` to remain a positive integer. Do not make the default unlimited:
+  endpoint-only profiles, an unrestricted local backend, injected clients, and
+  aggregate work across several backends may have no narrower PromptKit limit.
+- Continue defaulting `concurrency.stage_workers.extract` to the effective
+  `total_llm`, making its default 16 as part of the same change. Preserve an
+  explicit lower extract-worker setting when an operator wants less queued or
+  concurrent extraction work.
+- Define effective provider concurrency as the intersection of the Notarius
+  application-wide limit, the selected PromptKit backend limit when present,
+  and the work made available by stage execution. A Notarius limit of 16 does
+  not narrow a backend already limited to 16, while a local backend limited to
+  4 remains bounded at 4.
+- Treat the default as an application-wide safety ceiling across profiles,
+  backends, modules, retries, and validators. A run that intentionally needs
+  the combined capacity of several backends may configure a higher
+  `total_llm` and an appropriate extract-worker count explicitly.
+- Retain the existing configuration and environment override surfaces. Update
+  canonical configuration, operations, and internal documentation together
+  when the default changes.
+- Reconsider decoupling the extract-worker default from `total_llm` only after
+  mixed-backend workloads demonstrate a need for a high global emergency
+  ceiling with a lower default work-production rate.
+
 ## Shared Normalization And Quality Work
 
 ### Generic LLM-Assisted Deduplication

docs/roadmap/implementation.md

@@ -0,0 +1,586 @@
+# PromptKit v0.5 Implementation Plan
+
+## Objective
+
+Implement the target state in
+[PromptKit v0.5 Integration And LLM Profile Policy](promptkit.md). Each numbered
+stage is intended to be one implementation prompt for a GPT-5.6-Terra coding
+agent. Complete stages in order and leave the repository buildable, tested, and
+internally coherent after every stage.
+
+Follow [Architecture](../policy/architecture.md),
+[Testing Policy](../policy/testing.md), and
+[Documentation Policy](../policy/documentation.md) throughout. Preserve
+unrelated user changes. Use `apply_patch` for source and documentation edits,
+run `gofmt` on changed Go files, and add only tests that protect the behaviors
+and risks assigned to that stage.
+
+Do not implement the separate deterministic session-ID or default-concurrency
+roadmap items as part of this plan. Do not perform paid or credentialed LLM
+calls.
+
+## Background Summary
+
+Notarius currently pins PromptKit v0.3.0, calls `Prepare` and then `Run` for one
+completion, validates profiles through a synthetic prompt, has no application
+fallback profile source, and accepts LLM profiles only at individual bindings
+or through the run-wide CLI override. PromptKit v0.5.0 is source-compatible
+with the current tree; a temporary v0.5.0 module override has already passed
+`go test ./...`.
+
+The implementation must nevertheless treat the upstream optional-parameter
+change as intentional: unset `temperature`, `max_tokens`, and `top_p` remain
+unset and are omitted from compatible provider requests. Do not restore the old
+implicit `top_p: 1` default.
+
+## Stage 1: Upgrade The PromptKit Dependency
+
+### Goal
+
+Establish a clean PromptKit v0.5.0 baseline before adopting its new APIs.
+
+### Work
+
+- Update `go.mod` and `go.sum` from PromptKit v0.3.0 to v0.5.0 and run
+  `go mod tidy`.
+- Change the PromptKit built-in profile-catalog marker in
+  `internal/framework/llm/promptkit_profile_fingerprint.go` to identify
+  v0.5.0. This deliberately invalidates LLM checkpoints tied to the prior
+  catalog identity.
+- Review PromptKit-facing compile errors or test failures against the v0.4.0
+  and v0.5.0 release guides. Do not adopt prepared execution, inspection, or
+  fallback profiles in this stage.
+- Replace the existing test assertion for one exact built-in fingerprint hash
+  with durable assertions that the fingerprint is deterministic, non-empty,
+  non-secret, and changes when a semantic profile source changes. Do not add a
+  new version-constant or exact-hash change detector.
+- Update `docs/integrations/pkg-promptkit.md` to pin and link v0.5.0 and state
+  the implemented dependency-level behavior: unset optional sampling controls
+  are provider defaults. Do not document later stages as implemented.
+- Update any other canonical text that explicitly claims the dependency is
+  v0.3.0, but defer descriptions of unimplemented v0.5 APIs.
+
+### Tests And Validation
+
+- `go test ./internal/framework/llm ./internal/cli`
+- `go test ./...`
+- `go vet ./...`
+- `go build ./cmd/notarius`
+- `rg -n 'promptkit v0\.3\.0|promptkit@v0\.3\.0|PromptKit v0\.3\.0' .`
+- `git diff --check`
+
+### Completion Criteria
+
+- The repository directly pins v0.5.0 and all default offline checks pass.
+- The profile-source fingerprint identifies the new upstream catalog without a
+  brittle literal-hash test.
+- Current documentation no longer identifies v0.3.0 as the supported version.
+
+## Stage 2: Execute One Frozen Prepared Snapshot
+
+### Goal
+
+Make Notarius debug details and generation use one exact PromptKit preparation.
+
+### Work
+
+- Refactor `PromptKitClient.CompleteStructured` to call
+  `PrepareExecution`, immediately defer `Discard`, obtain a caller-owned
+  `Details` value, and execute with `RunPrepared`.
+- Preserve the existing Notarius request mapping, cancellation precedence,
+  validation classification, raw structured bytes, response decoding,
+  profile recording, usage reporting, and credential redaction.
+- Ensure every preparation, execution, validation, empty-result, and decode
+  error retains useful Notarius prompt context without exposing prepared handle
+  state or secrets.
+- Use `errors.As` to obtain `*promptkit.CapacityError` on admission rejection.
+  Preserve `contracts.ErrLLMCapacityExceeded` as the stable classification and
+  add a nonblank backend ID only to safe application-owned diagnostic context.
+  Do not expose `promptkit.CapacityError` outside the LLM adapter.
+- Update `docs/internal/llm.md` and the implemented-mechanics portion of
+  `docs/integrations/pkg-promptkit.md` to describe the single frozen execution
+  snapshot and structured capacity adaptation.
+
+### Tests And Validation
+
+- Adapt existing PromptKit client tests to the prepared-execution path.
+- Retain or add one behavioral test proving that the debug prompt details match
+  the request actually passed to generation when a backing prompt source could
+  otherwise change between independent preparations. Test the resulting
+  snapshot consistency, not a private helper call count.
+- Retain capacity tests proving `errors.Is` reaches
+  `contracts.ErrLLMCapacityExceeded`, the selected backend can appear in safe
+  diagnostic context, and provider calls are not made after rejected
+  admission.
+- Run `go test ./internal/framework/llm` and
+  `go test -race ./internal/framework/llm`.
+- Run `go test ./...` and `git diff --check`.
+
+### Completion Criteria
+
+- `CompleteStructured` no longer calls independent `Prepare` and `Run`
+  operations for one request.
+- Debug prompt material and generation result originate from the same frozen
+  PromptKit snapshot.
+- Capacity remains a provider-neutral Notarius error classification.
+
+## Stage 3: Replace Synthetic Profile Validation With Inspection
+
+### Goal
+
+Validate profiles through PromptKit's exact profile-inspection boundary and
+centralize engine profile-source construction.
+
+### Work
+
+- Introduce a small provider-adapter-owned profile inspection or validation
+  function in `internal/framework/llm`. Its public internal signature must use
+  Notarius-owned configuration and result/error types rather than returning
+  PromptKit types to the CLI.
+- Share the code that applies `profile_dir`, `profile_file`, and registered
+  backend options between the production PromptKit engine and the inspection
+  engine. Preserve the mutual-exclusion and local-backend rules.
+- Change CLI explicit-profile preflight to use `Engine.InspectProfile` through
+  that LLM boundary.
+- Remove `profileCheckPromptID`, `profileCheckPromptFS`, the `testing/fstest`
+  production dependency, and the synthetic `Prepare` request.
+- Preserve distinct, useful errors for an absent profile, invalid profile,
+  unknown backend registration, cancellation, and invalid profile source.
+- Do not require `api_key_env` to be populated during configuration validation.
+  Inspection may report credential requirements internally, but actual
+  preparation remains responsible for credential availability before a model
+  call.
+- Update current-behavior sections in `docs/internal/cli.md` and
+  `docs/internal/llm.md`. Keep field definitions in `docs/config.md`.
+
+### Tests And Validation
+
+- Replace synthetic-prompt tests with profile inspection tests covering:
+  configured local backend success; missing local backend failure; absent
+  profile; malformed profile; and an otherwise valid profile whose credential
+  environment variable is intentionally unset.
+- Prove validation performs no provider HTTP call and remains offline.
+- Run `go test ./internal/framework/llm ./internal/cli` and `go test ./...`.
+- Run `git diff --check`.
+
+### Completion Criteria
+
+- No production synthetic profile-check prompt remains.
+- Profile validation uses the same ordinary profile source and backend
+  registrations as execution.
+- Configuration validation succeeds for structurally valid profiles without
+  reading credential values.
+
+## Stage 4: Add Application Fallback Profile Asset Plumbing
+
+### Goal
+
+Allow module families to register application-owned fallback profile YAML
+without placing domain policy in generic LLM code.
+
+### Work
+
+- Extend `internal/framework/llm.AssetRegistry` with a separate fallback
+  profile source collection, registration method, flattened filesystem, and
+  safe content digest.
+- Reuse the existing asset-source path validation and flattening behavior where
+  appropriate. Reject invalid roots, unreadable assets, and duplicate flattened
+  paths. Do not parse PromptKit profile YAML in Notarius.
+- Add `promptkit.WithFallbackProfileFS` to production engine options only when
+  at least one fallback profile source is registered.
+- Supply the identical assembled fallback source to the profile-inspection
+  engine. Adjust CLI composition so pipeline-aware profile validation can use
+  the production LLM asset registry without exposing PromptKit types.
+- Extend profile-source checkpoint identity to include the exact fallback
+  profile asset digest in addition to the PromptKit catalog marker and operator
+  source. Keep the resulting fingerprint hash-only and path/content/credential
+  free.
+- Keep operator source precedence owned by PromptKit. Do not implement profile
+  merging or duplicate PromptKit source resolution in Notarius.
+- Update `docs/internal/llm.md` only for the new implemented generic asset and
+  fingerprint mechanics. No domain fallback exists until Stage 5.
+
+### Tests And Validation
+
+- Add focused AssetRegistry tests for successful flattening, invalid roots,
+  duplicate paths, and hash changes when fallback bytes change.
+- Add adapter-level tests showing that the fallback filesystem reaches both
+  execution construction and inspection construction.
+- Extend checkpoint tests to prove fallback content changes profile-source
+  identity without exposing raw YAML or paths. Use relational comparisons, not
+  a fixed hash literal.
+- Run `go test ./internal/framework/llm ./internal/cli` and `go test ./...`.
+- Run `git diff --check`.
+
+### Completion Criteria
+
+- Generic plumbing can carry application fallback profiles while remaining
+  unaware of D&D IDs or model settings.
+- Inspection, execution, and checkpoint identity use the same fallback asset
+  source.
+
+## Stage 5: Adopt The D&D `dnd-extraction` Fallback
+
+### Goal
+
+Give the D&D module family one stable embedded workload profile that operators
+can replace.
+
+### Work
+
+- Add a D&D-owned embedded PromptKit profile asset with ID `dnd-extraction`
+  under `internal/modules/dnd`. Use the exact baseline defined in
+  `promptkit.md`: OpenRouter, `openai/gpt-5.6-luna`, no explicit reasoning
+  effort, a 240-second timeout, flex service tier, and no selected temperature,
+  token limit, or `top_p`. The omitted reasoning value intentionally allows
+  OpenAI's backend to apply its `medium` default.
+- Register the profile filesystem from the D&D registrar through the generic
+  fallback profile asset boundary. Keep D&D policy out of
+  `internal/framework/llm` and the CLI composition root.
+- Change every maintained D&D LLM prompt definition—including scene chunking,
+  all D&D extractors, and NPC normalization—from the model-named default to
+  `default_profile: dnd-extraction`.
+- Add an integration-level profile-resolution test proving that:
+  - the fallback resolves when no operator source defines the ID;
+  - a valid operator profile with the same ID wins completely; and
+  - an invalid matching operator profile fails rather than falling through.
+- Test through Notarius's assembled production assets and PromptKit boundary;
+  do not duplicate every upstream source-precedence case.
+- Update the implemented profile ownership and prompt-default behavior in
+  `docs/internal/dnd.md`, `docs/internal/llm.md`, and
+  `docs/integrations/pkg-promptkit.md`. Defer the complete operator walkthrough
+  and examples to Stage 10.
+
+### Tests And Validation
+
+- Run focused D&D prompt preparation tests and the production composition
+  tests.
+- Run `go test ./internal/modules/dnd/... ./internal/framework/llm
+  ./internal/cli`.
+- Run `go test ./...`.
+- Verify `rg -n 'default_profile: gemini-2-flash' internal/modules/dnd`
+  returns no matches.
+- Run `git diff --check`.
+
+### Completion Criteria
+
+- All maintained D&D prompts use the application-owned logical profile ID.
+- The fallback works without an operator profile and remains authoritatively
+  overridable by a matching valid operator definition.
+
+## Stage 6: Introduce Module Execution-Class Metadata
+
+### Goal
+
+Make each production module's ability to use an LLM statically discoverable
+without yet changing profile inheritance.
+
+### Work
+
+- Add `ExecutionClass contracts.ExecutionClass` to `pipeline.ModuleSpec` and
+  preserve it through normalization, cloning, catalogs, registries, JSON/debug
+  views, and lookup helpers.
+- In this transitional stage only, allow an omitted execution class to
+  normalize to deterministic so existing test-only fixtures can be migrated in
+  Stage 7 without breaking the repository midway.
+- Explicitly classify every production module:
+  - D&D scene chunking, every D&D extractor, and D&D NPC normalization as
+    `llm_backed`;
+  - all other current production input, chunk, merge, normalize, and output
+    modules as `deterministic`.
+- Update production module specification tests and production catalog tests to
+  assert the semantic class alongside stage, artifact kind, and capabilities.
+- Add catalog lookup support needed by later resolution to retrieve a selected
+  module's execution class by stage and key without constructing it.
+- Do not implement pipeline-level profile inheritance or reject deterministic
+  profiles yet.
+- Update `docs/internal/modules.md` and `docs/internal/dnd.md` to identify
+  execution class as registered module metadata, while noting only implemented
+  uses.
+
+### Tests And Validation
+
+- Run module registration/spec tests across generic, Seriatim, and D&D
+  families.
+- Run `go test ./internal/framework/pipeline ./internal/modules/...`.
+- Run `go test ./...` and `git diff --check`.
+
+### Completion Criteria
+
+- Every production module has an explicit correct execution class.
+- Catalog consumers can retrieve that class without a concrete module
+  instance.
+- Test-only omitted classes remain the only temporary compatibility behavior.
+
+## Stage 7: Enforce Execution Metadata And Remove Runtime Probing
+
+### Goal
+
+Finish the execution-class contract so missing metadata cannot cause future
+profile drift.
+
+### Work
+
+- Update every framework, CLI, and integration test module specification to
+  declare an explicit execution class appropriate to the fake behavior.
+- Change module-spec validation so an empty or unsupported execution class is a
+  registration error. Remove the transitional deterministic default from
+  Stage 6.
+- Replace the chunk runner's special `ChunkExecutionClassProvider` probe with
+  specification-derived behavior. Remove the now-redundant provider interface,
+  implementation methods, and tests when they have no remaining consumer.
+- Ensure chunk producer provenance remains unchanged: it records a non-empty
+  effective binding profile for an LLM-backed chunker, while a deterministic
+  chunker records no profile. A profile selected only through the prompt
+  default remains represented by PromptKit's actual-profile manifest rather
+  than being invented as an explicit chunk binding.
+- Review helper constructors and fixtures for opportunities to set execution
+  class once without obscuring the class under test. Do not introduce an
+  elaborate test-spec framework.
+- Update internal documentation if the removal changes any described runtime
+  mechanics.
+
+### Tests And Validation
+
+- Add or retain focused registration tests for missing and invalid execution
+  classes.
+- Retain chunk-plan provenance tests for LLM-backed and deterministic
+  chunkers.
+- Run `go test ./internal/framework/pipeline ./internal/modules/...`.
+- Run `go test ./...`, `go vet ./...`, and `git diff --check`.
+
+### Completion Criteria
+
+- No registered module specification relies on an implicit execution class.
+- Pipeline metadata, not a concrete runtime type assertion, owns module
+  execution classification.
+
+## Stage 8: Resolve Programmatic Pipeline Profile Defaults
+
+### Goal
+
+Implement profile inheritance and precedence inside the pipeline resolver
+before exposing the field through YAML configuration.
+
+### Work
+
+- Add an optional trimmed `LLMProfile` field to
+  `pipeline.PipelineProfile`. Add a non-empty runtime override field to
+  `pipeline.ResolveOptions` so all precedence decisions occur in the resolver
+  rather than through pre-resolution mutation.
+- After module selection, `--only` filtering, default validator-chain
+  selection, and validator compatibility resolution, apply effective profiles
+  to every selected input, chunk, extract, merge, normalize, output, and
+  validator binding according to the precedence in `promptkit.md`.
+- Apply profiles only when the selected module or validator execution class is
+  `llm_backed`.
+- Reject a binding-specific `llm_profile` on any deterministic module or
+  validator. Do not reject or inspect an unused pipeline default when no
+  selected LLM-backed binding consumes it.
+- Leave an LLM-backed binding empty when no CLI, binding, or pipeline profile is
+  selected so PromptKit can use the prompt's `default_profile`.
+- Store the effective values on resolved bindings before digest construction.
+  Do not add a second inheritance decision to execution.
+- Ensure semantically equivalent repeated binding profiles and one inherited
+  default produce the same resolved pipeline digest. Ensure any changed
+  effective profile changes the digest.
+- Do not modify file configuration or CLI parsing in this stage.
+
+### Tests And Validation
+
+- Add pipeline package tests for the complete precedence matrix:
+  runtime override; binding-specific exception; pipeline default; prompt
+  fallback; and deterministic bindings.
+- Cover default and explicitly configured validator chains, all relevant stage
+  categories, `--only` lane selection, unused defaults, deterministic-profile
+  rejection, and semantic digest equivalence.
+- Prefer table-driven package-level tests over assertions on private traversal
+  helpers.
+- Run `go test ./internal/framework/pipeline` and `go test ./...`.
+- Run `git diff --check`.
+
+### Completion Criteria
+
+- Programmatic pipelines resolve one canonical effective profile policy.
+- Only LLM-backed resolved bindings can contain a profile.
+- Runtime override, binding, pipeline, and prompt precedence is unambiguous and
+  digest-stable.
+
+## Stage 9: Expose Pipeline Defaults Through Configuration And CLI
+
+### Goal
+
+Make the profile-default workflow available to operators while preserving
+validation and override behavior.
+
+### Work
+
+- Add optional `pipelines.<id>.llm_profile` support to the version 4 file
+  configuration model. Use presence-aware decoding so an explicitly set blank
+  value is rejected, while omission remains valid.
+- Preserve the field through file application, configuration cloning,
+  effective configuration, and programmatic profile copies without aliasing or
+  trimming drift.
+- Remove `applyLLMProfileOverride`. Pass the CLI override through the resolver's
+  runtime-override input so deterministic bindings are never populated.
+- Update effective profile-ID collection to cover every selected LLM-backed
+  module stage and LLM-backed validator, including future LLM-backed input and
+  output modules. Do not inspect deterministic or unselected profiles.
+- Ensure `run`, `config validate --pipeline`, resume/checkpoint identity, and
+  relevant dry preflight paths all use the same resolved effective profiles.
+- Preserve `--llm-profile` as the highest-precedence non-empty run-wide
+  override and preserve binding-specific profiles as exceptions when no CLI
+  override is present.
+- Do not increment the configuration version.
+- Update current configuration and CLI contracts in `docs/config.md` and
+  `docs/cli.md` in the same stage. Link to operations for the deployment
+  workflow rather than duplicating it prematurely.
+
+### Tests And Validation
+
+- Add file-config tests for omission, trimming, explicit blank rejection,
+  unknown-key behavior, cloning, and round-trip application.
+- Add effective-config and CLI contract tests for precedence, LLM-only
+  application, inherited-profile inspection failure before factory execution,
+  `--only`, and digest changes.
+- Retain offline operation and do not require credentials for
+  `config validate --pipeline`.
+- Run `go test ./internal/core/config ./internal/framework/pipeline
+  ./internal/cli`.
+- Run `go test ./...`, `go vet ./...`, and `git diff --check`.
+
+### Completion Criteria
+
+- Operators can select `dnd-extraction` once per pipeline.
+- Configuration and CLI paths share the resolver's precedence policy.
+- Unknown effective profiles fail preflight, while deterministic and unused
+  profiles do not cause spurious inspection.
+
+## Stage 10: Complete Operator Documentation, Examples, And Decision Record
+
+### Goal
+
+Make the implemented workflow understandable, copyable, and maintainable
+without duplicating canonical facts.
+
+### Work
+
+- Create an ADR using the next sequential number for the durable decision to
+  use workload-oriented pipeline defaults with operator-overridable application
+  fallback profiles. Record context, decision, alternatives, and consequences;
+  do not turn the ADR into a field reference or implementation log.
+- Complete `docs/config.md` as the canonical owner of profile-source fields,
+  `pipelines.<id>.llm_profile`, validation, and precedence.
+- Complete `docs/operations.md` with an operator workflow that distinguishes
+  Notarius embedded prompts, Notarius fallback profiles, PromptKit built-ins,
+  and deployment filesystem profiles. Include production/development/local use
+  of the same `dnd-extraction` ID, credential handling, absolute-path guidance,
+  and the fact that current relative profile paths use the process working
+  directory rather than the configuration file's directory.
+- Complete `docs/integrations/pkg-promptkit.md` with the v0.5.0 boundary,
+  prepared execution, inspection, fallback and ordinary source precedence,
+  optional provider controls, capacity adaptation, and compatibility policy.
+- Update `docs/internal/configuration.md`, `docs/internal/pipeline.md`,
+  `docs/internal/cli.md`, `docs/internal/llm.md`, `docs/internal/modules.md`, and
+  `docs/internal/dnd.md` only for their owned implementation details. Link to
+  canonical configuration, operations, and upstream format contracts rather
+  than restating them.
+- Keep exactly the existing two D&D configuration examples. Add
+  `llm_profile: dnd-extraction` to the minimal and complete pipelines and remove
+  the now-redundant model-named binding override from the complete example.
+- Add one secret-free maintained operator profile at
+  `examples/profiles/dnd-extraction.yml`. It should be a complete valid profile
+  for the same logical ID and may mirror the embedded baseline; its purpose is
+  to demonstrate file ownership and format, not claim automatic environment
+  detection. Link it from the configuration and operations documentation.
+- If the complete example selects the external profile file, use a path that
+  is valid for the documented repository-root invocation and explicitly note
+  the working-directory rule. Keep the minimal example dependent only on the
+  embedded fallback.
+- Add or extend maintained-example validation so both configuration examples
+  and the profile YAML are checked without generation or credentials.
+- Remove the now-implemented `Pipeline-Level LLM Profile Defaults` section from
+  `docs/roadmap/future.md`. Preserve the unrelated deterministic session and
+  concurrency items.
+- Do not delete `promptkit.md` or this implementation plan during the feature
+  implementation; retire them only after post-implementation review.
+
+### Tests And Validation
+
+- Run maintained example/configuration tests and relevant CLI help/parser
+  tests.
+- Run `go test ./...`.
+- Run `rg -n 'gemini-2-flash' examples docs` and review every remaining match
+  for intentional model-policy or historical context.
+- Run `rg -n 'v0\.3\.0|profileCheckPrompt|applyLLMProfileOverride' .` and resolve
+  stale production or current-documentation matches.
+- Verify all new links and `git diff --check`.
+
+### Completion Criteria
+
+- Every current fact has one canonical documentation owner.
+- Operators can distinguish and deploy all profile layers without reading Go
+  source.
+- Both maintained configurations and the maintained external profile are valid,
+  secret-free, and tested offline.
+- Implemented profile work no longer remains in `future.md`.
+
+## Stage 11: Final Verification And Quality Review
+
+### Goal
+
+Verify the complete migration as one integrated change and correct only defects
+or omissions found during that review.
+
+### Work
+
+- Review the final diff against every acceptance criterion in `promptkit.md`.
+- Confirm provider-specific PromptKit types remain inside the LLM integration
+  boundary and D&D policy remains inside the D&D module family.
+- Confirm execution and inspection receive identical ordinary, fallback, and
+  backend configuration.
+- Confirm no paths, profile YAML, endpoints, credentials, or prepared handle
+  state leak into fingerprints or ordinary diagnostics.
+- Confirm all production module specs have explicit correct execution classes
+  and every resolved deterministic binding is profile-free.
+- Confirm prompt default, pipeline default, binding override, and CLI override
+  behavior through representative assembled configurations.
+- Review tests for redundancy and remove obsolete synthetic-prompt,
+  runtime-probe, exact-hash, or duplicated upstream-behavior tests superseded by
+  stronger contract tests.
+- Perform an optional manual D&D quality comparison if credentials and an
+  evaluation transcript are deliberately supplied. Record no private input or
+  credential material, and do not make this comparison a completion gate.
+
+### Validation Commands
+
+```sh
+gofmt -w <changed-go-files>
+go test ./...
+go test -race ./internal/framework/llm ./internal/core/config ./internal/framework/pipeline ./internal/cli
+go vet ./...
+go build ./cmd/notarius
+git diff --check
+```
+
+Also run focused stale-contract searches:
+
+```sh
+rg -n 'gitea.maximumdirect.net/eric/promptkit v0\.3\.0|PromptKit v0\.3\.0' .
+rg -n 'default_profile: gemini-2-flash|profileCheckPrompt|applyLLMProfileOverride' internal docs examples
+```
+
+Review any matches rather than deleting intentional historical references
+blindly.
+
+### Completion Criteria
+
+- All automated checks pass offline and without real credentials.
+- The implemented behavior matches `promptkit.md` with no known architecture,
+  provenance, checkpoint, profile-precedence, or documentation gap.
+- Any optional live evaluation is clearly separate from correctness testing.
+
+## Open Questions
+
+None. The roadmap decisions are sufficient to implement every stage without an
+additional product or architecture choice.

docs/roadmap/promptkit.md

@@ -0,0 +1,318 @@
+# PromptKit v0.5 Integration And LLM Profile Policy
+
+## Purpose
+
+This roadmap defines the target state for upgrading Notarius from PromptKit
+v0.3.0 to v0.5.0 and adopting the upstream runtime and profile facilities that
+directly improve Notarius. It also defines the application policy for stable,
+domain-oriented LLM profile names, operator overrides, pipeline inheritance,
+profile validation, provider defaults, checkpoint identity, and documentation.
+
+The ordered work needed to reach this state belongs in
+[the implementation plan](implementation.md). Current behavior remains defined
+by the canonical documentation outside `docs/roadmap/` until the corresponding
+work is implemented.
+
+## Background
+
+Notarius currently pins PromptKit v0.3.0. Its adapter prepares a request once
+for debug material and then independently runs the original request, causing
+PromptKit to prepare the same logical call a second time. The CLI validates an
+explicit profile by preparing a synthetic prompt. PromptKit profile selection
+can be repeated on individual module bindings or replaced for one invocation
+with `--llm-profile`, but a configured pipeline cannot yet declare one inherited
+profile policy.
+
+PromptKit v0.4.0 and v0.5.0 add the upstream boundaries needed to improve these
+areas:
+
+- [v0.4.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/releases/v0.4.0.md)
+  adds opaque prepared executions, exact profile and prompt inspection, and a
+  typed backend-capacity error;
+- [v0.5.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/releases/v0.5.0.md)
+  adds application fallback profile filesystems and stops sending unset
+  optional sampling controls as framework-selected provider values; and
+- the [v0.5.0 format contract](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.5.0/docs/formats.md)
+  defines the resulting profile-source and execution-setting precedence.
+
+A source-compatibility test of the current Notarius repository against
+PromptKit v0.5.0 completed successfully. The work is therefore primarily an
+intentional runtime and configuration migration rather than a repair for a
+breaking Go API change.
+
+## Goals
+
+- Pin and document PromptKit v0.5.0 as Notarius's supported upstream contract.
+- Execute the exact prepared request snapshot whose safe details are recorded
+  in Notarius debug material.
+- Validate configured PromptKit profiles through the upstream inspection API
+  without synthetic prompts, provider calls, or credential-value access.
+- Give Notarius an application-owned, operator-overridable
+  `dnd-extraction` profile fallback.
+- Let a pipeline choose one default LLM profile without repeating that ID on
+  every LLM-backed binding.
+- Apply profile inheritance and run-wide overrides only where the resolved
+  module or validator can use an LLM.
+- Preserve accurate checkpoint invalidation, effective profile provenance,
+  redaction, cancellation, concurrency, and provider-neutral module contracts.
+- Provide operators with one clear deployment pattern for production,
+  development, and local profile definitions.
+
+## Target End State
+
+### PromptKit Runtime Boundary
+
+Notarius depends on PromptKit v0.5.0 and uses its public APIs rather than
+reimplementing source or execution resolution.
+
+For each structured completion, the adapter:
+
+1. builds one PromptKit run request from the provider-neutral Notarius request;
+2. calls `PrepareExecution` once;
+3. immediately arranges an idempotent `Discard` for every unexecuted handle;
+4. obtains credential-redacted `Details` for debug and response metadata; and
+5. calls `RunPrepared` so generation uses that exact frozen snapshot.
+
+The debug prompt and successful result therefore describe the same selected
+profile, rendered messages, input bytes, session, output contract, and effective
+settings even when a filesystem-backed source changes concurrently. PromptKit
+handle types remain private to `internal/framework/llm`.
+
+PromptKit admission failures continue to match Notarius's provider-neutral
+`ErrLLMCapacityExceeded` contract. When PromptKit supplies a `CapacityError`,
+the adapter obtains the normalized backend ID through `errors.As` and may add it
+to safe application-owned diagnostics without parsing upstream error wording.
+The backend ID does not become a provider-specific module contract.
+
+### Optional Provider Controls
+
+Notarius accepts PromptKit v0.5.0's new behavior for `temperature`,
+`max_tokens`, and `top_p`: an unset setting is omitted from compatible provider
+requests and the provider chooses its own default. Notarius does not restore
+PromptKit's former implicit `top_p: 1` value globally.
+
+An operator who requires a particular value specifies it in the selected
+PromptKit profile. The application fallback described below intentionally
+leaves these controls unset. A human-reviewed D&D extraction comparison should
+be performed after the upgrade, but paid or nondeterministic model output is
+not part of the default automated test suite.
+
+### Profile Inspection
+
+Pipeline-aware configuration validation uses `Engine.InspectProfile` for every
+effective explicit profile ID. It verifies that the profile exists, parses and
+validates, resolves its backend and target, and is compatible with the engine's
+registered backends. It does not create a synthetic prompt, load prompt inputs,
+contact a provider, or require credential values to exist in the validation
+process environment.
+
+Credential availability is execution-time state. PromptKit preparation still
+enforces the selected profile's credential contract before generation. This
+keeps `notarius config validate` useful in build and deployment validation
+environments where secrets are deliberately absent.
+
+PromptKit construction for inspection and execution uses one shared internal
+profile-source and backend-option path. The CLI does not expose PromptKit public
+types across the Notarius LLM boundary merely to perform inspection.
+
+`InspectPrompt` is not adopted merely because it exists. It remains available
+for a later, separately defined module-to-prompt interface preflight if a
+concrete validation requirement justifies that additional contract.
+
+### Application And Operator Profile Sources
+
+Notarius embeds one ordinary PromptKit YAML profile with the stable ID
+`dnd-extraction`. It is an application fallback registered through
+`WithFallbackProfileFS`, is owned by the D&D module family, and initially
+preserves the current effective D&D baseline:
+
+- backend: PromptKit's built-in `openrouter` backend;
+- model: `openai/gpt-5.6-luna`;
+- reasoning effort: unset, allowing OpenAI's backend to apply its default of
+  `medium`;
+- generation timeout: 240 seconds;
+- service tier: `flex`; and
+- no application-selected `temperature`, `max_tokens`, or `top_p`.
+
+All maintained D&D LLM prompt definitions use `dnd-extraction` as their
+`default_profile`. The ID communicates workload intent rather than a provider,
+model, or environment. Changing the embedded fallback is an intentional
+Notarius execution-policy change and participates in checkpoint identity.
+
+Effective profile definitions resolve in PromptKit's order:
+
+1. programmatic in-memory profiles used by tests or explicit consumers;
+2. the operator source configured by `promptkit.profile_file` or
+   `promptkit.profile_dir`;
+3. the Notarius application fallback source; and
+4. PromptKit's embedded built-in catalog.
+
+Only an absent ID falls through to the next source. A matching profile is a
+complete definition: fields are not merged with a lower-precedence definition,
+and a malformed matching operator profile fails rather than silently selecting
+the application fallback.
+
+Production, development, and local deployments should normally provide
+different complete definitions for the same `dnd-extraction` ID. An operator
+source is optional because the application fallback keeps the maintained D&D
+workflow usable, but a deployment that needs an intentional model or backend
+policy should configure its own definition.
+
+### Domain Ownership And Asset Assembly
+
+The D&D fallback profile remains under `internal/modules/dnd` and is registered
+by the D&D registrar, consistent with ADR-0004. Generic LLM plumbing knows how
+to collect and flatten application fallback profile filesystems but contains no
+D&D model or policy knowledge.
+
+The shared asset registry detects invalid roots, unreadable sources, and
+duplicate flattened paths. PromptKit remains responsible for strict profile
+YAML parsing, duplicate profile-ID detection, source precedence, and effective
+target resolution. The same assembled fallback source is supplied to runtime
+execution and CLI profile inspection.
+
+### Explicit Module Execution Metadata
+
+Every registered input, chunk, extract, merge, normalize, and output module
+declares one required execution class: `deterministic` or `llm_backed`.
+Validator registrations continue to declare the same distinction through their
+validator specifications.
+
+The registered specification is authoritative for configuration resolution.
+Current production classifications are:
+
+- the D&D scene chunker, all D&D extractors, and the D&D NPC normalizer are
+  LLM-backed;
+- the Seriatim input adapter, generic chunker, all current mergers, all other
+  current normalizers, and the JSON output encoder are deterministic; and
+- current validators retain their declared classifications.
+
+Missing or unsupported execution metadata is a registration error. Explicitly
+assigning `llm_profile` to a deterministic module or validator is a pipeline
+resolution error. The framework does not infer execution class by inspecting
+domain package names or concrete implementation types at runtime.
+
+The module specification replaces the chunk runner's special runtime
+execution-class probe. Effective resolved bindings already express the result:
+only LLM-backed bindings may retain a non-empty profile.
+
+### Pipeline-Level Profile Default
+
+Configuration version 4 gains one optional non-empty pipeline field:
+
+```yaml
+pipelines:
+  dnd-session:
+    llm_profile: dnd-extraction
+```
+
+No configuration-version increment is required because the field is additive
+and existing files remain valid. An explicitly present blank value is invalid.
+
+For every selected LLM-backed module and validator, the effective profile uses
+this precedence:
+
+1. non-empty run-wide `--llm-profile` override;
+2. binding-specific `llm_profile`;
+3. pipeline-level `llm_profile`; and
+4. the prompt definition's `default_profile`, represented by an empty effective
+   Notarius binding profile.
+
+The run-wide override and inherited pipeline default never attach to a
+deterministic binding. Binding-specific exceptions remain available when one
+operation needs a different cost, latency, quality, backend, or reasoning
+policy.
+
+Inheritance is resolved after module and validator selection, including
+`--only` lane filtering, but before effective-pipeline validation, digest
+construction, explicit-profile inspection, checkpoint construction,
+preparation, execution, or provenance capture. Only profiles used by selected
+LLM-backed bindings are inspected. An unused pipeline default in a pipeline
+with no selected LLM-backed work does not require an otherwise unused profile
+to exist.
+
+The resolved pipeline contains effective binding profiles rather than a second
+runtime inheritance mechanism. Two pipelines that differ only by spelling the
+same effective policy once as a pipeline default and once on every LLM-backed
+binding have the same semantic resolved digest. Changing an effective profile
+changes the digest and applicable checkpoint identity.
+
+### Provenance And Checkpoints
+
+The PromptKit profile-source checkpoint fingerprint covers:
+
+- the PromptKit v0.5.0 built-in profile catalog identity;
+- exact application fallback profile asset content; and
+- exact configured operator profile YAML content, when present.
+
+The existing local-backend target fingerprint remains separate and continues
+to exclude scheduling-only concurrency limits. Fingerprints contain hashes and
+stable markers, not profile contents, filesystem paths, endpoints, credentials,
+or other secrets.
+
+Changing the PromptKit version, application fallback, operator profile, or
+effective pipeline profile makes incompatible LLM checkpoints ineligible for
+reuse. The dependency upgrade is expected to invalidate checkpoints produced
+under v0.3.0.
+
+Successful run manifests continue to record only profiles actually selected by
+PromptKit, including their effective model, backend, and reasoning metadata.
+Debug output reports the same effective execution snapshot used for generation.
+
+### Operator Documentation And Examples
+
+Canonical documentation clearly distinguishes:
+
+- Notarius prompt and schema assets embedded in the application;
+- Notarius application fallback profiles embedded in the application;
+- PromptKit's own embedded built-in profiles; and
+- operator profile files on the deployment filesystem.
+
+The configuration reference owns the pipeline field, profile-source fields,
+validation rules, and precedence. Operations owns deployment layout, working
+directory behavior, credentials, and environment-specific profile management.
+The PromptKit integration document owns the pinned upstream contract and
+source-precedence boundary. Internal documents describe asset registration,
+resolution, inspection, prepared execution, fingerprinting, and tests without
+duplicating user-facing field definitions.
+
+The maintained examples continue to include only the minimal and complete D&D
+configurations. They use the stable `dnd-extraction` policy, and one maintained
+PromptKit profile file under `examples/` demonstrates an operator override.
+Examples remain secret-free and are validated without live provider calls.
+
+## Out Of Scope
+
+- Implementing the separate deterministic prompt-session identity roadmap
+  item.
+- Changing the default `concurrency.total_llm` value; PromptKit's retained
+  OpenRouter capacity of 16 remains relevant to that separate item.
+- Adding model evaluation as a deterministic or CI correctness gate.
+- Automatically selecting production, development, or local environments.
+  Deployment configuration chooses the operator profile source.
+- Profile inheritance, partial profile merging, or cross-profile aliases.
+- Exposing PromptKit types to modules, validators, durable output contracts, or
+  public configuration structures.
+- Adopting `InspectPrompt` without a separately justified prompt-interface
+  validation contract.
+
+## Acceptance Criteria
+
+- Notarius builds and its offline test suite passes with PromptKit v0.5.0.
+- Every structured completion executes the exact snapshot used for safe debug
+  prompt details.
+- Profile preflight uses profile inspection and no synthetic prompt.
+- The embedded `dnd-extraction` fallback resolves without an operator source,
+  and a matching valid operator profile replaces it completely.
+- Every production module has explicit, correct execution metadata.
+- Pipeline, binding, CLI, and prompt-default precedence behaves as defined for
+  modules and validators, while deterministic bindings remain profile-free.
+- Effective profiles participate in pipeline digests, profile inspection,
+  checkpoint identity, debug records, and run provenance at the appropriate
+  boundaries.
+- The dependency and application fallback changes invalidate incompatible old
+  checkpoints without exposing profile or credential content.
+- Canonical documentation and maintained examples accurately describe and
+  exercise the implemented operator workflow.
+- Default tests remain deterministic, offline, credential-free, and focused on
+  Notarius-owned behavior rather than duplicating PromptKit's upstream suite.

examples/dnd-complete.config.yml

@@ -1,4 +1,6 @@
-version: 3
+version: 4
+promptkit:
+  profile_file: ./examples/profiles/dnd-extraction.yml
 concurrency:
   total_llm: 2
   stage_workers:
@@ -16,6 +18,7 @@ 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:
@@ -28,6 +31,15 @@ pipelines:
       module: json
       options:
         include_chunk_map: true
+        evidence_context:
+          enabled: true
+          window_units: 3
+          lanes:
+            - item-events
+            - npcs
+            - spells
+            - combat-turns
+            - npc-interactions
     steps:
       # Establish session-wide reference artifacts alongside independent item events.
       - id: describe-session
@@ -45,7 +57,6 @@ pipelines:
             merge: appendorder
             normalize:
               module: dnd/npcs
-              llm_profile: gemini-2-flash
               retries: 2
           scene-descriptions:
             extract:

examples/dnd-minimal.config.yml

@@ -1,6 +1,7 @@
-version: 3
+version: 4
 pipelines:
   dnd-session:
+    llm_profile: dnd-extraction
     input: seriatim
     artifacts:
       spells:

examples/profiles/dnd-extraction.yml

@@ -0,0 +1,5 @@
+id: dnd-extraction
+backend: openrouter
+model: openai/gpt-5.6-luna
+timeout_seconds: 240
+service_tier: flex

go.mod

@@ -3,7 +3,7 @@ module gitea.maximumdirect.net/eric/notarius
 go 1.25.5
 
 require (
-	gitea.maximumdirect.net/eric/scriptorium v0.11.1
+	gitea.maximumdirect.net/eric/promptkit v0.5.0
 	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2
 	gopkg.in/yaml.v3 v3.0.1
 )

go.sum

@@ -1,13 +1,9 @@
-gitea.maximumdirect.net/eric/scriptorium v0.11.0 h1:rjvbt9FTaWHxYlHq7QlUzmMVUt3QdbTmeCkmH81N//o=
-gitea.maximumdirect.net/eric/scriptorium v0.11.0/go.mod h1:FQ5lEuNxmrQyNgIomkpZdxvfTC0jWjbXYuq3tbJWF64=
-gitea.maximumdirect.net/eric/scriptorium v0.11.1 h1:zBKtB3+fP8FcHGI8DJD99CiTL6crAGitBhWtE+xYJHc=
-gitea.maximumdirect.net/eric/scriptorium v0.11.1/go.mod h1:FQ5lEuNxmrQyNgIomkpZdxvfTC0jWjbXYuq3tbJWF64=
+gitea.maximumdirect.net/eric/promptkit v0.5.0 h1:jnpazLyyNhWrB2xzwwtUkNUfktkTdkENTwuSPnKiYrc=
+gitea.maximumdirect.net/eric/promptkit v0.5.0/go.mod h1:R95NM6fbMDGDC0/UomgnSBP6ui2ns+8SZb8bESNvrDQ=
 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=
-golang.org/x/text v0.14.0 h1:ScX5w1eTa3QqT8oi6+ziP7dTV1S2+ALU0bI+0zXKWiQ=
-golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
 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

@@ -213,11 +213,12 @@ func assembledSpellPipeline(t *testing.T, options assembledSpellPipelineOptions)
 	components := productionTestComponents(t)
 	extractor := &assembledSpellExtractor{unknownSpell: options.unknownSpell}
 	if err := pipeline.RegisterExtractor[dnd.SpellList](components.registries.Extractors, pipeline.ModuleSpec{
-		Key:          assembledSpellExtractorKey,
-		Stage:        pipeline.StageExtract,
-		Requires:     []string{"chunks", "source.transcript"},
-		Provides:     []string{"dnd.spell_casts"},
-		ArtifactKind: dnd.SpellListKind,
+		Key:            assembledSpellExtractorKey,
+		Stage:          pipeline.StageExtract,
+		ExecutionClass: contracts.ExecutionClassDeterministic,
+		Requires:       []string{"chunks", "source.transcript"},
+		Provides:       []string{"dnd.spell_casts"},
+		ArtifactKind:   dnd.SpellListKind,
 	}, func() (contracts.Extractor[dnd.SpellList], error) {
 		return extractor, nil
 	}); err != nil {

internal/cli/catalog.go

@@ -21,15 +21,16 @@ type productionComponents struct {
 
 func newProductionComponents() (productionComponents, error) {
 	registries := pipeline.Registries{
-		Inputs:          pipeline.NewInputAdapterRegistry(),
-		Chunkers:        pipeline.NewChunkerRegistry(),
-		ArtifactCodecs:  pipeline.NewArtifactCodecRegistry(),
-		Extractors:      pipeline.NewExtractorRegistry(),
-		Mergers:         pipeline.NewMergerRegistry(),
-		Normalizers:     pipeline.NewNormalizerRegistry(),
-		Validators:      pipeline.NewValidatorRegistry(),
-		ValidatorChains: pipeline.NewValidatorChainRegistry(),
-		Outputs:         pipeline.NewOutputEncoderRegistry(),
+		Inputs:           pipeline.NewInputAdapterRegistry(),
+		Chunkers:         pipeline.NewChunkerRegistry(),
+		ArtifactCodecs:   pipeline.NewArtifactCodecRegistry(),
+		ArtifactEvidence: pipeline.NewArtifactEvidenceRegistry(),
+		Extractors:       pipeline.NewExtractorRegistry(),
+		Mergers:          pipeline.NewMergerRegistry(),
+		Normalizers:      pipeline.NewNormalizerRegistry(),
+		Validators:       pipeline.NewValidatorRegistry(),
+		ValidatorChains:  pipeline.NewValidatorChainRegistry(),
+		Outputs:          pipeline.NewOutputEncoderRegistry(),
 	}
 	assets := llm.NewAssetRegistry()
 	registrars := []struct {
@@ -88,29 +89,31 @@ func effectiveRegistries(opts Options) (pipeline.Registries, error) {
 
 func catalogFromRegistries(registries pipeline.Registries) pipeline.ModuleCatalog {
 	return pipeline.ModuleCatalog{
-		Inputs:          registries.Inputs,
-		Chunkers:        registries.Chunkers,
-		ArtifactCodecs:  registries.ArtifactCodecs,
-		Extractors:      registries.Extractors,
-		Mergers:         registries.Mergers,
-		Normalizers:     registries.Normalizers,
-		Validators:      registries.Validators,
-		ValidatorChains: registries.ValidatorChains,
-		Outputs:         registries.Outputs,
+		Inputs:           registries.Inputs,
+		Chunkers:         registries.Chunkers,
+		ArtifactCodecs:   registries.ArtifactCodecs,
+		ArtifactEvidence: registries.ArtifactEvidence,
+		Extractors:       registries.Extractors,
+		Mergers:          registries.Mergers,
+		Normalizers:      registries.Normalizers,
+		Validators:       registries.Validators,
+		ValidatorChains:  registries.ValidatorChains,
+		Outputs:          registries.Outputs,
 	}
 }
 
 func registriesFromCatalog(catalog pipeline.ModuleCatalog) pipeline.Registries {
 	return pipeline.Registries{
-		Inputs:          catalog.Inputs,
-		Chunkers:        catalog.Chunkers,
-		ArtifactCodecs:  catalog.ArtifactCodecs,
-		Extractors:      catalog.Extractors,
-		Mergers:         catalog.Mergers,
-		Normalizers:     catalog.Normalizers,
-		Validators:      catalog.Validators,
-		ValidatorChains: catalog.ValidatorChains,
-		Outputs:         catalog.Outputs,
+		Inputs:           catalog.Inputs,
+		Chunkers:         catalog.Chunkers,
+		ArtifactCodecs:   catalog.ArtifactCodecs,
+		ArtifactEvidence: catalog.ArtifactEvidence,
+		Extractors:       catalog.Extractors,
+		Mergers:          catalog.Mergers,
+		Normalizers:      catalog.Normalizers,
+		Validators:       catalog.Validators,
+		ValidatorChains:  catalog.ValidatorChains,
+		Outputs:          catalog.Outputs,
 	}
 }
 
@@ -118,6 +121,7 @@ func isEmptyCatalog(catalog pipeline.ModuleCatalog) bool {
 	return catalog.Inputs == nil &&
 		catalog.Chunkers == nil &&
 		catalog.ArtifactCodecs == nil &&
+		catalog.ArtifactEvidence == nil &&
 		catalog.Extractors == nil &&
 		catalog.Mergers == nil &&
 		catalog.Normalizers == nil &&
@@ -130,6 +134,7 @@ func isEmptyRegistries(registries pipeline.Registries) bool {
 	return registries.Inputs == nil &&
 		registries.Chunkers == nil &&
 		registries.ArtifactCodecs == nil &&
+		registries.ArtifactEvidence == nil &&
 		registries.Extractors == nil &&
 		registries.Mergers == nil &&
 		registries.Normalizers == nil &&
@@ -138,24 +143,13 @@ func isEmptyRegistries(registries pipeline.Registries) bool {
 		registries.Outputs == nil
 }
 
-func productionLLMClientFactory(ctx context.Context, cfg config.Config, profileID string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
-	if err := ctx.Err(); err != nil {
-		return nil, nil, err
-	}
-	assets, err := productionPromptAssets()
-	if err != nil {
-		return nil, nil, err
-	}
-	return buildProductionLLMClient(ctx, cfg, profileID, assets)
-}
-
 func productionLLMClientFactoryWithAssets(assets *llm.AssetRegistry) LLMClientFactory {
-	return func(ctx context.Context, cfg config.Config, profileID string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
-		return buildProductionLLMClient(ctx, cfg, profileID, assets)
+	return func(ctx context.Context, cfg config.Config, profileID string, overrides LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		return buildProductionLLMClient(ctx, cfg, profileID, overrides, assets)
 	}
 }
 
-func buildProductionLLMClient(ctx context.Context, cfg config.Config, profileID string, assets *llm.AssetRegistry) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+func buildProductionLLMClient(ctx context.Context, cfg config.Config, profileID string, overrides LLMRuntimeOverrides, assets *llm.AssetRegistry) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 	if err := ctx.Err(); err != nil {
 		return nil, nil, err
 	}
@@ -163,14 +157,16 @@ func buildProductionLLMClient(ctx context.Context, cfg config.Config, profileID
 		return nil, nil, fmt.Errorf("production asset registry must not be nil")
 	}
 	recorder := llm.NewLLMProfileRecorder()
-	client, err := llm.NewScriptoriumClient(llm.ScriptoriumClientConfig{
-		ProfileDir:  cfg.Scriptorium.ProfileDir,
-		ProfileFile: cfg.Scriptorium.ProfileFile,
-		Assets:      assets,
-		Recorder:    recorder,
+	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,
 	})
 	if err != nil {
-		return nil, nil, fmt.Errorf("create Scriptorium-backed LLM client: %w", err)
+		return nil, nil, fmt.Errorf("create PromptKit-backed LLM client: %w", err)
 	}
 	scheduler, err := llm.NewScheduler(cfg.Concurrency.TotalLLM)
 	if err != nil {

internal/cli/command_contract_test.go

@@ -154,6 +154,23 @@ func TestConfigValidateResolvesPipelineAndChecksSelection(t *testing.T) {
 	}
 }
 
+func TestConfigValidatePipelineDefaultProfileIsOffline(t *testing.T) {
+	configPath := writeCommandConfigContent(t, `version: 4
+pipelines:
+  demo:
+    llm_profile: dnd-extraction
+    input: seriatim
+    artifacts:
+      spells:
+        extract: dnd/spells
+`)
+	var stdout, stderr bytes.Buffer
+	code := RunWithOptions([]string{"config", "validate", "--config", configPath, "--pipeline", "demo"}, &stdout, &stderr, Options{})
+	if code != 0 || !strings.Contains(stdout.String(), "valid for pipeline \"demo\"") || stderr.Len() != 0 {
+		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+	}
+}
+
 func TestPipelinesListSortsNormalizedIDsInTextAndJSON(t *testing.T) {
 	configPath := writeCommandConfig(t, " zeta ", "alpha")
 	options := commandContractOptions(t)
@@ -214,13 +231,13 @@ func commandContractOptionsWithLookup(t *testing.T, lookup func(string) (string,
 
 func writeCommandConfig(t *testing.T, firstID, secondID string) string {
 	t.Helper()
-	content := fmt.Sprintf("version: 3\npipelines:\n  %q:\n    input: seriatim\n  %q:\n    input: seriatim\n", firstID, secondID)
+	content := fmt.Sprintf("version: 4\npipelines:\n  %q:\n    input: seriatim\n  %q:\n    input: seriatim\n", firstID, secondID)
 	return writeCommandConfigContent(t, content)
 }
 
 func writeResolvableCommandConfig(t *testing.T) string {
 	t.Helper()
-	return writeCommandConfigContent(t, `version: 3
+	return writeCommandConfigContent(t, `version: 4
 pipelines:
   demo:
     input: seriatim

internal/cli/dnd_scene_descriptions_contract_test.go

@@ -6,8 +6,10 @@ import (
 	"fmt"
 	"reflect"
 	"strings"
+	"sync"
 	"testing"
 
+	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
@@ -28,7 +30,7 @@ func TestProductionSceneDescriptionWorkflow(t *testing.T) {
 		Output: pipeline.Binding("json"),
 		Artifacts: map[string]pipeline.ArtifactLaneProfile{
 			"scene-descriptions": {
-				Extract:   pipeline.Binding(sceneextract.Key),
+				Extract:   pipeline.ModuleBinding{Module: sceneextract.Key, LLMProfile: "scene-description-profile"},
 				Normalize: pipeline.Binding(scenenormalize.Key),
 			},
 		},
@@ -45,7 +47,8 @@ func TestProductionSceneDescriptionWorkflow(t *testing.T) {
 		t.Fatalf("resolved references = %#v / %#v, want no generated or required references", lane.ExtractReferences, lane.NormalizeReferences)
 	}
 
-	prepared, err := pipeline.Prepare(effective.ResolvedPipeline, components.registries, pipeline.ModuleDependencies{LLM: sceneDescriptionLLM{}})
+	llmClient := &sceneDescriptionLLM{}
+	prepared, err := pipeline.Prepare(effective.ResolvedPipeline, components.registries, pipeline.ModuleDependencies{LLM: llmClient})
 	if err != nil {
 		t.Fatalf("Prepare() error = %v", err)
 	}
@@ -60,6 +63,14 @@ func TestProductionSceneDescriptionWorkflow(t *testing.T) {
 	if output.Manifest.ValidationStatus != "approved" || len(output.Rejected) != 0 || len(output.NormalizeOutputs) != 1 {
 		t.Fatalf("run output = %#v, want one approved normalized artifact", output)
 	}
+	wantProfiles := []artifacts.LLMProfileManifest{{
+		ID:       "scene-description-profile",
+		Provider: "promptkit",
+		Model:    "deterministic",
+	}}
+	if !reflect.DeepEqual(output.Manifest.LLMProfiles, wantProfiles) {
+		t.Fatalf("manifest LLM profiles = %#v, want %#v", output.Manifest.LLMProfiles, wantProfiles)
+	}
 	normalizedOutput := output.NormalizeOutputs[0]
 	if normalizedOutput.NormalizerKey != scenenormalize.Key || normalizedOutput.Artifact.Kind != dnd.SceneDescriptionListKind || normalizedOutput.Artifact.Schema.ID != scenecodec.SchemaID || normalizedOutput.Artifact.Schema.Name != scenecodec.SchemaName || normalizedOutput.Artifact.Schema.Version != scenecodec.SchemaVersion {
 		t.Fatalf("normalized output = %#v, want registered durable scene-description schema", normalizedOutput)
@@ -85,9 +96,12 @@ func TestProductionSceneDescriptionWorkflow(t *testing.T) {
 	}
 }
 
-type sceneDescriptionLLM struct{}
+type sceneDescriptionLLM struct {
+	mu      sync.Mutex
+	profile *artifacts.LLMProfileManifest
+}
 
-func (sceneDescriptionLLM) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
+func (client *sceneDescriptionLLM) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
 	if err := ctx.Err(); err != nil {
 		return contracts.StructuredCompletionResponse{}, err
 	}
@@ -107,5 +121,27 @@ func (sceneDescriptionLLM) CompleteStructured(ctx context.Context, req contracts
 	if err := json.Unmarshal([]byte(content), out); err != nil {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("populate structured response: %w", err)
 	}
-	return contracts.StructuredCompletionResponse{Content: []byte(content), Provider: "test", Model: "deterministic", ProfileID: req.ProfileID}, nil
+	profile := artifacts.LLMProfileManifest{
+		ID:       req.ProfileID,
+		Provider: "promptkit",
+		Model:    "deterministic",
+	}
+	client.mu.Lock()
+	client.profile = &profile
+	client.mu.Unlock()
+	return contracts.StructuredCompletionResponse{
+		Content:   []byte(content),
+		Provider:  profile.Provider,
+		Model:     profile.Model,
+		ProfileID: profile.ID,
+	}, nil
+}
+
+func (client *sceneDescriptionLLM) LLMProfileManifests() []artifacts.LLMProfileManifest {
+	client.mu.Lock()
+	defer client.mu.Unlock()
+	if client.profile == nil {
+		return nil
+	}
+	return []artifacts.LLMProfileManifest{*client.profile}
 }

internal/cli/example_contract_test.go

@@ -94,6 +94,37 @@ func TestMaintainedConfigurationExampleSet(t *testing.T) {
 	if got := strings.Join(names, ","); got != "dnd-complete.config.yml,dnd-minimal.config.yml" {
 		t.Fatalf("maintained configuration examples = %q, want only the minimal and complete D&D examples", got)
 	}
+
+	profileEntries, err := os.ReadDir(repositoryPath("examples", "profiles"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	names = names[:0]
+	for _, entry := range profileEntries {
+		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)
+	}
+}
+
+func TestMaintainedExamplesValidateEffectiveProfilesOffline(t *testing.T) {
+	t.Chdir(repositoryPath())
+	t.Setenv("OPENROUTER_API_KEY", "")
+	for _, example := range maintainedExampleFiles(t) {
+		t.Run(example.name, func(t *testing.T) {
+			var stdout, stderr strings.Builder
+			code := RunWithOptions([]string{
+				"config", "validate", "--config", example.path, "--pipeline", "dnd-session",
+			}, &stdout, &stderr, Options{})
+			if code != 0 || stderr.Len() != 0 || !strings.Contains(stdout.String(), `valid for pipeline "dnd-session"`) {
+				t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+			}
+		})
+	}
 }
 
 func exampleStepLaneIDs(resolved pipeline.ResolvedPipeline) []string {

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: 3
+	content := fmt.Sprintf(`version: 4
 cache:
   chunk_plans:
     mode: bypass
@@ -51,7 +51,7 @@ pipelines:
 	options := Options{
 		Catalog:    catalogFromRegistries(components.registries),
 		Registries: components.registries,
-		LLMClientFactory: func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		LLMClientFactory: func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 			llmConstructed = true
 			return nil, nil, errors.New("LLM client must not be constructed")
 		},

internal/cli/production_contract_test.go

@@ -6,6 +6,8 @@ import (
 	"errors"
 	"fmt"
 	"io/fs"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"path/filepath"
 	"reflect"
@@ -13,13 +15,16 @@ import (
 	"sort"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"testing"
+	"testing/fstest"
 	"time"
 
 	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/chunkmap"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/llm"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 	"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd"
 	"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/chunk/scenes"
@@ -33,6 +38,7 @@ import (
 	itemeventnormalize "gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/normalize/itemevents"
 	spellnormalize "gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/normalize/spells"
 	"gitea.maximumdirect.net/eric/notarius/internal/modules/generic/normalize/noop"
+	"gitea.maximumdirect.net/eric/promptkit"
 )
 
 func TestProductionCatalogCoversMaintainedConfigurations(t *testing.T) {
@@ -160,9 +166,38 @@ func TestProductionCatalogCoversMaintainedConfigurations(t *testing.T) {
 	assertProductionContains(t, "production prompt assets", assetNames, requiredAssets)
 
 	catalog := catalogFromRegistries(registries)
+	for _, test := range []struct {
+		stage pipeline.ModuleStage
+		key   string
+		want  contracts.ExecutionClass
+	}{
+		{stage: pipeline.StageInput, key: "seriatim", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageChunk, key: "generic", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageChunk, key: "dnd/scenes", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/spells", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/npcs", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/combat-turns", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/item-events", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/npc-interactions", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageExtract, key: "dnd/scene-descriptions", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageMerge, key: "appendorder", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "noop", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "dnd/spells", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "dnd/npcs", want: contracts.ExecutionClassLLMBacked},
+		{stage: pipeline.StageNormalize, key: "dnd/combat-turns", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "dnd/item-events", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "dnd/npc-interactions", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageNormalize, key: "dnd/scene-descriptions", want: contracts.ExecutionClassDeterministic},
+		{stage: pipeline.StageOutput, key: "json", want: contracts.ExecutionClassDeterministic},
+	} {
+		got, ok := catalog.ExecutionClass(test.stage, test.key)
+		if !ok || got != test.want {
+			t.Fatalf("production execution class for %s/%s = %q, %t; want %q, true", test.stage, test.key, got, ok, test.want)
+		}
+	}
 	converted := registriesFromCatalog(catalog)
-	if converted.ArtifactCodecs != registries.ArtifactCodecs || converted.ValidatorChains != registries.ValidatorChains {
-		t.Fatal("catalog/registry conversion did not preserve codec and validator-chain registries")
+	if converted.ArtifactCodecs != registries.ArtifactCodecs || converted.ArtifactEvidence != registries.ArtifactEvidence || converted.ValidatorChains != registries.ValidatorChains {
+		t.Fatal("catalog/registry conversion did not preserve artifact and validator registries")
 	}
 	codecSpec, ok := catalog.ArtifactCodecs.Spec(dnd.SpellListKind)
 	if !ok || codecSpec.Kind != dnd.SpellListKind || codecSpec.Schema.ID != spellcodec.SchemaID {
@@ -195,6 +230,76 @@ func TestDefaultCLICompositionValidatesRepresentativeConfiguration(t *testing.T)
 	}
 }
 
+func TestProductionAssetsResolveDNDExtractionProfile(t *testing.T) {
+	components := productionTestComponents(t)
+	newEngine := func(profileFile string) (*promptkit.Engine, error) {
+		t.Helper()
+		options, err := components.assets.PromptKitOptions()
+		if err != nil {
+			return nil, err
+		}
+		if profileFile != "" {
+			options = append(options, promptkit.WithProfileFile(profileFile))
+		}
+		return promptkit.NewEngine(promptkit.Config{}, options...)
+	}
+
+	t.Run("fallback", func(t *testing.T) {
+		engine, err := newEngine("")
+		if err != nil {
+			t.Fatal(err)
+		}
+		inspection, err := engine.InspectProfile(context.Background(), "dnd-extraction")
+		if err != nil {
+			t.Fatalf("InspectProfile() error = %v, want fallback profile", err)
+		}
+		params := inspection.EffectiveModelParams
+		if params.BackendID != "openrouter" || params.Model != "openai/gpt-5.6-luna" || params.TimeoutSeconds != 240 || params.ServiceTier != "flex" {
+			t.Fatalf("fallback profile parameters = %#v", params)
+		}
+		if params.ReasoningEffort != "" || params.Temperature != 0 || params.MaxTokens != 0 || params.TopP != 0 {
+			t.Fatalf("fallback profile selected optional provider controls: %#v", params)
+		}
+	})
+
+	t.Run("valid operator profile wins", func(t *testing.T) {
+		profilePath := filepath.Join(t.TempDir(), "profiles.yaml")
+		if err := os.WriteFile(profilePath, []byte(`id: dnd-extraction
+endpoint: http://operator.example.test/v1
+model: operator-model
+timeout_seconds: 75
+`), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		engine, err := newEngine(profilePath)
+		if err != nil {
+			t.Fatal(err)
+		}
+		inspection, err := engine.InspectProfile(context.Background(), "dnd-extraction")
+		if err != nil {
+			t.Fatalf("InspectProfile() error = %v, want operator profile", err)
+		}
+		params := inspection.EffectiveModelParams
+		if params.BackendID != "" || params.Endpoint != "http://operator.example.test/v1" || params.Model != "operator-model" || params.TimeoutSeconds != 75 || params.ServiceTier != "" {
+			t.Fatalf("operator profile parameters = %#v, want complete replacement", params)
+		}
+	})
+
+	t.Run("invalid operator profile does not fall through", func(t *testing.T) {
+		profilePath := filepath.Join(t.TempDir(), "profiles.yaml")
+		if err := os.WriteFile(profilePath, []byte("id: dnd-extraction\nendpoint: http://operator.example.test/v1\nmodel: operator-model\nunknown: value\n"), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		engine, err := newEngine(profilePath)
+		if err == nil {
+			_, err = engine.InspectProfile(context.Background(), "dnd-extraction")
+		}
+		if err == nil {
+			t.Fatal("operator profile error = nil, want failure instead of fallback")
+		}
+	})
+}
+
 func TestProductionPromptAssetsPrepareWithoutProviderCredentials(t *testing.T) {
 	components := productionTestComponents(t)
 	cfg := config.Default()
@@ -321,7 +426,7 @@ func TestProductionSpellNormalizerRejectsInvalidCatalogReferencesBeforeExecution
 		options := Options{
 			Catalog:    catalogFromRegistries(components.registries),
 			Registries: components.registries,
-			LLMClientFactory: func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+			LLMClientFactory: func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 				llmConstructed = true
 				return nil, nil, errors.New("LLM client must not be constructed")
 			},
@@ -369,31 +474,150 @@ func setNormalizeSpellCatalogSource(t *testing.T, resolved *pipeline.ResolvedPip
 	resolved.Steps[0].ArtifactLanes[0].NormalizeReferences.Bindings = bindings
 }
 
-func TestProductionLLMClientFactoriesBuildOfflineRuntime(t *testing.T) {
+func TestProductionLLMClientFactoryBuildsOfflineRuntime(t *testing.T) {
 	components := productionTestComponents(t)
-	factories := []struct {
-		name    string
-		factory LLMClientFactory
-	}{
-		{name: "default production assets", factory: productionLLMClientFactory},
-		{name: "provided production assets", factory: productionLLMClientFactoryWithAssets(components.assets)},
+	client, manifests, err := productionLLMClientFactoryWithAssets(components.assets)(context.Background(), config.Default(), "test-profile", LLMRuntimeOverrides{})
+	if err != nil {
+		t.Fatalf("build production LLM runtime: %v", err)
 	}
-	for _, tt := range factories {
-		t.Run(tt.name, func(t *testing.T) {
-			client, manifests, err := tt.factory(context.Background(), config.Default(), "test-profile")
-			if err != nil {
-				t.Fatalf("build production LLM runtime: %v", err)
-			}
-			if client == nil {
-				t.Fatal("production LLM runtime returned a nil client")
-			}
-			if len(manifests) != 0 {
-				t.Fatalf("eager profile manifests = %#v, want none", manifests)
-			}
-			if _, ok := client.(contracts.LLMProfileManifestProvider); !ok {
-				t.Fatalf("production LLM client %T does not provide profile manifests", client)
-			}
-		})
+	if client == nil {
+		t.Fatal("production LLM runtime returned a nil client")
+	}
+	if len(manifests) != 0 {
+		t.Fatalf("eager profile manifests = %#v, want none", manifests)
+	}
+	fingerprintProvider, ok := client.(llm.CheckpointFingerprintProvider)
+	if !ok {
+		t.Fatalf("production LLM client %T does not provide one profile-source checkpoint fingerprint", client)
+	}
+	fingerprints, err := fingerprintProvider.LLMCheckpointFingerprints()
+	if err != nil || len(fingerprints) != 1 {
+		t.Fatalf("production LLM checkpoint fingerprints = %#v, error = %v, want one profile-source identity", fingerprints, err)
+	}
+	if _, ok := client.(contracts.LLMProfileManifestProvider); !ok {
+		t.Fatalf("production LLM client %T does not provide profile manifests", client)
+	}
+}
+
+func TestNormalizeOptionsSharesProductionProfileAssetsWithDefaultRuntime(t *testing.T) {
+	opts, err := normalizeOptions(Options{
+		Catalog: pipeline.ModuleCatalog{Inputs: pipeline.NewInputAdapterRegistry()},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if opts.promptKitAssets == nil || opts.LLMClientFactory == nil {
+		t.Fatalf("normalized options = %#v, want shared profile assets and default runtime factory", opts)
+	}
+	if err := validateExplicitPromptKitProfiles(context.Background(), config.Default(), []string{"dnd-extraction"}, opts.promptKitAssets); err != nil {
+		t.Fatalf("inspect application fallback profile: %v", err)
+	}
+
+	client, _, err := opts.LLMClientFactory(context.Background(), config.Default(), "dnd-extraction", LLMRuntimeOverrides{})
+	if err != nil {
+		t.Fatalf("build default runtime: %v", err)
+	}
+	fingerprintProvider, ok := client.(llm.CheckpointFingerprintProvider)
+	if !ok {
+		t.Fatalf("default runtime client %T does not provide checkpoint fingerprints", client)
+	}
+	runtimeFingerprints, err := fingerprintProvider.LLMCheckpointFingerprints()
+	if err != nil {
+		t.Fatal(err)
+	}
+	directClient, err := llm.NewPromptKitClient(llm.PromptKitClientConfig{Assets: opts.promptKitAssets})
+	if err != nil {
+		t.Fatal(err)
+	}
+	inspectionFingerprints, err := directClient.LLMCheckpointFingerprints()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !reflect.DeepEqual(runtimeFingerprints, inspectionFingerprints) {
+		t.Fatalf("runtime profile fingerprints = %#v, inspection profile fingerprints = %#v", runtimeFingerprints, inspectionFingerprints)
+	}
+}
+
+func TestProductionLLMClientFactoryUsesConfiguredLocalBackend(t *testing.T) {
+	var providerCalls atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		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")
+		_, _ = w.Write([]byte(`{
+			"choices": [{"message": {"role": "assistant", "content": "{\"ok\":true}"}}],
+			"usage": {"prompt_tokens": 3, "completion_tokens": 4, "total_tokens": 7}
+		}`))
+	}))
+	defer server.Close()
+
+	profilePath := filepath.Join(t.TempDir(), "profiles.yml")
+	if err := os.WriteFile(profilePath, []byte(`id: local-profile
+backend: local
+model: local-model
+`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	assets := llm.NewAssetRegistry()
+	if err := assets.RegisterPromptFS(fstest.MapFS{
+		"production.local.yaml": {Data: []byte(`id: production.local
+version: "v1"
+inputs:
+  - name: transcript
+    required: true
+messages:
+  - role: user
+    content: '{{ input "transcript" }}'
+output:
+  format: json
+  validation_mode: json
+`)},
+	}, "."); err != nil {
+		t.Fatalf("register prompt assets: %v", err)
+	}
+
+	cfg := config.Default()
+	cfg.PromptKit.ProfileFile = profilePath
+	cfg.PromptKit.LocalBackend = &config.PromptKitLocalBackendConfig{
+		Endpoint:         server.URL + "/v1",
+		ConcurrencyLimit: 2,
+	}
+	client, manifests, err := productionLLMClientFactoryWithAssets(assets)(
+		context.Background(),
+		cfg,
+		"local-profile",
+		LLMRuntimeOverrides{},
+	)
+	if err != nil {
+		t.Fatalf("build production LLM runtime: %v", err)
+	}
+	if len(manifests) != 0 {
+		t.Fatalf("eager profile manifests = %#v, want none", manifests)
+	}
+
+	var out map[string]any
+	_, err = client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
+		PromptID:  "production.local",
+		ProfileID: "local-profile",
+		Inputs: contracts.LLMInputSet{
+			"transcript": contracts.NewLLMInputMaterial("transcript", "text/plain", []byte("local request"), "", ""),
+		},
+	}, &out)
+	if err != nil {
+		t.Fatalf("CompleteStructured() error = %v, want nil", err)
+	}
+	if providerCalls.Load() != 1 {
+		t.Fatalf("provider calls = %d, want 1", providerCalls.Load())
+	}
+	provider, ok := client.(contracts.LLMProfileManifestProvider)
+	if !ok {
+		t.Fatalf("production client %T does not provide profile manifests", client)
+	}
+	recorded := provider.LLMProfileManifests()
+	if len(recorded) != 1 || recorded[0].BackendID != promptkit.BackendLocal {
+		t.Fatalf("production profile manifests = %#v, want local backend", recorded)
 	}
 }
 
@@ -401,14 +625,15 @@ func TestProductionLLMClientFactoriesRejectInvalidConstruction(t *testing.T) {
 	t.Run("canceled context", func(t *testing.T) {
 		ctx, cancel := context.WithCancel(context.Background())
 		cancel()
-		client, manifests, err := productionLLMClientFactory(ctx, config.Default(), "test-profile")
+		components := productionTestComponents(t)
+		client, manifests, err := productionLLMClientFactoryWithAssets(components.assets)(ctx, config.Default(), "test-profile", LLMRuntimeOverrides{})
 		if !errors.Is(err, context.Canceled) || client != nil || len(manifests) != 0 {
 			t.Fatalf("client=%T manifests=%#v error=%v, want canceled construction", client, manifests, err)
 		}
 	})
 
 	t.Run("nil assets", func(t *testing.T) {
-		client, manifests, err := productionLLMClientFactoryWithAssets(nil)(context.Background(), config.Default(), "test-profile")
+		client, manifests, err := productionLLMClientFactoryWithAssets(nil)(context.Background(), config.Default(), "test-profile", LLMRuntimeOverrides{})
 		if err == nil || !strings.Contains(err.Error(), "asset registry must not be nil") || client != nil || len(manifests) != 0 {
 			t.Fatalf("client=%T manifests=%#v error=%v, want nil-assets failure", client, manifests, err)
 		}
@@ -418,7 +643,7 @@ func TestProductionLLMClientFactoriesRejectInvalidConstruction(t *testing.T) {
 		components := productionTestComponents(t)
 		cfg := config.Default()
 		cfg.Concurrency.TotalLLM = 0
-		client, manifests, err := productionLLMClientFactoryWithAssets(components.assets)(context.Background(), cfg, "test-profile")
+		client, manifests, err := productionLLMClientFactoryWithAssets(components.assets)(context.Background(), cfg, "test-profile", LLMRuntimeOverrides{})
 		if err == nil || !strings.Contains(err.Error(), "create LLM scheduler") || !strings.Contains(err.Error(), "greater than zero") || client != nil || len(manifests) != 0 {
 			t.Fatalf("client=%T manifests=%#v error=%v, want scheduler-construction failure", client, manifests, err)
 		}
@@ -583,7 +808,7 @@ func maintainedExampleFiles(t *testing.T) []maintainedExample {
 
 func productionSpellCatalogContractConfig(t *testing.T) string {
 	t.Helper()
-	return fmt.Sprintf(`version: 3
+	return fmt.Sprintf(`version: 4
 cache:
   chunk_plans:
     mode: bypass
@@ -648,9 +873,10 @@ func productionCLIOptions(t *testing.T) Options {
 
 func productionOptionsFromComponents(components productionComponents) Options {
 	return Options{
-		Catalog:    catalogFromRegistries(components.registries),
-		Registries: components.registries,
-		LookupEnv:  emptyLookup,
+		Catalog:         catalogFromRegistries(components.registries),
+		Registries:      components.registries,
+		LookupEnv:       emptyLookup,
+		promptKitAssets: components.assets,
 	}
 }
 
@@ -673,14 +899,14 @@ func productionRunOptions(t *testing.T, fake *productionFakeLLMClient) Options {
 	options.Now = func() time.Time { return time.Unix(1700000000, 0).UTC() }
 	options.RunIDGenerator = func(time.Time) (string, error) { return productionRunID, nil }
 	options.UserCacheDir = func() (string, error) { return "", errors.New("user cache must not be used") }
-	options.LLMClientFactory = func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+	options.LLMClientFactory = func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 		return fake, nil, nil
 	}
 	return options
 }
 
 func productionRunConfig(outputRoot, chunkModule string) string {
-	return fmt.Sprintf(`version: 3
+	return fmt.Sprintf(`version: 4
 output:
   directory: %q
 cache:

internal/cli/promptkit_profiles.go

@@ -0,0 +1,46 @@
+package cli
+
+import (
+	"context"
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/llm"
+)
+
+func validateExplicitPromptKitProfiles(ctx context.Context, cfg config.Config, profileIDs []string, assets *llm.AssetRegistry) error {
+	if len(profileIDs) == 0 {
+		return nil
+	}
+	inspector, err := llm.NewPromptKitProfileInspector(llm.PromptKitProfileInspectorConfig{
+		Source: promptKitProfileSourceConfig(cfg),
+		Assets: assets,
+	})
+	if err != nil {
+		return fmt.Errorf("load PromptKit profiles: %w", err)
+	}
+	for _, profileID := range profileIDs {
+		if _, err := inspector.InspectProfile(ctx, profileID); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func promptKitProfileSourceConfig(cfg config.Config) llm.PromptKitProfileSourceConfig {
+	return llm.PromptKitProfileSourceConfig{
+		ProfileDir:   cfg.PromptKit.ProfileDir,
+		ProfileFile:  cfg.PromptKit.ProfileFile,
+		LocalBackend: mapPromptKitLocalBackend(cfg.PromptKit.LocalBackend),
+	}
+}
+
+func mapPromptKitLocalBackend(cfg *config.PromptKitLocalBackendConfig) *llm.PromptKitLocalBackendConfig {
+	if cfg == nil {
+		return nil
+	}
+	return &llm.PromptKitLocalBackendConfig{
+		Endpoint:         cfg.Endpoint,
+		ConcurrencyLimit: cfg.ConcurrencyLimit,
+	}
+}

internal/cli/promptkit_profiles_test.go

@@ -0,0 +1,159 @@
+package cli
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/llm"
+)
+
+func TestExplicitPromptKitProfileValidationInspectsProfilesWithoutGeneration(t *testing.T) {
+	var providerCalls atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		providerCalls.Add(1)
+	}))
+	defer server.Close()
+
+	writeProfile := func(t *testing.T, name, content string) string {
+		t.Helper()
+		profilePath := filepath.Join(t.TempDir(), name+".yaml")
+		if err := os.WriteFile(profilePath, []byte(content), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		return profilePath
+	}
+	localProfile := "id: local-profile\nbackend: local\nmodel: local-model\n"
+	credentialProfile := `id: credential-profile
+endpoint: ` + server.URL + `/v1
+model: credential-model
+api_key_env: NOTARIUS_PROMPTKIT_PROFILE_INSPECTION_TEST_KEY
+`
+	t.Setenv("NOTARIUS_PROMPTKIT_PROFILE_INSPECTION_TEST_KEY", "")
+
+	tests := []struct {
+		name         string
+		profilePath  string
+		profileID    string
+		profileDir   bool
+		localBackend bool
+		canceled     bool
+		wantErr      []string
+		rejectErr    []string
+	}{
+		{
+			name:         "configured local backend",
+			profilePath:  writeProfile(t, "local-profile", localProfile),
+			profileID:    "local-profile",
+			profileDir:   true,
+			localBackend: true,
+		},
+		{
+			name:        "missing local backend registration",
+			profilePath: writeProfile(t, "local-profile", localProfile),
+			profileID:   "local-profile",
+			wantErr:     []string{`PromptKit profile "local-profile" is invalid or unreadable`},
+		},
+		{
+			name:         "absent profile",
+			profilePath:  writeProfile(t, "local-profile", localProfile),
+			profileID:    "absent-profile",
+			localBackend: true,
+			wantErr:      []string{`PromptKit profile "absent-profile" is not configured`},
+		},
+		{
+			name:        "malformed profile",
+			profilePath: writeProfile(t, "malformed-profile", "id: malformed-profile\nbackend: [\n"),
+			profileID:   "malformed-profile",
+			wantErr:     []string{`PromptKit profile "malformed-profile" is invalid or unreadable`},
+			rejectErr:   []string{"malformed-profile.yaml", "backend: ["},
+		},
+		{
+			name:        "invalid profile source",
+			profilePath: filepath.Join(t.TempDir(), "missing-profile.yaml"),
+			profileID:   "missing-profile",
+			wantErr:     []string{"load PromptKit profiles", "profile configuration is invalid or unreadable"},
+		},
+		{
+			name:        "credential environment intentionally unset",
+			profilePath: writeProfile(t, "credential-profile", credentialProfile),
+			profileID:   "credential-profile",
+		},
+		{
+			name:         "canceled inspection",
+			profilePath:  writeProfile(t, "local-profile", localProfile),
+			profileID:    "local-profile",
+			localBackend: true,
+			canceled:     true,
+			wantErr:      []string{"context canceled"},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := config.Default()
+			if tt.profileDir {
+				cfg.PromptKit.ProfileDir = filepath.Dir(tt.profilePath)
+			} else {
+				cfg.PromptKit.ProfileFile = tt.profilePath
+			}
+			if tt.localBackend {
+				cfg.PromptKit.LocalBackend = &config.PromptKitLocalBackendConfig{
+					Endpoint:         server.URL + "/v1",
+					ConcurrencyLimit: 2,
+				}
+			}
+			ctx := context.Background()
+			if tt.canceled {
+				var cancel context.CancelFunc
+				ctx, cancel = context.WithCancel(ctx)
+				cancel()
+			}
+			err := validateExplicitPromptKitProfiles(ctx, cfg, []string{tt.profileID}, nil)
+			if len(tt.wantErr) == 0 {
+				if err != nil {
+					t.Fatalf("validateExplicitPromptKitProfiles() error = %v, want nil", err)
+				}
+				return
+			}
+			if err == nil {
+				t.Fatal("validateExplicitPromptKitProfiles() error = nil, want failure")
+			}
+			if tt.canceled && !errors.Is(err, context.Canceled) {
+				t.Fatalf("canceled inspection error = %v, want context canceled", err)
+			}
+			for _, want := range tt.wantErr {
+				if !strings.Contains(err.Error(), want) {
+					t.Fatalf("validation error = %q, want %q", err, want)
+				}
+			}
+			for _, rejected := range append(tt.rejectErr, tt.profilePath) {
+				if rejected != "" && strings.Contains(err.Error(), rejected) {
+					t.Fatalf("validation error = %q, must not expose %q", err, rejected)
+				}
+			}
+		})
+	}
+	if providerCalls.Load() != 0 {
+		t.Fatalf("provider calls during profile inspection = %d, want 0", providerCalls.Load())
+	}
+}
+
+func TestExplicitPromptKitProfileValidationUsesFallbackAssets(t *testing.T) {
+	assets := llm.NewAssetRegistry()
+	if err := assets.RegisterFallbackProfileFS(fstest.MapFS{
+		"profiles/fallback.yaml": {Data: []byte("id: fallback-profile\nendpoint: http://promptkit.test/v1\nmodel: fallback-model\n")},
+	}, "profiles"); err != nil {
+		t.Fatalf("RegisterFallbackProfileFS() error = %v, want nil", err)
+	}
+	if err := validateExplicitPromptKitProfiles(context.Background(), config.Default(), []string{"fallback-profile"}, assets); err != nil {
+		t.Fatalf("validateExplicitPromptKitProfiles() error = %v, want nil", err)
+	}
+}

internal/cli/recompute_execution_contract_test.go

@@ -116,7 +116,7 @@ func (h *recomputeTestHarness) options() Options {
 	for _, key := range []string{"test/extract/producer", "test/extract/unrelated", "test/extract/middle", "test/extract/dependent"} {
 		moduleKey := key
 		spec := pipeline.ModuleSpec{
-			Key: moduleKey, Stage: pipeline.StageExtract, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind,
+			Key: moduleKey, Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind,
 			ReferenceSlots: []contracts.ReferenceSlot{{Name: "upstream", AcceptedMediaTypes: []string{"application/json"}, AcceptedArtifactKinds: []contracts.ArtifactKind{stateTestArtifactKind}}},
 		}
 		if err := pipeline.RegisterExtractor(opts.Registries.Extractors, spec, func() (contracts.Extractor[stateTestArtifact], error) {
@@ -125,7 +125,7 @@ func (h *recomputeTestHarness) options() Options {
 			panic(err)
 		}
 	}
-	if err := opts.Registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "test/recompute-output", Stage: pipeline.StageOutput, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) {
+	if err := opts.Registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "test/recompute-output", Stage: pipeline.StageOutput, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) {
 		return recomputeTestOutput{}, nil
 	}); err != nil {
 		panic(err)
@@ -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: 3
+	config := fmt.Sprintf(`version: 4
 output:
   directory: %q
 cache:

internal/cli/reference_contract_test.go

@@ -217,7 +217,7 @@ func TestReferenceMaterializationSeparatesCLIAndConfigPathOrigins(t *testing.T)
 	workingDir := t.TempDir()
 	cfg := referenceContractConfig()
 	configPath := filepath.Join(configDir, "config.yml")
-	if err := os.WriteFile(configPath, []byte("version: 3\n"), 0o600); err != nil {
+	if err := os.WriteFile(configPath, []byte("version: 4\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
 	if err := os.WriteFile(filepath.Join(configDir, "required.txt"), []byte("config reference"), 0o600); err != nil {
@@ -364,21 +364,21 @@ func referenceContractCatalog(t *testing.T, includeBetaMerger, includeBetaNormal
 			t.Fatal(err)
 		}
 	}
-	register(registries.Inputs.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "reference/input", Stage: pipeline.StageInput, Provides: []string{"source"}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.InputAdapter, error) { return stateTestInput{}, nil }))
-	register(registries.Chunkers.RegisterWithSpec(pipeline.ModuleSpec{Key: "reference/chunk", Stage: pipeline.StageChunk, 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(registries.Inputs.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "reference/input", Stage: pipeline.StageInput, ExecutionClass: contracts.ExecutionClassDeterministic, Provides: []string{"source"}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.InputAdapter, error) { return stateTestInput{}, nil }))
+	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, 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-beta", Stage: pipeline.StageExtract, 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, 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 }))
+	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-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 {
-		register(pipeline.RegisterMerger(registries.Mergers, pipeline.ModuleSpec{Key: "reference/shared-merge", Stage: pipeline.StageMerge, Requires: []string{"artifact"}, Provides: []string{"merged"}, ArtifactKind: referenceContractKindBeta, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "beta-merge"}, {Name: "required-merge", Required: true}}}, func() (contracts.Merger[stateTestArtifact], error) { return stateTestMerger{}, 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: referenceContractKindBeta, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "beta-merge"}, {Name: "required-merge", Required: true}}}, func() (contracts.Merger[stateTestArtifact], error) { return stateTestMerger{}, nil }))
 	}
-	register(pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "reference/shared-normalize", Stage: pipeline.StageNormalize, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: referenceContractKindAlpha, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "alpha-normalize"}, {Name: "required-normalize", Required: true}}}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{}, nil }))
+	register(pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "reference/shared-normalize", Stage: pipeline.StageNormalize, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: referenceContractKindAlpha, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "alpha-normalize"}, {Name: "required-normalize", Required: true}}}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{}, nil }))
 	if includeBetaNormalizer {
-		register(pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "reference/shared-normalize", Stage: pipeline.StageNormalize, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: referenceContractKindBeta, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "beta-normalize"}, {Name: "required-normalize", Required: true}}}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{}, nil }))
+		register(pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "reference/shared-normalize", Stage: pipeline.StageNormalize, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: referenceContractKindBeta, ReferenceSlots: []contracts.ReferenceSlot{{Name: "shared"}, {Name: "beta-normalize"}, {Name: "required-normalize", Required: true}}}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{}, nil }))
 	}
-	register(registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "reference/output", Stage: pipeline.StageOutput, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) { return stateTestOutput{}, nil }))
+	register(registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "reference/output", Stage: pipeline.StageOutput, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) { return stateTestOutput{}, nil }))
 	return catalogFromRegistries(registries)
 }
 

internal/cli/run.go

@@ -5,6 +5,7 @@ import (
 	"crypto/sha256"
 	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"flag"
 	"fmt"
 	"io"
@@ -22,13 +23,14 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/chunkplan"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	frameworkdebug "gitea.maximumdirect.net/eric/notarius/internal/framework/debug"
+	frameworkllm "gitea.maximumdirect.net/eric/notarius/internal/framework/llm"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
 const defaultConfigPath = "/usr/local/etc/notarius/config.yml"
 const usage = `Usage:
   notarius help
-  notarius run <pipeline-id> --input path/to/source.json [--config path/to/config.yml] [--output-dir path] [--chunk_cache auto|bypass|refresh] [--resume] [--recompute-step step-id] [--debug [--debug-dir path]] [--only lane-a,lane-b] [--session-id id] [--reference selector=path] [--without-reference selector]
+  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]
 `
@@ -44,9 +46,14 @@ type Options struct {
 	ChunkPlanStoreFactory pipeline.ChunkPlanStoreFactory
 	DebugRecorderFactory  func(string) (pipeline.DebugRecorder, error)
 	DebugTerminalFactory  func(*debugbundle.SummaryWriter) DebugTerminalWriter
+	promptKitAssets       *frameworkllm.AssetRegistry
 }
 
-type LLMClientFactory func(ctx context.Context, cfg config.Config, profileID string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error)
+type LLMRuntimeOverrides struct {
+	ReasoningEffort *string
+}
+
+type LLMClientFactory func(ctx context.Context, cfg config.Config, profileID string, overrides LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error)
 
 // Run executes the command-line interface and returns a process exit code.
 func Run(args []string, stdout, stderr io.Writer) int {
@@ -115,12 +122,17 @@ func normalizeOptions(opts Options) (Options, error) {
 		}
 		opts.Registries = components.registries
 		opts.Catalog = catalogFromRegistries(components.registries)
-		if opts.LLMClientFactory == nil {
-			opts.LLMClientFactory = productionLLMClientFactoryWithAssets(components.assets)
-		}
+		opts.promptKitAssets = components.assets
 	}
 	if opts.LLMClientFactory == nil {
-		opts.LLMClientFactory = productionLLMClientFactory
+		if opts.promptKitAssets == nil {
+			assets, err := productionPromptAssets()
+			if err != nil {
+				return Options{}, err
+			}
+			opts.promptKitAssets = assets
+		}
+		opts.LLMClientFactory = productionLLMClientFactoryWithAssets(opts.promptKitAssets)
 	}
 	return opts, nil
 }
@@ -132,16 +144,20 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	inputPath := fs.String("input", "", "source input file path")
 	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
 	outputDir := fs.String("output-dir", "", "output directory")
+	machineOutput := fs.Bool("json", false, "write the successful run result as JSON")
 	debug := fs.Bool("debug", false, "write a debug bundle")
 	debugDir := fs.String("debug-dir", "", "debug bundle directory")
 	llmProfile := fs.String("llm-profile", "", "LLM profile override")
+	reasoningEffort := singleValueFlag{name: "--reasoning-effort"}
+	clearReasoningEffort := fs.Bool("clear-reasoning-effort", false, "clear the LLM profile reasoning effort")
 	resume := fs.Bool("resume", false, "reuse compatible recorded checkpoints")
-	recomputeStep := singleValueFlag{}
+	recomputeStep := singleValueFlag{name: "--recompute-step"}
 	chunkCache := chunkCacheFlag{}
 	sessionID := sessionIDFlag{}
 	referenceFlags := stringListFlag{}
 	withoutReferenceFlags := stringListFlag{}
 	fs.Var(&sessionID, "session-id", "prompt session identifier")
+	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(&withoutReferenceFlags, "without-reference", "unbind a reference, using the same selector forms as --reference")
@@ -187,6 +203,22 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		fmt.Fprintln(stderr, "notarius: --session-id must not be empty")
 		return 2
 	}
+	if reasoningEffort.set && *clearReasoningEffort {
+		fmt.Fprintln(stderr, "notarius: --reasoning-effort cannot be combined with --clear-reasoning-effort")
+		return 2
+	}
+	if reasoningEffort.set && strings.TrimSpace(reasoningEffort.value) == "" {
+		fmt.Fprintln(stderr, "notarius: --reasoning-effort must not be empty")
+		return 2
+	}
+	runtimeOverrides := LLMRuntimeOverrides{}
+	if reasoningEffort.set {
+		value := strings.TrimSpace(reasoningEffort.value)
+		runtimeOverrides.ReasoningEffort = &value
+	} else if *clearReasoningEffort {
+		value := ""
+		runtimeOverrides.ReasoningEffort = &value
+	}
 	only, err := parseOnly(*onlyRaw)
 	if err != nil {
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
@@ -277,17 +309,18 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		debugRecorder = pipeline.SynchronizedDebugRecorder(debugRecorder)
 	}
 	invocation := debugbundle.Invocation{
-		Operation:          "run",
-		PipelineID:         pipelineID,
-		InputPath:          strings.TrimSpace(*inputPath),
-		ConfigPath:         loadedConfigPath,
-		ConfigSource:       configSource(*configPath),
-		OnlyLanes:          append([]string(nil), only...),
-		ChunkCacheOverride: chunkCache.explicitValue(),
-		Resume:             *resume,
-		RecomputeStep:      strings.TrimSpace(recomputeStep.value),
-		RunID:              runID,
-		StartedAt:          startedAt,
+		Operation:               "run",
+		PipelineID:              pipelineID,
+		InputPath:               strings.TrimSpace(*inputPath),
+		ConfigPath:              loadedConfigPath,
+		ConfigSource:            configSource(*configPath),
+		OnlyLanes:               append([]string(nil), only...),
+		ChunkCacheOverride:      chunkCache.explicitValue(),
+		ReasoningEffortOverride: runtimeOverrides.ReasoningEffort,
+		Resume:                  *resume,
+		RecomputeStep:           strings.TrimSpace(recomputeStep.value),
+		RunID:                   runID,
+		StartedAt:               startedAt,
 	}
 	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))
@@ -313,7 +346,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		return failPipelineCommand(stderr, commandState, terminalWriter, err)
 	}
 	profileIDs := effectiveLLMProfileIDs(effective.ResolvedPipeline)
-	if err := validateExplicitScriptoriumProfiles(context.Background(), effective.Config, profileIDs); err != nil {
+	if err := validateExplicitPromptKitProfiles(context.Background(), effective.Config, profileIDs, opts.promptKitAssets); err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, err)
 	}
 	workingDir, err := os.Getwd()
@@ -361,10 +394,17 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	if len(profileIDs) == 1 {
 		factoryProfileID = profileIDs[0]
 	}
-	llmClient, llmProfiles, err := opts.LLMClientFactory(ctx, effective.Config, factoryProfileID)
+	llmClient, llmProfiles, err := opts.LLMClientFactory(ctx, effective.Config, factoryProfileID, runtimeOverrides)
 	if err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("create LLM client for profile %q: %w", factoryProfileID, err))
 	}
+	var llmFingerprints []checkpoint.Fingerprint
+	if effective.Config.Cache.Checkpoints.Enabled {
+		llmFingerprints, err = llmCheckpointFingerprints(llmClient)
+		if err != nil {
+			return failPipelineCommand(stderr, commandState, terminalWriter, fmt.Errorf("prepare LLM checkpoint identity: %w", err))
+		}
+	}
 	llmClient = pipeline.WithDebugLLMRecording(llmClient, debugRecorder)
 	prepared, err := pipeline.Prepare(effective.ResolvedPipeline, registries, pipeline.ModuleDependencies{LLM: llmClient})
 	if err != nil {
@@ -378,7 +418,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	if err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, err)
 	}
-	checkpointRecorder, checkpointLoader, err := checkpointHandlersForRun(effective.Config.Cache.Checkpoints, opts, effective.ResolvedPipeline, prepared.CheckpointFingerprints(), rawInput, only, llmProfiles, strings.TrimSpace(*llmProfile), strings.TrimSpace(sessionID.value), *resume)
+	checkpointRecorder, checkpointLoader, err := checkpointHandlersForRun(effective.Config.Cache.Checkpoints, opts, effective.ResolvedPipeline, prepared.CheckpointFingerprints(), llmFingerprints, rawInput, only, llmProfiles, strings.TrimSpace(*llmProfile), strings.TrimSpace(sessionID.value), runtimeOverrides, *resume)
 	if err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, err)
 	}
@@ -415,6 +455,17 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	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)
+		if err != nil {
+			return failPipelineCommand(stderr, commandState, terminalWriter, err)
+		}
+		encodedResult, err = encodeRunResult(result)
+		if err != nil {
+			return failPipelineCommand(stderr, commandState, terminalWriter, err)
+		}
+	}
 	if err := writeOutputFiles(runOutputDir, output.OutputFiles); err != nil {
 		return failPipelineCommand(stderr, commandState, terminalWriter, err)
 	}
@@ -422,9 +473,15 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		return writePipelineCommandFailure(stderr, commandState, primaryErr, persistenceErr)
 	}
 
-	fmt.Fprintf(stdout, "pipeline %q complete: outputs=%d rejected=%d output=%s\n", effective.PipelineID, len(output.NormalizeOutputs), len(output.Rejected), runOutputDir)
-	if debugPath != "" {
-		fmt.Fprintf(stdout, "debug=%s\n", debugPath)
+	if *machineOutput {
+		if err := writeRunResult(stdout, encodedResult); err != nil {
+			return writePipelineCommandFailure(stderr, commandState, errors.New("write run result"), nil)
+		}
+	} else {
+		fmt.Fprintf(stdout, "pipeline %q complete: outputs=%d rejected=%d output=%s\n", effective.PipelineID, len(output.NormalizeOutputs), len(output.Rejected), runOutputDir)
+		if debugPath != "" {
+			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))
@@ -462,11 +519,13 @@ func checkpointHandlersForRun(
 	opts Options,
 	resolved pipeline.ResolvedPipeline,
 	componentFingerprints []pipeline.CheckpointFingerprint,
+	llmFingerprints []checkpoint.Fingerprint,
 	rawInput []byte,
 	only []string,
 	llmProfiles []artifacts.LLMProfileManifest,
 	llmProfileOverride string,
 	sessionID string,
+	runtimeOverrides LLMRuntimeOverrides,
 	resume bool,
 ) (pipeline.CheckpointRecorder, pipeline.CheckpointLoader, error) {
 	if !settings.Enabled {
@@ -476,13 +535,17 @@ func checkpointHandlersForRun(
 		return pipeline.NoopCheckpointRecorder(), pipeline.NoopCheckpointLoader(), nil
 	}
 	identity, err := checkpoint.NewIdentity(checkpoint.IdentityInput{
-		Pipeline:               resolved,
-		InputKey:               resolved.Input.Module,
-		RawInputDigest:         rawInputDigest(rawInput),
-		SelectedLanes:          only,
-		RuntimeOverrides:       runtimeOverrideFingerprints(llmProfileOverride, sessionID),
-		References:             pipeline.ReferenceProvenance(resolved),
-		ProvenanceFingerprints: append(llmProfileFingerprints(llmProfiles), checkpointIdentityFingerprints(componentFingerprints)...),
+		Pipeline:         resolved,
+		InputKey:         resolved.Input.Module,
+		RawInputDigest:   rawInputDigest(rawInput),
+		SelectedLanes:    only,
+		RuntimeOverrides: runtimeOverrideFingerprints(llmProfileOverride, sessionID, runtimeOverrides),
+		References:       pipeline.ReferenceProvenance(resolved),
+		ProvenanceFingerprints: combineCheckpointFingerprints(
+			llmProfileFingerprints(llmProfiles),
+			llmFingerprints,
+			checkpointIdentityFingerprints(componentFingerprints),
+		),
 	})
 	if err != nil {
 		return nil, nil, fmt.Errorf("create checkpoint identity: %w", err)
@@ -508,6 +571,30 @@ func checkpointHandlersForRun(
 	return recorder, loader, nil
 }
 
+func llmCheckpointFingerprints(client contracts.StructuredLLMClient) ([]checkpoint.Fingerprint, error) {
+	provider, ok := client.(frameworkllm.CheckpointFingerprintProvider)
+	if !ok {
+		return nil, nil
+	}
+	values, err := provider.LLMCheckpointFingerprints()
+	if err != nil {
+		return nil, err
+	}
+	out := make([]checkpoint.Fingerprint, 0, len(values))
+	for _, value := range values {
+		out = append(out, checkpoint.Fingerprint{Name: value.Name, Value: value.Value})
+	}
+	return out, nil
+}
+
+func combineCheckpointFingerprints(sources ...[]checkpoint.Fingerprint) []checkpoint.Fingerprint {
+	var out []checkpoint.Fingerprint
+	for _, source := range sources {
+		out = append(out, source...)
+	}
+	return out
+}
+
 func recomputePolicy(resolved pipeline.ResolvedPipeline, requestedStep string) (pipeline.CheckpointExecutionPolicy, error) {
 	requestedStep = strings.TrimSpace(requestedStep)
 	if requestedStep == "" {
@@ -613,7 +700,7 @@ func rawInputDigest(data []byte) string {
 	return "sha256:" + hex.EncodeToString(sum[:])
 }
 
-func runtimeOverrideFingerprints(llmProfileOverride string, sessionID string) []checkpoint.Fingerprint {
+func runtimeOverrideFingerprints(llmProfileOverride string, sessionID string, runtimeOverrides LLMRuntimeOverrides) []checkpoint.Fingerprint {
 	var values []checkpoint.Fingerprint
 	if strings.TrimSpace(llmProfileOverride) != "" {
 		values = append(values, checkpoint.Fingerprint{Name: "llm_profile_override", Value: strings.TrimSpace(llmProfileOverride)})
@@ -621,6 +708,13 @@ func runtimeOverrideFingerprints(llmProfileOverride string, sessionID string) []
 	if strings.TrimSpace(sessionID) != "" {
 		values = append(values, checkpoint.Fingerprint{Name: "session_id", Value: strings.TrimSpace(sessionID)})
 	}
+	if runtimeOverrides.ReasoningEffort != nil {
+		value := strings.TrimSpace(*runtimeOverrides.ReasoningEffort)
+		if value == "" {
+			value = "<cleared>"
+		}
+		values = append(values, checkpoint.Fingerprint{Name: "reasoning_effort_override", Value: value})
+	}
 	return values
 }
 
@@ -777,7 +871,7 @@ func reorderRunArgs(args []string) []string {
 
 func runFlagTakesValue(arg string) bool {
 	switch arg {
-	case "--config", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
+	case "--config", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--reasoning-effort", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
 		return true
 	default:
 		return false
@@ -837,11 +931,11 @@ func chunkPlanStoreForRun(cfg config.ChunkPlanCacheConfig, opts Options) (pipeli
 
 func validateRunFlagValues(args []string) error {
 	for i, arg := range args {
-		if arg != "--session-id" {
+		if arg != "--session-id" && arg != "--reasoning-effort" {
 			continue
 		}
 		if i+1 >= len(args) || strings.HasPrefix(args[i+1], "-") {
-			return fmt.Errorf("flag needs an argument: --session-id")
+			return fmt.Errorf("flag needs an argument: %s", arg)
 		}
 	}
 	return nil
@@ -864,11 +958,22 @@ func effectiveLLMProfileIDs(resolved pipeline.ResolvedPipeline) []string {
 			seen[id] = struct{}{}
 		}
 	}
-	add(resolved.Chunk)
+	if resolved.InputExecutionClass == contracts.ExecutionClassLLMBacked {
+		add(resolved.Input)
+	}
+	if resolved.ChunkExecutionClass == contracts.ExecutionClassLLMBacked {
+		add(resolved.Chunk)
+	}
 	for _, lane := range resolved.AllArtifactLanes() {
-		add(lane.Extract)
-		add(lane.Merge)
-		add(lane.Normalize)
+		if lane.ExtractExecutionClass == contracts.ExecutionClassLLMBacked {
+			add(lane.Extract)
+		}
+		if lane.MergeExecutionClass == contracts.ExecutionClassLLMBacked {
+			add(lane.Merge)
+		}
+		if lane.NormalizeExecutionClass == contracts.ExecutionClassLLMBacked {
+			add(lane.Normalize)
+		}
 	}
 	for _, chain := range resolved.ValidatorChains {
 		for _, validator := range chain.Validators {
@@ -877,6 +982,9 @@ func effectiveLLMProfileIDs(resolved pipeline.ResolvedPipeline) []string {
 			}
 		}
 	}
+	if resolved.OutputExecutionClass == contracts.ExecutionClassLLMBacked {
+		add(resolved.Output)
+	}
 	ids := make([]string, 0, len(seen))
 	for id := range seen {
 		ids = append(ids, id)
@@ -960,7 +1068,7 @@ func runConfigValidate(args []string, stdout, stderr io.Writer, opts Options) in
 			fmt.Fprintf(stderr, "notarius: %v\n", err)
 			return 1
 		}
-		if err := validateExplicitScriptoriumProfiles(context.Background(), effective.Config, effectiveLLMProfileIDs(effective.ResolvedPipeline)); err != nil {
+		if err := validateExplicitPromptKitProfiles(context.Background(), effective.Config, effectiveLLMProfileIDs(effective.ResolvedPipeline), opts.promptKitAssets); err != nil {
 			fmt.Fprintf(stderr, "notarius: %v\n", err)
 			return 1
 		}
@@ -1125,6 +1233,7 @@ type sessionIDFlag struct {
 }
 
 type singleValueFlag struct {
+	name  string
 	value string
 	set   bool
 }
@@ -1138,7 +1247,7 @@ func (flag *singleValueFlag) String() string {
 
 func (flag *singleValueFlag) Set(value string) error {
 	if flag.set {
-		return fmt.Errorf("--recompute-step may be specified only once")
+		return fmt.Errorf("%s may be specified only once", flag.name)
 	}
 	flag.value = value
 	flag.set = true

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/framework/checkpoint"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
@@ -241,12 +242,14 @@ func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 	t.Run("one effective profile reaches the factory and modules", func(t *testing.T) {
 		roots := newStateTestRoots(t)
 		profileDir := writeRunContractProfiles(t, "override-profile")
-		prependRunContractConfig(t, roots, fmt.Sprintf("scriptorium:\n  profile_dir: %q\n", profileDir))
+		prependRunContractConfig(t, roots, fmt.Sprintf("promptkit:\n  profile_dir: %q\n", profileDir))
 		harness := newStateTestHarness()
 		var factoryProfiles []string
 		opts := harness.options()
-		opts.LLMClientFactory = func(_ context.Context, _ config.Config, profileID string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		var factoryOverrides []LLMRuntimeOverrides
+		opts.LLMClientFactory = func(_ context.Context, _ config.Config, profileID string, overrides LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 			factoryProfiles = append(factoryProfiles, profileID)
+			factoryOverrides = append(factoryOverrides, overrides)
 			return nil, nil, nil
 		}
 		var stdout, stderr bytes.Buffer
@@ -257,6 +260,9 @@ func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 		if len(factoryProfiles) != 1 || factoryProfiles[0] != "override-profile" {
 			t.Fatalf("factory profiles = %#v, want one override profile", factoryProfiles)
 		}
+		if len(factoryOverrides) != 1 || factoryOverrides[0].ReasoningEffort != nil {
+			t.Fatalf("factory overrides = %#v, want inherited reasoning", factoryOverrides)
+		}
 		harness.mu.Lock()
 		profiles := append([]string(nil), harness.moduleProfiles...)
 		harness.mu.Unlock()
@@ -270,16 +276,16 @@ func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 		}
 	})
 
-	t.Run("validator profile remains distinct", func(t *testing.T) {
+	t.Run("runtime override applies to validators", func(t *testing.T) {
 		roots := newStateTestRoots(t)
 		profileDir := writeRunContractProfiles(t, "override-profile", "validator-profile")
-		prependRunContractConfig(t, roots, fmt.Sprintf("scriptorium:\n  profile_dir: %q\n", profileDir))
+		prependRunContractConfig(t, roots, fmt.Sprintf("promptkit:\n  profile_dir: %q\n", profileDir))
 		harness := newStateTestHarness()
 		var validatorProfiles []string
 		opts := harness.options()
 		registerRunContractValidator(t, &opts, &validatorProfiles)
 		factoryProfiles := []string{}
-		opts.LLMClientFactory = func(_ context.Context, _ config.Config, profileID string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		opts.LLMClientFactory = func(_ context.Context, _ config.Config, profileID string, _ LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 			factoryProfiles = append(factoryProfiles, profileID)
 			return nil, nil, nil
 		}
@@ -288,21 +294,21 @@ func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 		if code != 0 || stderr.Len() != 0 {
 			t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 		}
-		if len(factoryProfiles) != 1 || factoryProfiles[0] != "" {
-			t.Fatalf("factory profiles = %#v, want one call without a unique profile", factoryProfiles)
+		if len(factoryProfiles) != 1 || factoryProfiles[0] != "override-profile" {
+			t.Fatalf("factory profiles = %#v, want one override profile", factoryProfiles)
 		}
-		if len(validatorProfiles) != 1 || validatorProfiles[0] != "validator-profile" {
-			t.Fatalf("validator profiles = %#v, want configured validator profile", validatorProfiles)
+		if len(validatorProfiles) != 1 || validatorProfiles[0] != "override-profile" {
+			t.Fatalf("validator profiles = %#v, want runtime override", validatorProfiles)
 		}
 	})
 
 	t.Run("unknown profile is rejected without factory access", func(t *testing.T) {
 		roots := newStateTestRoots(t)
 		profileDir := writeRunContractProfiles(t, "override-profile")
-		prependRunContractConfig(t, roots, fmt.Sprintf("scriptorium:\n  profile_dir: %q\n", profileDir))
+		prependRunContractConfig(t, roots, fmt.Sprintf("promptkit:\n  profile_dir: %q\n", profileDir))
 		factoryCalls := 0
 		opts := newStateTestHarness().options()
-		opts.LLMClientFactory = func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		opts.LLMClientFactory = func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 			factoryCalls++
 			return nil, nil, nil
 		}
@@ -312,28 +318,175 @@ func TestRunLLMProfileOverrideAndValidationUseInjectedBoundaries(t *testing.T) {
 			t.Fatalf("code=%d stdout=%q stderr=%q factoryCalls=%d", code, stdout.String(), stderr.String(), factoryCalls)
 		}
 	})
+
+	t.Run("pipeline default is rejected before factory access", func(t *testing.T) {
+		roots := newStateTestRoots(t)
+		profileDir := writeRunContractProfiles(t, "configured-profile")
+		prependRunContractConfig(t, roots, fmt.Sprintf("promptkit:\n  profile_dir: %q\n", profileDir))
+		replaceStateTestConfigLine(t, roots.config, "  sample:\n", "  sample:\n    llm_profile: missing-profile\n")
+		factoryCalls := 0
+		opts := newStateTestHarness().options()
+		opts.LLMClientFactory = func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+			factoryCalls++
+			return nil, nil, nil
+		}
+		var stdout, stderr bytes.Buffer
+		code := RunWithOptions([]string{"run", "sample", "--config", roots.config, "--input", roots.input, "--chunk_cache", "bypass"}, &stdout, &stderr, opts)
+		if code != 1 || !strings.Contains(stderr.String(), "not configured") || factoryCalls != 0 || stdout.Len() != 0 {
+			t.Fatalf("code=%d stdout=%q stderr=%q factoryCalls=%d", code, stdout.String(), stderr.String(), factoryCalls)
+		}
+	})
+}
+
+func TestRunReasoningEffortOverrideReachesFactory(t *testing.T) {
+	tests := []struct {
+		name      string
+		flags     []string
+		wantValue string
+		wantSet   bool
+	}{
+		{name: "inherit"},
+		{name: "replace", flags: []string{"--reasoning-effort", " focused "}, wantValue: "focused", wantSet: true},
+		{name: "clear", flags: []string{"--clear-reasoning-effort"}, wantSet: true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			roots := newStateTestRoots(t)
+			opts := newStateTestHarness().options()
+			var got []LLMRuntimeOverrides
+			opts.LLMClientFactory = func(_ context.Context, _ config.Config, _ string, overrides LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+				got = append(got, overrides)
+				return nil, nil, nil
+			}
+			args := append([]string{"run", "sample", "--config", roots.config, "--input", roots.input, "--chunk_cache", "bypass"}, tt.flags...)
+			var stdout, stderr bytes.Buffer
+			if code := RunWithOptions(args, &stdout, &stderr, opts); code != 0 || stderr.Len() != 0 {
+				t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+			}
+			if len(got) != 1 {
+				t.Fatalf("factory overrides = %#v, want one call", got)
+			}
+			if !tt.wantSet {
+				if got[0].ReasoningEffort != nil {
+					t.Fatalf("reasoning effort = %q, want inherit", *got[0].ReasoningEffort)
+				}
+				return
+			}
+			if got[0].ReasoningEffort == nil || *got[0].ReasoningEffort != tt.wantValue {
+				t.Fatalf("reasoning effort = %#v, want %q", got[0].ReasoningEffort, tt.wantValue)
+			}
+		})
+	}
+}
+
+func TestRunReasoningEffortOverrideRejectsInvalidSyntax(t *testing.T) {
+	tests := []struct {
+		name      string
+		flags     []string
+		wantError string
+	}{
+		{
+			name:      "mutually exclusive controls",
+			flags:     []string{"--reasoning-effort", "focused", "--clear-reasoning-effort"},
+			wantError: "cannot be combined",
+		},
+		{
+			name:      "empty replacement",
+			flags:     []string{"--reasoning-effort", "  "},
+			wantError: "must not be empty",
+		},
+		{
+			name:      "duplicate replacement",
+			flags:     []string{"--reasoning-effort", "low", "--reasoning-effort", "high"},
+			wantError: "may be specified only once",
+		},
+		{
+			name:      "missing replacement",
+			flags:     []string{"--reasoning-effort"},
+			wantError: "flag needs an argument",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			roots := newStateTestRoots(t)
+			args := append([]string{"run", "sample", "--config", roots.config, "--input", roots.input}, tt.flags...)
+			var stdout, stderr bytes.Buffer
+			code := RunWithOptions(args, &stdout, &stderr, newStateTestHarness().options())
+			if code != 2 || stdout.Len() != 0 || !strings.Contains(stderr.String(), tt.wantError) {
+				t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+			}
+			assertNoRunState(t, roots)
+		})
+	}
+}
+
+func TestReasoningEffortOverrideSeparatesCheckpointIdentities(t *testing.T) {
+	replacement := " focused "
+	cleared := ""
+	states := []struct {
+		name      string
+		overrides LLMRuntimeOverrides
+		wantValue string
+		wantSet   bool
+	}{
+		{name: "inherit"},
+		{name: "replace", overrides: LLMRuntimeOverrides{ReasoningEffort: &replacement}, wantValue: "focused", wantSet: true},
+		{name: "clear", overrides: LLMRuntimeOverrides{ReasoningEffort: &cleared}, wantValue: "<cleared>", wantSet: true},
+	}
+	digests := make(map[string]string, len(states))
+	for _, state := range states {
+		fingerprints := runtimeOverrideFingerprints("", "", state.overrides)
+		var value string
+		var found bool
+		for _, fingerprint := range fingerprints {
+			if fingerprint.Name == "reasoning_effort_override" {
+				value, found = fingerprint.Value, true
+			}
+		}
+		if found != state.wantSet || (found && value != state.wantValue) {
+			t.Fatalf("%s fingerprint found=%t value=%q, want found=%t value=%q", state.name, found, value, state.wantSet, state.wantValue)
+		}
+		identity, err := checkpoint.NewIdentity(checkpoint.IdentityInput{
+			Pipeline:         pipeline.ResolvedPipeline{ID: "sample", Digest: "sha256:pipeline", Input: pipeline.Binding("test/input")},
+			RawInputDigest:   "sha256:input",
+			RuntimeOverrides: fingerprints,
+		})
+		if err != nil {
+			t.Fatal(err)
+		}
+		digests[state.name] = identity.Digest
+	}
+	if digests["inherit"] == digests["replace"] || digests["inherit"] == digests["clear"] || digests["replace"] == digests["clear"] {
+		t.Fatalf("checkpoint identity digests are not distinct: %#v", digests)
+	}
 }
 
 func TestEffectiveLLMProfileIDsAreSortedDeduplicatedAndLLMOnly(t *testing.T) {
 	resolved := pipeline.ResolvedPipeline{
-		Input: pipeline.ModuleBinding{LLMProfile: "input-profile"},
-		Chunk: pipeline.ModuleBinding{LLMProfile: " zeta "},
+		Input:               pipeline.ModuleBinding{LLMProfile: "input-profile"},
+		InputExecutionClass: contracts.ExecutionClassLLMBacked,
+		Chunk:               pipeline.ModuleBinding{LLMProfile: " zeta "},
+		ChunkExecutionClass: contracts.ExecutionClassLLMBacked,
 		Steps: []pipeline.ResolvedPipelineStep{{
 			ID: "default",
 			ArtifactLanes: []pipeline.ResolvedArtifactLane{{
-				Extract:   pipeline.ModuleBinding{LLMProfile: "alpha"},
-				Merge:     pipeline.ModuleBinding{LLMProfile: "zeta"},
-				Normalize: pipeline.ModuleBinding{LLMProfile: " gamma "},
+				Extract:                 pipeline.ModuleBinding{LLMProfile: "alpha"},
+				ExtractExecutionClass:   contracts.ExecutionClassLLMBacked,
+				Merge:                   pipeline.ModuleBinding{LLMProfile: "deterministic-merge"},
+				MergeExecutionClass:     contracts.ExecutionClassDeterministic,
+				Normalize:               pipeline.ModuleBinding{LLMProfile: " gamma "},
+				NormalizeExecutionClass: contracts.ExecutionClassLLMBacked,
 			}},
 		}},
 		ValidatorChains: []pipeline.ResolvedValidatorChain{{Validators: []pipeline.ResolvedValidator{
 			{Binding: pipeline.ModuleBinding{LLMProfile: "deterministic-profile"}, ExecutionClass: contracts.ExecutionClassDeterministic},
 			{Binding: pipeline.ModuleBinding{LLMProfile: "beta"}, ExecutionClass: contracts.ExecutionClassLLMBacked},
 		}}},
-		Output: pipeline.ModuleBinding{LLMProfile: "output-profile"},
+		Output:               pipeline.ModuleBinding{LLMProfile: "output-profile"},
+		OutputExecutionClass: contracts.ExecutionClassLLMBacked,
 	}
 	got := effectiveLLMProfileIDs(resolved)
-	want := []string{"alpha", "beta", "gamma", "zeta"}
+	want := []string{"alpha", "beta", "gamma", "input-profile", "output-profile", "zeta"}
 	if strings.Join(got, ",") != strings.Join(want, ",") {
 		t.Fatalf("effective profiles = %#v, want %#v", got, want)
 	}
@@ -376,7 +529,7 @@ func TestRunFactoryAndPreparationFailuresAreProcessFailures(t *testing.T) {
 	t.Run("LLM factory", func(t *testing.T) {
 		roots := newStateTestRoots(t)
 		opts := newStateTestHarness().options()
-		opts.LLMClientFactory = func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+		opts.LLMClientFactory = func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 			return nil, nil, errors.New("injected LLM factory failure")
 		}
 		var stdout, stderr bytes.Buffer
@@ -397,7 +550,7 @@ func TestRunFactoryAndPreparationFailuresAreProcessFailures(t *testing.T) {
 			t.Fatal(err)
 		}
 		opts := newStateTestHarness().options()
-		if err := pipeline.RegisterExtractorBuilder(opts.Registries.Extractors, pipeline.ModuleSpec{Key: "test/failing-extract", Stage: pipeline.StageExtract, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.Extractor[stateTestArtifact], error) {
+		if err := pipeline.RegisterExtractorBuilder(opts.Registries.Extractors, pipeline.ModuleSpec{Key: "test/failing-extract", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.Extractor[stateTestArtifact], error) {
 			return nil, errors.New("injected extractor construction failure")
 		}); err != nil {
 			t.Fatal(err)

internal/cli/run_result.go

@@ -0,0 +1,111 @@
+package cli
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
+)
+
+const runResultSchemaVersion = "notarius.run-result.v1"
+
+type runResult struct {
+	SchemaVersion         string `json:"schema_version"`
+	RunID                 string `json:"run_id"`
+	PipelineID            string `json:"pipeline_id"`
+	OutputDirectory       string `json:"output_directory"`
+	IndexFile             string `json:"index_file,omitempty"`
+	NormalizedOutputCount int    `json:"normalized_output_count"`
+	RejectedOutputCount   int    `json:"rejected_output_count"`
+	WarningCount          int    `json:"warning_count"`
+	ValidationStatus      string `json:"validation_status"`
+	DebugDirectory        string `json:"debug_directory,omitempty"`
+}
+
+func newRunResult(resolved pipeline.ResolvedPipeline, output pipeline.RunOutput, outputDirectory, debugDirectory string) (runResult, error) {
+	if strings.TrimSpace(output.Manifest.RunID) == "" {
+		return runResult{}, fmt.Errorf("run result requires a run ID")
+	}
+	if strings.TrimSpace(resolved.ID) == "" {
+		return runResult{}, fmt.Errorf("run result requires a resolved pipeline ID")
+	}
+	if strings.TrimSpace(output.Manifest.PipelineID) == "" {
+		return runResult{}, fmt.Errorf("run result requires a manifest pipeline ID")
+	}
+	if output.Manifest.PipelineID != resolved.ID {
+		return runResult{}, fmt.Errorf("run result pipeline ID does not match resolved pipeline")
+	}
+	if strings.TrimSpace(output.Manifest.ValidationStatus) == "" {
+		return runResult{}, fmt.Errorf("run result requires a validation status")
+	}
+	if strings.TrimSpace(outputDirectory) == "" {
+		return runResult{}, fmt.Errorf("run result requires an output directory")
+	}
+
+	absOutputDirectory, err := filepath.Abs(outputDirectory)
+	if err != nil {
+		return runResult{}, fmt.Errorf("make output directory absolute: %w", err)
+	}
+
+	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,
+	}
+
+	if strings.TrimSpace(debugDirectory) != "" {
+		absDebugDirectory, err := filepath.Abs(debugDirectory)
+		if err != nil {
+			return runResult{}, fmt.Errorf("make debug directory absolute: %w", err)
+		}
+		result.DebugDirectory = absDebugDirectory
+	}
+
+	if resolved.Output.Module == pipeline.DefaultOutputModule {
+		indexCount := 0
+		for _, file := range output.OutputFiles {
+			if file.Name == "index.json" {
+				indexCount++
+			}
+		}
+		if indexCount != 1 {
+			return runResult{}, fmt.Errorf("production JSON output must contain exactly one index.json file")
+		}
+		result.IndexFile = "index.json"
+	}
+
+	return result, nil
+}
+
+func encodeRunResult(result runResult) ([]byte, error) {
+	encoded, err := json.Marshal(result)
+	if err != nil {
+		return nil, fmt.Errorf("encode run result: %w", err)
+	}
+	return append(encoded, '\n'), nil
+}
+
+func writeRunResult(writer io.Writer, content []byte) error {
+	for len(content) > 0 {
+		written, err := writer.Write(content)
+		if written < 0 || written > len(content) {
+			return io.ErrShortWrite
+		}
+		content = content[written:]
+		if err != nil {
+			return err
+		}
+		if written == 0 {
+			return io.ErrShortWrite
+		}
+	}
+	return nil
+}

internal/cli/run_result_command_test.go

@@ -0,0 +1,199 @@
+package cli
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	alwaysreject "gitea.maximumdirect.net/eric/notarius/internal/modules/generic/validate/always_reject"
+)
+
+func TestMaintainedMinimalInvocationEmitsRunResult(t *testing.T) {
+	outputRoot := filepath.Join(t.TempDir(), "output")
+	var stdout, stderr strings.Builder
+	code := RunWithOptions([]string{
+		"run", "dnd-session",
+		"--config", repositoryPath("examples", "dnd-minimal.config.yml"),
+		"--input", repositoryPath("examples", "seriatim-minimal-transcript.json"),
+		"--only", "spells", "--chunk_cache", "bypass", "--output-dir", outputRoot, "--json",
+	}, &stdout, &stderr, productionRunOptions(t, &productionFakeLLMClient{}))
+	if code != 0 || stderr.Len() != 0 {
+		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+	}
+
+	receipt := decodeRunResultDocument(t, stdout.String())
+	if got := receipt["schema_version"]; got != "notarius.run-result.v1" {
+		t.Fatalf("schema_version = %q", got)
+	}
+	if got := receipt["run_id"]; got != productionRunID {
+		t.Fatalf("run_id = %q", got)
+	}
+	if got := receipt["pipeline_id"]; got != "dnd-session" {
+		t.Fatalf("pipeline_id = %q", got)
+	}
+	if got := receipt["index_file"]; got != "index.json" {
+		t.Fatalf("index_file = %q", got)
+	}
+	if got := receipt["normalized_output_count"]; got != float64(1) {
+		t.Fatalf("normalized_output_count = %v", got)
+	}
+	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["validation_status"]; got != "approved" {
+		t.Fatalf("validation_status = %q", got)
+	}
+
+	outputDirectory, ok := receipt["output_directory"].(string)
+	if !ok || !filepath.IsAbs(outputDirectory) || outputDirectory != filepath.Join(outputRoot, productionRunID) {
+		t.Fatalf("output_directory = %q", receipt["output_directory"])
+	}
+	indexFile := receipt["index_file"].(string)
+	assertFile(t, filepath.Join(outputDirectory, indexFile))
+}
+
+func TestRunResultReportsWarningsAndDebugBundle(t *testing.T) {
+	roots := newStateTestRoots(t)
+	harness := newStateTestHarness()
+	harness.chunkWarnings = []contracts.Warning{{Scope: "chunk", ReasonCode: "contract-warning", Message: "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)") {
+		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)
+	}
+	debugDirectory, ok := receipt["debug_directory"].(string)
+	if !ok || !filepath.IsAbs(debugDirectory) || debugDirectory != onlyChildDir(t, roots.debug) {
+		t.Fatalf("debug_directory = %q", receipt["debug_directory"])
+	}
+	if strings.Contains(stdout.String(), "complete:") || strings.Contains(stdout.String(), "debug=") {
+		t.Fatalf("machine stdout contains human reporting: %q", stdout.String())
+	}
+}
+
+func TestRunResultReportsSuccessfulRejection(t *testing.T) {
+	roots := newStateTestRoots(t)
+	configBytes, err := os.ReadFile(roots.config)
+	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"))
+	if err := os.WriteFile(roots.config, configBytes, 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	harness := newStateTestHarness()
+	opts := harness.options()
+	if err := alwaysreject.RegisterTyped[stateTestArtifact](opts.Registries.Validators, stateTestArtifactKind); err != nil {
+		t.Fatal(err)
+	}
+	opts.Catalog = catalogFromRegistries(opts.Registries)
+
+	var stdout, stderr bytes.Buffer
+	code := RunWithOptions([]string{
+		"run", "sample", "--config", roots.config, "--input", roots.input,
+		"--chunk_cache", "bypass", "--json",
+	}, &stdout, &stderr, opts)
+	if code != 0 || stderr.Len() != 0 {
+		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+	}
+	receipt := decodeRunResultDocument(t, stdout.String())
+	if got := receipt["normalized_output_count"]; got != float64(0) {
+		t.Fatalf("normalized_output_count = %v", got)
+	}
+	if got := receipt["rejected_output_count"]; got != float64(1) {
+		t.Fatalf("rejected_output_count = %v", got)
+	}
+	if got := receipt["validation_status"]; got != "rejected" {
+		t.Fatalf("validation_status = %q", got)
+	}
+}
+
+func TestRunResultIsAbsentForSyntaxAndRuntimeFailures(t *testing.T) {
+	t.Run("syntax", func(t *testing.T) {
+		var stdout, stderr bytes.Buffer
+		code := RunWithOptions([]string{"run", "sample", "--json"}, &stdout, &stderr, newStateTestHarness().options())
+		if code != 2 || stdout.Len() != 0 || stderr.Len() == 0 {
+			t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+		}
+	})
+
+	t.Run("runtime", func(t *testing.T) {
+		roots := newStateTestRoots(t)
+		harness := newStateTestHarness()
+		harness.extractErr = errors.New("injected extraction failure")
+		var stdout, stderr bytes.Buffer
+		code := RunWithOptions([]string{
+			"run", "sample", "--config", roots.config, "--input", roots.input,
+			"--chunk_cache", "bypass", "--json",
+		}, &stdout, &stderr, harness.options())
+		if code != 1 || stdout.Len() != 0 || stderr.Len() == 0 {
+			t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
+		}
+	})
+}
+
+func TestRunResultDeliveryFailureRetainsPublishedBundles(t *testing.T) {
+	roots := newStateTestRoots(t)
+	writerErr := errors.New("result writer sentinel")
+	stdout := &resultDeliveryWriter{err: writerErr}
+	var stderr bytes.Buffer
+	code := RunWithOptions([]string{
+		"run", "sample", "--config", roots.config, "--input", roots.input,
+		"--chunk_cache", "bypass", "--debug", "--json",
+	}, stdout, &stderr, newStateTestHarness().options())
+	if code != 1 || !strings.Contains(stderr.String(), "write run result") || strings.Contains(stderr.String(), writerErr.Error()) {
+		t.Fatalf("code=%d stderr=%q", code, stderr.String())
+	}
+	if stdout.accepted.Len() != 0 {
+		t.Fatalf("accepted stdout = %q", stdout.accepted.String())
+	}
+	assertStateTestOutput(t, roots.output)
+	debugBundle := onlyChildDir(t, roots.debug)
+	report := readStateTestRunReport(t, debugBundle)
+	if !report.Succeeded {
+		t.Fatalf("debug report = %#v, want successful persisted run", report)
+	}
+	if strings.Contains(readAllFiles(t, debugBundle), writerErr.Error()) {
+		t.Fatalf("debug bundle contains result writer error")
+	}
+}
+
+func decodeRunResultDocument(t *testing.T, stdout string) map[string]any {
+	t.Helper()
+	if strings.Count(stdout, "\n") != 1 {
+		t.Fatalf("stdout = %q, want one JSON document", stdout)
+	}
+	var receipt map[string]any
+	if err := json.Unmarshal([]byte(stdout), &receipt); err != nil {
+		t.Fatalf("decode run result: %v; stdout=%q", err, stdout)
+	}
+	return receipt
+}
+
+type resultDeliveryWriter struct {
+	err      error
+	accepted bytes.Buffer
+}
+
+func (w *resultDeliveryWriter) Write(content []byte) (int, error) {
+	if w.err != nil {
+		return 0, w.err
+	}
+	return w.accepted.Write(content)
+}

internal/cli/run_result_test.go

@@ -0,0 +1,200 @@
+package cli
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"io"
+	"path/filepath"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
+)
+
+func TestRunResultEncodesRequiredFieldsAndCounts(t *testing.T) {
+	result, err := newRunResult(testResolvedPipeline(pipeline.DefaultOutputModule), testRunOutput(), "relative-output", "relative-debug")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	encoded, err := encodeRunResult(result)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if encoded[len(encoded)-1] != '\n' || bytes.Count(encoded, []byte{'\n'}) != 1 {
+		t.Fatalf("encoded result is not one newline-terminated object: %q", encoded)
+	}
+
+	var decoded map[string]any
+	if err := json.Unmarshal(encoded, &decoded); err != nil {
+		t.Fatal(err)
+	}
+	if got := decoded["schema_version"]; got != runResultSchemaVersion {
+		t.Fatalf("schema_version = %q", got)
+	}
+	if got := decoded["run_id"]; got != "run-123" {
+		t.Fatalf("run_id = %q", got)
+	}
+	if got := decoded["pipeline_id"]; got != "sample" {
+		t.Fatalf("pipeline_id = %q", got)
+	}
+	if got := decoded["validation_status"]; got != "rejected" {
+		t.Fatalf("validation_status = %q", got)
+	}
+	if got := decoded["index_file"]; got != "index.json" {
+		t.Fatalf("index_file = %q", got)
+	}
+	if got := decoded["normalized_output_count"]; got != float64(2) {
+		t.Fatalf("normalized_output_count = %v", got)
+	}
+	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["output_directory"]; got != filepath.Join(mustWorkingDirectory(t), "relative-output") {
+		t.Fatalf("output_directory = %q", got)
+	}
+	if got := decoded["debug_directory"]; got != filepath.Join(mustWorkingDirectory(t), "relative-debug") {
+		t.Fatalf("debug_directory = %q", got)
+	}
+}
+
+func TestRunResultRejectsInvalidRequiredValues(t *testing.T) {
+	tests := []struct {
+		name      string
+		resolved  pipeline.ResolvedPipeline
+		output    pipeline.RunOutput
+		directory string
+	}{
+		{name: "blank run ID", resolved: testResolvedPipeline(pipeline.DefaultOutputModule), output: testRunOutputWithout(func(output *pipeline.RunOutput) { output.Manifest.RunID = " " }), directory: "output"},
+		{name: "blank resolved pipeline ID", resolved: pipeline.ResolvedPipeline{Output: pipeline.ModuleBinding{Module: pipeline.DefaultOutputModule}}, output: testRunOutput(), directory: "output"},
+		{name: "blank manifest pipeline ID", resolved: testResolvedPipeline(pipeline.DefaultOutputModule), output: testRunOutputWithout(func(output *pipeline.RunOutput) { output.Manifest.PipelineID = "" }), directory: "output"},
+		{name: "mismatched pipeline IDs", resolved: testResolvedPipeline(pipeline.DefaultOutputModule), output: testRunOutputWithout(func(output *pipeline.RunOutput) { output.Manifest.PipelineID = "other" }), directory: "output"},
+		{name: "blank validation status", resolved: testResolvedPipeline(pipeline.DefaultOutputModule), output: testRunOutputWithout(func(output *pipeline.RunOutput) { output.Manifest.ValidationStatus = " " }), directory: "output"},
+		{name: "blank output directory", resolved: testResolvedPipeline(pipeline.DefaultOutputModule), output: testRunOutput(), directory: " "},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if _, err := newRunResult(tt.resolved, tt.output, tt.directory, ""); err == nil {
+				t.Fatal("newRunResult() succeeded")
+			}
+		})
+	}
+}
+
+func TestRunResultOmitsIndexFileForOtherOutputModules(t *testing.T) {
+	result, err := newRunResult(testResolvedPipeline("test/output"), testRunOutputWithout(func(output *pipeline.RunOutput) {
+		output.OutputFiles = nil
+	}), "output", "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if result.IndexFile != "" {
+		t.Fatalf("index_file = %q", result.IndexFile)
+	}
+	encoded, err := encodeRunResult(result)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var decoded map[string]any
+	if err := json.Unmarshal(encoded, &decoded); err != nil {
+		t.Fatal(err)
+	}
+	if _, ok := decoded["index_file"]; ok {
+		t.Fatalf("encoded non-JSON result contains index_file: %s", encoded)
+	}
+	if _, ok := decoded["debug_directory"]; ok {
+		t.Fatalf("encoded result without debug capture contains debug_directory: %s", encoded)
+	}
+}
+
+func TestRunResultRequiresOneProductionIndexFile(t *testing.T) {
+	tests := []struct {
+		name  string
+		files []contracts.OutputFile
+	}{
+		{name: "missing", files: nil},
+		{name: "duplicate", files: []contracts.OutputFile{{Name: "index.json"}, {Name: "index.json"}}},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			output := testRunOutput()
+			output.OutputFiles = tt.files
+			if _, err := newRunResult(testResolvedPipeline(pipeline.DefaultOutputModule), output, "output", ""); err == nil {
+				t.Fatal("newRunResult() succeeded")
+			}
+		})
+	}
+}
+
+func TestWriteRunResultCompletesAndReportsWriterFailure(t *testing.T) {
+	content := []byte("result\n")
+	var target bytes.Buffer
+	if err := writeRunResult(partialResultWriter{writer: &target, limit: 2}, content); err != nil {
+		t.Fatal(err)
+	}
+	if got := target.String(); got != string(content) {
+		t.Fatalf("written result = %q", got)
+	}
+
+	writerErr := errors.New("result writer failed")
+	if err := writeRunResult(failingResultWriter{err: writerErr}, content); !errors.Is(err, writerErr) {
+		t.Fatalf("writeRunResult() error = %v", err)
+	}
+	if err := writeRunResult(zeroResultWriter{}, content); !errors.Is(err, io.ErrShortWrite) {
+		t.Fatalf("zero-progress error = %v", err)
+	}
+}
+
+func testResolvedPipeline(outputModule string) pipeline.ResolvedPipeline {
+	return pipeline.ResolvedPipeline{ID: "sample", Output: pipeline.ModuleBinding{Module: outputModule}}
+}
+
+func testRunOutput() pipeline.RunOutput {
+	return 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"}},
+	}
+}
+
+func testRunOutputWithout(change func(*pipeline.RunOutput)) pipeline.RunOutput {
+	output := testRunOutput()
+	change(&output)
+	return output
+}
+
+func mustWorkingDirectory(t *testing.T) string {
+	t.Helper()
+	workingDirectory, err := filepath.Abs(".")
+	if err != nil {
+		t.Fatal(err)
+	}
+	return workingDirectory
+}
+
+type partialResultWriter struct {
+	writer io.Writer
+	limit  int
+}
+
+func (w partialResultWriter) Write(content []byte) (int, error) {
+	if len(content) > w.limit {
+		content = content[:w.limit]
+	}
+	return w.writer.Write(content)
+}
+
+type failingResultWriter struct{ err error }
+
+func (w failingResultWriter) Write([]byte) (int, error) { return 0, w.err }
+
+type zeroResultWriter struct{}
+
+func (zeroResultWriter) Write([]byte) (int, error) { return 0, nil }

internal/cli/scriptorium_profiles.go

@@ -1,68 +0,0 @@
-package cli
-
-import (
-	"context"
-	"errors"
-	"fmt"
-	"testing/fstest"
-
-	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
-	"gitea.maximumdirect.net/eric/scriptorium"
-)
-
-const profileCheckPromptID = "notarius.profile.check"
-
-var profileCheckPromptFS = fstest.MapFS{
-	"prompts/profile-check.yaml": &fstest.MapFile{Data: []byte(`id: notarius.profile.check
-version: "1.0.0"
-default_profile: mistral-small-3
-inputs:
-  - name: transcript
-    required: true
-messages:
-  - role: user
-    content: "{{input \"transcript\"}}"
-output:
-  format: text
-  validation_mode: none
-  repair_attempts: 0
-`)},
-}
-
-func validateExplicitScriptoriumProfiles(ctx context.Context, cfg config.Config, profileIDs []string) error {
-	if len(profileIDs) == 0 {
-		return nil
-	}
-	engine, err := newProfileValidationEngine(cfg)
-	if err != nil {
-		return fmt.Errorf("load Scriptorium profiles: %w", err)
-	}
-	for _, profileID := range profileIDs {
-		if _, err := engine.Prepare(ctx, scriptorium.RunRequest{
-			PromptID:  profileCheckPromptID,
-			ProfileID: profileID,
-			Inputs: map[string]scriptorium.ArtifactRef{
-				"transcript": scriptorium.Inline("profile check"),
-			},
-		}); err != nil {
-			if errors.Is(err, scriptorium.ErrProfileNotFound) {
-				return fmt.Errorf("Scriptorium profile %q is not configured", profileID)
-			}
-			return fmt.Errorf("validate Scriptorium profile %q: %w", profileID, err)
-		}
-	}
-	return nil
-}
-
-func newProfileValidationEngine(cfg config.Config) (*scriptorium.Engine, error) {
-	opts := []scriptorium.Option{
-		scriptorium.WithPromptFS(profileCheckPromptFS, "prompts"),
-	}
-	if cfg.Scriptorium.ProfileFile != "" {
-		opts = append(opts, scriptorium.WithProfileFile(cfg.Scriptorium.ProfileFile))
-	}
-	return scriptorium.NewEngine(scriptorium.Config{
-		PromptDir:  "unused",
-		ProfileDir: cfg.Scriptorium.ProfileDir,
-	}, opts...)
-}

internal/cli/spell_catalog_identity_contract_test.go

@@ -214,8 +214,9 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 		t.Fatal(err)
 	}
 	fingerprints := prepared.CheckpointFingerprints()
+	llmFingerprints := []checkpoint.Fingerprint{{Name: "promptkit_profile_source", Value: "sha256:profile-source-one"}}
 	settings := config.CheckpointCacheConfig{Enabled: true, Directory: t.TempDir()}
-	recorder, _, err := checkpointHandlersForRun(settings, Options{}, materialized, fingerprints, []byte("same input"), nil, nil, "", "", false)
+	recorder, _, err := checkpointHandlersForRun(settings, Options{}, materialized, fingerprints, llmFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -241,7 +242,7 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 		t.Fatal(err)
 	}
 
-	_, sameLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, fingerprints, []byte("same input"), nil, nil, "", "", true)
+	_, sameLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, fingerprints, llmFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, true)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -253,7 +254,7 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 	}
 	changed := replaceCheckpointFingerprintValue(t, fingerprints, normalizeSpellCatalogFingerprintName(), "sha256:changed-effective-catalog")
 	assertOnlyCheckpointFingerprintChanged(t, fingerprints, changed, normalizeSpellCatalogFingerprintName())
-	_, changedLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, changed, []byte("same input"), nil, nil, "", "", true)
+	_, changedLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, changed, llmFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, true)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -265,13 +266,25 @@ func TestChangedSemanticSpellCatalogFingerprintCannotResumeRecordedCheckpoint(t
 	}
 	changedMapping := replaceCheckpointFingerprintValue(t, fingerprints, extractSpellMappingFingerprintName(), "dnd.spells.extract_mapping.v3")
 	assertOnlyCheckpointFingerprintChanged(t, fingerprints, changedMapping, extractSpellMappingFingerprintName())
-	_, mappingLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, changedMapping, []byte("same input"), nil, nil, "", "", true)
+	_, mappingLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, changedMapping, llmFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, true)
 	if err != nil {
 		t.Fatal(err)
 	}
 	if _, decision := mappingLoader.Source(materialized.Input.Module); decision.Reused {
 		t.Fatalf("changed mapping policy decision = %#v, want cold miss", decision)
 	}
+
+	changedLLMFingerprints := []checkpoint.Fingerprint{{Name: "promptkit_profile_source", Value: "sha256:profile-source-two"}}
+	_, profileLoader, err := checkpointHandlersForRun(settings, Options{}, materialized, fingerprints, changedLLMFingerprints, []byte("same input"), nil, nil, "", "", LLMRuntimeOverrides{}, true)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, decision := profileLoader.Source(materialized.Input.Module); decision.Reused {
+		t.Fatalf("changed PromptKit profile source decision = %#v, want cold miss", decision)
+	}
+	if _, decision := profileLoader.Normalize("spells", spellnormalize.Key, normalizeDependencies); decision.Reused {
+		t.Fatalf("changed PromptKit profile normalize decision = %#v, want cold miss", decision)
+	}
 }
 
 func normalizeSpellCatalogFingerprintName() string {

internal/cli/state_hardening_test.go

@@ -635,7 +635,7 @@ func newStateTestRoots(t *testing.T) stateTestRoots {
 		t.Fatal(err)
 	}
 	roots.config = filepath.Join(base, "config.yml")
-	config := fmt.Sprintf("version: 3\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: 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)
 	if err := os.WriteFile(roots.config, []byte(config), 0o600); err != nil {
 		t.Fatal(err)
 	}
@@ -823,22 +823,22 @@ func (h *stateTestHarness) options() Options {
 	if err := pipeline.RegisterArtifactCodec(registries.ArtifactCodecs, stateTestCodec{}); err != nil {
 		panic(err)
 	}
-	if err := registries.Inputs.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "test/input", Stage: pipeline.StageInput, Provides: []string{"source"}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.InputAdapter, error) { return stateTestInput{}, nil }); err != nil {
+	if err := registries.Inputs.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "test/input", Stage: pipeline.StageInput, ExecutionClass: contracts.ExecutionClassDeterministic, Provides: []string{"source"}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.InputAdapter, error) { return stateTestInput{}, nil }); err != nil {
 		panic(err)
 	}
-	if err := registries.Chunkers.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "test/chunk", Stage: pipeline.StageChunk, Requires: []string{"source"}, Provides: []string{"chunks"}, ReferenceSlots: []contracts.ReferenceSlot{{Name: "cache-reference"}}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.Chunker, error) { return stateTestChunker{h}, nil }); err != nil {
+	if err := registries.Chunkers.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "test/chunk", Stage: pipeline.StageChunk, ExecutionClass: contracts.ExecutionClassLLMBacked, Requires: []string{"source"}, Provides: []string{"chunks"}, ReferenceSlots: []contracts.ReferenceSlot{{Name: "cache-reference"}}}, func(map[string]any) error { return nil }, func(pipeline.BuildRequest) (contracts.Chunker, error) { return stateTestChunker{h}, nil }); err != nil {
 		panic(err)
 	}
-	if err := pipeline.RegisterExtractor(registries.Extractors, pipeline.ModuleSpec{Key: "test/extract", Stage: pipeline.StageExtract, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Extractor[stateTestArtifact], error) { return stateTestExtractor{h}, nil }); err != nil {
+	if err := pipeline.RegisterExtractor(registries.Extractors, pipeline.ModuleSpec{Key: "test/extract", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassLLMBacked, Requires: []string{"chunks"}, Provides: []string{"artifact"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Extractor[stateTestArtifact], error) { return stateTestExtractor{h}, nil }); err != nil {
 		panic(err)
 	}
-	if err := pipeline.RegisterMerger(registries.Mergers, pipeline.ModuleSpec{Key: "test/merge", Stage: pipeline.StageMerge, Requires: []string{"artifact"}, Provides: []string{"merged"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Merger[stateTestArtifact], error) { return stateTestMerger{harness: h}, nil }); err != nil {
+	if err := pipeline.RegisterMerger(registries.Mergers, pipeline.ModuleSpec{Key: "test/merge", Stage: pipeline.StageMerge, ExecutionClass: contracts.ExecutionClassLLMBacked, Requires: []string{"artifact"}, Provides: []string{"merged"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Merger[stateTestArtifact], error) { return stateTestMerger{harness: h}, nil }); err != nil {
 		panic(err)
 	}
-	if err := pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "test/normalize", Stage: pipeline.StageNormalize, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{harness: h}, nil }); err != nil {
+	if err := pipeline.RegisterNormalizer(registries.Normalizers, pipeline.ModuleSpec{Key: "test/normalize", Stage: pipeline.StageNormalize, ExecutionClass: contracts.ExecutionClassLLMBacked, Requires: []string{"merged"}, Provides: []string{"normalized"}, ArtifactKind: stateTestArtifactKind}, func() (contracts.Normalizer[stateTestArtifact], error) { return stateTestNormalizer{harness: h}, nil }); err != nil {
 		panic(err)
 	}
-	if err := registries.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "test/output", Stage: pipeline.StageOutput, Requires: []string{"normalized"}, Provides: []string{"output"}}, func() (contracts.OutputEncoder, error) {
+	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
 	}); err != nil {
 		panic(err)
@@ -848,7 +848,7 @@ func (h *stateTestHarness) options() Options {
 		defer h.mu.Unlock()
 		h.runIDCalls++
 		return fmt.Sprintf("run-%d-%032x", startedAt.UnixNano(), h.runIDCalls), nil
-	}, UserCacheDir: func() (string, error) { return "", errors.New("unexpected user cache lookup") }, LLMClientFactory: func(context.Context, config.Config, string) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
+	}, UserCacheDir: func() (string, error) { return "", errors.New("unexpected user cache lookup") }, LLMClientFactory: func(context.Context, config.Config, string, LLMRuntimeOverrides) (contracts.StructuredLLMClient, []artifacts.LLMProfileManifest, error) {
 		return nil, nil, nil
 	}}
 }

internal/core/artifacts/artifacts.go

@@ -1,6 +1,7 @@
 package artifacts
 
 import (
+	"strings"
 	"time"
 )
 
@@ -26,9 +27,32 @@ type ValidatorManifest struct {
 }
 
 type LLMProfileManifest struct {
-	ID       string `json:"id"`
-	Provider string `json:"provider,omitempty"`
-	Model    string `json:"model,omitempty"`
+	ID              string `json:"id"`
+	Provider        string `json:"provider,omitempty"`
+	Model           string `json:"model,omitempty"`
+	BackendID       string `json:"backend_id,omitempty"`
+	ReasoningEffort string `json:"reasoning_effort,omitempty"`
+}
+
+// Normalized returns the canonical representation used for manifest identity
+// and publication.
+func (profile LLMProfileManifest) Normalized() LLMProfileManifest {
+	profile.ID = strings.TrimSpace(profile.ID)
+	profile.Provider = strings.TrimSpace(profile.Provider)
+	profile.Model = strings.TrimSpace(profile.Model)
+	profile.BackendID = strings.TrimSpace(profile.BackendID)
+	profile.ReasoningEffort = strings.TrimSpace(profile.ReasoningEffort)
+	return profile
+}
+
+// IdentityKey returns an opaque, deterministic key for the effective profile.
+func (profile LLMProfileManifest) IdentityKey() string {
+	profile = profile.Normalized()
+	return profile.ID + "\x00" +
+		profile.Provider + "\x00" +
+		profile.Model + "\x00" +
+		profile.BackendID + "\x00" +
+		profile.ReasoningEffort
 }
 
 type ReferenceProvenance struct {

internal/core/artifacts/artifacts_test.go

@@ -53,7 +53,13 @@ func TestRunManifestIncludesPipelineAndArtifactLaneFields(t *testing.T) {
 		PipelineID:     "pipeline-1",
 		PipelineDigest: "sha256:abc123",
 		LLMProfiles: []LLMProfileManifest{
-			{ID: "default", Provider: "scriptorium", Model: "model-a"},
+			{
+				ID:              "default",
+				Provider:        "promptkit",
+				Model:           "model-a",
+				BackendID:       "openrouter",
+				ReasoningEffort: "high",
+			},
 		},
 		ArtifactLanes: []ArtifactLaneManifest{
 			{
@@ -101,7 +107,13 @@ func TestRunManifestIncludesPipelineAndArtifactLaneFields(t *testing.T) {
 	if !ok {
 		t.Fatalf("llm_profiles[0] = %#v, want object", profiles[0])
 	}
-	assertHasKeys(t, profile, "id", "provider", "model")
+	assertHasKeys(t, profile, "id", "provider", "model", "backend_id", "reasoning_effort")
+	if profile["provider"] != "promptkit" {
+		t.Fatalf("llm_profiles[0].provider = %#v, want promptkit", profile["provider"])
+	}
+	if profile["backend_id"] != "openrouter" || profile["reasoning_effort"] != "high" {
+		t.Fatalf("llm_profiles[0] = %#v, want backend and reasoning provenance", profile)
+	}
 
 	lanes, ok := got["artifact_lanes"].([]any)
 	if !ok {

internal/core/config/config.go

@@ -4,10 +4,10 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
 )
 
-const SupportedFileConfigVersion = 3
+const SupportedFileConfigVersion = 4
 
 type Config struct {
-	Scriptorium ScriptoriumConfig                   `json:"scriptorium,omitempty"`
+	PromptKit   PromptKitConfig                     `json:"promptkit,omitempty"`
 	Pipelines   map[string]pipeline.PipelineProfile `json:"pipelines"`
 	Concurrency ConcurrencyConfig                   `json:"concurrency"`
 	Output      OutputConfig                        `json:"output"`
@@ -15,9 +15,15 @@ type Config struct {
 	Debug       DebugConfig                         `json:"debug"`
 }
 
-type ScriptoriumConfig struct {
-	ProfileDir  string `json:"profile_dir,omitempty"`
-	ProfileFile string `json:"profile_file,omitempty"`
+type PromptKitConfig struct {
+	ProfileDir   string                       `json:"profile_dir,omitempty"`
+	ProfileFile  string                       `json:"profile_file,omitempty"`
+	LocalBackend *PromptKitLocalBackendConfig `json:"local_backend,omitempty"`
+}
+
+type PromptKitLocalBackendConfig struct {
+	Endpoint         string `json:"endpoint"`
+	ConcurrencyLimit int    `json:"concurrency_limit"`
 }
 
 type ConcurrencyConfig struct {
@@ -66,6 +72,10 @@ func Default() Config {
 
 func cloneConfig(in Config) Config {
 	out := in
+	if in.PromptKit.LocalBackend != nil {
+		localBackend := *in.PromptKit.LocalBackend
+		out.PromptKit.LocalBackend = &localBackend
+	}
 	out.Concurrency.StageWorkers = cloneIntMap(in.Concurrency.StageWorkers)
 	out.Pipelines = make(map[string]pipeline.PipelineProfile, len(in.Pipelines))
 	for key, profile := range in.Pipelines {

internal/core/config/effective_config.go

@@ -42,12 +42,10 @@ func (c Config) Resolve(input ResolveInput) (EffectiveConfig, error) {
 	}
 	profile = clonePipelineProfile(profile)
 	profile.ID = pipelineID
-	if override := strings.TrimSpace(input.LLMProfileOverride); override != "" {
-		applyLLMProfileOverride(&profile, override)
-	}
 
 	resolved, err := pipeline.ResolvePipeline(profile, pipeline.ResolveOptions{
 		Only:               input.Only,
+		LLMProfileOverride: input.LLMProfileOverride,
 		ReferenceOverrides: append([]pipeline.ReferenceBinding(nil), input.ReferenceOverrides...),
 		ReferenceUnbinds:   append([]pipeline.ReferenceUnbind(nil), input.ReferenceUnbinds...),
 	}, input.Catalog)
@@ -65,22 +63,6 @@ func (c Config) Resolve(input ResolveInput) (EffectiveConfig, error) {
 	}, nil
 }
 
-func applyLLMProfileOverride(profile *pipeline.PipelineProfile, profileID string) {
-	profile.Chunk.LLMProfile = profileID
-	apply := func(artifacts map[string]pipeline.ArtifactLaneProfile) {
-		for laneID, lane := range artifacts {
-			lane.Extract.LLMProfile = profileID
-			lane.Merge.LLMProfile = profileID
-			lane.Normalize.LLMProfile = profileID
-			artifacts[laneID] = lane
-		}
-	}
-	apply(profile.Artifacts)
-	for index := range profile.Steps {
-		apply(profile.Steps[index].Artifacts)
-	}
-}
-
 func lookupPipelineProfile(profiles map[string]pipeline.PipelineProfile, pipelineID string) (pipeline.PipelineProfile, bool) {
 	pipelineID = strings.TrimSpace(pipelineID)
 	for rawID, profile := range profiles {

internal/core/config/effective_config_contract_test.go

@@ -84,6 +84,56 @@ func TestEffectiveConfigMaterializesDefaultBindingsThroughCatalog(t *testing.T)
 	}
 }
 
+func TestEffectiveConfigPreservesPromptKitProfileSource(t *testing.T) {
+	tests := []struct {
+		name          string
+		profileSource PromptKitConfig
+	}{
+		{name: "profile directory", profileSource: PromptKitConfig{ProfileDir: "./profiles"}},
+		{name: "profile file", profileSource: PromptKitConfig{ProfileFile: "./profiles.yml"}},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := configForEffectiveTests(t, effectiveProfile())
+			cfg.PromptKit = tt.profileSource
+			effective, err := cfg.Resolve(ResolveInput{PipelineID: "main", Catalog: effectiveCatalog(t)})
+			if err != nil {
+				t.Fatalf("Resolve() error = %v", err)
+			}
+			if effective.Config.PromptKit != cfg.PromptKit {
+				t.Fatalf("effective PromptKit config = %#v, want %#v", effective.Config.PromptKit, cfg.PromptKit)
+			}
+		})
+	}
+}
+
+func TestEffectiveConfigOwnsPromptKitLocalBackend(t *testing.T) {
+	cfg := configForEffectiveTests(t, effectiveProfile())
+	cfg.PromptKit.LocalBackend = &PromptKitLocalBackendConfig{
+		Endpoint:         "http://localhost:8000/v1",
+		ConcurrencyLimit: 2,
+	}
+	effective, err := cfg.Resolve(ResolveInput{PipelineID: "main", Catalog: effectiveCatalog(t)})
+	if err != nil {
+		t.Fatalf("Resolve() error = %v", err)
+	}
+	if effective.Config.PromptKit.LocalBackend == nil {
+		t.Fatal("effective local backend = nil")
+	}
+	if effective.Config.PromptKit.LocalBackend == cfg.PromptKit.LocalBackend {
+		t.Fatal("effective local backend aliases input config")
+	}
+
+	cfg.PromptKit.LocalBackend.Endpoint = "http://changed-input.example/v1"
+	if effective.Config.PromptKit.LocalBackend.Endpoint != "http://localhost:8000/v1" {
+		t.Fatalf("input mutation changed effective config: %#v", effective.Config.PromptKit.LocalBackend)
+	}
+	effective.Config.PromptKit.LocalBackend.ConcurrencyLimit = 9
+	if cfg.PromptKit.LocalBackend.ConcurrencyLimit != 2 {
+		t.Fatalf("effective mutation changed input config: %#v", cfg.PromptKit.LocalBackend)
+	}
+}
+
 func TestEffectiveConfigResolutionFailuresRetainContext(t *testing.T) {
 	tests := []struct {
 		name   string
@@ -154,7 +204,7 @@ func TestEffectiveConfigResolutionFailuresRetainContext(t *testing.T) {
 	}
 }
 
-func TestEffectiveConfigLLMProfileOverrideChangesDigestWithoutOverridingValidators(t *testing.T) {
+func TestEffectiveConfigLLMProfileOverrideChangesDigestAndOverridesValidators(t *testing.T) {
 	profile := effectiveProfile()
 	profile.Chunk.LLMProfile = "chunk-profile"
 	lane := profile.Artifacts["lane"]
@@ -187,8 +237,27 @@ func TestEffectiveConfigLLMProfileOverrideChangesDigestWithoutOverridingValidato
 		t.Fatalf("pipeline profile override was not applied: %#v", resolved)
 	}
 	validators := findEffectiveValidatorChain(resolved, pipeline.StageExtract, "lane")
-	if len(validators.Validators) != 1 || validators.Validators[0].Binding.LLMProfile != "validator-profile" {
-		t.Fatalf("validator profile was overridden: %#v", validators)
+	if len(validators.Validators) != 1 || validators.Validators[0].Binding.LLMProfile != "override-profile" {
+		t.Fatalf("validator profile = %#v, want runtime override", validators)
+	}
+}
+
+func TestEffectiveConfigPipelineLLMProfileIsInheritedWithoutMutatingConfig(t *testing.T) {
+	profile := effectiveProfile()
+	profile.LLMProfile = " configured-profile "
+	effective, err := resolveEffectiveProfile(t, profile, ResolveInput{})
+	if err != nil {
+		t.Fatalf("Resolve() error = %v", err)
+	}
+	if got := effective.Config.Pipelines["main"].LLMProfile; got != " configured-profile " {
+		t.Fatalf("effective config pipeline llm profile = %q, want preserved programmatic value", got)
+	}
+	resolved := effective.ResolvedPipeline
+	if got := resolved.Chunk.LLMProfile; got != "configured-profile" {
+		t.Fatalf("resolved chunk profile = %q, want inherited profile", got)
+	}
+	if got := resolved.Steps[0].ArtifactLanes[0].Extract.LLMProfile; got != "configured-profile" {
+		t.Fatalf("resolved extract profile = %q, want inherited profile", got)
 	}
 }
 
@@ -435,7 +504,7 @@ func effectiveCatalog(t *testing.T) pipeline.ModuleCatalog {
 	if err := pipeline.RegisterArtifactCodec(catalog.ArtifactCodecs, effectiveCodec{}); err != nil {
 		t.Fatal(err)
 	}
-	if err := catalog.Inputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "input", Stage: pipeline.StageInput, Provides: []string{"source"}}, func() (contracts.InputAdapter, error) {
+	if err := catalog.Inputs.RegisterWithSpec(pipeline.ModuleSpec{Key: "input", Stage: pipeline.StageInput, ExecutionClass: contracts.ExecutionClassDeterministic, Provides: []string{"source"}}, func() (contracts.InputAdapter, error) {
 		return effectiveInput{key: "input"}, nil
 	}); err != nil {
 		t.Fatal(err)
@@ -443,6 +512,7 @@ func effectiveCatalog(t *testing.T) pipeline.ModuleCatalog {
 	chunkSpec := pipeline.ModuleSpec{
 		Key:            pipeline.DefaultChunkModule,
 		Stage:          pipeline.StageChunk,
+		ExecutionClass: contracts.ExecutionClassLLMBacked,
 		Requires:       []string{"source"},
 		Provides:       []string{"chunk"},
 		ReferenceSlots: []contracts.ReferenceSlot{{Name: "chunk-ref"}},
@@ -453,32 +523,32 @@ func effectiveCatalog(t *testing.T) pipeline.ModuleCatalog {
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := catalog.Chunkers.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "needs-capability", Stage: pipeline.StageChunk, Requires: []string{"missing"}}, chunkOptions, func(pipeline.BuildRequest) (contracts.Chunker, error) {
+	if err := catalog.Chunkers.RegisterBuilderWithSpec(pipeline.ModuleSpec{Key: "needs-capability", Stage: pipeline.StageChunk, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"missing"}}, chunkOptions, func(pipeline.BuildRequest) (contracts.Chunker, error) {
 		return effectiveChunker{key: "needs-capability"}, nil
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := pipeline.RegisterExtractor(catalog.Extractors, pipeline.ModuleSpec{Key: "extract", Stage: pipeline.StageExtract, ArtifactKind: effectiveArtifactKind, Requires: []string{"chunk"}, Provides: []string{"candidate"}}, func() (contracts.Extractor[effectiveArtifact], error) {
+	if err := pipeline.RegisterExtractor(catalog.Extractors, pipeline.ModuleSpec{Key: "extract", Stage: pipeline.StageExtract, ExecutionClass: contracts.ExecutionClassLLMBacked, ArtifactKind: effectiveArtifactKind, Requires: []string{"chunk"}, Provides: []string{"candidate"}}, func() (contracts.Extractor[effectiveArtifact], error) {
 		return effectiveExtractor{key: "extract"}, nil
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := pipeline.RegisterMerger(catalog.Mergers, pipeline.ModuleSpec{Key: pipeline.DefaultMergeModule, Stage: pipeline.StageMerge, ArtifactKind: effectiveArtifactKind, Requires: []string{"candidate"}, Provides: []string{"merged"}}, func() (contracts.Merger[effectiveArtifact], error) {
+	if err := pipeline.RegisterMerger(catalog.Mergers, pipeline.ModuleSpec{Key: pipeline.DefaultMergeModule, Stage: pipeline.StageMerge, ExecutionClass: contracts.ExecutionClassLLMBacked, ArtifactKind: effectiveArtifactKind, Requires: []string{"candidate"}, Provides: []string{"merged"}}, func() (contracts.Merger[effectiveArtifact], error) {
 		return effectiveMerger{key: pipeline.DefaultMergeModule}, nil
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := pipeline.RegisterMerger(catalog.Mergers, pipeline.ModuleSpec{Key: "other-merge", Stage: pipeline.StageMerge, ArtifactKind: "other-kind", Requires: []string{"candidate"}, Provides: []string{"merged"}}, func() (contracts.Merger[effectiveArtifact], error) {
+	if err := pipeline.RegisterMerger(catalog.Mergers, pipeline.ModuleSpec{Key: "other-merge", Stage: pipeline.StageMerge, ExecutionClass: contracts.ExecutionClassDeterministic, ArtifactKind: "other-kind", Requires: []string{"candidate"}, Provides: []string{"merged"}}, func() (contracts.Merger[effectiveArtifact], error) {
 		return effectiveMerger{key: "other-merge"}, nil
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := pipeline.RegisterNormalizer(catalog.Normalizers, pipeline.ModuleSpec{Key: pipeline.DefaultNormalizeModule, Stage: pipeline.StageNormalize, ArtifactKind: effectiveArtifactKind, Requires: []string{"merged"}, Provides: []string{"normalized"}}, func() (contracts.Normalizer[effectiveArtifact], error) {
+	if err := pipeline.RegisterNormalizer(catalog.Normalizers, pipeline.ModuleSpec{Key: pipeline.DefaultNormalizeModule, Stage: pipeline.StageNormalize, ExecutionClass: contracts.ExecutionClassLLMBacked, ArtifactKind: effectiveArtifactKind, Requires: []string{"merged"}, Provides: []string{"normalized"}}, func() (contracts.Normalizer[effectiveArtifact], error) {
 		return effectiveNormalizer{key: pipeline.DefaultNormalizeModule}, nil
 	}); err != nil {
 		t.Fatal(err)
 	}
-	if err := catalog.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: pipeline.DefaultOutputModule, Stage: pipeline.StageOutput, Requires: []string{"normalized"}}, func() (contracts.OutputEncoder, error) {
+	if err := catalog.Outputs.RegisterWithSpec(pipeline.ModuleSpec{Key: pipeline.DefaultOutputModule, Stage: pipeline.StageOutput, ExecutionClass: contracts.ExecutionClassDeterministic, Requires: []string{"normalized"}}, func() (contracts.OutputEncoder, error) {
 		return effectiveOutput{key: pipeline.DefaultOutputModule}, nil
 	}); err != nil {
 		t.Fatal(err)

internal/core/config/env_contract_test.go

@@ -10,7 +10,7 @@ import (
 )
 
 func TestPrecedenceFileValuesOverrideBuiltInDefaults(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 concurrency:
   total_llm: 4
   stage_workers:
@@ -35,7 +35,7 @@ debug:
 }
 
 func TestPrecedenceOperationalEnvironmentOverridesFileValues(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 concurrency:
   total_llm: 2
   stage_workers:
@@ -81,21 +81,21 @@ func TestPrecedenceExtractWorkersFollowEffectiveConcurrencyUnlessExplicit(t *tes
 	}{
 		{
 			name:       "default follows environment total",
-			file:       "version: 3\n",
+			file:       "version: 4\n",
 			env:        map[string]string{"NOTARIUS_TOTAL_LLM_CONCURRENCY": "5"},
 			wantTotal:  5,
 			wantWorker: 5,
 		},
 		{
 			name:       "file worker is retained",
-			file:       "version: 3\nconcurrency:\n  total_llm: 3\n  stage_workers:\n    extract: 2\n",
+			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: 3\nconcurrency:\n  total_llm: 2\n",
+			file: "version: 4\nconcurrency:\n  total_llm: 2\n",
 			env: map[string]string{
 				"NOTARIUS_TOTAL_LLM_CONCURRENCY": "6",
 				"NOTARIUS_STAGE_WORKERS_EXTRACT": "4",
@@ -118,7 +118,7 @@ func TestPrecedenceExtractWorkersFollowEffectiveConcurrencyUnlessExplicit(t *tes
 }
 
 func TestPrecedenceEmptyFileCacheDirectoriesDeferPerUserResolution(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 cache:
   chunk_plans:
     directory: ""

internal/core/config/file_config.go

@@ -14,7 +14,7 @@ import (
 
 type FileConfig struct {
 	Version     int                            `yaml:"version"`
-	Scriptorium *FileScriptoriumConfig         `yaml:"scriptorium,omitempty"`
+	PromptKit   *FilePromptKitConfig           `yaml:"promptkit,omitempty"`
 	Pipelines   map[string]FilePipelineProfile `yaml:"pipelines,omitempty"`
 	Concurrency *FileConcurrencyConfig         `yaml:"concurrency,omitempty"`
 	Output      *FileOutputConfig              `yaml:"output,omitempty"`
@@ -22,27 +22,35 @@ type FileConfig struct {
 	Debug       *FileDebugConfig               `yaml:"debug,omitempty"`
 }
 
-type FileScriptoriumConfig struct {
-	ProfileDir  *string `yaml:"profile_dir,omitempty"`
-	ProfileFile *string `yaml:"profile_file,omitempty"`
+type FilePromptKitConfig struct {
+	ProfileDir   *string                          `yaml:"profile_dir,omitempty"`
+	ProfileFile  *string                          `yaml:"profile_file,omitempty"`
+	LocalBackend *FilePromptKitLocalBackendConfig `yaml:"local_backend,omitempty"`
+}
+
+type FilePromptKitLocalBackendConfig struct {
+	Endpoint         *string `yaml:"endpoint,omitempty"`
+	ConcurrencyLimit *int    `yaml:"concurrency_limit,omitempty"`
 }
 
 type FilePipelineProfile struct {
-	Input        fileModuleBinding                  `yaml:"input"`
-	Chunk        *fileModuleBinding                 `yaml:"chunk,omitempty"`
-	Artifacts    map[string]FileArtifactLaneProfile `yaml:"artifacts,omitempty"`
-	Steps        []FilePipelineStepProfile          `yaml:"steps,omitempty"`
-	Output       *fileModuleBinding                 `yaml:"output,omitempty"`
-	References   map[string]fileReferenceSource     `yaml:"references,omitempty"`
-	artifactsSet bool                               `yaml:"-"`
-	stepsSet     bool                               `yaml:"-"`
+	LLMProfile    *string                            `yaml:"llm_profile,omitempty"`
+	Input         fileModuleBinding                  `yaml:"input"`
+	Chunk         *fileModuleBinding                 `yaml:"chunk,omitempty"`
+	Artifacts     map[string]FileArtifactLaneProfile `yaml:"artifacts,omitempty"`
+	Steps         []FilePipelineStepProfile          `yaml:"steps,omitempty"`
+	Output        *fileModuleBinding                 `yaml:"output,omitempty"`
+	References    map[string]fileReferenceSource     `yaml:"references,omitempty"`
+	artifactsSet  bool                               `yaml:"-"`
+	stepsSet      bool                               `yaml:"-"`
+	llmProfileSet bool                               `yaml:"-"`
 }
 
 func (p *FilePipelineProfile) UnmarshalYAML(node *yaml.Node) error {
 	type plainFilePipelineProfile FilePipelineProfile
 	var decoded plainFilePipelineProfile
 	seen, err := decodeKnownMapping(node, &decoded, map[string]struct{}{
-		"input": {}, "chunk": {}, "artifacts": {}, "steps": {}, "output": {}, "references": {},
+		"llm_profile": {}, "input": {}, "chunk": {}, "artifacts": {}, "steps": {}, "output": {}, "references": {},
 	}, "pipeline profile")
 	if err != nil {
 		return err
@@ -50,6 +58,7 @@ func (p *FilePipelineProfile) UnmarshalYAML(node *yaml.Node) error {
 	*p = FilePipelineProfile(decoded)
 	_, p.artifactsSet = seen["artifacts"]
 	_, p.stepsSet = seen["steps"]
+	_, p.llmProfileSet = seen["llm_profile"]
 	return nil
 }
 
@@ -251,6 +260,9 @@ func (b *fileModuleBinding) UnmarshalYAML(node *yaml.Node) error {
 					return err
 				}
 				b.LLMProfile = strings.TrimSpace(llmProfile)
+				if b.LLMProfile == "" {
+					return fmt.Errorf("llm_profile must not be empty when set")
+				}
 			case "retries":
 				var retries int
 				if err := valueNode.Decode(&retries); err != nil {
@@ -325,6 +337,9 @@ func ParseFileConfigYAML(data []byte) (FileConfig, error) {
 	if header.Version == 2 {
 		return FileConfig{}, fmt.Errorf("config version 2 is no longer supported; migrate the file using the version 2-to-3 migration in docs/config.md")
 	}
+	if header.Version == 3 {
+		return FileConfig{}, fmt.Errorf("config version 3 is no longer supported; change \"version: 3\" to \"version: 4\" and rename \"scriptorium:\" to \"promptkit:\"")
+	}
 	if header.Version != SupportedFileConfigVersion {
 		return FileConfig{}, fmt.Errorf("unsupported config version %d (supported version is %d)", header.Version, SupportedFileConfigVersion)
 	}
@@ -450,25 +465,46 @@ func (c *Config) applyFileConfigWithLookup(fileCfg FileConfig, lookup func(strin
 		}
 	}
 
-	if fileCfg.Scriptorium != nil {
-		if fileCfg.Scriptorium.ProfileDir != nil {
-			value := strings.TrimSpace(*fileCfg.Scriptorium.ProfileDir)
+	if fileCfg.PromptKit != nil {
+		if fileCfg.PromptKit.ProfileDir != nil {
+			value := strings.TrimSpace(*fileCfg.PromptKit.ProfileDir)
 			if value == "" {
-				return fmt.Errorf("scriptorium.profile_dir must not be empty when set")
+				return fmt.Errorf("promptkit.profile_dir must not be empty when set")
 			}
-			c.Scriptorium.ProfileDir = value
+			c.PromptKit.ProfileDir = value
 		}
-		if fileCfg.Scriptorium.ProfileFile != nil {
-			value := strings.TrimSpace(*fileCfg.Scriptorium.ProfileFile)
+		if fileCfg.PromptKit.ProfileFile != nil {
+			value := strings.TrimSpace(*fileCfg.PromptKit.ProfileFile)
 			if value == "" {
-				return fmt.Errorf("scriptorium.profile_file must not be empty when set")
+				return fmt.Errorf("promptkit.profile_file must not be empty when set")
 			}
-			c.Scriptorium.ProfileFile = value
+			c.PromptKit.ProfileFile = value
+		}
+		if fileCfg.PromptKit.LocalBackend != nil {
+			if fileCfg.PromptKit.LocalBackend.Endpoint == nil {
+				return fmt.Errorf("promptkit.local_backend.endpoint must not be empty when set")
+			}
+			endpoint := strings.TrimSpace(*fileCfg.PromptKit.LocalBackend.Endpoint)
+			if endpoint == "" {
+				return fmt.Errorf("promptkit.local_backend.endpoint must not be empty when set")
+			}
+			localBackend := PromptKitLocalBackendConfig{Endpoint: endpoint}
+			if fileCfg.PromptKit.LocalBackend.ConcurrencyLimit != nil {
+				localBackend.ConcurrencyLimit = *fileCfg.PromptKit.LocalBackend.ConcurrencyLimit
+			}
+			c.PromptKit.LocalBackend = &localBackend
 		}
 	}
 
 	for _, pipelineID := range pipelineIDs {
 		filePipeline := fileCfg.Pipelines[rawPipelineIDs[pipelineID]]
+		llmProfile := ""
+		if filePipeline.llmProfileSet || filePipeline.LLMProfile != nil {
+			if filePipeline.LLMProfile == nil || strings.TrimSpace(*filePipeline.LLMProfile) == "" {
+				return fmt.Errorf("pipeline %q llm_profile must not be empty when set", pipelineID)
+			}
+			llmProfile = strings.TrimSpace(*filePipeline.LLMProfile)
+		}
 		hasSteps := filePipeline.stepsSet || filePipeline.Steps != nil
 		laneIDs, rawLaneIDs, err := normalizedMapKeys(filePipeline.Artifacts, fmt.Sprintf("pipeline %q artifact lane id", pipelineID))
 		if err != nil {
@@ -476,6 +512,7 @@ func (c *Config) applyFileConfigWithLookup(fileCfg FileConfig, lookup func(strin
 		}
 		profile := pipeline.PipelineProfile{
 			ID:         pipelineID,
+			LLMProfile: llmProfile,
 			Input:      filePipeline.Input.toPipelineBinding(),
 			Artifacts:  make(map[string]pipeline.ArtifactLaneProfile, len(filePipeline.Artifacts)),
 			References: fileReferenceSourcesToPipeline(filePipeline.References),

internal/core/config/file_config_contract_test.go

@@ -1,6 +1,8 @@
 package config
 
 import (
+	"encoding/json"
+	"fmt"
 	"os"
 	"path/filepath"
 	"reflect"
@@ -34,8 +36,8 @@ func TestDefaultReturnsDocumentedValuesAndIndependentMaps(t *testing.T) {
 	}
 }
 
-func TestFileConfigMinimalVersion3AppliesOverDefaults(t *testing.T) {
-	file := parseFileConfig(t, "version: 3\n")
+func TestFileConfigMinimalVersion4AppliesOverDefaults(t *testing.T) {
+	file := parseFileConfig(t, "version: 4\n")
 	cfg := Default()
 	if err := cfg.ApplyFileConfig(file); err != nil {
 		t.Fatal(err)
@@ -48,6 +50,259 @@ func TestFileConfigMinimalVersion3AppliesOverDefaults(t *testing.T) {
 	}
 }
 
+func TestFilePipelineLLMProfileIsPresenceAwareAndDetached(t *testing.T) {
+	const pipelineYAML = `version: 4
+pipelines:
+  main:
+    %s
+    input: input
+    artifacts:
+      lane:
+        extract: extract
+`
+
+	t.Run("omitted", func(t *testing.T) {
+		file := parseFileConfig(t, fmt.Sprintf(pipelineYAML, ""))
+		if file.Pipelines["main"].LLMProfile != nil || file.Pipelines["main"].llmProfileSet {
+			t.Fatalf("parsed pipeline profile = %#v, want omitted llm profile", file.Pipelines["main"])
+		}
+		cfg := Default()
+		if err := cfg.ApplyFileConfig(file); err != nil {
+			t.Fatal(err)
+		}
+		if got := cfg.Pipelines["main"].LLMProfile; got != "" {
+			t.Fatalf("pipeline llm profile = %q, want empty", got)
+		}
+	})
+
+	t.Run("trimmed and detached", func(t *testing.T) {
+		file := parseFileConfig(t, fmt.Sprintf(pipelineYAML, "llm_profile: ' configured-profile '"))
+		cfg := Default()
+		if err := cfg.ApplyFileConfig(file); err != nil {
+			t.Fatal(err)
+		}
+		if got := cfg.Pipelines["main"].LLMProfile; got != "configured-profile" {
+			t.Fatalf("pipeline llm profile = %q, want trimmed value", got)
+		}
+		*file.Pipelines["main"].LLMProfile = "changed-profile"
+		if got := cfg.Pipelines["main"].LLMProfile; got != "configured-profile" {
+			t.Fatalf("effective config aliases parsed file: %q", got)
+		}
+		if got := cloneConfig(cfg).Pipelines["main"].LLMProfile; got != "configured-profile" {
+			t.Fatalf("cloned pipeline llm profile = %q", got)
+		}
+		data, err := json.Marshal(cfg)
+		if err != nil {
+			t.Fatal(err)
+		}
+		var roundTripped Config
+		if err := json.Unmarshal(data, &roundTripped); err != nil {
+			t.Fatal(err)
+		}
+		if got := roundTripped.Pipelines["main"].LLMProfile; got != "configured-profile" {
+			t.Fatalf("round-tripped pipeline llm profile = %q", got)
+		}
+	})
+
+	for _, value := range []string{"''", "'   '", "null"} {
+		t.Run("explicit empty "+value, func(t *testing.T) {
+			file := parseFileConfig(t, fmt.Sprintf(pipelineYAML, "llm_profile: "+value))
+			cfg := Default()
+			err := cfg.ApplyFileConfig(file)
+			if err == nil || !strings.Contains(err.Error(), `pipeline "main" llm_profile must not be empty`) {
+				t.Fatalf("ApplyFileConfig() error = %v, want explicit-empty rejection", err)
+			}
+		})
+	}
+}
+
+func TestFileModuleBindingRejectsExplicitEmptyLLMProfile(t *testing.T) {
+	const configYAML = `version: 4
+pipelines:
+  main:
+    input:
+      module: input
+      llm_profile: %s
+    artifacts:
+      lane:
+        extract: extract
+`
+	for _, value := range []string{"''", "'   '", "null"} {
+		t.Run(value, func(t *testing.T) {
+			_, err := ParseFileConfigYAML([]byte(fmt.Sprintf(configYAML, value)))
+			if err == nil || !strings.Contains(err.Error(), "llm_profile must not be empty when set") {
+				t.Fatalf("ParseFileConfigYAML() error = %v, want explicit-empty binding profile rejection", err)
+			}
+		})
+	}
+}
+
+func TestFilePromptKitProfileSourcesSurviveConfigBoundaries(t *testing.T) {
+	tests := []struct {
+		name string
+		yaml string
+		want PromptKitConfig
+	}{
+		{
+			name: "profile directory",
+			yaml: "version: 4\npromptkit:\n  profile_dir: ' ./profiles '\n",
+			want: PromptKitConfig{ProfileDir: "./profiles"},
+		},
+		{
+			name: "profile file",
+			yaml: "version: 4\npromptkit:\n  profile_file: ' ./profiles.yml '\n",
+			want: PromptKitConfig{ProfileFile: "./profiles.yml"},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := applyFileConfig(t, tt.yaml)
+			if cfg.PromptKit != tt.want {
+				t.Fatalf("PromptKit config = %#v, want %#v", cfg.PromptKit, tt.want)
+			}
+			if got := cloneConfig(cfg).PromptKit; got != tt.want {
+				t.Fatalf("cloned PromptKit config = %#v, want %#v", got, tt.want)
+			}
+			if got := cfg.Redacted().PromptKit; got != tt.want {
+				t.Fatalf("redacted PromptKit config = %#v, want %#v", got, tt.want)
+			}
+
+			data, err := json.Marshal(cfg)
+			if err != nil {
+				t.Fatalf("json.Marshal() error = %v", err)
+			}
+			var payload map[string]json.RawMessage
+			if err := json.Unmarshal(data, &payload); err != nil {
+				t.Fatalf("json.Unmarshal() error = %v", err)
+			}
+			if _, ok := payload["promptkit"]; !ok {
+				t.Fatalf("runtime JSON keys = %v, want promptkit", payload)
+			}
+			if _, ok := payload["scriptorium"]; ok {
+				t.Fatalf("runtime JSON keys = %v, must not contain removed section", payload)
+			}
+		})
+	}
+}
+
+func TestFilePromptKitLocalBackendSurvivesConfigBoundaries(t *testing.T) {
+	tests := []struct {
+		name            string
+		concurrencyYAML string
+		wantConcurrency int
+	}{
+		{name: "omitted concurrency defaults to zero"},
+		{name: "positive concurrency is preserved", concurrencyYAML: "    concurrency_limit: 2\n", wantConcurrency: 2},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			file := parseFileConfig(t, "version: 4\npromptkit:\n  local_backend:\n    endpoint: ' http://localhost:8000/v1 '\n"+tt.concurrencyYAML)
+			cfg := Default()
+			if err := cfg.ApplyFileConfig(file); err != nil {
+				t.Fatalf("ApplyFileConfig() error = %v", err)
+			}
+			want := PromptKitLocalBackendConfig{
+				Endpoint:         "http://localhost:8000/v1",
+				ConcurrencyLimit: tt.wantConcurrency,
+			}
+			if cfg.PromptKit.LocalBackend == nil || *cfg.PromptKit.LocalBackend != want {
+				t.Fatalf("local backend config = %#v, want %#v", cfg.PromptKit.LocalBackend, want)
+			}
+
+			*file.PromptKit.LocalBackend.Endpoint = "http://changed.example/v1"
+			if file.PromptKit.LocalBackend.ConcurrencyLimit != nil {
+				*file.PromptKit.LocalBackend.ConcurrencyLimit = 99
+			}
+			if *cfg.PromptKit.LocalBackend != want {
+				t.Fatalf("effective config aliases parsed file model: %#v", cfg.PromptKit.LocalBackend)
+			}
+
+			cloned := cloneConfig(cfg)
+			if cloned.PromptKit.LocalBackend == cfg.PromptKit.LocalBackend || *cloned.PromptKit.LocalBackend != want {
+				t.Fatalf("cloned local backend = %#v, want detached %#v", cloned.PromptKit.LocalBackend, want)
+			}
+			cloned.PromptKit.LocalBackend.Endpoint = "http://clone.example/v1"
+			if *cfg.PromptKit.LocalBackend != want {
+				t.Fatalf("mutating clone changed source config: %#v", cfg.PromptKit.LocalBackend)
+			}
+
+			redacted := cfg.Redacted()
+			if redacted.PromptKit.LocalBackend == cfg.PromptKit.LocalBackend || *redacted.PromptKit.LocalBackend != want {
+				t.Fatalf("redacted local backend = %#v, want detached %#v", redacted.PromptKit.LocalBackend, want)
+			}
+
+			data, err := json.Marshal(cfg)
+			if err != nil {
+				t.Fatalf("json.Marshal() error = %v", err)
+			}
+			var payload struct {
+				PromptKit map[string]json.RawMessage `json:"promptkit"`
+			}
+			if err := json.Unmarshal(data, &payload); err != nil {
+				t.Fatalf("json.Unmarshal() error = %v", err)
+			}
+			localJSON, ok := payload.PromptKit["local_backend"]
+			if !ok {
+				t.Fatalf("runtime PromptKit JSON keys = %v, want local_backend", payload.PromptKit)
+			}
+			var localPayload map[string]json.RawMessage
+			if err := json.Unmarshal(localJSON, &localPayload); err != nil {
+				t.Fatalf("unmarshal local_backend JSON: %v", err)
+			}
+			if _, ok := localPayload["endpoint"]; !ok {
+				t.Fatalf("runtime local_backend JSON keys = %v, want endpoint", localPayload)
+			}
+			if _, ok := localPayload["concurrency_limit"]; !ok {
+				t.Fatalf("runtime local_backend JSON keys = %v, want concurrency_limit", localPayload)
+			}
+		})
+	}
+}
+
+func TestFilePromptKitLocalBackendRequiresEndpoint(t *testing.T) {
+	for _, tt := range []struct {
+		name string
+		yaml string
+	}{
+		{name: "missing", yaml: "version: 4\npromptkit:\n  local_backend: {}\n"},
+		{name: "empty", yaml: "version: 4\npromptkit:\n  local_backend:\n    endpoint: ''\n"},
+		{name: "blank", yaml: "version: 4\npromptkit:\n  local_backend:\n    endpoint: '   '\n"},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			file := parseFileConfig(t, tt.yaml)
+			cfg := Default()
+			err := cfg.ApplyFileConfig(file)
+			if err == nil || !strings.Contains(err.Error(), "promptkit.local_backend.endpoint") {
+				t.Fatalf("ApplyFileConfig() error = %v, want endpoint field context", err)
+			}
+		})
+	}
+}
+
+func TestFilePromptKitExplicitEmptyProfileSourcesAreRejected(t *testing.T) {
+	for _, field := range []string{"profile_dir", "profile_file"} {
+		t.Run(field, func(t *testing.T) {
+			file := parseFileConfig(t, "version: 4\npromptkit:\n  "+field+": ''\n")
+			cfg := Default()
+			err := cfg.ApplyFileConfig(file)
+			if err == nil || !strings.Contains(err.Error(), "promptkit."+field+" must not be empty") {
+				t.Fatalf("ApplyFileConfig() error = %v, want explicit-empty rejection", err)
+			}
+		})
+	}
+}
+
+func TestFilePromptKitProfileSourcesRemainMutuallyExclusive(t *testing.T) {
+	cfg := applyFileConfig(t, `version: 4
+promptkit:
+  profile_dir: ./profiles
+  profile_file: ./profiles.yml
+`)
+	if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "promptkit profile_dir and profile_file are mutually exclusive") {
+		t.Fatalf("Validate() error = %v, want mutually exclusive profile sources", err)
+	}
+}
+
 func TestFileConfigMissingVersionIsReportedBeforeFieldDecoding(t *testing.T) {
 	_, err := ParseFileConfigYAML([]byte("workspace:\n  directory: /tmp/old\n"))
 	if err == nil || !strings.Contains(err.Error(), "config version is required") {
@@ -55,6 +310,15 @@ func TestFileConfigMissingVersionIsReportedBeforeFieldDecoding(t *testing.T) {
 	}
 }
 
+func TestFileConfigVersion3ReportsPromptKitMigration(t *testing.T) {
+	_, err := ParseFileConfigYAML([]byte("version: 3\nscriptorium:\n  profile_dir: ./profiles\n"))
+	if err == nil ||
+		!strings.Contains(err.Error(), `change "version: 3" to "version: 4"`) ||
+		!strings.Contains(err.Error(), `rename "scriptorium:" to "promptkit:"`) {
+		t.Fatalf("version 3 error = %v, want actionable version and section migration", err)
+	}
+}
+
 func TestFileConfigRejectsUnknownCurrentAndRemovedFields(t *testing.T) {
 	tests := []struct {
 		name string
@@ -63,14 +327,19 @@ func TestFileConfigRejectsUnknownCurrentAndRemovedFields(t *testing.T) {
 	}{
 		{
 			name: "removed diagnostics",
-			yaml: "version: 3\ndiagnostics: {}\n",
+			yaml: "version: 4\ndiagnostics: {}\n",
 			want: "field diagnostics not found",
 		},
 		{
 			name: "removed llm profiles",
-			yaml: "version: 3\nllm_profiles: {}\n",
+			yaml: "version: 4\nllm_profiles: {}\n",
 			want: "field llm_profiles not found",
 		},
+		{
+			name: "removed scriptorium section",
+			yaml: "version: 4\nscriptorium: {}\n",
+			want: "field scriptorium not found",
+		},
 		{
 			name: "version 2 migration",
 			yaml: "version: 2\nworkspace:\n  directory: /tmp/old\n",
@@ -78,29 +347,34 @@ func TestFileConfigRejectsUnknownCurrentAndRemovedFields(t *testing.T) {
 		},
 		{
 			name: "pipeline field",
-			yaml: "version: 3\npipelines:\n  main:\n    unknown: true\n",
+			yaml: "version: 4\npipelines:\n  main:\n    unknown: true\n",
 			want: "field unknown not found",
 		},
 		{
 			name: "lane field",
-			yaml: "version: 3\npipelines:\n  main:\n    artifacts:\n      spells:\n        unknown: true\n",
+			yaml: "version: 4\npipelines:\n  main:\n    artifacts:\n      spells:\n        unknown: true\n",
 			want: "field unknown not found",
 		},
 		{
 			name: "module binding field",
-			yaml: "version: 3\npipelines:\n  main:\n    input:\n      module: seriatim\n      unknown: true\n",
+			yaml: "version: 4\npipelines:\n  main:\n    input:\n      module: seriatim\n      unknown: true\n",
 			want: "field unknown not found in module binding",
 		},
 		{
 			name: "checkpoint field",
-			yaml: "version: 3\ncache:\n  checkpoints:\n    unknown: true\n",
+			yaml: "version: 4\ncache:\n  checkpoints:\n    unknown: true\n",
 			want: "field unknown not found",
 		},
 		{
 			name: "checkpoint enabled type",
-			yaml: "version: 3\ncache:\n  checkpoints:\n    enabled: definitely\n",
+			yaml: "version: 4\ncache:\n  checkpoints:\n    enabled: definitely\n",
 			want: "cannot unmarshal",
 		},
+		{
+			name: "local backend field",
+			yaml: "version: 4\npromptkit:\n  local_backend:\n    endpoint: http://localhost:8000/v1\n    unknown: true\n",
+			want: "field unknown not found",
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
@@ -113,7 +387,7 @@ func TestFileConfigRejectsUnknownCurrentAndRemovedFields(t *testing.T) {
 }
 
 func TestFileConfigModuleBindingsPreserveFormsAndValidatorPresence(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 pipelines:
   main:
     input: seriatim
@@ -160,7 +434,7 @@ pipelines:
 }
 
 func TestFileConfigReferencePrecedenceIsRetained(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 pipelines:
   main:
     input: seriatim
@@ -224,7 +498,7 @@ pipelines:
 }
 
 func TestFileConfigStageLocalValidatorsPreserveOrderAndFields(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
+	cfg := applyFileConfig(t, `version: 4
 pipelines:
   main:
     input: seriatim
@@ -277,8 +551,8 @@ pipelines:
 }
 
 func TestFileConfigStateSectionsApplyIndependently(t *testing.T) {
-	cfg := applyFileConfig(t, `version: 3
-scriptorium:
+	cfg := applyFileConfig(t, `version: 4
+promptkit:
   profile_dir: ./profiles
 concurrency:
   total_llm: 7
@@ -294,15 +568,15 @@ cache:
 debug:
   directory: ./debug
 `)
-	if cfg.Scriptorium.ProfileDir != "./profiles" || cfg.Scriptorium.ProfileFile != "" {
-		t.Fatalf("scriptorium = %#v", cfg.Scriptorium)
+	if cfg.PromptKit.ProfileDir != "./profiles" || cfg.PromptKit.ProfileFile != "" {
+		t.Fatalf("promptkit = %#v", cfg.PromptKit)
 	}
 	if cfg.Concurrency.TotalLLM != 7 || cfg.Concurrency.StageWorkers["extract"] != 7 {
 		t.Fatalf("concurrency = %#v", cfg.Concurrency)
 	}
 	if cfg.Output.Directory != "./output" || cfg.Cache.ChunkPlans.Directory != "plans" || cfg.Cache.ChunkPlans.Mode != pipeline.ChunkCacheBypass ||
 		!cfg.Cache.Checkpoints.Enabled || cfg.Cache.Checkpoints.Directory != "checkpoints" || cfg.Debug.Directory != "./debug" {
-		t.Fatalf("state sections = %#v, %#v, %#v, %#v", cfg.Output, cfg.Cache, cfg.Debug, cfg.Scriptorium)
+		t.Fatalf("state sections = %#v, %#v, %#v, %#v", cfg.Output, cfg.Cache, cfg.Debug, cfg.PromptKit)
 	}
 	if cfg.Output.Directory == cfg.Cache.ChunkPlans.Directory || cfg.Cache.ChunkPlans.Directory == cfg.Cache.Checkpoints.Directory || cfg.Cache.Checkpoints.Directory == cfg.Debug.Directory {
 		t.Fatal("state roots were coupled")
@@ -310,11 +584,11 @@ debug:
 }
 
 func TestFileConfigCheckpointEnabledCanBeExplicitlyDisabled(t *testing.T) {
-	cfg := applyFileConfig(t, "version: 3\ncache:\n  checkpoints:\n    enabled: true\n")
+	cfg := applyFileConfig(t, "version: 4\ncache:\n  checkpoints:\n    enabled: true\n")
 	if !cfg.Cache.Checkpoints.Enabled || !cloneConfig(cfg).Cache.Checkpoints.Enabled {
 		t.Fatalf("enabled checkpoint config was not retained: %#v", cfg.Cache.Checkpoints)
 	}
-	file := parseFileConfig(t, "version: 3\ncache:\n  checkpoints:\n    enabled: false\n")
+	file := parseFileConfig(t, "version: 4\ncache:\n  checkpoints:\n    enabled: false\n")
 	if err := cfg.ApplyFileConfig(file); err != nil {
 		t.Fatal(err)
 	}
@@ -331,17 +605,17 @@ func TestFileConfigRejectsTrimmedKeyCollisions(t *testing.T) {
 	}{
 		{
 			name: "pipeline ids",
-			yaml: "version: 3\npipelines:\n  main: {}\n  ' main ': {}\n",
+			yaml: "version: 4\npipelines:\n  main: {}\n  ' main ': {}\n",
 			want: "pipeline id \"main\" is duplicated after trimming",
 		},
 		{
 			name: "lane ids",
-			yaml: "version: 3\npipelines:\n  main:\n    artifacts:\n      spells: {}\n      ' spells ': {}\n",
+			yaml: "version: 4\npipelines:\n  main:\n    artifacts:\n      spells: {}\n      ' spells ': {}\n",
 			want: "artifact lane id \"spells\" is duplicated after trimming",
 		},
 		{
 			name: "reference slots",
-			yaml: "version: 3\npipelines:\n  main:\n    references:\n      slot: ./one.txt\n      ' slot ': ./two.txt\n",
+			yaml: "version: 4\npipelines:\n  main:\n    references:\n      slot: ./one.txt\n      ' slot ': ./two.txt\n",
 			want: "reference slot \"slot\" is duplicated after trimming",
 		},
 	}
@@ -358,7 +632,7 @@ func TestFileConfigRejectsTrimmedKeyCollisions(t *testing.T) {
 }
 
 func TestFileConfigParsesOrderedStepsAndReferenceSources(t *testing.T) {
-	file := parseFileConfig(t, `version: 3
+	file := parseFileConfig(t, `version: 4
 pipelines:
   session:
     input: seriatim
@@ -397,7 +671,7 @@ func TestFileConfigRejectsAmbiguousReferenceSourceForms(t *testing.T) {
 		"artifact: {step: 1, lane: b}",
 		"1",
 	} {
-		_, err := ParseFileConfigYAML([]byte("version: 3\npipelines:\n  p:\n    input: text\n    references:\n      slot: " + source + "\n"))
+		_, err := ParseFileConfigYAML([]byte("version: 4\npipelines:\n  p:\n    input: text\n    references:\n      slot: " + source + "\n"))
 		if err == nil {
 			t.Fatalf("ParseFileConfigYAML(%q) error = nil", source)
 		}
@@ -412,12 +686,12 @@ func TestFileConfigRejectsEmptyAndAmbiguousPipelineShapes(t *testing.T) {
 	}{
 		{
 			name: "empty steps",
-			yaml: "version: 3\npipelines:\n  p:\n    input: text\n    steps: []\n",
+			yaml: "version: 4\npipelines:\n  p:\n    input: text\n    steps: []\n",
 			want: "at least one ordered step",
 		},
 		{
 			name: "both forms",
-			yaml: "version: 3\npipelines:\n  p:\n    input: text\n    artifacts: {}\n    steps: []\n",
+			yaml: "version: 4\npipelines:\n  p:\n    input: text\n    artifacts: {}\n    steps: []\n",
 			want: "both artifacts and steps",
 		},
 	}

internal/core/config/redaction_test.go

@@ -19,23 +19,28 @@ func TestRedactedResolvedPipelinePayloadRedactsEveryBinding(t *testing.T) {
 	}
 
 	resolved := pipeline.ResolvedPipeline{
-		ID:              "redaction-test",
-		Digest:          "sha256:safe-digest",
-		Input:           bindings["input"],
-		Chunk:           bindings["chunk"],
-		ChunkReferences: redactionTestReferenceTarget(pipeline.StageChunk, "", "chunk-reference-content"),
+		ID:                  "redaction-test",
+		Digest:              "sha256:safe-digest",
+		Input:               bindings["input"],
+		InputExecutionClass: contracts.ExecutionClassDeterministic,
+		Chunk:               bindings["chunk"],
+		ChunkExecutionClass: contracts.ExecutionClassLLMBacked,
+		ChunkReferences:     redactionTestReferenceTarget(pipeline.StageChunk, "", "chunk-reference-content"),
 		Steps: []pipeline.ResolvedPipelineStep{{
 			ID: "default",
 			ArtifactLanes: []pipeline.ResolvedArtifactLane{{
-				ID:                  "safe-lane",
-				ArtifactKind:        "safe/artifact",
-				Extract:             bindings["extract"],
-				Merge:               bindings["merge"],
-				Normalize:           bindings["normalize"],
-				Validators:          []pipeline.ModuleBinding{bindings["lane-validator"]},
-				ExtractReferences:   redactionTestReferenceTarget(pipeline.StageExtract, "safe-lane", "extract-reference-content"),
-				MergeReferences:     redactionTestReferenceTarget(pipeline.StageMerge, "safe-lane", "merge-reference-content"),
-				NormalizeReferences: redactionTestReferenceTarget(pipeline.StageNormalize, "safe-lane", "normalize-reference-content"),
+				ID:                      "safe-lane",
+				ArtifactKind:            "safe/artifact",
+				Extract:                 bindings["extract"],
+				ExtractExecutionClass:   contracts.ExecutionClassLLMBacked,
+				Merge:                   bindings["merge"],
+				MergeExecutionClass:     contracts.ExecutionClassDeterministic,
+				Normalize:               bindings["normalize"],
+				NormalizeExecutionClass: contracts.ExecutionClassLLMBacked,
+				Validators:              []pipeline.ModuleBinding{bindings["lane-validator"]},
+				ExtractReferences:       redactionTestReferenceTarget(pipeline.StageExtract, "safe-lane", "extract-reference-content"),
+				MergeReferences:         redactionTestReferenceTarget(pipeline.StageMerge, "safe-lane", "merge-reference-content"),
+				NormalizeReferences:     redactionTestReferenceTarget(pipeline.StageNormalize, "safe-lane", "normalize-reference-content"),
 			}},
 		}},
 		ValidatorChains: []pipeline.ResolvedValidatorChain{{
@@ -49,7 +54,8 @@ func TestRedactedResolvedPipelinePayloadRedactsEveryBinding(t *testing.T) {
 				ArtifactKind:   "safe/artifact",
 			}},
 		}},
-		Output: bindings["output"],
+		Output:               bindings["output"],
+		OutputExecutionClass: contracts.ExecutionClassDeterministic,
 	}
 	effective := EffectiveConfig{
 		Config: Config{Pipelines: map[string]pipeline.PipelineProfile{
@@ -88,6 +94,11 @@ func TestRedactedResolvedPipelinePayloadRedactsEveryBinding(t *testing.T) {
 			t.Fatalf("resolved pipeline summary does not retain %q: %s", safe, text)
 		}
 	}
+	for _, executionClass := range []string{"input_execution_class\":\"deterministic", "chunk_execution_class\":\"llm_backed", "extract_execution_class\":\"llm_backed", "merge_execution_class\":\"deterministic", "normalize_execution_class\":\"llm_backed", "output_execution_class\":\"deterministic"} {
+		if !strings.Contains(text, executionClass) {
+			t.Fatalf("resolved pipeline summary does not retain %q: %s", executionClass, text)
+		}
+	}
 
 	payload.Input.Options["safe"] = "mutated"
 	nested := payload.Input.Options["nested"].([]any)[0].([]any)[0].(map[string]any)

internal/core/config/validation.go

@@ -2,6 +2,7 @@ package config
 
 import (
 	"fmt"
+	"net/url"
 	"sort"
 	"strings"
 
@@ -10,7 +11,7 @@ import (
 
 func (c Config) Validate() error {
 	c.Concurrency.recomputeStageWorkerDefaults()
-	if err := validateScriptorium(c.Scriptorium); err != nil {
+	if err := validatePromptKit(c.PromptKit); err != nil {
 		return err
 	}
 	if err := validateStateSurfaces(c); err != nil {
@@ -49,9 +50,30 @@ func validateStageWorkers(cfg ConcurrencyConfig) error {
 	return nil
 }
 
-func validateScriptorium(cfg ScriptoriumConfig) error {
+func validatePromptKit(cfg PromptKitConfig) error {
 	if strings.TrimSpace(cfg.ProfileDir) != "" && strings.TrimSpace(cfg.ProfileFile) != "" {
-		return fmt.Errorf("scriptorium profile_dir and profile_file are mutually exclusive")
+		return fmt.Errorf("promptkit profile_dir and profile_file are mutually exclusive")
+	}
+	if cfg.LocalBackend == nil {
+		return nil
+	}
+	endpoint := strings.TrimSpace(cfg.LocalBackend.Endpoint)
+	if endpoint == "" {
+		return fmt.Errorf("promptkit.local_backend.endpoint must not be empty when set")
+	}
+	parsed, err := url.Parse(endpoint)
+	if err != nil ||
+		(!strings.EqualFold(parsed.Scheme, "http") && !strings.EqualFold(parsed.Scheme, "https")) ||
+		!parsed.IsAbs() ||
+		parsed.Hostname() == "" ||
+		parsed.User != nil ||
+		parsed.RawQuery != "" ||
+		parsed.ForceQuery ||
+		strings.Contains(endpoint, "#") {
+		return fmt.Errorf("promptkit.local_backend.endpoint must be an absolute HTTP or HTTPS URL with a host and no user information, query, or fragment")
+	}
+	if cfg.LocalBackend.ConcurrencyLimit < 0 {
+		return fmt.Errorf("promptkit.local_backend.concurrency_limit must not be negative")
 	}
 	return nil
 }

internal/core/config/validation_contract_test.go

@@ -87,10 +87,77 @@ func TestValidateConcurrencyRules(t *testing.T) {
 	}
 }
 
-func TestValidateScriptoriumSourcesAreMutuallyExclusive(t *testing.T) {
+func TestValidatePromptKitSourcesAreMutuallyExclusive(t *testing.T) {
 	cfg := Default()
-	cfg.Scriptorium = ScriptoriumConfig{ProfileDir: "./profiles", ProfileFile: "./profile.yml"}
-	assertValidationContains(t, cfg, "scriptorium profile_dir and profile_file are mutually exclusive")
+	cfg.PromptKit = PromptKitConfig{ProfileDir: "./profiles", ProfileFile: "./profile.yml"}
+	assertValidationContains(t, cfg, "promptkit profile_dir and profile_file are mutually exclusive")
+}
+
+func TestValidatePromptKitLocalBackendEndpoints(t *testing.T) {
+	tests := []struct {
+		name          string
+		endpoint      string
+		profileSource PromptKitConfig
+	}{
+		{
+			name:          "HTTP endpoint with path and profile directory",
+			endpoint:      "http://localhost:8000/v1",
+			profileSource: PromptKitConfig{ProfileDir: "./profiles"},
+		},
+		{
+			name:          "case-insensitive HTTPS endpoint and profile file",
+			endpoint:      "HTTPS://inference.example.test/api",
+			profileSource: PromptKitConfig{ProfileFile: "./profiles.yml"},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := Default()
+			cfg.PromptKit = tt.profileSource
+			cfg.PromptKit.LocalBackend = &PromptKitLocalBackendConfig{
+				Endpoint:         tt.endpoint,
+				ConcurrencyLimit: 2,
+			}
+			if err := cfg.Validate(); err != nil {
+				t.Fatalf("Validate() error = %v", err)
+			}
+		})
+	}
+}
+
+func TestValidatePromptKitLocalBackendRejectsInvalidValues(t *testing.T) {
+	tests := []struct {
+		name             string
+		endpoint         string
+		concurrencyLimit int
+		want             string
+	}{
+		{name: "blank endpoint", endpoint: "   ", want: "promptkit.local_backend.endpoint"},
+		{name: "relative URL", endpoint: "localhost:8000/v1", want: "promptkit.local_backend.endpoint"},
+		{name: "unsupported scheme", endpoint: "ftp://localhost/model", want: "promptkit.local_backend.endpoint"},
+		{name: "missing host", endpoint: "http:///v1", want: "promptkit.local_backend.endpoint"},
+		{name: "user information", endpoint: "http://user:secret@localhost/v1", want: "promptkit.local_backend.endpoint"},
+		{name: "query", endpoint: "http://localhost/v1?model=example", want: "promptkit.local_backend.endpoint"},
+		{name: "empty query", endpoint: "http://localhost/v1?", want: "promptkit.local_backend.endpoint"},
+		{name: "fragment", endpoint: "http://localhost/v1#model", want: "promptkit.local_backend.endpoint"},
+		{name: "empty fragment", endpoint: "http://localhost/v1#", want: "promptkit.local_backend.endpoint"},
+		{
+			name:             "negative concurrency",
+			endpoint:         "http://localhost:8000/v1",
+			concurrencyLimit: -1,
+			want:             "promptkit.local_backend.concurrency_limit",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := Default()
+			cfg.PromptKit.LocalBackend = &PromptKitLocalBackendConfig{
+				Endpoint:         tt.endpoint,
+				ConcurrencyLimit: tt.concurrencyLimit,
+			}
+			assertValidationContains(t, cfg, tt.want)
+		})
+	}
 }
 
 func TestValidateStateSurfaceRules(t *testing.T) {

internal/core/debugbundle/bundle_test.go

@@ -2,6 +2,7 @@ package debugbundle
 
 import (
 	"bytes"
+	"encoding/json"
 	"os"
 	"path/filepath"
 	"strings"
@@ -137,6 +138,47 @@ func TestSummaryWriterWritesEverySummaryArtifact(t *testing.T) {
 		}
 	}
 }
+
+func TestWriteInvocationPreservesReasoningEffortOverrideStates(t *testing.T) {
+	replacement := "focused"
+	cleared := ""
+	tests := []struct {
+		name      string
+		override  *string
+		wantValue string
+		wantSet   bool
+	}{
+		{name: "inherit"},
+		{name: "replace", override: &replacement, wantValue: "focused", wantSet: true},
+		{name: "clear", override: &cleared, wantSet: true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			bundle, err := Allocate(t.TempDir(), testBundleRunID, time.Unix(0, 42))
+			if err != nil {
+				t.Fatal(err)
+			}
+			if err := bundle.Summary().WriteInvocation(Invocation{
+				Operation:               "run",
+				ReasoningEffortOverride: tt.override,
+			}); err != nil {
+				t.Fatal(err)
+			}
+			data, err := os.ReadFile(filepath.Join(bundle.SummaryRoot(), ArtifactInvocationMetadata))
+			if err != nil {
+				t.Fatal(err)
+			}
+			var payload map[string]any
+			if err := json.Unmarshal(data, &payload); err != nil {
+				t.Fatal(err)
+			}
+			value, found := payload["reasoning_effort_override"]
+			if found != tt.wantSet || (found && value != tt.wantValue) {
+				t.Fatalf("reasoning override found=%t value=%#v, want found=%t value=%q; JSON=%s", found, value, tt.wantSet, tt.wantValue, data)
+			}
+		})
+	}
+}
 func TestSummaryWriterInternalWritesConfineArtifacts(t *testing.T) {
 	bundle, err := Allocate(t.TempDir(), testBundleRunID, time.Unix(0, 42))
 	if err != nil {

internal/core/debugbundle/summary.go

@@ -29,18 +29,19 @@ type RedactedResolvedPipelinePayload interface {
 	RedactedResolvedPipelinePayload() pipeline.ResolvedPipeline
 }
 type Invocation struct {
-	Operation          string    `json:"operation"`
-	PipelineID         string    `json:"pipeline_id,omitempty"`
-	PipelineDigest     string    `json:"pipeline_digest,omitempty"`
-	Resume             bool      `json:"resume,omitempty"`
-	RecomputeStep      string    `json:"recompute_step,omitempty"`
-	InputPath          string    `json:"input_path,omitempty"`
-	ConfigPath         string    `json:"config_path,omitempty"`
-	ConfigSource       string    `json:"config_source,omitempty"`
-	OnlyLanes          []string  `json:"only_lanes,omitempty"`
-	ChunkCacheOverride string    `json:"chunk_cache_override,omitempty"`
-	RunID              string    `json:"run_id"`
-	StartedAt          time.Time `json:"started_at"`
+	Operation               string    `json:"operation"`
+	PipelineID              string    `json:"pipeline_id,omitempty"`
+	PipelineDigest          string    `json:"pipeline_digest,omitempty"`
+	Resume                  bool      `json:"resume,omitempty"`
+	RecomputeStep           string    `json:"recompute_step,omitempty"`
+	InputPath               string    `json:"input_path,omitempty"`
+	ConfigPath              string    `json:"config_path,omitempty"`
+	ConfigSource            string    `json:"config_source,omitempty"`
+	OnlyLanes               []string  `json:"only_lanes,omitempty"`
+	ChunkCacheOverride      string    `json:"chunk_cache_override,omitempty"`
+	ReasoningEffortOverride *string   `json:"reasoning_effort_override,omitempty"`
+	RunID                   string    `json:"run_id"`
+	StartedAt               time.Time `json:"started_at"`
 }
 type RunReport struct {
 	RunID            string `json:"run_id"`
@@ -68,6 +69,10 @@ func (w *SummaryWriter) WriteInvocation(payload Invocation) error {
 	if payload.StartedAt.IsZero() {
 		payload.StartedAt = w.createdAt
 	}
+	if payload.ReasoningEffortOverride != nil {
+		value := *payload.ReasoningEffortOverride
+		payload.ReasoningEffortOverride = &value
+	}
 	return w.writeJSON(ArtifactInvocationMetadata, payload)
 }
 func (w *SummaryWriter) WriteRedactedEffectiveConfig(payload RedactedSummaryPayload) error {

internal/framework/contracts/contracts.go

@@ -43,6 +43,7 @@ type LLMDebugPrompt struct {
 	PromptVersion        string            `json:"prompt_version,omitempty"`
 	PromptHash           string            `json:"prompt_hash,omitempty"`
 	SelectedProfileID    string            `json:"selected_profile_id,omitempty"`
+	SelectedBackendID    string            `json:"selected_backend_id,omitempty"`
 	SessionID            string            `json:"session_id,omitempty"`
 	RenderedPromptHash   string            `json:"rendered_prompt_hash,omitempty"`
 	Messages             []LLMDebugMessage `json:"messages,omitempty"`
@@ -266,10 +267,6 @@ const (
 	ExecutionClassLLMBacked     ExecutionClass = "llm_backed"
 )
 
-type ChunkExecutionClassProvider interface {
-	ExecutionClass() ExecutionClass
-}
-
 type ValidationResult struct {
 	Approved               bool      `json:"approved"`
 	ReasonCode             string    `json:"reason_code,omitempty"`
@@ -292,6 +289,7 @@ type OutputRequest struct {
 	LLMProfile       string                `json:"llm_profile,omitempty"`
 	Metadata         map[string]any        `json:"metadata,omitempty"`
 	ChunkMap         *SerializedArtifact   `json:"chunk_map,omitempty"`
+	EvidenceContext  *SerializedArtifact   `json:"evidence_context,omitempty"`
 }
 
 type OutputFile struct {

internal/framework/contracts/errors.go

@@ -5,3 +5,7 @@ import "errors"
 // ErrInvalidStructuredOutput identifies a provider response that cannot satisfy
 // the caller's declared structured-output contract.
 var ErrInvalidStructuredOutput = errors.New("invalid structured output")
+
+// ErrLLMCapacityExceeded identifies backend admission exhaustion before model
+// generation begins.
+var ErrLLMCapacityExceeded = errors.New("LLM capacity exceeded")

internal/framework/evidencecontext/assets/schemas/source_evidence_context.v1.json

@@ -0,0 +1,66 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "notarius.source.evidence_context",
+  "title": "notarius_source_evidence_context_v1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["source_id", "source_digest", "window_units", "selected_lanes", "contexts"],
+  "properties": {
+    "source_id": {"type": "string", "minLength": 1},
+    "source_digest": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
+    "window_units": {"type": "integer", "minimum": 0},
+    "selected_lanes": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "minLength": 1}
+    },
+    "contexts": {
+      "type": "array",
+      "items": {"$ref": "#/$defs/context"}
+    }
+  },
+  "$defs": {
+    "source_ref": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["source_id", "start_unit_id", "end_unit_id"],
+      "properties": {
+        "source_id": {"type": "string", "minLength": 1},
+        "start_unit_id": {"type": "integer", "minimum": 1},
+        "end_unit_id": {"type": "integer", "minimum": 1}
+      }
+    },
+    "unit": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["id", "kind", "text", "ref"],
+      "properties": {
+        "id": {"type": "integer", "minimum": 1},
+        "kind": {"type": "string", "minLength": 1},
+        "text": {"type": "string", "minLength": 1},
+        "ref": {"$ref": "#/$defs/source_ref"},
+        "metadata": {"type": "object", "additionalProperties": true}
+      }
+    },
+    "evidence_ref": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["lane_id", "source_ref"],
+      "properties": {
+        "lane_id": {"type": "string", "minLength": 1},
+        "source_ref": {"$ref": "#/$defs/source_ref"}
+      }
+    },
+    "context": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["context_ref", "evidence_refs", "units"],
+      "properties": {
+        "context_ref": {"$ref": "#/$defs/source_ref"},
+        "evidence_refs": {"type": "array", "minItems": 1, "items": {"$ref": "#/$defs/evidence_ref"}},
+        "units": {"type": "array", "minItems": 1, "items": {"$ref": "#/$defs/unit"}}
+      }
+    }
+  }
+}

internal/framework/evidencecontext/build.go

@@ -0,0 +1,196 @@
+package evidencecontext
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+)
+
+type contribution struct {
+	laneID   string
+	ref      source.SourceRef
+	startPos int
+	endPos   int
+}
+
+type expandedRange struct {
+	startPos      int
+	endPos        int
+	contributions []contribution
+}
+
+// Build validates accepted direct references, expands them by source-document
+// position, and returns their deterministic context union.
+func Build(request BuildRequest) (Document, error) {
+	if request.WindowUnits < 0 {
+		return Document{}, fmt.Errorf("window_units must not be negative")
+	}
+	lanes, err := normalizeSelectedLanes(request.SelectedLanes)
+	if err != nil {
+		return Document{}, err
+	}
+	if err := source.ValidateDocument(request.Source); err != nil {
+		return Document{}, fmt.Errorf("validate source document: %w", err)
+	}
+	digest, err := source.DigestDocument(request.Source)
+	if err != nil {
+		return Document{}, fmt.Errorf("digest source document: %w", err)
+	}
+	if digest != request.Source.Digest {
+		return Document{}, fmt.Errorf("source digest does not match source document digest")
+	}
+
+	selected := make(map[string]struct{}, len(lanes))
+	for _, laneID := range lanes {
+		selected[laneID] = struct{}{}
+	}
+	index := source.NewDocumentIndex(request.Source)
+	seen := make(map[evidenceKey]struct{})
+	contributions := make([]contribution, 0)
+	for laneIndex, laneEvidence := range request.LaneEvidence {
+		laneID := strings.TrimSpace(laneEvidence.LaneID)
+		if _, ok := selected[laneID]; !ok {
+			return Document{}, fmt.Errorf("lane evidence[%d] lane %q is not selected", laneIndex, laneID)
+		}
+		for refIndex, ref := range laneEvidence.SourceRefs {
+			if err := index.ValidateRef(ref); err != nil {
+				return Document{}, fmt.Errorf("lane %q source reference[%d]: %w", laneID, refIndex, err)
+			}
+			key := evidenceKey{laneID: laneID, ref: ref}
+			if _, exists := seen[key]; exists {
+				continue
+			}
+			seen[key] = struct{}{}
+			startPos, _ := index.Position(ref.StartUnitID)
+			endPos, _ := index.Position(ref.EndUnitID)
+			contributions = append(contributions, contribution{laneID: laneID, ref: ref, startPos: expandStart(startPos, request.WindowUnits), endPos: expandEnd(endPos, len(request.Source.Units), request.WindowUnits)})
+		}
+	}
+
+	sort.Slice(contributions, func(i, j int) bool { return lessContribution(contributions[i], contributions[j]) })
+	document := Document{
+		SourceID:      request.Source.ID,
+		SourceDigest:  digest,
+		WindowUnits:   request.WindowUnits,
+		SelectedLanes: lanes,
+		Contexts:      make([]Context, 0),
+	}
+	for _, rangeValue := range mergeRanges(contributions) {
+		context, err := buildContext(request.Source, rangeValue)
+		if err != nil {
+			return Document{}, err
+		}
+		document.Contexts = append(document.Contexts, context)
+	}
+	canonical, err := canonicalize(document)
+	if err != nil {
+		return Document{}, fmt.Errorf("validate evidence context: %w", err)
+	}
+	return clone(canonical)
+}
+
+type evidenceKey struct {
+	laneID string
+	ref    source.SourceRef
+}
+
+func normalizeSelectedLanes(values []string) ([]string, error) {
+	if len(values) == 0 {
+		return nil, fmt.Errorf("selected_lanes must not be empty")
+	}
+	seen := make(map[string]struct{}, len(values))
+	lanes := make([]string, 0, len(values))
+	for index, raw := range values {
+		laneID := strings.TrimSpace(raw)
+		if laneID == "" {
+			return nil, fmt.Errorf("selected_lanes[%d] must not be empty", index)
+		}
+		if _, exists := seen[laneID]; exists {
+			return nil, fmt.Errorf("selected_lanes lane %q is duplicated", laneID)
+		}
+		seen[laneID] = struct{}{}
+		lanes = append(lanes, laneID)
+	}
+	sort.Strings(lanes)
+	return lanes, nil
+}
+
+func expandStart(position, window int) int {
+	if window > position {
+		return 0
+	}
+	return position - window
+}
+
+func expandEnd(position, length, window int) int {
+	last := length - 1
+	if window > last-position {
+		return last
+	}
+	return position + window
+}
+
+func lessContribution(left, right contribution) bool {
+	if left.startPos != right.startPos {
+		return left.startPos < right.startPos
+	}
+	if left.endPos != right.endPos {
+		return left.endPos < right.endPos
+	}
+	return lessEvidenceRef(EvidenceRef{LaneID: left.laneID, SourceRef: left.ref}, EvidenceRef{LaneID: right.laneID, SourceRef: right.ref})
+}
+
+func mergeRanges(values []contribution) []expandedRange {
+	if len(values) == 0 {
+		return nil
+	}
+	ranges := make([]expandedRange, 0, len(values))
+	for _, value := range values {
+		if len(ranges) == 0 || value.startPos > ranges[len(ranges)-1].endPos+1 {
+			ranges = append(ranges, expandedRange{startPos: value.startPos, endPos: value.endPos, contributions: []contribution{value}})
+			continue
+		}
+		current := &ranges[len(ranges)-1]
+		if value.endPos > current.endPos {
+			current.endPos = value.endPos
+		}
+		current.contributions = append(current.contributions, value)
+	}
+	return ranges
+}
+
+func buildContext(document *source.SourceDocument, value expandedRange) (Context, error) {
+	evidenceRefs := make([]EvidenceRef, 0, len(value.contributions))
+	for _, contribution := range value.contributions {
+		evidenceRefs = append(evidenceRefs, EvidenceRef{LaneID: contribution.laneID, SourceRef: contribution.ref})
+	}
+	sort.Slice(evidenceRefs, func(i, j int) bool { return lessEvidenceRef(evidenceRefs[i], evidenceRefs[j]) })
+	units := make([]source.SourceUnit, 0, value.endPos-value.startPos+1)
+	for position := value.startPos; position <= value.endPos; position++ {
+		unit, err := cloneSourceUnit(document.Units[position])
+		if err != nil {
+			return Context{}, fmt.Errorf("clone source unit at position %d: %w", position, err)
+		}
+		units = append(units, unit)
+	}
+	return Context{
+		ContextRef:   source.SourceRef{SourceID: document.ID, StartUnitID: units[0].ID, EndUnitID: units[len(units)-1].ID},
+		EvidenceRefs: evidenceRefs,
+		Units:        units,
+	}, nil
+}
+
+func lessEvidenceRef(left, right EvidenceRef) bool {
+	if left.LaneID != right.LaneID {
+		return left.LaneID < right.LaneID
+	}
+	if left.SourceRef.SourceID != right.SourceRef.SourceID {
+		return left.SourceRef.SourceID < right.SourceRef.SourceID
+	}
+	if left.SourceRef.StartUnitID != right.SourceRef.StartUnitID {
+		return left.SourceRef.StartUnitID < right.SourceRef.StartUnitID
+	}
+	return left.SourceRef.EndUnitID < right.SourceRef.EndUnitID
+}

internal/framework/evidencecontext/codec.go

@@ -0,0 +1,329 @@
+package evidencecontext
+
+import (
+	"bytes"
+	"embed"
+	"encoding/json"
+	"fmt"
+	"io"
+	"regexp"
+	"strings"
+	"sync"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"github.com/santhosh-tekuri/jsonschema/v6"
+)
+
+//go:embed assets/schemas/source_evidence_context.v1.json
+var schemaAssets embed.FS
+
+var digestPattern = regexp.MustCompile(`^sha256:[0-9a-f]{64}$`)
+
+var (
+	loadSchemaOnce sync.Once
+	loadedSchema   []byte
+	compiledSchema *jsonschema.Schema
+	loadSchemaErr  error
+)
+
+// Codec owns strict serialization for the durable evidence-context contract.
+type Codec struct{}
+
+func New() *Codec { return &Codec{} }
+
+func (c *Codec) Kind() contracts.ArtifactKind { return ArtifactKind }
+
+func (c *Codec) Schema() contracts.ArtifactSchema {
+	raw, err := c.schemaBytes()
+	if err != nil {
+		return contracts.ArtifactSchema{}
+	}
+	return contracts.ArtifactSchema{ID: SchemaID, Name: SchemaName, Version: SchemaVersion, JSONSchema: raw}
+}
+
+func (c *Codec) MediaType() string { return MediaType }
+
+// Serialize builds and encodes the framework-owned serialized artifact.
+func Serialize(request BuildRequest) (contracts.SerializedArtifact, error) {
+	value, err := Build(request)
+	if err != nil {
+		return contracts.SerializedArtifact{}, err
+	}
+	codec := New()
+	content, err := codec.Encode(value)
+	if err != nil {
+		return contracts.SerializedArtifact{}, err
+	}
+	return contracts.SerializedArtifact{Kind: ArtifactKind, Schema: codec.Schema(), MediaType: MediaType, Content: content}, nil
+}
+
+func (c *Codec) Encode(value Document) ([]byte, error) {
+	if _, err := c.schemaBytes(); err != nil {
+		return nil, err
+	}
+	canonical, err := canonicalize(value)
+	if err != nil {
+		return nil, fmt.Errorf("encode evidence context: %w", err)
+	}
+	content, err := json.Marshal(canonical)
+	if err != nil {
+		return nil, fmt.Errorf("encode evidence context: %w", err)
+	}
+	if err := validateSchemaInstance(content); err != nil {
+		return nil, fmt.Errorf("encode evidence context: %w", err)
+	}
+	return content, nil
+}
+
+func (c *Codec) Decode(content []byte) (Document, error) {
+	if _, err := c.schemaBytes(); err != nil {
+		return Document{}, err
+	}
+	if err := validateSchemaInstance(content); err != nil {
+		return Document{}, fmt.Errorf("decode evidence context: %w", err)
+	}
+	decoder := json.NewDecoder(bytes.NewReader(content))
+	decoder.DisallowUnknownFields()
+	var value Document
+	if err := decoder.Decode(&value); err != nil {
+		return Document{}, fmt.Errorf("decode evidence context: %w", err)
+	}
+	var trailing any
+	if err := decoder.Decode(&trailing); err != io.EOF {
+		return Document{}, fmt.Errorf("decode evidence context: multiple JSON values")
+	}
+	canonical, err := canonicalize(value)
+	if err != nil {
+		return Document{}, fmt.Errorf("decode evidence context: %w", err)
+	}
+	return clone(canonical)
+}
+
+func (c *Codec) schemaBytes() ([]byte, error) {
+	loadSchemaOnce.Do(loadAndCompileSchema)
+	if loadSchemaErr != nil {
+		return nil, loadSchemaErr
+	}
+	return append([]byte(nil), loadedSchema...), nil
+}
+
+func loadAndCompileSchema() {
+	raw, err := schemaAssets.ReadFile("assets/schemas/source_evidence_context.v1.json")
+	if err != nil {
+		loadSchemaErr = fmt.Errorf("read source evidence context schema: %w", err)
+		return
+	}
+	var identity struct {
+		ID       string   `json:"$id"`
+		Title    string   `json:"title"`
+		Type     string   `json:"type"`
+		Required []string `json:"required"`
+	}
+	if err := json.Unmarshal(raw, &identity); err != nil {
+		loadSchemaErr = fmt.Errorf("decode source evidence context schema: %w", err)
+		return
+	}
+	if identity.ID != SchemaID || identity.Title != SchemaName || identity.Type != "object" || !hasRequiredFields(identity.Required) {
+		loadSchemaErr = fmt.Errorf("source evidence context schema identity or required fields are invalid")
+		return
+	}
+	schemaDocument, err := jsonschema.UnmarshalJSON(bytes.NewReader(raw))
+	if err != nil {
+		loadSchemaErr = fmt.Errorf("parse source evidence context schema: %w", err)
+		return
+	}
+	compiler := jsonschema.NewCompiler()
+	if err := compiler.AddResource("source-evidence-context-schema.json", schemaDocument); err != nil {
+		loadSchemaErr = fmt.Errorf("load source evidence context schema: %w", err)
+		return
+	}
+	compiled, err := compiler.Compile("source-evidence-context-schema.json")
+	if err != nil {
+		loadSchemaErr = fmt.Errorf("compile source evidence context schema: %w", err)
+		return
+	}
+	loadedSchema = append([]byte(nil), raw...)
+	compiledSchema = compiled
+}
+
+func validateSchemaInstance(content []byte) error {
+	instance, err := jsonschema.UnmarshalJSON(bytes.NewReader(content))
+	if err != nil {
+		return fmt.Errorf("payload is not valid JSON: %w", err)
+	}
+	if err := compiledSchema.Validate(instance); err != nil {
+		return fmt.Errorf("payload does not conform to source evidence context schema: %w", err)
+	}
+	return nil
+}
+
+func hasRequiredFields(required []string) bool {
+	want := map[string]bool{"source_id": true, "source_digest": true, "window_units": true, "selected_lanes": true, "contexts": true}
+	for _, field := range required {
+		delete(want, field)
+	}
+	return len(want) == 0
+}
+
+func canonicalize(value Document) (Document, error) {
+	owned, err := clone(value)
+	if err != nil {
+		return Document{}, err
+	}
+	value = owned
+	if err := requireIdentity("source_id", value.SourceID); err != nil {
+		return Document{}, err
+	}
+	if !digestPattern.MatchString(value.SourceDigest) {
+		return Document{}, fmt.Errorf("source_digest must be a sha256 digest")
+	}
+	if value.WindowUnits < 0 {
+		return Document{}, fmt.Errorf("window_units must not be negative")
+	}
+	if err := validateSelectedLanes(value.SelectedLanes); err != nil {
+		return Document{}, err
+	}
+	if value.Contexts == nil {
+		value.Contexts = make([]Context, 0)
+	}
+	selected := make(map[string]struct{}, len(value.SelectedLanes))
+	for _, laneID := range value.SelectedLanes {
+		selected[laneID] = struct{}{}
+	}
+	seenUnits := make(map[int]struct{})
+	for contextIndex := range value.Contexts {
+		context, err := canonicalizeContext(value.SourceID, selected, seenUnits, value.Contexts[contextIndex], contextIndex)
+		if err != nil {
+			return Document{}, err
+		}
+		value.Contexts[contextIndex] = context
+	}
+	return value, nil
+}
+
+func validateSelectedLanes(lanes []string) error {
+	if len(lanes) == 0 {
+		return fmt.Errorf("selected_lanes must not be empty")
+	}
+	for index, laneID := range lanes {
+		if err := requireIdentity(fmt.Sprintf("selected_lanes[%d]", index), laneID); err != nil {
+			return err
+		}
+		if index > 0 && lanes[index-1] >= laneID {
+			return fmt.Errorf("selected_lanes must be unique and in lexical order")
+		}
+	}
+	return nil
+}
+
+func canonicalizeContext(sourceID string, selected map[string]struct{}, seenUnits map[int]struct{}, value Context, contextIndex int) (Context, error) {
+	prefix := fmt.Sprintf("contexts[%d]", contextIndex)
+	if len(value.EvidenceRefs) == 0 {
+		return Context{}, fmt.Errorf("%s.evidence_refs must not be empty", prefix)
+	}
+	if len(value.Units) == 0 {
+		return Context{}, fmt.Errorf("%s.units must not be empty", prefix)
+	}
+	if err := validateRefIdentity(sourceID, value.ContextRef, prefix+".context_ref"); err != nil {
+		return Context{}, err
+	}
+	positions := make(map[int]int, len(value.Units))
+	for unitIndex := range value.Units {
+		unit, err := cloneSourceUnit(value.Units[unitIndex])
+		if err != nil {
+			return Context{}, fmt.Errorf("%s.units[%d]: %w", prefix, unitIndex, err)
+		}
+		if unit.ID <= 0 || strings.TrimSpace(unit.Kind) == "" || strings.TrimSpace(unit.Text) == "" {
+			return Context{}, fmt.Errorf("%s.units[%d] has invalid required fields", prefix, unitIndex)
+		}
+		if err := validateRefIdentity(sourceID, unit.Ref, fmt.Sprintf("%s.units[%d].ref", prefix, unitIndex)); err != nil {
+			return Context{}, err
+		}
+		if unit.Ref.StartUnitID != unit.ID || unit.Ref.EndUnitID != unit.ID {
+			return Context{}, fmt.Errorf("%s.units[%d].ref must identify unit id %d", prefix, unitIndex, unit.ID)
+		}
+		if _, exists := positions[unit.ID]; exists {
+			return Context{}, fmt.Errorf("%s.units contains duplicate unit id %d", prefix, unit.ID)
+		}
+		if _, exists := seenUnits[unit.ID]; exists {
+			return Context{}, fmt.Errorf("contexts contain duplicate unit id %d", unit.ID)
+		}
+		positions[unit.ID] = unitIndex
+		seenUnits[unit.ID] = struct{}{}
+		value.Units[unitIndex] = unit
+	}
+	if value.ContextRef.StartUnitID != value.Units[0].ID || value.ContextRef.EndUnitID != value.Units[len(value.Units)-1].ID {
+		return Context{}, fmt.Errorf("%s.context_ref must identify the first and last units", prefix)
+	}
+	for evidenceIndex := range value.EvidenceRefs {
+		evidence := value.EvidenceRefs[evidenceIndex]
+		if _, ok := selected[evidence.LaneID]; !ok {
+			return Context{}, fmt.Errorf("%s.evidence_refs[%d].lane_id is not selected", prefix, evidenceIndex)
+		}
+		if err := requireIdentity(fmt.Sprintf("%s.evidence_refs[%d].lane_id", prefix, evidenceIndex), evidence.LaneID); err != nil {
+			return Context{}, err
+		}
+		if err := validateRefIdentity(sourceID, evidence.SourceRef, fmt.Sprintf("%s.evidence_refs[%d].source_ref", prefix, evidenceIndex)); err != nil {
+			return Context{}, err
+		}
+		start, startOK := positions[evidence.SourceRef.StartUnitID]
+		end, endOK := positions[evidence.SourceRef.EndUnitID]
+		if !startOK || !endOK || start > end {
+			return Context{}, fmt.Errorf("%s.evidence_refs[%d].source_ref is outside context units", prefix, evidenceIndex)
+		}
+		if evidenceIndex > 0 && !lessEvidenceRef(value.EvidenceRefs[evidenceIndex-1], evidence) {
+			return Context{}, fmt.Errorf("%s.evidence_refs must be unique and in deterministic order", prefix)
+		}
+	}
+	return value, nil
+}
+
+func validateRefIdentity(sourceID string, ref source.SourceRef, field string) error {
+	if ref.SourceID != sourceID {
+		return fmt.Errorf("%s.source_id does not match source_id", field)
+	}
+	if ref.StartUnitID <= 0 || ref.EndUnitID <= 0 {
+		return fmt.Errorf("%s endpoints must be positive", field)
+	}
+	return nil
+}
+
+func requireIdentity(field, value string) error {
+	if strings.TrimSpace(value) == "" || strings.TrimSpace(value) != value {
+		return fmt.Errorf("%s must be a non-empty trimmed string", field)
+	}
+	return nil
+}
+
+func clone(value Document) (Document, error) {
+	value.SelectedLanes = append([]string(nil), value.SelectedLanes...)
+	if value.Contexts == nil {
+		value.Contexts = make([]Context, 0)
+	} else {
+		contexts := make([]Context, len(value.Contexts))
+		for contextIndex, context := range value.Contexts {
+			contexts[contextIndex].ContextRef = context.ContextRef
+			contexts[contextIndex].EvidenceRefs = append([]EvidenceRef(nil), context.EvidenceRefs...)
+			contexts[contextIndex].Units = make([]source.SourceUnit, len(context.Units))
+			for unitIndex, unit := range context.Units {
+				cloned, err := cloneSourceUnit(unit)
+				if err != nil {
+					return Document{}, fmt.Errorf("clone contexts[%d].units[%d]: %w", contextIndex, unitIndex, err)
+				}
+				contexts[contextIndex].Units[unitIndex] = cloned
+			}
+		}
+		value.Contexts = contexts
+	}
+	return value, nil
+}
+
+func cloneSourceUnit(unit source.SourceUnit) (source.SourceUnit, error) {
+	metadata, err := source.CloneMetadata(unit.Metadata)
+	if err != nil {
+		return source.SourceUnit{}, fmt.Errorf("clone metadata: %w", err)
+	}
+	unit.Metadata = metadata
+	return unit, nil
+}

internal/framework/evidencecontext/evidencecontext_test.go

@@ -0,0 +1,350 @@
+package evidencecontext
+
+import (
+	"bytes"
+	"encoding/json"
+	"math"
+	"os"
+	"reflect"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+)
+
+func TestBuildExpandsAndMergesEvidenceByDocumentPosition(t *testing.T) {
+	for _, test := range []struct {
+		name      string
+		window    int
+		evidence  []LaneEvidence
+		wantUnits [][]int
+		wantRefs  [][]EvidenceRef
+	}{
+		{
+			name:      "zero window",
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}},
+			wantUnits: [][]int{{3}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(3, 3)}}},
+		},
+		{
+			name:      "non monotonic ids use positions and clip boundaries",
+			window:    1,
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}},
+			wantUnits: [][]int{{10, 3, 30}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(3, 3)}}},
+		},
+		{
+			name:      "separate gaps stay separate",
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(10, 10), ref(50, 50)}}},
+			wantUnits: [][]int{{10}, {50}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(10, 10)}}, {{LaneID: "npcs", SourceRef: ref(50, 50)}}},
+		},
+		{
+			name:      "overlapping windows merge",
+			window:    1,
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3), ref(30, 30)}}},
+			wantUnits: [][]int{{10, 3, 30, 7}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(3, 3)}, {LaneID: "npcs", SourceRef: ref(30, 30)}}},
+		},
+		{
+			name:      "contiguous windows merge",
+			window:    1,
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(10, 10), ref(7, 7)}}},
+			wantUnits: [][]int{{10, 3, 30, 7, 50}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(7, 7)}, {LaneID: "npcs", SourceRef: ref(10, 10)}}},
+		},
+		{
+			name: "duplicate contributions retain unique lane attribution",
+			evidence: []LaneEvidence{
+				{LaneID: "spells", SourceRefs: []source.SourceRef{ref(30, 30), ref(30, 30)}},
+				{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(30, 30)}},
+			},
+			wantUnits: [][]int{{30}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(30, 30)}, {LaneID: "spells", SourceRef: ref(30, 30)}}},
+		},
+		{
+			name:      "empty contributions retain explicit empty contexts",
+			evidence:  []LaneEvidence{{LaneID: "npcs"}},
+			wantUnits: [][]int{},
+			wantRefs:  [][]EvidenceRef{},
+		},
+		{
+			name:      "largest window clips without overflow",
+			window:    math.MaxInt,
+			evidence:  []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(30, 30)}}},
+			wantUnits: [][]int{{10, 3, 30, 7, 50}},
+			wantRefs:  [][]EvidenceRef{{{LaneID: "npcs", SourceRef: ref(30, 30)}}},
+		},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			document := testDocument(t)
+			got, err := Build(BuildRequest{Source: document, WindowUnits: test.window, SelectedLanes: []string{"spells", "npcs"}, LaneEvidence: test.evidence})
+			if err != nil {
+				t.Fatalf("Build() error = %v", err)
+			}
+			if want := []string{"npcs", "spells"}; !reflect.DeepEqual(got.SelectedLanes, want) {
+				t.Fatalf("SelectedLanes = %#v, want %#v", got.SelectedLanes, want)
+			}
+			if got.WindowUnits != test.window || got.SourceID != document.ID || got.SourceDigest != document.Digest {
+				t.Fatalf("Build() identity = %#v, want source and window identity", got)
+			}
+			if actual := contextUnitIDs(got.Contexts); !reflect.DeepEqual(actual, test.wantUnits) {
+				t.Fatalf("context unit ids = %#v, want %#v", actual, test.wantUnits)
+			}
+			if actual := contextEvidenceRefs(got.Contexts); !reflect.DeepEqual(actual, test.wantRefs) {
+				t.Fatalf("context evidence refs = %#v, want %#v", actual, test.wantRefs)
+			}
+		})
+	}
+}
+
+func TestBuildIsStableAndOwnsSourceAndInputs(t *testing.T) {
+	document := testDocument(t)
+	refs := []source.SourceRef{ref(30, 30), ref(3, 3)}
+	request := BuildRequest{
+		Source:        document,
+		WindowUnits:   1,
+		SelectedLanes: []string{"spells", "npcs"},
+		LaneEvidence:  []LaneEvidence{{LaneID: "spells", SourceRefs: refs}, {LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}},
+	}
+	first, err := Build(request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	secondRequest := request
+	secondRequest.LaneEvidence = []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}, {LaneID: "spells", SourceRefs: []source.SourceRef{ref(3, 3), ref(30, 30)}}}
+	second, err := Build(secondRequest)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !reflect.DeepEqual(first, second) {
+		t.Fatalf("Build() order differs:\nfirst: %#v\nsecond: %#v", first, second)
+	}
+	first.SelectedLanes[0] = "changed"
+	first.Contexts[0].Units[0].Metadata["nested"].(map[string]any)["value"] = "changed"
+	if document.Units[0].Metadata["nested"].(map[string]any)["value"] != "original" {
+		t.Fatal("Build() returned metadata aliases to source document")
+	}
+	document.Units[0].Metadata["nested"].(map[string]any)["value"] = "later"
+	if second.Contexts[0].Units[0].Metadata["nested"].(map[string]any)["value"] != "original" {
+		t.Fatal("Build() retained metadata aliases to source document")
+	}
+	refs[0].StartUnitID = 999
+	if !containsEvidenceRef(second.Contexts[0].EvidenceRefs, ref(30, 30)) {
+		t.Fatal("Build() retained source-reference input aliases")
+	}
+}
+
+func TestBuildRejectsInvalidInputs(t *testing.T) {
+	for _, test := range []struct {
+		name   string
+		mutate func(*BuildRequest)
+		want   string
+	}{
+		{name: "negative window", mutate: func(request *BuildRequest) { request.WindowUnits = -1 }, want: "window_units"},
+		{name: "blank selected lane", mutate: func(request *BuildRequest) { request.SelectedLanes = []string{" "} }, want: "selected_lanes"},
+		{name: "duplicate selected lane", mutate: func(request *BuildRequest) { request.SelectedLanes = []string{"npcs", " npcs "} }, want: "duplicated"},
+		{name: "unselected contribution", mutate: func(request *BuildRequest) {
+			request.LaneEvidence = []LaneEvidence{{LaneID: "other", SourceRefs: []source.SourceRef{ref(3, 3)}}}
+		}, want: "not selected"},
+		{name: "source digest mismatch", mutate: func(request *BuildRequest) { request.Source.Digest = "sha256:" + strings.Repeat("0", 64) }, want: "does not match"},
+		{name: "invalid reference", mutate: func(request *BuildRequest) {
+			request.LaneEvidence = []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(99, 99)}}}
+		}, want: "source reference[0]"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			request := BuildRequest{Source: testDocument(t), SelectedLanes: []string{"npcs"}, LaneEvidence: []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}}}
+			test.mutate(&request)
+			if _, err := Build(request); err == nil || !strings.Contains(err.Error(), test.want) {
+				t.Fatalf("Build() error = %v, want %q", err, test.want)
+			}
+		})
+	}
+}
+
+func TestCodecRoundTripsCompactFixtureAndOwnsDecodedValues(t *testing.T) {
+	fixture, err := os.ReadFile("testdata/source_evidence_context.v1.json")
+	if err != nil {
+		t.Fatal(err)
+	}
+	codec := New()
+	value, err := codec.Decode(fixture)
+	if err != nil {
+		t.Fatalf("Decode(fixture) error = %v", err)
+	}
+	encoded, err := codec.Encode(value)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !bytes.Equal(encoded, bytes.TrimSpace(fixture)) {
+		t.Fatalf("fixture does not use canonical encoding\nwant: %s\n got: %s", fixture, encoded)
+	}
+	value.Contexts[0].Units[0].Text = "changed"
+	decoded, err := codec.Decode(fixture)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if decoded.Contexts[0].Units[0].Text != "The party meets Rowan." {
+		t.Fatal("Decode() reused mutable document storage")
+	}
+	built, err := Build(BuildRequest{Source: testDocument(t), WindowUnits: 1, SelectedLanes: []string{"npcs"}, LaneEvidence: []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	content, err := codec.Encode(built)
+	if err != nil {
+		t.Fatal(err)
+	}
+	first, err := codec.Decode(content)
+	if err != nil {
+		t.Fatal(err)
+	}
+	second, err := codec.Decode(content)
+	if err != nil {
+		t.Fatal(err)
+	}
+	first.Contexts[0].Units[0].Metadata["nested"].(map[string]any)["value"] = "changed"
+	if second.Contexts[0].Units[0].Metadata["nested"].(map[string]any)["value"] != "original" {
+		t.Fatal("Decode() returned metadata aliases")
+	}
+}
+
+func TestCodecRejectsInvalidDurableBoundaries(t *testing.T) {
+	value, err := Build(BuildRequest{Source: testDocument(t), SelectedLanes: []string{"npcs"}, LaneEvidence: []LaneEvidence{{LaneID: "npcs", SourceRefs: []source.SourceRef{ref(3, 3)}}}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, test := range []struct {
+		name   string
+		mutate func(*Document)
+	}{
+		{name: "unsorted lanes", mutate: func(value *Document) { value.SelectedLanes = []string{"z", "a"} }},
+		{name: "context range mismatch", mutate: func(value *Document) { value.Contexts[0].ContextRef.EndUnitID = 999 }},
+		{name: "mismatched evidence source", mutate: func(value *Document) { value.Contexts[0].EvidenceRefs[0].SourceRef.SourceID = "other" }},
+		{name: "invalid evidence range", mutate: func(value *Document) {
+			value.Contexts[0].EvidenceRefs[0].SourceRef.StartUnitID = 10
+		}},
+		{name: "duplicate context unit", mutate: func(value *Document) { value.Contexts = append(value.Contexts, value.Contexts[0]) }},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			candidate, err := clone(value)
+			if err != nil {
+				t.Fatal(err)
+			}
+			test.mutate(&candidate)
+			if _, err := New().Encode(candidate); err == nil {
+				t.Fatal("Encode() error = nil, want durable model rejection")
+			}
+		})
+	}
+	content, err := New().Encode(value)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, test := range []struct {
+		name   string
+		mutate func(map[string]any)
+	}{
+		{name: "missing contexts", mutate: func(value map[string]any) { delete(value, "contexts") }},
+		{name: "null contexts", mutate: func(value map[string]any) { value["contexts"] = nil }},
+		{name: "unknown fixed field", mutate: func(value map[string]any) { value["unknown"] = true }},
+		{name: "missing units", mutate: func(value map[string]any) { delete(contextObject(value, 0), "units") }},
+		{name: "null evidence refs", mutate: func(value map[string]any) { contextObject(value, 0)["evidence_refs"] = nil }},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			raw := decodeJSON(t, content)
+			test.mutate(raw)
+			mutated, err := json.Marshal(raw)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if _, err := New().Decode(mutated); err == nil {
+				t.Fatal("Decode() error = nil, want strict payload rejection")
+			}
+		})
+	}
+	if _, err := New().Decode(append(content, []byte(" {}")...)); err == nil {
+		t.Fatal("Decode() error = nil, want trailing JSON rejection")
+	}
+}
+
+func TestSerializeUsesFixedArtifactIdentity(t *testing.T) {
+	artifact, err := Serialize(BuildRequest{Source: testDocument(t), SelectedLanes: []string{"npcs"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if artifact.Kind != ArtifactKind || artifact.MediaType != MediaType || artifact.Schema.ID != SchemaID || artifact.Schema.Name != SchemaName || artifact.Schema.Version != SchemaVersion {
+		t.Fatalf("Serialize() = %#v, want fixed artifact identity", artifact)
+	}
+	decoded, err := New().Decode(artifact.Content)
+	if err != nil || len(decoded.Contexts) != 0 || decoded.Contexts == nil {
+		t.Fatalf("Decode(Serialize()) = %#v, %v; want explicit empty contexts", decoded, err)
+	}
+}
+
+func testDocument(t *testing.T) *source.SourceDocument {
+	t.Helper()
+	document := &source.SourceDocument{
+		ID: "session", Kind: "transcript", Format: "application/json",
+		Units: []source.SourceUnit{
+			{ID: 10, Kind: "segment", Text: "first", Ref: ref(10, 10), Metadata: map[string]any{"nested": map[string]any{"value": "original"}}},
+			{ID: 3, Kind: "segment", Text: "second", Ref: ref(3, 3)},
+			{ID: 30, Kind: "segment", Text: "third", Ref: ref(30, 30)},
+			{ID: 7, Kind: "segment", Text: "fourth", Ref: ref(7, 7)},
+			{ID: 50, Kind: "segment", Text: "fifth", Ref: ref(50, 50)},
+		},
+	}
+	digest, err := source.DigestDocument(document)
+	if err != nil {
+		t.Fatal(err)
+	}
+	document.Digest = digest
+	return document
+}
+
+func ref(start, end int) source.SourceRef {
+	return source.SourceRef{SourceID: "session", StartUnitID: start, EndUnitID: end}
+}
+
+func contextUnitIDs(contexts []Context) [][]int {
+	values := make([][]int, len(contexts))
+	for index, context := range contexts {
+		values[index] = make([]int, len(context.Units))
+		for unitIndex, unit := range context.Units {
+			values[index][unitIndex] = unit.ID
+		}
+	}
+	return values
+}
+
+func contextEvidenceRefs(contexts []Context) [][]EvidenceRef {
+	values := make([][]EvidenceRef, len(contexts))
+	for index, context := range contexts {
+		values[index] = append([]EvidenceRef(nil), context.EvidenceRefs...)
+	}
+	return values
+}
+
+func containsEvidenceRef(values []EvidenceRef, want source.SourceRef) bool {
+	for _, value := range values {
+		if value.SourceRef == want {
+			return true
+		}
+	}
+	return false
+}
+
+func decodeJSON(t *testing.T, content []byte) map[string]any {
+	t.Helper()
+	decoder := json.NewDecoder(bytes.NewReader(content))
+	decoder.UseNumber()
+	var value map[string]any
+	if err := decoder.Decode(&value); err != nil {
+		t.Fatal(err)
+	}
+	return value
+}
+
+func contextObject(value map[string]any, index int) map[string]any {
+	return value["contexts"].([]any)[index].(map[string]any)
+}

internal/framework/evidencecontext/model.go

@@ -0,0 +1,50 @@
+// Package evidencecontext owns the durable source evidence-context contract.
+package evidencecontext
+
+import (
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+const (
+	ArtifactKind  contracts.ArtifactKind = "source/evidence-context"
+	SchemaID                             = "notarius.source.evidence_context"
+	SchemaName                           = "notarius_source_evidence_context_v1"
+	SchemaVersion                        = "v1"
+	MediaType                            = "application/json"
+)
+
+// Document is the durable union of direct evidence and surrounding source
+// context selected for one accepted source document.
+type Document struct {
+	SourceID      string    `json:"source_id"`
+	SourceDigest  string    `json:"source_digest"`
+	WindowUnits   int       `json:"window_units"`
+	SelectedLanes []string  `json:"selected_lanes"`
+	Contexts      []Context `json:"contexts"`
+}
+
+type Context struct {
+	ContextRef   source.SourceRef    `json:"context_ref"`
+	EvidenceRefs []EvidenceRef       `json:"evidence_refs"`
+	Units        []source.SourceUnit `json:"units"`
+}
+
+type EvidenceRef struct {
+	LaneID    string           `json:"lane_id"`
+	SourceRef source.SourceRef `json:"source_ref"`
+}
+
+// LaneEvidence attributes direct source references to one selected lane.
+type LaneEvidence struct {
+	LaneID     string             `json:"lane_id"`
+	SourceRefs []source.SourceRef `json:"source_refs"`
+}
+
+// BuildRequest supplies accepted source material and direct lane evidence.
+type BuildRequest struct {
+	Source        *source.SourceDocument
+	WindowUnits   int
+	SelectedLanes []string
+	LaneEvidence  []LaneEvidence
+}

internal/framework/evidencecontext/testdata/source_evidence_context.v1.json

@@ -0,0 +1 @@
+{"source_id":"session-alpha","source_digest":"sha256:0000000000000000000000000000000000000000000000000000000000000000","window_units":0,"selected_lanes":["npcs"],"contexts":[{"context_ref":{"source_id":"session-alpha","start_unit_id":13,"end_unit_id":13},"evidence_refs":[{"lane_id":"npcs","source_ref":{"source_id":"session-alpha","start_unit_id":13,"end_unit_id":13}}],"units":[{"id":13,"kind":"transcript_segment","text":"The party meets Rowan.","ref":{"source_id":"session-alpha","start_unit_id":13,"end_unit_id":13}}]}]}

internal/framework/llm/asset_registry.go

@@ -12,7 +12,7 @@ import (
 	"strings"
 	"time"
 
-	"gitea.maximumdirect.net/eric/scriptorium"
+	"gitea.maximumdirect.net/eric/promptkit"
 )
 
 type AssetSource struct {
@@ -21,8 +21,9 @@ type AssetSource struct {
 }
 
 type AssetRegistry struct {
-	prompts []AssetSource
-	schemas []AssetSource
+	prompts          []AssetSource
+	schemas          []AssetSource
+	fallbackProfiles []AssetSource
 }
 
 type AssetHashPart struct {
@@ -58,6 +59,20 @@ func (r *AssetRegistry) RegisterSchemaFS(fsys fs.FS, root string) error {
 	return nil
 }
 
+// RegisterFallbackProfileFS registers profile assets that PromptKit uses only
+// when an operator-configured source does not provide a matching profile.
+func (r *AssetRegistry) RegisterFallbackProfileFS(fsys fs.FS, root string) error {
+	if r == nil {
+		return fmt.Errorf("asset registry must not be nil")
+	}
+	source, err := newAssetSource(fsys, root)
+	if err != nil {
+		return fmt.Errorf("register fallback profile assets: %w", err)
+	}
+	r.fallbackProfiles = append(r.fallbackProfiles, source)
+	return nil
+}
+
 func (r *AssetRegistry) PromptFS() (fs.FS, error) {
 	if r == nil {
 		return nil, fmt.Errorf("asset registry must not be nil")
@@ -72,19 +87,76 @@ func (r *AssetRegistry) SchemaFS() (fs.FS, error) {
 	return flattenAssetSources(r.schemas)
 }
 
-func (r *AssetRegistry) ScriptoriumOptions() ([]scriptorium.Option, error) {
+func (r *AssetRegistry) FallbackProfileFS() (fs.FS, error) {
+	if r == nil {
+		return nil, fmt.Errorf("asset registry must not be nil")
+	}
+	return flattenAssetSources(r.fallbackProfiles)
+}
+
+// FallbackProfileDigest returns a deterministic, non-secret identity for the
+// flattened fallback profile assets.
+func (r *AssetRegistry) FallbackProfileDigest() (string, error) {
+	_, digest, _, err := r.fallbackProfileAssets()
+	return digest, err
+}
+
+func (r *AssetRegistry) PromptKitOptions() ([]promptkit.Option, error) {
+	options, _, err := r.promptKitOptions()
+	return options, err
+}
+
+func (r *AssetRegistry) promptKitOptions() ([]promptkit.Option, string, error) {
 	promptFS, err := r.PromptFS()
 	if err != nil {
-		return nil, fmt.Errorf("prepare prompt assets: %w", err)
+		return nil, "", fmt.Errorf("prepare prompt assets: %w", err)
 	}
 	schemaFS, err := r.SchemaFS()
 	if err != nil {
-		return nil, fmt.Errorf("prepare schema assets: %w", err)
+		return nil, "", fmt.Errorf("prepare schema assets: %w", err)
 	}
-	return []scriptorium.Option{
-		scriptorium.WithPromptFS(promptFS, "."),
-		scriptorium.WithSchemaFS(schemaFS, "."),
-	}, nil
+	options := []promptkit.Option{
+		promptkit.WithPromptFS(promptFS, "."),
+		promptkit.WithSchemaFS(schemaFS, "."),
+	}
+	fallbackFS, fallbackDigest, hasFallback, err := r.fallbackProfileAssets()
+	if err != nil {
+		return nil, "", err
+	}
+	if hasFallback {
+		options = append(options, promptkit.WithFallbackProfileFS(fallbackFS, "."))
+	}
+	return options, fallbackDigest, nil
+}
+
+func (r *AssetRegistry) promptKitFallbackProfileOption() (promptkit.Option, bool, error) {
+	fallbackFS, _, hasFallback, err := r.fallbackProfileAssets()
+	if err != nil {
+		return nil, false, err
+	}
+	if !hasFallback {
+		return nil, false, nil
+	}
+	return promptkit.WithFallbackProfileFS(fallbackFS, "."), true, nil
+}
+
+func (r *AssetRegistry) fallbackProfileAssets() (fs.FS, string, bool, error) {
+	if r == nil {
+		return nil, "", false, fmt.Errorf("asset registry must not be nil")
+	}
+	if len(r.fallbackProfiles) == 0 {
+		empty := sha256.Sum256([]byte("notarius:fallback-profile-assets:empty"))
+		return nil, "sha256:" + hex.EncodeToString(empty[:]), false, nil
+	}
+	fallbackFS, err := r.FallbackProfileFS()
+	if err != nil {
+		return nil, "", false, fmt.Errorf("prepare fallback profile assets: %w", err)
+	}
+	digest, err := hashAssetFilesystem(fallbackFS)
+	if err != nil {
+		return nil, "", false, err
+	}
+	return fallbackFS, digest, true, nil
 }
 
 func HashAssets(parts []AssetHashPart) (string, error) {
@@ -117,6 +189,28 @@ func HashAssets(parts []AssetHashPart) (string, error) {
 	return "sha256:" + hex.EncodeToString(hash.Sum(nil)), nil
 }
 
+func hashAssetFilesystem(fsys fs.FS) (string, error) {
+	var parts []AssetHashPart
+	err := fs.WalkDir(fsys, ".", func(name string, entry fs.DirEntry, walkErr error) error {
+		if walkErr != nil {
+			return walkErr
+		}
+		if entry.IsDir() {
+			return nil
+		}
+		parts = append(parts, AssetHashPart{FS: fsys, Path: name})
+		return nil
+	})
+	if err != nil {
+		return "", fmt.Errorf("walk assets for digest: %w", err)
+	}
+	if len(parts) > 0 {
+		return HashAssets(parts)
+	}
+	empty := sha256.Sum256([]byte("notarius:fallback-profile-assets:empty"))
+	return "sha256:" + hex.EncodeToString(empty[:]), nil
+}
+
 func newAssetSource(fsys fs.FS, root string) (AssetSource, error) {
 	if fsys == nil {
 		return AssetSource{}, fmt.Errorf("filesystem must not be nil")
@@ -248,6 +342,9 @@ func (m assetMapFS) dirEntries(dir string) []fs.DirEntry {
 		children[childName] = entry
 	}
 	if len(children) == 0 {
+		if dir == "." {
+			return []fs.DirEntry{}
+		}
 		return nil
 	}
 	names := make([]string, 0, len(children))

internal/framework/llm/asset_registry_test.go

@@ -2,12 +2,13 @@ package llm
 
 import (
 	"context"
+	"io/fs"
 	"strings"
 	"testing"
 	"testing/fstest"
 	"time"
 
-	"gitea.maximumdirect.net/eric/scriptorium"
+	"gitea.maximumdirect.net/eric/promptkit"
 )
 
 func TestAssetRegistryCombinesPromptAndSchemaSources(t *testing.T) {
@@ -22,12 +23,12 @@ func TestAssetRegistryCombinesPromptAndSchemaSources(t *testing.T) {
 		"root/schemas/out.json": {Data: []byte(`{"type":"object"}`)},
 	}, "root")
 
-	engine := newAssetTestEngine(t, registry)
-	prepared, err := engine.Prepare(context.Background(), scriptorium.RunRequest{
+	engine := newPromptKitAssetTestEngine(t, registry)
+	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
 		PromptID:  "asset.test",
 		ProfileID: "asset-test-profile",
-		Inputs: map[string]scriptorium.ArtifactRef{
-			"transcript": scriptorium.Inline(`{"ok":true}`),
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline(`{"ok":true}`),
 		},
 	})
 	if err != nil {
@@ -50,12 +51,12 @@ func TestAssetRegistryPrepareFailsForMissingPromptAsset(t *testing.T) {
 		"out.json": {Data: []byte(`{"type":"object"}`)},
 	}, ".")
 
-	engine := newAssetTestEngine(t, registry)
-	_, err := engine.Prepare(context.Background(), scriptorium.RunRequest{
+	engine := newPromptKitAssetTestEngine(t, registry)
+	_, err := engine.Prepare(context.Background(), promptkit.RunRequest{
 		PromptID:  "asset.test",
 		ProfileID: "asset-test-profile",
-		Inputs: map[string]scriptorium.ArtifactRef{
-			"transcript": scriptorium.Inline(`{"ok":true}`),
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline(`{"ok":true}`),
 		},
 	})
 	if err == nil || !strings.Contains(err.Error(), "content_file") {
@@ -74,12 +75,12 @@ func TestAssetRegistryPrepareFailsForMissingSchemaAsset(t *testing.T) {
 		"present.json": {Data: []byte(`{"type":"object"}`)},
 	}, ".")
 
-	engine := newAssetTestEngine(t, registry)
-	_, err := engine.Prepare(context.Background(), scriptorium.RunRequest{
+	engine := newPromptKitAssetTestEngine(t, registry)
+	_, err := engine.Prepare(context.Background(), promptkit.RunRequest{
 		PromptID:  "asset.test",
 		ProfileID: "asset-test-profile",
-		Inputs: map[string]scriptorium.ArtifactRef{
-			"transcript": scriptorium.Inline(`{"ok":true}`),
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline(`{"ok":true}`),
 		},
 	})
 	if err == nil || !strings.Contains(err.Error(), "missing.json") {
@@ -98,6 +99,89 @@ func TestAssetRegistryRejectsDuplicateAssetPaths(t *testing.T) {
 	}
 }
 
+func TestAssetRegistryCombinesFallbackProfileSources(t *testing.T) {
+	registry := NewAssetRegistry()
+	if err := registry.RegisterFallbackProfileFS(fstest.MapFS{
+		"first/profiles/one.yaml": {Data: []byte("id: one\nmodel: first\n")},
+	}, "first/profiles"); err != nil {
+		t.Fatalf("RegisterFallbackProfileFS() error = %v, want nil", err)
+	}
+	if err := registry.RegisterFallbackProfileFS(fstest.MapFS{
+		"second/two.yaml": {Data: []byte("id: two\nmodel: second\n")},
+	}, "second"); err != nil {
+		t.Fatalf("RegisterFallbackProfileFS() error = %v, want nil", err)
+	}
+
+	fallbackFS, err := registry.FallbackProfileFS()
+	if err != nil {
+		t.Fatalf("FallbackProfileFS() error = %v, want nil", err)
+	}
+	for _, name := range []string{"one.yaml", "two.yaml"} {
+		if _, err := fs.ReadFile(fallbackFS, name); err != nil {
+			t.Fatalf("FallbackProfileFS().ReadFile(%q) error = %v, want nil", name, err)
+		}
+	}
+}
+
+func TestAssetRegistryRejectsInvalidFallbackProfileRoot(t *testing.T) {
+	registry := NewAssetRegistry()
+	err := registry.RegisterFallbackProfileFS(fstest.MapFS{}, "../profiles")
+	if err == nil || !strings.Contains(err.Error(), "invalid path") {
+		t.Fatalf("RegisterFallbackProfileFS() error = %v, want invalid root error", err)
+	}
+}
+
+func TestAssetRegistryRejectsUnreadableFallbackProfileAssets(t *testing.T) {
+	registry := NewAssetRegistry()
+	if err := registry.RegisterFallbackProfileFS(unreadableAssetFS{}, "."); err != nil {
+		t.Fatalf("RegisterFallbackProfileFS() error = %v, want nil", err)
+	}
+	_, err := registry.FallbackProfileFS()
+	if err == nil || !strings.Contains(err.Error(), "permission denied") {
+		t.Fatalf("FallbackProfileFS() error = %v, want unreadable asset error", err)
+	}
+}
+
+func TestAssetRegistryRejectsDuplicateFallbackProfilePaths(t *testing.T) {
+	registry := NewAssetRegistry()
+	if err := registry.RegisterFallbackProfileFS(fstest.MapFS{"first/profile.yaml": {Data: []byte("id: first\n")}}, "first"); err != nil {
+		t.Fatal(err)
+	}
+	if err := registry.RegisterFallbackProfileFS(fstest.MapFS{"second/profile.yaml": {Data: []byte("id: second\n")}}, "second"); err != nil {
+		t.Fatal(err)
+	}
+	_, err := registry.FallbackProfileFS()
+	if err == nil || !strings.Contains(err.Error(), "duplicate asset path") {
+		t.Fatalf("FallbackProfileFS() error = %v, want duplicate path error", err)
+	}
+}
+
+func TestAssetRegistryFallbackProfileDigestTracksContentWithoutLeakingIt(t *testing.T) {
+	digestFor := func(content string) string {
+		t.Helper()
+		registry := NewAssetRegistry()
+		if err := registry.RegisterFallbackProfileFS(fstest.MapFS{
+			"profiles/profile.yaml": {Data: []byte(content)},
+		}, "profiles"); err != nil {
+			t.Fatal(err)
+		}
+		digest, err := registry.FallbackProfileDigest()
+		if err != nil {
+			t.Fatal(err)
+		}
+		return digest
+	}
+
+	first := digestFor("id: fallback\nmodel: model-one\n")
+	second := digestFor("id: fallback\nmodel: model-two\n")
+	if first == second {
+		t.Fatalf("fallback profile digests = %q and %q, want content change", first, second)
+	}
+	if !strings.HasPrefix(first, "sha256:") || strings.Contains(first, "model-one") || strings.Contains(first, "profile.yaml") {
+		t.Fatalf("fallback profile digest leaked source details: %q", first)
+	}
+}
+
 func TestAssetRegistryCombinesNamespacedPromptSources(t *testing.T) {
 	registry := NewAssetRegistry()
 	mustRegisterPromptFS(t, registry, fstest.MapFS{
@@ -145,18 +229,18 @@ func TestHashAssetsOmitsRawAssetContent(t *testing.T) {
 	}
 }
 
-func newAssetTestEngine(t *testing.T, registry *AssetRegistry) *scriptorium.Engine {
+func newPromptKitAssetTestEngine(t *testing.T, registry *AssetRegistry) *promptkit.Engine {
 	t.Helper()
-	options, err := registry.ScriptoriumOptions()
+	options, err := registry.PromptKitOptions()
 	if err != nil {
-		t.Fatalf("ScriptoriumOptions() error = %v, want nil", err)
+		t.Fatalf("PromptKitOptions() error = %v, want nil", err)
 	}
-	options = append(options, scriptorium.WithProfiles(scriptorium.OpenAICompatibleProfile(scriptorium.OpenAICompatibleProfileConfig{
+	options = append(options, promptkit.WithProfiles(promptkit.OpenAICompatibleProfile(promptkit.OpenAICompatibleProfileConfig{
 		ID:       "asset-test-profile",
 		Endpoint: "http://127.0.0.1:1/v1",
 		Model:    "asset-test-model",
 	})))
-	engine, err := scriptorium.NewEngine(scriptorium.Config{Timeout: time.Second}, options...)
+	engine, err := promptkit.NewEngine(promptkit.Config{Timeout: time.Second}, options...)
 	if err != nil {
 		t.Fatalf("NewEngine() error = %v, want nil", err)
 	}
@@ -196,3 +280,9 @@ output:
   repair_attempts: 0
 `
 }
+
+type unreadableAssetFS struct{}
+
+func (unreadableAssetFS) Open(name string) (fs.File, error) {
+	return nil, &fs.PathError{Op: "open", Path: name, Err: fs.ErrPermission}
+}

internal/framework/llm/checkpoint_fingerprint.go

@@ -0,0 +1,14 @@
+package llm
+
+// CheckpointFingerprint is a stable, non-secret semantic identity contributed
+// by the LLM runtime before pipeline execution.
+type CheckpointFingerprint struct {
+	Name  string
+	Value string
+}
+
+// CheckpointFingerprintProvider exposes LLM-runtime identities that must
+// participate in checkpoint composition.
+type CheckpointFingerprintProvider interface {
+	LLMCheckpointFingerprints() ([]CheckpointFingerprint, error)
+}

internal/framework/llm/promptkit_client.go

@@ -3,6 +3,7 @@ package llm
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net/http"
 	"regexp"
@@ -13,24 +14,36 @@ import (
 
 	"gitea.maximumdirect.net/eric/notarius/internal/core/artifacts"
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
-	"gitea.maximumdirect.net/eric/scriptorium"
+	"gitea.maximumdirect.net/eric/promptkit"
 )
 
-const scriptoriumProviderName = "scriptorium"
+const promptKitProviderName = "promptkit"
 
-type ScriptoriumClientConfig struct {
-	ProfileDir    string
-	ProfileFile   string
-	Assets        *AssetRegistry
-	Timeout       time.Duration
-	HTTPClient    *http.Client
-	EngineOptions []scriptorium.Option
-	Recorder      *LLMProfileRecorder
+type PromptKitLocalBackendConfig struct {
+	Endpoint         string
+	ConcurrencyLimit int
 }
 
-type ScriptoriumClient struct {
-	engine   *scriptorium.Engine
-	recorder *LLMProfileRecorder
+type PromptKitClientConfig struct {
+	ProfileDir      string
+	ProfileFile     string
+	LocalBackend    *PromptKitLocalBackendConfig
+	Assets          *AssetRegistry
+	Timeout         time.Duration
+	HTTPClient      *http.Client
+	EngineOptions   []promptkit.Option
+	Recorder        *LLMProfileRecorder
+	ReasoningEffort *string
+}
+
+type PromptKitClient struct {
+	engine                *promptkit.Engine
+	recorder              *LLMProfileRecorder
+	profileDir            string
+	profileFile           string
+	localEndpoint         string
+	fallbackProfileDigest string
+	reasoningEffort       *string
 }
 
 type LLMProfileRecorder struct {
@@ -38,49 +51,66 @@ type LLMProfileRecorder struct {
 	profiles map[string]artifacts.LLMProfileManifest
 }
 
-var _ contracts.StructuredLLMClient = (*ScriptoriumClient)(nil)
-var _ contracts.LLMProfileManifestProvider = (*ScriptoriumClient)(nil)
+var _ contracts.StructuredLLMClient = (*PromptKitClient)(nil)
+var _ contracts.LLMProfileManifestProvider = (*PromptKitClient)(nil)
 
-func NewScriptoriumClient(cfg ScriptoriumClientConfig) (*ScriptoriumClient, error) {
+func PromptKitLocalBackendOption(cfg PromptKitLocalBackendConfig) promptkit.Option {
+	return promptkit.WithBackend(promptkit.LocalBackend(cfg.Endpoint, cfg.ConcurrencyLimit))
+}
+
+func NewPromptKitClient(cfg PromptKitClientConfig) (*PromptKitClient, error) {
 	if cfg.Assets == nil {
-		return nil, fmt.Errorf("scriptorium client assets must not be nil")
+		return nil, fmt.Errorf("PromptKit client assets must not be nil")
 	}
-	if strings.TrimSpace(cfg.ProfileDir) != "" && strings.TrimSpace(cfg.ProfileFile) != "" {
-		return nil, fmt.Errorf("scriptorium profile_dir and profile_file are mutually exclusive")
-	}
-	options, err := cfg.Assets.ScriptoriumOptions()
+	profileSource, profileOptions, err := promptKitProfileSourceEngineOptions(PromptKitProfileSourceConfig{
+		ProfileDir:   cfg.ProfileDir,
+		ProfileFile:  cfg.ProfileFile,
+		LocalBackend: cfg.LocalBackend,
+	})
 	if err != nil {
 		return nil, err
 	}
-	if profileFile := strings.TrimSpace(cfg.ProfileFile); profileFile != "" {
-		options = append(options, scriptorium.WithProfileFile(profileFile))
+	options, fallbackProfileDigest, err := cfg.Assets.promptKitOptions()
+	if err != nil {
+		return nil, err
 	}
+	options = append(options, profileOptions...)
 	options = append(options, cfg.EngineOptions...)
 
-	engine, err := scriptorium.NewEngine(scriptorium.Config{
-		ProfileDir: strings.TrimSpace(cfg.ProfileDir),
+	engine, err := promptkit.NewEngine(promptkit.Config{
+		ProfileDir: profileSource.ProfileDir,
 		Timeout:    cfg.Timeout,
 		HTTPClient: cfg.HTTPClient,
 	}, options...)
 	if err != nil {
-		return nil, fmt.Errorf("create Scriptorium engine: %w", err)
+		return nil, fmt.Errorf("create PromptKit engine: %w", err)
 	}
 	recorder := cfg.Recorder
 	if recorder == nil {
 		recorder = NewLLMProfileRecorder()
 	}
-	return &ScriptoriumClient{
-		engine:   engine,
-		recorder: recorder,
+	var reasoningEffort *string
+	if cfg.ReasoningEffort != nil {
+		value := *cfg.ReasoningEffort
+		reasoningEffort = &value
+	}
+	return &PromptKitClient{
+		engine:                engine,
+		recorder:              recorder,
+		profileDir:            profileSource.ProfileDir,
+		profileFile:           profileSource.ProfileFile,
+		localEndpoint:         profileSource.localEndpoint(),
+		fallbackProfileDigest: fallbackProfileDigest,
+		reasoningEffort:       reasoningEffort,
 	}, nil
 }
 
-func (c *ScriptoriumClient) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
+func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
 	if c == nil {
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("scriptorium client must not be nil")
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("PromptKit client must not be nil")
 	}
 	if c.engine == nil {
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("scriptorium client engine must not be nil")
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("PromptKit client engine must not be nil")
 	}
 	if err := validateOutputTarget(out); err != nil {
 		return contracts.StructuredCompletionResponse{}, err
@@ -89,54 +119,85 @@ func (c *ScriptoriumClient) CompleteStructured(ctx context.Context, req contract
 	if promptID == "" {
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("structured completion prompt_id must not be empty")
 	}
+	sessionID := strings.TrimSpace(req.SessionID)
+	var execution *promptkit.ExecutionTargetOverride
+	if c.reasoningEffort != nil {
+		reasoningEffort := *c.reasoningEffort
+		execution = &promptkit.ExecutionTargetOverride{
+			ReasoningEffort: &reasoningEffort,
+		}
+	}
 
-	runReq := scriptorium.RunRequest{
+	runReq := promptkit.RunRequest{
 		PromptID:      promptID,
 		PromptVersion: strings.TrimSpace(req.PromptVersion),
 		ProfileID:     strings.TrimSpace(req.ProfileID),
-		Inputs:        scriptoriumInputs(req.Inputs),
-		Vars:          scriptoriumVars(req),
-		Metadata:      scriptoriumMetadata(req),
+		SessionID:     sessionID,
+		Inputs:        promptKitInputs(req.Inputs),
+		Vars:          promptKitVars(req, sessionID),
+		Execution:     execution,
 	}
-	prepared, err := c.engine.Prepare(ctx, runReq)
+	prepared, err := c.engine.PrepareExecution(ctx, runReq)
 	if err != nil {
 		if ctxErr := ctx.Err(); ctxErr != nil {
 			return contracts.StructuredCompletionResponse{}, ctxErr
 		}
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("prepare Scriptorium prompt %q: %w", promptID, redactScriptoriumError(err))
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("prepare PromptKit prompt %q: %w", promptID, redactPromptKitError(err))
 	}
-	result, err := c.engine.Run(ctx, runReq)
+	defer prepared.Discard()
+	preparedDetails := prepared.Details()
+	result, err := c.engine.RunPrepared(ctx, prepared)
 	if err != nil {
 		if ctxErr := ctx.Err(); ctxErr != nil {
 			return contracts.StructuredCompletionResponse{}, ctxErr
 		}
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run Scriptorium prompt %q: %w", promptID, redactScriptoriumError(err))
+		if errors.Is(err, promptkit.ErrCapacityExceeded) {
+			var capacityErr *promptkit.CapacityError
+			if errors.As(err, &capacityErr) && strings.TrimSpace(capacityErr.BackendID) != "" {
+				return contracts.StructuredCompletionResponse{}, fmt.Errorf(
+					"run PromptKit prompt %q on backend %q: %w: %v",
+					promptID,
+					strings.TrimSpace(capacityErr.BackendID),
+					contracts.ErrLLMCapacityExceeded,
+					redactPromptKitError(err),
+				)
+			}
+			return contracts.StructuredCompletionResponse{}, fmt.Errorf(
+				"run PromptKit prompt %q: %w: %v",
+				promptID,
+				contracts.ErrLLMCapacityExceeded,
+				redactPromptKitError(err),
+			)
+		}
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run PromptKit prompt %q: %w", promptID, redactPromptKitError(err))
 	}
 	if result == nil {
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run Scriptorium prompt %q: %w: empty result", promptID, contracts.ErrInvalidStructuredOutput)
+		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run PromptKit prompt %q: %w: empty result", promptID, contracts.ErrInvalidStructuredOutput)
 	}
-	response := c.responseFromResult(result, prepared)
-	if result.Validation.Status == scriptorium.ValidationFailed || !result.Validation.IsValid {
-		return response, fmt.Errorf("run Scriptorium prompt %q: %w: validation failed: %s", promptID, contracts.ErrInvalidStructuredOutput, strings.Join(result.Validation.Errors, "; "))
+	response := c.responseFromResult(result, &preparedDetails)
+	if result.Validation.Status == promptkit.ValidationFailed || !result.Validation.IsValid {
+		return response, fmt.Errorf("run PromptKit prompt %q: %w: validation failed: %s", promptID, contracts.ErrInvalidStructuredOutput, strings.Join(result.Validation.Errors, "; "))
 	}
 	if len(strings.TrimSpace(string(response.Content))) == 0 {
-		return response, fmt.Errorf("run Scriptorium prompt %q: %w: empty structured output", promptID, contracts.ErrInvalidStructuredOutput)
+		return response, fmt.Errorf("run PromptKit prompt %q: %w: empty structured output", promptID, contracts.ErrInvalidStructuredOutput)
 	}
 	if err := json.Unmarshal(response.Content, out); err != nil {
-		return response, fmt.Errorf("decode Scriptorium structured output for prompt %q: %w: %w", promptID, contracts.ErrInvalidStructuredOutput, err)
+		return response, fmt.Errorf("decode PromptKit structured output for prompt %q: %w: %w", promptID, contracts.ErrInvalidStructuredOutput, err)
 	}
 	return response, nil
 }
 
-func (c *ScriptoriumClient) responseFromResult(result *scriptorium.RunResult, prepared *scriptorium.PreparedRun) contracts.StructuredCompletionResponse {
+func (c *PromptKitClient) responseFromResult(result *promptkit.RunResult, prepared *promptkit.PreparedRun) contracts.StructuredCompletionResponse {
 	content := result.Artifact.Body
 	if len(content) == 0 {
 		content = []byte(result.RawOutput)
 	}
 	profile := artifacts.LLMProfileManifest{
-		ID:       strings.TrimSpace(result.SelectedProfileID),
-		Provider: scriptoriumProviderName,
-		Model:    firstNonEmpty(result.ModelName, result.EffectiveModelParams.Model),
+		ID:              strings.TrimSpace(result.SelectedProfileID),
+		Provider:        promptKitProviderName,
+		Model:           firstNonEmpty(result.ModelName, result.EffectiveModelParams.Model),
+		BackendID:       strings.TrimSpace(result.SelectedBackendID),
+		ReasoningEffort: strings.TrimSpace(result.EffectiveModelParams.ReasoningEffort),
 	}
 	if c.recorder != nil {
 		c.recorder.Record(profile)
@@ -149,17 +210,17 @@ func (c *ScriptoriumClient) responseFromResult(result *scriptorium.RunResult, pr
 		PromptTokens:     result.Usage.PromptTokens,
 		CompletionTokens: result.Usage.CompletionTokens,
 		TotalTokens:      result.Usage.TotalTokens,
-		Debug:            scriptoriumDebugMaterial(prepared, result),
+		Debug:            promptKitDebugMaterial(prepared, result),
 	}
 }
 
-func scriptoriumDebugMaterial(prepared *scriptorium.PreparedRun, result *scriptorium.RunResult) *contracts.LLMDebugMaterial {
+func promptKitDebugMaterial(prepared *promptkit.PreparedRun, result *promptkit.RunResult) *contracts.LLMDebugMaterial {
 	material := &contracts.LLMDebugMaterial{}
 	if prepared != nil {
-		material.Prompt = scriptoriumDebugPrompt(prepared)
+		material.Prompt = promptKitDebugPrompt(prepared)
 	}
 	if result != nil {
-		material.Response = scriptoriumDebugResponse(result)
+		material.Response = promptKitDebugResponse(result)
 	}
 	if material.Prompt == nil && material.Response == nil {
 		return nil
@@ -167,7 +228,7 @@ func scriptoriumDebugMaterial(prepared *scriptorium.PreparedRun, result *scripto
 	return material
 }
 
-func scriptoriumDebugPrompt(prepared *scriptorium.PreparedRun) *contracts.LLMDebugPrompt {
+func promptKitDebugPrompt(prepared *promptkit.PreparedRun) *contracts.LLMDebugPrompt {
 	if prepared == nil {
 		return nil
 	}
@@ -184,6 +245,7 @@ func scriptoriumDebugPrompt(prepared *scriptorium.PreparedRun) *contracts.LLMDeb
 		PromptVersion:        prepared.PromptVersion,
 		PromptHash:           prepared.PromptHash,
 		SelectedProfileID:    prepared.SelectedProfileID,
+		SelectedBackendID:    prepared.SelectedBackendID,
 		SessionID:            prepared.SessionID,
 		RenderedPromptHash:   prepared.RenderedPromptHash,
 		Messages:             messages,
@@ -194,7 +256,7 @@ func scriptoriumDebugPrompt(prepared *scriptorium.PreparedRun) *contracts.LLMDeb
 	}
 }
 
-func scriptoriumDebugResponse(result *scriptorium.RunResult) *contracts.LLMDebugResponse {
+func promptKitDebugResponse(result *promptkit.RunResult) *contracts.LLMDebugResponse {
 	if result == nil {
 		return nil
 	}
@@ -254,13 +316,28 @@ func cloneStringMap(values map[string]string) map[string]string {
 	return out
 }
 
-func (c *ScriptoriumClient) LLMProfileManifests() []artifacts.LLMProfileManifest {
+func (c *PromptKitClient) LLMProfileManifests() []artifacts.LLMProfileManifest {
 	if c == nil || c.recorder == nil {
 		return nil
 	}
 	return c.recorder.Manifests()
 }
 
+func (c *PromptKitClient) LLMCheckpointFingerprints() ([]CheckpointFingerprint, error) {
+	if c == nil {
+		return nil, nil
+	}
+	fingerprint, err := promptKitProfileFingerprint(c.profileDir, c.profileFile, c.fallbackProfileDigest)
+	if err != nil {
+		return nil, err
+	}
+	fingerprints := []CheckpointFingerprint{fingerprint}
+	if c.localEndpoint != "" {
+		fingerprints = append(fingerprints, promptKitLocalBackendFingerprint(c.localEndpoint))
+	}
+	return fingerprints, nil
+}
+
 func NewLLMProfileRecorder() *LLMProfileRecorder {
 	return &LLMProfileRecorder{profiles: map[string]artifacts.LLMProfileManifest{}}
 }
@@ -269,10 +346,8 @@ func (r *LLMProfileRecorder) Record(profile artifacts.LLMProfileManifest) {
 	if r == nil {
 		return
 	}
-	profile.ID = strings.TrimSpace(profile.ID)
-	profile.Provider = strings.TrimSpace(profile.Provider)
-	profile.Model = strings.TrimSpace(profile.Model)
-	key := profile.ID + "\x00" + profile.Provider + "\x00" + profile.Model
+	profile = profile.Normalized()
+	key := profile.IdentityKey()
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if r.profiles == nil {
@@ -302,11 +377,11 @@ func (r *LLMProfileRecorder) Manifests() []artifacts.LLMProfileManifest {
 	return out
 }
 
-func scriptoriumInputs(inputs contracts.LLMInputSet) map[string]scriptorium.ArtifactRef {
+func promptKitInputs(inputs contracts.LLMInputSet) map[string]promptkit.ArtifactRef {
 	if len(inputs) == 0 {
 		return nil
 	}
-	out := make(map[string]scriptorium.ArtifactRef, len(inputs))
+	out := make(map[string]promptkit.ArtifactRef, len(inputs))
 	for key, material := range inputs {
 		name := strings.TrimSpace(key)
 		if name == "" {
@@ -320,15 +395,15 @@ func scriptoriumInputs(inputs contracts.LLMInputSet) map[string]scriptorium.Arti
 			body = " "
 		}
 		if origin := strings.TrimSpace(material.OriginURI); origin != "" {
-			out[name] = scriptorium.InlineWithURI(origin, body)
+			out[name] = promptkit.InlineWithURI(origin, body)
 		} else {
-			out[name] = scriptorium.Inline(body)
+			out[name] = promptkit.Inline(body)
 		}
 	}
 	return out
 }
 
-func scriptoriumVars(req contracts.StructuredCompletionRequest) map[string]string {
+func promptKitVars(req contracts.StructuredCompletionRequest, sessionID string) map[string]string {
 	vars := make(map[string]string, len(req.Vars)+1)
 	for key, value := range req.Vars {
 		name := strings.TrimSpace(key)
@@ -337,7 +412,7 @@ func scriptoriumVars(req contracts.StructuredCompletionRequest) map[string]strin
 		}
 		vars[name] = fmt.Sprint(value)
 	}
-	if sessionID := strings.TrimSpace(req.SessionID); sessionID != "" {
+	if sessionID != "" {
 		vars["session_id"] = sessionID
 	}
 	if len(vars) == 0 {
@@ -346,20 +421,9 @@ func scriptoriumVars(req contracts.StructuredCompletionRequest) map[string]strin
 	return vars
 }
 
-func scriptoriumMetadata(req contracts.StructuredCompletionRequest) map[string]string {
-	metadata := map[string]string{}
-	if stageName := strings.TrimSpace(req.StageName); stageName != "" {
-		metadata["stage_name"] = stageName
-	}
-	if len(metadata) == 0 {
-		return nil
-	}
-	return metadata
-}
-
 var bearerTokenPattern = regexp.MustCompile(`(?i)Bearer\s+[A-Za-z0-9._~+/=-]+`)
 
-func redactScriptoriumError(err error) error {
+func redactPromptKitError(err error) error {
 	if err == nil {
 		return nil
 	}

internal/framework/llm/promptkit_profile_fingerprint.go

@@ -0,0 +1,110 @@
+package llm
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+const (
+	promptKitProfileFingerprintName      = "promptkit_profile_source"
+	promptKitLocalBackendFingerprintName = "promptkit_local_backend_target"
+	promptKitLocalBackendMarker          = "notarius:promptkit-local-backend:v1"
+	// The built-in profile catalog is compiled into this pinned PromptKit
+	// release. Update this identity when the dependency is upgraded.
+	promptKitBuiltinProfileCatalogID = "promptkit:v0.5.0:builtin-profiles"
+)
+
+func promptKitProfileFingerprint(profileDir, profileFile, fallbackProfileDigest string) (CheckpointFingerprint, error) {
+	hasher := sha256.New()
+	writeFingerprintPart(hasher, []byte(promptKitBuiltinProfileCatalogID))
+	writeFingerprintPart(hasher, []byte(strings.TrimSpace(fallbackProfileDigest)))
+
+	switch {
+	case strings.TrimSpace(profileFile) != "":
+		cleanProfileFile := strings.TrimSpace(profileFile)
+		data, err := os.ReadFile(cleanProfileFile)
+		if err != nil {
+			return CheckpointFingerprint{}, fmt.Errorf("read PromptKit profile file for checkpoint identity")
+		}
+		writeFingerprintPart(hasher, []byte(filepath.ToSlash(filepath.Base(cleanProfileFile))))
+		writeFingerprintPart(hasher, data)
+	case strings.TrimSpace(profileDir) != "":
+		files, err := promptKitProfileFiles(strings.TrimSpace(profileDir))
+		if err != nil {
+			return CheckpointFingerprint{}, err
+		}
+		for _, file := range files {
+			writeFingerprintPart(hasher, []byte(file.path))
+			writeFingerprintPart(hasher, file.digest)
+		}
+	}
+
+	return CheckpointFingerprint{
+		Name:  promptKitProfileFingerprintName,
+		Value: "sha256:" + hex.EncodeToString(hasher.Sum(nil)),
+	}, nil
+}
+
+func promptKitLocalBackendFingerprint(endpoint string) CheckpointFingerprint {
+	hasher := sha256.New()
+	writeFingerprintPart(hasher, []byte(promptKitLocalBackendMarker))
+	writeFingerprintPart(hasher, []byte(strings.TrimSpace(endpoint)))
+	return CheckpointFingerprint{
+		Name:  promptKitLocalBackendFingerprintName,
+		Value: "sha256:" + hex.EncodeToString(hasher.Sum(nil)),
+	}
+}
+
+type promptKitProfileFile struct {
+	path   string
+	digest []byte
+}
+
+func promptKitProfileFiles(root string) ([]promptKitProfileFile, error) {
+	var files []promptKitProfileFile
+	err := filepath.WalkDir(root, func(name string, entry fs.DirEntry, walkErr error) error {
+		if walkErr != nil {
+			return walkErr
+		}
+		if entry.IsDir() {
+			return nil
+		}
+		extension := filepath.Ext(entry.Name())
+		if extension != ".yaml" && extension != ".yml" {
+			return nil
+		}
+		data, err := os.ReadFile(name)
+		if err != nil {
+			return err
+		}
+		relativePath, err := filepath.Rel(root, name)
+		if err != nil {
+			return err
+		}
+		sum := sha256.Sum256(data)
+		files = append(files, promptKitProfileFile{
+			path:   filepath.ToSlash(relativePath),
+			digest: append([]byte(nil), sum[:]...),
+		})
+		return nil
+	})
+	if err != nil {
+		return nil, fmt.Errorf("read PromptKit profile directory for checkpoint identity")
+	}
+	sort.Slice(files, func(i, j int) bool {
+		return files[i].path < files[j].path
+	})
+	return files, nil
+}
+
+func writeFingerprintPart(hasher interface{ Write([]byte) (int, error) }, value []byte) {
+	length := []byte(fmt.Sprintf("%d:", len(value)))
+	_, _ = hasher.Write(length)
+	_, _ = hasher.Write(value)
+}

internal/framework/llm/promptkit_profiles.go

@@ -0,0 +1,146 @@
+package llm
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+type PromptKitProfileSourceConfig struct {
+	ProfileDir   string
+	ProfileFile  string
+	LocalBackend *PromptKitLocalBackendConfig
+}
+
+type PromptKitProfileInspectorConfig struct {
+	Source PromptKitProfileSourceConfig
+	Assets *AssetRegistry
+}
+
+func (c PromptKitProfileSourceConfig) localEndpoint() string {
+	if c.LocalBackend == nil {
+		return ""
+	}
+	return c.LocalBackend.Endpoint
+}
+
+type PromptKitProfileInspector struct {
+	engine *promptkit.Engine
+}
+
+type PromptKitProfileInspection struct {
+	ProfileID             string
+	BackendID             string
+	Model                 string
+	CredentialEnvironment string
+	CredentialRequired    bool
+}
+
+type PromptKitProfileInspectionError struct {
+	ProfileID string
+	err       error
+}
+
+func (e *PromptKitProfileInspectionError) Error() string {
+	switch {
+	case errors.Is(e.err, promptkit.ErrProfileNotFound):
+		return fmt.Sprintf("PromptKit profile %q is not configured", e.ProfileID)
+	case errors.Is(e.err, promptkit.ErrInvalidRequest):
+		return fmt.Sprintf("PromptKit profile ID %q is invalid", e.ProfileID)
+	case errors.Is(e.err, promptkit.ErrProfileLoad):
+		return fmt.Sprintf("PromptKit profile %q is invalid or unreadable", e.ProfileID)
+	default:
+		return fmt.Sprintf("PromptKit profile %q could not be inspected", e.ProfileID)
+	}
+}
+
+func (e *PromptKitProfileInspectionError) Unwrap() error {
+	return e.err
+}
+
+type promptKitProfileConfigurationError struct {
+	err error
+}
+
+func (e *promptKitProfileConfigurationError) Error() string {
+	return "PromptKit profile configuration is invalid or unreadable"
+}
+
+func (e *promptKitProfileConfigurationError) Unwrap() error {
+	return e.err
+}
+
+func NewPromptKitProfileInspector(cfg PromptKitProfileInspectorConfig) (*PromptKitProfileInspector, error) {
+	source, options, err := promptKitProfileSourceEngineOptions(cfg.Source)
+	if err != nil {
+		return nil, err
+	}
+	if cfg.Assets != nil {
+		fallbackOption, hasFallback, err := cfg.Assets.promptKitFallbackProfileOption()
+		if err != nil {
+			return nil, err
+		}
+		if hasFallback {
+			options = append(options, fallbackOption)
+		}
+	}
+	engine, err := promptkit.NewEngine(promptkit.Config{
+		PromptDir:  ".",
+		ProfileDir: source.ProfileDir,
+	}, options...)
+	if err != nil {
+		return nil, &promptKitProfileConfigurationError{err: err}
+	}
+	return &PromptKitProfileInspector{engine: engine}, nil
+}
+
+func (i *PromptKitProfileInspector) InspectProfile(ctx context.Context, profileID string) (PromptKitProfileInspection, error) {
+	if i == nil || i.engine == nil {
+		return PromptKitProfileInspection{}, fmt.Errorf("PromptKit profile inspector must not be nil")
+	}
+	profileID = strings.TrimSpace(profileID)
+	inspection, err := i.engine.InspectProfile(ctx, profileID)
+	if err != nil {
+		if ctxErr := ctx.Err(); ctxErr != nil {
+			return PromptKitProfileInspection{}, ctxErr
+		}
+		return PromptKitProfileInspection{}, &PromptKitProfileInspectionError{
+			ProfileID: profileID,
+			err:       err,
+		}
+	}
+	return PromptKitProfileInspection{
+		ProfileID:             inspection.ProfileID,
+		BackendID:             strings.TrimSpace(inspection.EffectiveModelParams.BackendID),
+		Model:                 strings.TrimSpace(inspection.EffectiveModelParams.Model),
+		CredentialEnvironment: strings.TrimSpace(inspection.EffectiveModelParams.APIKeyEnv),
+		CredentialRequired:    inspection.APIKeyRequired,
+	}, nil
+}
+
+func promptKitProfileSourceEngineOptions(cfg PromptKitProfileSourceConfig) (PromptKitProfileSourceConfig, []promptkit.Option, error) {
+	source := PromptKitProfileSourceConfig{
+		ProfileDir:  strings.TrimSpace(cfg.ProfileDir),
+		ProfileFile: strings.TrimSpace(cfg.ProfileFile),
+	}
+	if source.ProfileDir != "" && source.ProfileFile != "" {
+		return PromptKitProfileSourceConfig{}, nil, fmt.Errorf("PromptKit profile_dir and profile_file are mutually exclusive")
+	}
+	if cfg.LocalBackend != nil {
+		localBackend := *cfg.LocalBackend
+		localBackend.Endpoint = strings.TrimSpace(localBackend.Endpoint)
+		source.LocalBackend = &localBackend
+	}
+
+	var options []promptkit.Option
+	if source.ProfileFile != "" {
+		options = append(options, promptkit.WithProfileFile(source.ProfileFile))
+	}
+	if source.LocalBackend != nil {
+		options = append(options, PromptKitLocalBackendOption(*source.LocalBackend))
+	}
+	return source, options, nil
+}

internal/framework/llm/scheduled_client.go

@@ -53,3 +53,14 @@ func (c *scheduledClient) LLMProfileManifests() []artifacts.LLMProfileManifest {
 	}
 	return provider.LLMProfileManifests()
 }
+
+func (c *scheduledClient) LLMCheckpointFingerprints() ([]CheckpointFingerprint, error) {
+	if c == nil || c.client == nil {
+		return nil, nil
+	}
+	provider, ok := c.client.(CheckpointFingerprintProvider)
+	if !ok {
+		return nil, nil
+	}
+	return provider.LLMCheckpointFingerprints()
+}

internal/framework/llm/scheduled_client_test.go

@@ -75,6 +75,33 @@ func TestScheduledClientPropagatesSchedulerError(t *testing.T) {
 	}
 }
 
+func TestScheduledClientPreservesCheckpointFingerprints(t *testing.T) {
+	scheduler, err := NewScheduler(1)
+	if err != nil {
+		t.Fatal(err)
+	}
+	inner := &fingerprintedStructuredClient{
+		fingerprints: []CheckpointFingerprint{
+			{Name: "profile_source", Value: "sha256:one"},
+			{Name: "backend_target", Value: "sha256:two"},
+		},
+	}
+	client := NewScheduledClient(inner, scheduler)
+	provider, ok := client.(CheckpointFingerprintProvider)
+	if !ok {
+		t.Fatalf("scheduled client %T does not preserve checkpoint fingerprints", client)
+	}
+	got, err := provider.LLMCheckpointFingerprints()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(got) != len(inner.fingerprints) ||
+		got[0] != inner.fingerprints[0] ||
+		got[1] != inner.fingerprints[1] {
+		t.Fatalf("checkpoint fingerprints = %#v, want %#v", got, inner.fingerprints)
+	}
+}
+
 type blockingStructuredClient struct {
 	release     chan struct{}
 	inFlight    int32
@@ -110,6 +137,15 @@ type errorStructuredClient struct {
 	err error
 }
 
+type fingerprintedStructuredClient struct {
+	errorStructuredClient
+	fingerprints []CheckpointFingerprint
+}
+
+func (c *fingerprintedStructuredClient) LLMCheckpointFingerprints() ([]CheckpointFingerprint, error) {
+	return append([]CheckpointFingerprint(nil), c.fingerprints...), nil
+}
+
 func (c *errorStructuredClient) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
 	return contracts.StructuredCompletionResponse{}, c.err
 }

internal/framework/llm/scriptorium_client_test.go

@@ -1,372 +0,0 @@
-package llm
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"strings"
-	"sync"
-	"sync/atomic"
-	"testing"
-	"testing/fstest"
-	"time"
-
-	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
-	"gitea.maximumdirect.net/eric/scriptorium"
-)
-
-func TestScriptoriumClientMapsPromptRequestAndUnmarshalsOutput(t *testing.T) {
-	fake := &fakeScriptoriumLLM{content: `{"ok":true}`}
-	client := newTestScriptoriumClient(t, fake)
-
-	var out struct {
-		OK bool `json:"ok"`
-	}
-	resp, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		StageName:     "test-stage",
-		PromptID:      "adapter.test",
-		PromptVersion: "v1",
-		ProfileID:     "explicit-profile",
-		SessionID:     "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "sha256:source", "file:///source.json"),
-		},
-		Vars: map[string]any{"custom": "value"},
-	}, &out)
-	if err != nil {
-		t.Fatalf("CompleteStructured() error = %v, want nil", err)
-	}
-	if !out.OK {
-		t.Fatalf("decoded output OK = false, want true")
-	}
-	if resp.Provider != scriptoriumProviderName || resp.Model != "explicit-model" || resp.ProfileID != "explicit-profile" {
-		t.Fatalf("response metadata = %#v", resp)
-	}
-	if resp.PromptTokens != 11 || resp.CompletionTokens != 7 || resp.TotalTokens != 18 {
-		t.Fatalf("usage = %#v, want mapped token counts", resp)
-	}
-	if resp.Debug == nil || resp.Debug.Prompt == nil {
-		t.Fatalf("debug prompt = nil, want prepared prompt material")
-	}
-	if resp.Debug.Prompt.PromptID != "adapter.test" || resp.Debug.Prompt.SelectedProfileID != "explicit-profile" {
-		t.Fatalf("debug prompt metadata = %#v, want prompt/profile", resp.Debug.Prompt)
-	}
-	if len(resp.Debug.Prompt.Messages) != 1 || !strings.Contains(resp.Debug.Prompt.Messages[0].Content, `{"source":true}`) {
-		t.Fatalf("debug prompt messages = %#v, want rendered input content", resp.Debug.Prompt.Messages)
-	}
-	if resp.Debug.Response == nil || resp.Debug.Response.Content != `{"ok":true}` {
-		t.Fatalf("debug response = %#v, want raw response content", resp.Debug.Response)
-	}
-	debugJSON, err := json.Marshal(resp.Debug)
-	if err != nil {
-		t.Fatalf("marshal debug material: %v", err)
-	}
-	if strings.Contains(string(debugJSON), "secret-token") || strings.Contains(string(debugJSON), "sk-") {
-		t.Fatalf("debug material contains secret material: %s", debugJSON)
-	}
-	gotReq := fake.lastRequest()
-	if gotReq.Prompt.SessionID != "session-123" {
-		t.Fatalf("session id = %q, want session-123", gotReq.Prompt.SessionID)
-	}
-	if gotReq.Target.Model != "explicit-model" {
-		t.Fatalf("model = %q, want explicit-model", gotReq.Target.Model)
-	}
-	if len(gotReq.Prompt.Messages) != 1 || !strings.Contains(gotReq.Prompt.Messages[0].Content, `{"source":true}`) {
-		t.Fatalf("rendered messages = %#v, want transcript input content", gotReq.Prompt.Messages)
-	}
-	if gotReq.StructuredOutput == nil {
-		t.Fatalf("structured output = nil, want JSON schema")
-	}
-	manifests := client.LLMProfileManifests()
-	if len(manifests) != 1 || manifests[0].ID != "explicit-profile" || manifests[0].Model != "explicit-model" {
-		t.Fatalf("profile manifests = %#v", manifests)
-	}
-}
-
-func TestScriptoriumClientUsesPromptDefaultProfileWhenRequestProfileEmpty(t *testing.T) {
-	fake := &fakeScriptoriumLLM{content: `{"ok":true}`}
-	client := newTestScriptoriumClient(t, fake)
-
-	var out map[string]any
-	if _, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		PromptID:  "adapter.test",
-		SessionID: "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out); err != nil {
-		t.Fatalf("CompleteStructured() error = %v, want nil", err)
-	}
-	if got := fake.lastRequest().Target.Model; got != "default-model" {
-		t.Fatalf("model = %q, want prompt default profile model", got)
-	}
-}
-
-func TestScriptoriumClientValidationFailureReturnsError(t *testing.T) {
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{content: `{"bad":true}`})
-
-	var out map[string]any
-	resp, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		PromptID:  "adapter.test",
-		SessionID: "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out)
-	if err == nil || !errors.Is(err, contracts.ErrInvalidStructuredOutput) || !strings.Contains(err.Error(), "validation failed") {
-		t.Fatalf("CompleteStructured() error = %v, want validation failure", err)
-	}
-	if got := string(resp.Content); got != `{"bad":true}` {
-		t.Fatalf("response content = %q, want raw failed output", got)
-	}
-	if resp.Debug == nil || resp.Debug.Response == nil || resp.Debug.Response.Content != `{"bad":true}` {
-		t.Fatalf("debug response = %#v, want raw failed output", resp.Debug)
-	}
-	if resp.Debug.Prompt == nil || len(resp.Debug.Prompt.Messages) == 0 {
-		t.Fatalf("debug prompt = %#v, want prepared prompt material", resp.Debug.Prompt)
-	}
-}
-
-func TestScriptoriumClientDecodeFailureReturnsRawResponse(t *testing.T) {
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{content: `{"ok":true}`})
-
-	var out []any
-	resp, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		PromptID:  "adapter.test",
-		SessionID: "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out)
-	if err == nil || !errors.Is(err, contracts.ErrInvalidStructuredOutput) || !strings.Contains(err.Error(), "decode Scriptorium structured output") {
-		t.Fatalf("CompleteStructured() error = %v, want decode failure", err)
-	}
-	if got := string(resp.Content); got != `{"ok":true}` {
-		t.Fatalf("response content = %q, want raw decode-failed output", got)
-	}
-	if resp.Debug == nil || resp.Debug.Response == nil || resp.Debug.Response.Content != `{"ok":true}` {
-		t.Fatalf("debug response = %#v, want raw decode-failed output", resp.Debug)
-	}
-}
-
-func TestScriptoriumClientProviderFailureIncludesContextAndRedactsBearerToken(t *testing.T) {
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{err: errors.New("provider failed with Bearer secret-token")})
-
-	var out map[string]any
-	resp, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		PromptID:  "adapter.test",
-		SessionID: "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out)
-	if err == nil {
-		t.Fatalf("CompleteStructured() error = nil, want provider error")
-	}
-	if errors.Is(err, contracts.ErrInvalidStructuredOutput) {
-		t.Fatalf("provider error = %v, must not be classified as invalid structured output", err)
-	}
-	if !strings.Contains(err.Error(), `run Scriptorium prompt "adapter.test"`) {
-		t.Fatalf("error = %q, want operation context", err.Error())
-	}
-	if strings.Contains(err.Error(), "secret-token") || !strings.Contains(err.Error(), "Bearer [REDACTED]") {
-		t.Fatalf("error = %q, want redacted bearer token", err.Error())
-	}
-	if resp.Debug != nil {
-		t.Fatalf("debug material = %#v, want none for provider failure without result", resp.Debug)
-	}
-}
-
-func TestScriptoriumClientContextCancellationIsRespected(t *testing.T) {
-	ctx, cancel := context.WithCancel(context.Background())
-	cancel()
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{content: `{"ok":true}`})
-
-	var out map[string]any
-	_, err := client.CompleteStructured(ctx, contracts.StructuredCompletionRequest{
-		PromptID: "adapter.test",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out)
-	if !errors.Is(err, context.Canceled) || errors.Is(err, contracts.ErrInvalidStructuredOutput) {
-		t.Fatalf("CompleteStructured() error = %v, want context canceled", err)
-	}
-}
-
-func TestScriptoriumClientClassifiesEmptyStructuredCompletion(t *testing.T) {
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{allowEmpty: true})
-
-	var out map[string]any
-	_, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-		PromptID:  "adapter.test",
-		SessionID: "session-123",
-		Inputs: contracts.LLMInputSet{
-			"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-		},
-	}, &out)
-	if !errors.Is(err, contracts.ErrInvalidStructuredOutput) {
-		t.Fatalf("CompleteStructured() error = %v, want invalid structured output", err)
-	}
-}
-
-func TestScheduledScriptoriumClientBoundsConcurrentCalls(t *testing.T) {
-	fake := &fakeScriptoriumLLM{
-		content: `{"ok":true}`,
-		block:   make(chan struct{}),
-	}
-	client := newTestScriptoriumClient(t, fake)
-	scheduler, err := NewScheduler(1)
-	if err != nil {
-		t.Fatalf("NewScheduler() error = %v, want nil", err)
-	}
-	scheduled := NewScheduledClient(client, scheduler)
-
-	var wg sync.WaitGroup
-	for i := 0; i < 3; i++ {
-		wg.Add(1)
-		go func() {
-			defer wg.Done()
-			var out map[string]any
-			_, callErr := scheduled.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
-				PromptID:  "adapter.test",
-				SessionID: "session-123",
-				Inputs: contracts.LLMInputSet{
-					"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
-				},
-			}, &out)
-			if callErr != nil {
-				t.Errorf("CompleteStructured() error = %v, want nil", callErr)
-			}
-		}()
-	}
-	waitForAtomicAtLeast(t, &fake.calls, 1)
-	time.Sleep(20 * time.Millisecond)
-	if got := atomic.LoadInt32(&fake.maxInFlight); got > 1 {
-		t.Fatalf("max in-flight calls = %d, want <= 1", got)
-	}
-	close(fake.block)
-	wg.Wait()
-}
-
-func TestScriptoriumClientValidatesRequest(t *testing.T) {
-	client := newTestScriptoriumClient(t, &fakeScriptoriumLLM{content: `{"ok":true}`})
-	var out map[string]any
-	if _, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{}, &out); err == nil || !strings.Contains(err.Error(), "prompt_id") {
-		t.Fatalf("missing prompt id error = %v, want prompt_id validation", err)
-	}
-	if _, err := client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{PromptID: "adapter.test"}, nil); err == nil || !strings.Contains(err.Error(), "non-nil pointer") {
-		t.Fatalf("nil output error = %v, want output validation", err)
-	}
-}
-
-func newTestScriptoriumClient(t *testing.T, fake *fakeScriptoriumLLM) *ScriptoriumClient {
-	t.Helper()
-	registry := NewAssetRegistry()
-	if err := registry.RegisterPromptFS(fstest.MapFS{
-		"adapter.test.yaml": {Data: []byte(`id: adapter.test
-version: "v1"
-default_profile: default-profile
-session_id: "{{ .session_id }}"
-inputs:
-  - name: transcript
-    required: true
-    content_type: application/json
-messages:
-  - role: user
-    content: "Transcript: {{ input \"transcript\" }}"
-output:
-  format: json
-  validation_mode: json_schema
-  schema_path: adapter.schema.json
-  repair_attempts: 0
-`)},
-	}, "."); err != nil {
-		t.Fatalf("RegisterPromptFS() error = %v", err)
-	}
-	if err := registry.RegisterSchemaFS(fstest.MapFS{
-		"adapter.schema.json": {Data: []byte(`{"type":"object","required":["ok"],"properties":{"ok":{"type":"boolean"}}}`)},
-	}, "."); err != nil {
-		t.Fatalf("RegisterSchemaFS() error = %v", err)
-	}
-	client, err := NewScriptoriumClient(ScriptoriumClientConfig{
-		Assets: registry,
-		EngineOptions: []scriptorium.Option{
-			scriptorium.WithProfiles(
-				scriptorium.OpenAICompatibleProfile(scriptorium.OpenAICompatibleProfileConfig{
-					ID:       "default-profile",
-					Endpoint: "http://127.0.0.1:1/v1",
-					Model:    "default-model",
-				}),
-				scriptorium.OpenAICompatibleProfile(scriptorium.OpenAICompatibleProfileConfig{
-					ID:       "explicit-profile",
-					Endpoint: "http://127.0.0.1:1/v1",
-					Model:    "explicit-model",
-				}),
-			),
-			scriptorium.WithLLMClient(fake),
-		},
-	})
-	if err != nil {
-		t.Fatalf("NewScriptoriumClient() error = %v, want nil", err)
-	}
-	return client
-}
-
-type fakeScriptoriumLLM struct {
-	content     string
-	allowEmpty  bool
-	err         error
-	block       chan struct{}
-	mu          sync.Mutex
-	last        scriptorium.GenerateRequest
-	calls       int32
-	inFlight    int32
-	maxInFlight int32
-}
-
-func (f *fakeScriptoriumLLM) Generate(ctx context.Context, req scriptorium.GenerateRequest) (*scriptorium.GenerateResponse, error) {
-	f.mu.Lock()
-	f.last = req
-	f.mu.Unlock()
-	atomic.AddInt32(&f.calls, 1)
-	current := atomic.AddInt32(&f.inFlight, 1)
-	for {
-		seen := atomic.LoadInt32(&f.maxInFlight)
-		if current <= seen || atomic.CompareAndSwapInt32(&f.maxInFlight, seen, current) {
-			break
-		}
-	}
-	defer atomic.AddInt32(&f.inFlight, -1)
-	if f.block != nil {
-		select {
-		case <-f.block:
-		case <-ctx.Done():
-			return nil, ctx.Err()
-		}
-	}
-	if f.err != nil {
-		return nil, f.err
-	}
-	content := f.content
-	if content == "" && !f.allowEmpty {
-		content = `{"ok":true}`
-	}
-	if !f.allowEmpty && !json.Valid([]byte(content)) {
-		return nil, errors.New("test fake must return JSON content")
-	}
-	return &scriptorium.GenerateResponse{
-		Content: content,
-		Usage: scriptorium.TokenUsage{
-			PromptTokens:     11,
-			CompletionTokens: 7,
-			TotalTokens:      18,
-		},
-	}, nil
-}
-
-func (f *fakeScriptoriumLLM) lastRequest() scriptorium.GenerateRequest {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-	return f.last
-}

internal/framework/pipeline/artifact_evidence_registry.go

@@ -0,0 +1,148 @@
+package pipeline
+
+import (
+	"fmt"
+	"reflect"
+	"sort"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+// ArtifactEvidenceProjector returns the direct source references represented
+// by one normalized artifact value.
+type ArtifactEvidenceProjector[T any] func(T) []source.SourceRef
+
+// ArtifactEvidenceRegistry keeps the typed projection boundary private while
+// allowing preparation and execution to discover registered capabilities.
+type ArtifactEvidenceRegistry struct {
+	entries map[contracts.ArtifactKind]artifactEvidenceEntry
+}
+
+type artifactEvidenceEntry struct {
+	valueType reflect.Type
+	project   func(any) ([]source.SourceRef, error)
+}
+
+func NewArtifactEvidenceRegistry() *ArtifactEvidenceRegistry {
+	return &ArtifactEvidenceRegistry{entries: make(map[contracts.ArtifactKind]artifactEvidenceEntry)}
+}
+
+// RegisterArtifactEvidence registers one evidence projector for an artifact
+// kind. Projectors are invoked only after an exact Go-type check.
+func RegisterArtifactEvidence[T any](registry *ArtifactEvidenceRegistry, kind contracts.ArtifactKind, projector ArtifactEvidenceProjector[T]) error {
+	if registry == nil {
+		return fmt.Errorf("artifact evidence registry must not be nil")
+	}
+	if projector == nil {
+		return fmt.Errorf("artifact evidence projector must not be nil")
+	}
+	kind = normalizeArtifactKind(kind)
+	if kind == "" {
+		return fmt.Errorf("artifact evidence kind must not be empty")
+	}
+	if _, ok := registry.entries[kind]; ok {
+		return fmt.Errorf("artifact evidence %q is already registered", kind)
+	}
+	valueType := reflect.TypeFor[T]()
+	if registry.entries == nil {
+		registry.entries = make(map[contracts.ArtifactKind]artifactEvidenceEntry)
+	}
+	registry.entries[kind] = artifactEvidenceEntry{
+		valueType: valueType,
+		project: func(value any) ([]source.SourceRef, error) {
+			actualType := reflect.TypeOf(value)
+			if actualType != valueType {
+				return nil, newArtifactEvidenceTypeError(kind, valueType, actualType)
+			}
+			typed, ok := value.(T)
+			if !ok {
+				return nil, newArtifactEvidenceTypeError(kind, valueType, actualType)
+			}
+			return append([]source.SourceRef(nil), projector(typed)...), nil
+		},
+	}
+	return nil
+}
+
+func (r *ArtifactEvidenceRegistry) RegisteredKinds() []contracts.ArtifactKind {
+	if r == nil || len(r.entries) == 0 {
+		return nil
+	}
+	kinds := make([]contracts.ArtifactKind, 0, len(r.entries))
+	for kind := range r.entries {
+		kinds = append(kinds, kind)
+	}
+	sort.Slice(kinds, func(i, j int) bool { return kinds[i] < kinds[j] })
+	return kinds
+}
+
+// Project returns independently owned direct references for a registered
+// artifact value.
+func (r *ArtifactEvidenceRegistry) Project(kind contracts.ArtifactKind, value any) ([]source.SourceRef, error) {
+	entry, normalizedKind, err := r.entry(kind)
+	if err != nil {
+		return nil, err
+	}
+	refs, err := entry.project(value)
+	if err != nil {
+		return nil, fmt.Errorf("project artifact evidence %q: %w", normalizedKind, err)
+	}
+	return append([]source.SourceRef(nil), refs...), nil
+}
+
+func (r *ArtifactEvidenceRegistry) entry(kind contracts.ArtifactKind) (artifactEvidenceEntry, contracts.ArtifactKind, error) {
+	if r == nil {
+		return artifactEvidenceEntry{}, "", fmt.Errorf("artifact evidence registry must not be nil")
+	}
+	kind = normalizeArtifactKind(kind)
+	if kind == "" {
+		return artifactEvidenceEntry{}, "", fmt.Errorf("artifact evidence kind must not be empty")
+	}
+	entry, ok := r.entries[kind]
+	if !ok {
+		return artifactEvidenceEntry{}, kind, fmt.Errorf("artifact evidence %q is not registered", kind)
+	}
+	return entry, kind, nil
+}
+
+func (r *ArtifactEvidenceRegistry) valueType(kind contracts.ArtifactKind) (reflect.Type, bool) {
+	if r == nil {
+		return nil, false
+	}
+	entry, ok := r.entries[normalizeArtifactKind(kind)]
+	if !ok {
+		return nil, false
+	}
+	return entry.valueType, true
+}
+
+func newArtifactEvidenceTypeError(kind contracts.ArtifactKind, expected, actual reflect.Type) error {
+	actualName := "<nil>"
+	if actual != nil {
+		actualName = actual.String()
+	}
+	return fmt.Errorf("project artifact evidence %q: expected exact Go type %s, got %s", kind, expected, actualName)
+}
+
+func normalizeEvidenceLaneIDs(values []string) ([]string, error) {
+	if len(values) == 0 {
+		return nil, fmt.Errorf("evidence lane ids must not be empty")
+	}
+	seen := make(map[string]struct{}, len(values))
+	lanes := make([]string, 0, len(values))
+	for _, raw := range values {
+		lane := strings.TrimSpace(raw)
+		if lane == "" {
+			return nil, fmt.Errorf("evidence lane id must not be empty")
+		}
+		if _, ok := seen[lane]; ok {
+			return nil, fmt.Errorf("evidence lane id %q is duplicated", lane)
+		}
+		seen[lane] = struct{}{}
+		lanes = append(lanes, lane)
+	}
+	sort.Strings(lanes)
+	return lanes, nil
+}

internal/framework/pipeline/artifact_evidence_registry_test.go

@@ -0,0 +1,55 @@
+package pipeline
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+func TestArtifactEvidenceRegistryProjectsExactRegisteredTypesWithOwnedReferences(t *testing.T) {
+	registry := NewArtifactEvidenceRegistry()
+	input := codecNotes{Items: []string{"one"}}
+	refs := []source.SourceRef{{SourceID: "source", StartUnitID: 2, EndUnitID: 2}}
+	if err := RegisterArtifactEvidence(registry, "test/notes", func(codecNotes) []source.SourceRef { return refs }); err != nil {
+		t.Fatalf("RegisterArtifactEvidence() error = %v, want nil", err)
+	}
+	if err := RegisterArtifactEvidence(registry, "test/score", func(codecScore) []source.SourceRef { return nil }); err != nil {
+		t.Fatalf("RegisterArtifactEvidence() second kind error = %v, want nil", err)
+	}
+	if got, want := registry.RegisteredKinds(), []contracts.ArtifactKind{"test/notes", "test/score"}; !reflect.DeepEqual(got, want) {
+		t.Fatalf("RegisteredKinds() = %#v, want %#v", got, want)
+	}
+	projected, err := registry.Project(" test/notes ", input)
+	if err != nil {
+		t.Fatalf("Project() error = %v, want nil", err)
+	}
+	projected[0].StartUnitID = 99
+	if refs[0].StartUnitID != 2 {
+		t.Fatal("Project() retained projector reference storage")
+	}
+	if _, err := registry.Project("test/notes", codecNotesAlias(input)); err == nil || !strings.Contains(err.Error(), "expected exact Go type") {
+		t.Fatalf("Project() exact type error = %v, want exact type failure", err)
+	}
+}
+
+func TestArtifactEvidenceRegistryRejectsInvalidRegistration(t *testing.T) {
+	registry := NewArtifactEvidenceRegistry()
+	if err := RegisterArtifactEvidence[codecNotes](nil, "test/notes", func(codecNotes) []source.SourceRef { return nil }); err == nil || !strings.Contains(err.Error(), "must not be nil") {
+		t.Fatalf("nil registry error = %v, want failure", err)
+	}
+	if err := RegisterArtifactEvidence(registry, " ", func(codecNotes) []source.SourceRef { return nil }); err == nil || !strings.Contains(err.Error(), "must not be empty") {
+		t.Fatalf("blank kind error = %v, want failure", err)
+	}
+	if err := RegisterArtifactEvidence(registry, "test/notes", ArtifactEvidenceProjector[codecNotes](nil)); err == nil || !strings.Contains(err.Error(), "must not be nil") {
+		t.Fatalf("nil projector error = %v, want failure", err)
+	}
+	if err := RegisterArtifactEvidence(registry, "test/notes", func(codecNotes) []source.SourceRef { return nil }); err != nil {
+		t.Fatal(err)
+	}
+	if err := RegisterArtifactEvidence(registry, "test/notes", func(codecNotes) []source.SourceRef { return nil }); err == nil || !strings.Contains(err.Error(), "already registered") {
+		t.Fatalf("duplicate kind error = %v, want failure", err)
+	}
+}

internal/framework/pipeline/chunker_registry.go

@@ -24,10 +24,6 @@ func NewChunkerRegistry() *ChunkerRegistry {
 	}
 }
 
-func (r *ChunkerRegistry) Register(key string, constructor ChunkerConstructor) error {
-	return r.RegisterWithSpec(defaultModuleSpec(key, StageChunk), constructor)
-}
-
 func (r *ChunkerRegistry) RegisterWithSpec(spec ModuleSpec, constructor ChunkerConstructor) error {
 	if constructor == nil {
 		return fmt.Errorf("chunker constructor for %q must not be nil", strings.TrimSpace(spec.Key))

internal/framework/pipeline/chunker_registry_test.go

@@ -29,6 +29,10 @@ type registryBehaviorCase[M any] struct {
 	moduleKey        func(M) string
 }
 
+func testModuleSpec(key string, stage ModuleStage) ModuleSpec {
+	return ModuleSpec{Key: key, Stage: stage, ExecutionClass: contracts.ExecutionClassDeterministic}
+}
+
 func TestChunkerRegistryBehavior(t *testing.T) {
 	runRegistryBehaviorTests(t, registryBehaviorCase[contracts.Chunker]{
 		name:       "ChunkerRegistry",
@@ -39,7 +43,9 @@ func TestChunkerRegistryBehavior(t *testing.T) {
 			return NewChunkerRegistry()
 		},
 		register: func(registry any, key string, constructor func() (contracts.Chunker, error)) error {
-			return registry.(*ChunkerRegistry).Register(key, constructor)
+			return registry.(*ChunkerRegistry).RegisterWithSpec(ModuleSpec{
+				Key: key, Stage: StageChunk, ExecutionClass: contracts.ExecutionClassDeterministic,
+			}, constructor)
 		},
 		registerWithSpec: func(registry any, spec ModuleSpec, constructor func() (contracts.Chunker, error)) error {
 			return registry.(*ChunkerRegistry).RegisterWithSpec(spec, constructor)
@@ -55,7 +61,9 @@ func TestChunkerRegistryBehavior(t *testing.T) {
 		},
 		nilRegister: func(key string, constructor func() (contracts.Chunker, error)) error {
 			var registry *ChunkerRegistry
-			return registry.Register(key, constructor)
+			return registry.RegisterWithSpec(ModuleSpec{
+				Key: key, Stage: StageChunk, ExecutionClass: contracts.ExecutionClassDeterministic,
+			}, constructor)
 		},
 		nilBuild: func(key string) (contracts.Chunker, error) {
 			var registry *ChunkerRegistry
@@ -101,10 +109,11 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 	t.Run(testCase.name+"/metadata registration and lookup", func(t *testing.T) {
 		registry := testCase.newRegistry()
 		spec := ModuleSpec{
-			Key:      " " + testCase.key + " ",
-			Stage:    testCase.stage,
-			Provides: []string{" beta ", "alpha", "", "beta"},
-			Requires: []string{" source ", "source", ""},
+			Key:            " " + testCase.key + " ",
+			Stage:          testCase.stage,
+			ExecutionClass: contracts.ExecutionClassDeterministic,
+			Provides:       []string{" beta ", "alpha", "", "beta"},
+			Requires:       []string{" source ", "source", ""},
 		}
 		if err := testCase.registerWithSpec(registry, spec, testCase.constructor(testCase.key)); err != nil {
 			t.Fatalf("RegisterWithSpec() error = %v, want nil", err)
@@ -115,10 +124,11 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 			t.Fatal("Spec() ok = false, want true")
 		}
 		want := ModuleSpec{
-			Key:      testCase.key,
-			Stage:    testCase.stage,
-			Provides: []string{"alpha", "beta"},
-			Requires: []string{"source"},
+			Key:            testCase.key,
+			Stage:          testCase.stage,
+			ExecutionClass: contracts.ExecutionClassDeterministic,
+			Provides:       []string{"alpha", "beta"},
+			Requires:       []string{"source"},
 		}
 		if !reflect.DeepEqual(got, want) {
 			t.Fatalf("Spec() = %#v, want %#v", got, want)
@@ -134,7 +144,7 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 		}
 	})
 
-	t.Run(testCase.name+"/default spec from register", func(t *testing.T) {
+	t.Run(testCase.name+"/minimal explicit spec registration", func(t *testing.T) {
 		registry := testCase.newRegistry()
 		if err := testCase.register(registry, " "+testCase.key+" ", testCase.constructor(testCase.key)); err != nil {
 			t.Fatalf("Register() error = %v, want nil", err)
@@ -144,7 +154,7 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 		if !ok {
 			t.Fatal("Spec() ok = false, want true")
 		}
-		want := ModuleSpec{Key: testCase.key, Stage: testCase.stage}
+		want := ModuleSpec{Key: testCase.key, Stage: testCase.stage, ExecutionClass: contracts.ExecutionClassDeterministic}
 		if !reflect.DeepEqual(spec, want) {
 			t.Fatalf("Spec() = %#v, want %#v", spec, want)
 		}
@@ -152,7 +162,7 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 
 	t.Run(testCase.name+"/wrong stage rejection", func(t *testing.T) {
 		registry := testCase.newRegistry()
-		err := testCase.registerWithSpec(registry, ModuleSpec{Key: testCase.key, Stage: testCase.wrongStage}, testCase.constructor(testCase.key))
+		err := testCase.registerWithSpec(registry, ModuleSpec{Key: testCase.key, Stage: testCase.wrongStage, ExecutionClass: contracts.ExecutionClassDeterministic}, testCase.constructor(testCase.key))
 		if err == nil {
 			t.Fatal("RegisterWithSpec() error = nil, want error")
 		}
@@ -161,6 +171,22 @@ func runRegistryBehaviorTests[M any](t *testing.T, testCase registryBehaviorCase
 		}
 	})
 
+	t.Run(testCase.name+"/execution class rejection", func(t *testing.T) {
+		for _, spec := range []ModuleSpec{
+			{Key: testCase.key, Stage: testCase.stage},
+			{Key: testCase.key, Stage: testCase.stage, ExecutionClass: contracts.ExecutionClass("unsupported")},
+		} {
+			registry := testCase.newRegistry()
+			err := testCase.registerWithSpec(registry, spec, testCase.constructor(testCase.key))
+			if err == nil {
+				t.Fatal("RegisterWithSpec() error = nil, want execution class error")
+			}
+			if !strings.Contains(err.Error(), "execution class") {
+				t.Fatalf("RegisterWithSpec() error = %q, want execution class error", err.Error())
+			}
+		}
+	})
+
 	t.Run(testCase.name+"/key trimming", func(t *testing.T) {
 		registry := testCase.newRegistry()
 		if err := testCase.register(registry, " "+testCase.key+" ", testCase.constructor(testCase.key)); err != nil {

internal/framework/pipeline/evidence_output.go

@@ -0,0 +1,111 @@
+package pipeline
+
+import (
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/evidencecontext"
+)
+
+// debugEvidenceContextSummary intentionally contains only publication-safe
+// identifiers and aggregate counts. The evidence document itself can include
+// source text and must never be written to this debug envelope.
+type debugEvidenceContextSummary struct {
+	ArtifactKind  contracts.ArtifactKind `json:"artifact_kind"`
+	MediaType     string                 `json:"media_type"`
+	SchemaID      string                 `json:"schema_id"`
+	SchemaName    string                 `json:"schema_name"`
+	SchemaVersion string                 `json:"schema_version"`
+	SelectedLanes []string               `json:"selected_lanes"`
+	WindowUnits   int                    `json:"window_units"`
+	ContextCount  int                    `json:"context_count"`
+	UnitCount     int                    `json:"unit_count"`
+	SourceDigest  string                 `json:"source_digest"`
+}
+
+// buildOutputEvidenceContext projects the prepared output policy from accepted
+// normalized artifacts. It is intentionally separate from lane execution so
+// checkpointed normalized outputs use the same reconstruction path.
+func buildOutputEvidenceContext(prepared *PreparedPipeline, doc *source.SourceDocument, outputs []contracts.SerializedOutput) (*contracts.SerializedArtifact, *debugEvidenceContextSummary, error) {
+	if prepared == nil || prepared.evidencePlan == nil {
+		return nil, nil, nil
+	}
+	if doc == nil {
+		return nil, nil, fmt.Errorf("evidence context output: source document is unavailable")
+	}
+	if prepared.artifactCodecs == nil {
+		return nil, nil, fmt.Errorf("evidence context output: artifact codecs are unavailable")
+	}
+
+	byLane := make(map[string]contracts.SerializedOutput, len(outputs))
+	for _, output := range outputs {
+		laneID := strings.TrimSpace(output.LaneID)
+		if _, exists := byLane[laneID]; exists {
+			return nil, nil, fmt.Errorf("evidence context output: accepted normalized outputs contain duplicate lane %q", laneID)
+		}
+		byLane[laneID] = contracts.CloneSerializedOutput(output)
+	}
+
+	request := evidencecontext.BuildRequest{
+		Source:        doc,
+		WindowUnits:   prepared.evidencePlan.policy.WindowUnits,
+		SelectedLanes: append([]string(nil), prepared.evidencePlan.policy.LaneIDs...),
+		LaneEvidence:  make([]evidencecontext.LaneEvidence, 0, len(prepared.evidencePlan.lanes)),
+	}
+	for _, lane := range prepared.evidencePlan.lanes {
+		output, ok := byLane[lane.laneID]
+		if !ok {
+			continue
+		}
+		if output.SourceID != doc.ID {
+			return nil, nil, fmt.Errorf("evidence context output lane %q: accepted normalized output source is incompatible", lane.laneID)
+		}
+		if output.Artifact.Kind != lane.kind {
+			return nil, nil, fmt.Errorf("evidence context output lane %q: accepted normalized output artifact kind is incompatible", lane.laneID)
+		}
+		value, err := prepared.artifactCodecs.Decode(contracts.CloneSerializedArtifact(output.Artifact))
+		if err != nil {
+			return nil, nil, fmt.Errorf("evidence context output lane %q: accepted normalized artifact cannot be decoded", lane.laneID)
+		}
+		references, err := lane.project(value)
+		if err != nil {
+			return nil, nil, fmt.Errorf("evidence context output lane %q: accepted normalized artifact cannot be projected", lane.laneID)
+		}
+		request.LaneEvidence = append(request.LaneEvidence, evidencecontext.LaneEvidence{
+			LaneID:     lane.laneID,
+			SourceRefs: append([]source.SourceRef(nil), references...),
+		})
+	}
+
+	document, err := evidencecontext.Build(request)
+	if err != nil {
+		return nil, nil, fmt.Errorf("evidence context output: accepted evidence references are invalid")
+	}
+	content, err := evidencecontext.New().Encode(document)
+	if err != nil {
+		return nil, nil, fmt.Errorf("evidence context output: evidence context serialization failed")
+	}
+	artifact := &contracts.SerializedArtifact{
+		Kind:      evidencecontext.ArtifactKind,
+		Schema:    evidencecontext.New().Schema(),
+		MediaType: evidencecontext.MediaType,
+		Content:   content,
+	}
+	summary := debugEvidenceContextSummary{
+		ArtifactKind:  artifact.Kind,
+		MediaType:     artifact.MediaType,
+		SchemaID:      artifact.Schema.ID,
+		SchemaName:    artifact.Schema.Name,
+		SchemaVersion: artifact.Schema.Version,
+		SelectedLanes: append([]string(nil), document.SelectedLanes...),
+		WindowUnits:   document.WindowUnits,
+		ContextCount:  len(document.Contexts),
+		SourceDigest:  document.SourceDigest,
+	}
+	for _, context := range document.Contexts {
+		summary.UnitCount += len(context.Units)
+	}
+	return contracts.CloneSerializedArtifactPointer(artifact), &summary, nil
+}

internal/framework/pipeline/evidence_output_test.go

@@ -0,0 +1,252 @@
+package pipeline
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/evidencecontext"
+)
+
+type capturingEvidenceOutput struct {
+	requests []contracts.OutputRequest
+}
+
+func (*capturingEvidenceOutput) Key() string { return "capture/evidence-context" }
+
+func (output *capturingEvidenceOutput) Encode(_ context.Context, request contracts.OutputRequest) (contracts.OutputResult, error) {
+	request.EvidenceContext = contracts.CloneSerializedArtifactPointer(request.EvidenceContext)
+	output.requests = append(output.requests, request)
+	return contracts.OutputResult{}, nil
+}
+
+func installEvidencePlan(prepared *PreparedPipeline, window int, selected []string, project func(codecNotes) ([]source.SourceRef, error)) {
+	lanes := make([]preparedEvidenceLane, 0, len(selected))
+	for _, laneID := range selected {
+		for _, step := range prepared.Steps {
+			for _, lane := range step.lanes {
+				if lane.resolved.ID != laneID {
+					continue
+				}
+				lanes = append(lanes, preparedEvidenceLane{
+					laneID: laneID,
+					kind:   lane.resolved.ArtifactKind,
+					project: func(value any) ([]source.SourceRef, error) {
+						notes, ok := value.(codecNotes)
+						if !ok {
+							return nil, errors.New("unexpected artifact type")
+						}
+						return project(notes)
+					},
+				})
+			}
+		}
+	}
+	prepared.evidencePlan = &preparedEvidencePlan{policy: EvidenceContextPolicy{Enabled: true, WindowUnits: window, LaneIDs: append([]string(nil), selected...)}, lanes: lanes}
+}
+
+func setNormalizedNotes(prepared *PreparedPipeline, values map[string]codecNotes) {
+	for stepIndex := range prepared.Steps {
+		for laneIndex := range prepared.Steps[stepIndex].lanes {
+			lane := &prepared.Steps[stepIndex].lanes[laneIndex]
+			value, ok := values[lane.resolved.ID]
+			if !ok {
+				continue
+			}
+			lane.typed.normalize = func(context.Context, any, contracts.TypedNormalizeRequest[any]) (erasedTypedResult, error) {
+				return erasedTypedResult{Value: value}, nil
+			}
+		}
+	}
+}
+
+func decodeCapturedEvidence(t *testing.T, output *capturingEvidenceOutput) evidencecontext.Document {
+	t.Helper()
+	if len(output.requests) != 1 || output.requests[0].EvidenceContext == nil {
+		t.Fatalf("output requests = %#v, want one evidence context", output.requests)
+	}
+	value, err := evidencecontext.New().Decode(output.requests[0].EvidenceContext.Content)
+	if err != nil {
+		t.Fatalf("Decode(evidence context): %v", err)
+	}
+	return value
+}
+
+func TestRunnerBuildsEvidenceContextFromSelectedNormalizedOutputs(t *testing.T) {
+	prepared := preparedOrderedPipeline(t, 3, orderedLaneSpec{id: "alpha", profile: "notes"}, orderedLaneSpec{id: "beta", profile: "notes"})
+	encoder := &capturingEvidenceOutput{}
+	prepared.output = encoder
+	setNormalizedNotes(prepared, map[string]codecNotes{"alpha": {Items: []string{"one"}}, "beta": {Items: []string{"two"}}})
+	installEvidencePlan(prepared, 1, []string{"alpha", "beta", "inactive"}, func(notes codecNotes) ([]source.SourceRef, error) {
+		switch notes.Items[0] {
+		case "one":
+			return []source.SourceRef{{SourceID: "source", StartUnitID: 1, EndUnitID: 1}}, nil
+		case "two":
+			return []source.SourceRef{{SourceID: "source", StartUnitID: 2, EndUnitID: 2}}, nil
+		default:
+			return nil, nil
+		}
+	})
+	debug := newCapturedDebugRecorder()
+	if _, err := New().Run(context.Background(), RunInput{Prepared: prepared, RawInput: []byte("input"), Debug: debug}); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+
+	value := decodeCapturedEvidence(t, encoder)
+	if !reflect.DeepEqual(value.SelectedLanes, []string{"alpha", "beta", "inactive"}) || len(value.Contexts) != 1 || len(value.Contexts[0].Units) != 3 {
+		t.Fatalf("evidence context = %#v, want selected union", value)
+	}
+	if got := value.Contexts[0].EvidenceRefs; len(got) != 2 || got[0].LaneID != "alpha" || got[1].LaneID != "beta" {
+		t.Fatalf("evidence refs = %#v, want both selected lanes", got)
+	}
+	debugJSON := string(debug.json["output/evidence-context.json"])
+	if strings.Contains(debugJSON, "text-1") || strings.Contains(debugJSON, "metadata") || !strings.Contains(debugJSON, `"artifact_kind":"source/evidence-context"`) || !strings.Contains(debugJSON, `"schema_id":"notarius.source.evidence_context"`) || !strings.Contains(debugJSON, `"context_count":1`) || !strings.Contains(debugJSON, `"unit_count":3`) {
+		t.Fatalf("evidence debug envelope = %s, want only allowlisted summary", debugJSON)
+	}
+}
+
+func TestRunnerEvidenceContextOmitsAbsentAndRejectedLanes(t *testing.T) {
+	prepared := preparedOrderedPipeline(t, 2, orderedLaneSpec{id: "present", profile: "notes"}, orderedLaneSpec{id: "rejected", profile: "notes"})
+	encoder := &capturingEvidenceOutput{}
+	prepared.output = encoder
+	setNormalizedNotes(prepared, map[string]codecNotes{"present": {Items: []string{"present"}}, "rejected": {Items: []string{"rejected"}}})
+	prepared.Steps[1].lanes[0].mergeValidators.validators = []preparedValidator{{
+		resolved: ResolvedValidator{Binding: Binding("reject"), Target: ValidatorTargetTyped, ArtifactKind: "test/notes"},
+		typedValidate: func(context.Context, any, typedValidationTarget) (contracts.ValidationResult, error) {
+			return contracts.ValidationResult{Approved: false, ReasonCode: "rejected", Message: "not accepted"}, nil
+		},
+	}}
+	installEvidencePlan(prepared, 0, []string{"absent", "present", "rejected"}, func(notes codecNotes) ([]source.SourceRef, error) {
+		if len(notes.Items) > 0 && notes.Items[0] == "present" {
+			return []source.SourceRef{{SourceID: "source", StartUnitID: 1, EndUnitID: 1}}, nil
+		}
+		return []source.SourceRef{{SourceID: "source", StartUnitID: 2, EndUnitID: 2}}, nil
+	})
+	result, err := New().Run(context.Background(), RunInput{Prepared: prepared, RawInput: []byte("input")})
+	if err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	if len(result.Rejected) != 1 || result.Rejected[0].LaneID != "rejected" {
+		t.Fatalf("rejections = %#v, want rejected lane unchanged", result.Rejected)
+	}
+	value := decodeCapturedEvidence(t, encoder)
+	if len(value.Contexts) != 1 || len(value.Contexts[0].EvidenceRefs) != 1 || value.Contexts[0].EvidenceRefs[0].LaneID != "present" {
+		t.Fatalf("evidence context = %#v, want present lane only", value)
+	}
+}
+
+func TestRunnerEvidenceContextFailurePreventsOutputEncoding(t *testing.T) {
+	prepared := preparedConcurrentPipeline(t, 1)
+	encoder := &capturingEvidenceOutput{}
+	prepared.output = encoder
+	setNormalizedNotes(prepared, map[string]codecNotes{"notes": {Items: []string{"invalid"}}})
+	installEvidencePlan(prepared, 0, []string{"notes"}, func(codecNotes) ([]source.SourceRef, error) {
+		return []source.SourceRef{{SourceID: "source", StartUnitID: 99, EndUnitID: 99}}, nil
+	})
+	result, err := New().Run(context.Background(), RunInput{Prepared: prepared, RawInput: []byte("input")})
+	if err == nil || !strings.Contains(err.Error(), "accepted evidence references are invalid") {
+		t.Fatalf("Run() error = %v, want evidence context failure", err)
+	}
+	if len(encoder.requests) != 0 || result.Manifest.ValidationStatus != "failed" || len(result.Rejected) != 0 {
+		t.Fatalf("output requests = %#v manifest = %#v rejected = %#v, want failed run before output encoding", encoder.requests, result.Manifest, result.Rejected)
+	}
+}
+
+func TestBuildOutputEvidenceContextRejectsIncompatibleAcceptedOutputs(t *testing.T) {
+	prepared := preparedConcurrentPipeline(t, 1)
+	doc := prepared.input.(*typedTestInput).doc
+	installEvidencePlan(prepared, 0, []string{"notes"}, func(codecNotes) ([]source.SourceRef, error) {
+		return []source.SourceRef{{SourceID: doc.ID, StartUnitID: 1, EndUnitID: 1}}, nil
+	})
+	artifact, err := prepared.artifactCodecs.Encode("test/notes", codecNotes{Items: []string{"valid"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	valid := contracts.SerializedOutput{LaneID: "notes", SourceID: doc.ID, Artifact: artifact}
+	for _, test := range []struct {
+		name    string
+		outputs []contracts.SerializedOutput
+		mutate  func(*contracts.SerializedOutput)
+		want    string
+	}{
+		{name: "duplicate lane", outputs: []contracts.SerializedOutput{valid, valid}, want: "duplicate lane"},
+		{name: "foreign source", outputs: []contracts.SerializedOutput{valid}, mutate: func(output *contracts.SerializedOutput) { output.SourceID = "other" }, want: "source is incompatible"},
+		{name: "wrong artifact kind", outputs: []contracts.SerializedOutput{valid}, mutate: func(output *contracts.SerializedOutput) { output.Artifact.Kind = "test/score" }, want: "artifact kind is incompatible"},
+		{name: "invalid artifact payload", outputs: []contracts.SerializedOutput{valid}, mutate: func(output *contracts.SerializedOutput) { output.Artifact.Content = []byte("not JSON") }, want: "cannot be decoded"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			outputs := append([]contracts.SerializedOutput(nil), test.outputs...)
+			for index := range outputs {
+				outputs[index] = contracts.CloneSerializedOutput(outputs[index])
+			}
+			if test.mutate != nil {
+				test.mutate(&outputs[0])
+			}
+			_, _, err := buildOutputEvidenceContext(prepared, doc, outputs)
+			if err == nil || !strings.Contains(err.Error(), test.want) {
+				t.Fatalf("buildOutputEvidenceContext() error = %v, want %q", err, test.want)
+			}
+			if strings.Contains(err.Error(), "not JSON") {
+				t.Fatalf("buildOutputEvidenceContext() exposed artifact payload: %v", err)
+			}
+		})
+	}
+}
+
+func TestEvidenceContextOutputRequestOwnsArtifactBytes(t *testing.T) {
+	artifact := &contracts.SerializedArtifact{Content: []byte("original"), Metadata: map[string]any{"source": "original"}}
+	request := contracts.OutputRequest{EvidenceContext: contracts.CloneSerializedArtifactPointer(artifact)}
+	request.EvidenceContext.Content[0] = 'X'
+	request.EvidenceContext.Metadata["source"] = "changed"
+	if string(artifact.Content) != "original" || artifact.Metadata["source"] != "original" {
+		t.Fatalf("output request evidence context aliases source artifact: %#v", artifact)
+	}
+}
+
+func TestRunnerEvidenceContextRebuildsFromAcceptedCheckpoint(t *testing.T) {
+	prepared := preparedConcurrentPipeline(t, 1)
+	encoder := &capturingEvidenceOutput{}
+	prepared.output = encoder
+	doc := prepared.input.(*typedTestInput).doc
+	lane := prepared.Steps[0].lanes[0]
+	stored, err := checkpointArtifact(lane.typed.codec, lane.resolved.ID, lane.resolved.Normalize.Module, doc.ID, codecNotes{Items: []string{"stored"}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	loader := newAcceptedCheckpointLoader()
+	key := CheckpointLaneKey(lane.resolved.StepID, lane.resolved.ID)
+	loader.accepted[key] = NormalizeCheckpoint{Output: stored}
+	loader.acceptedDecision[key] = NewCheckpointDecision(CheckpointDecisionReused, CheckpointReasonAcceptedArtifactReused)
+	installEvidencePlan(prepared, 0, []string{"notes"}, func(notes codecNotes) ([]source.SourceRef, error) {
+		if !reflect.DeepEqual(notes.Items, []string{"stored"}) {
+			return nil, errors.New("checkpoint artifact was not projected")
+		}
+		return []source.SourceRef{{SourceID: doc.ID, StartUnitID: 1, EndUnitID: 1}}, nil
+	})
+	policy := CheckpointExecutionPolicy{RequireReusableLanes: map[string]struct{}{key: {}}}
+	if _, err := New().Run(context.Background(), RunInput{Prepared: prepared, RawInput: []byte("input"), Checkpoint: loader, CheckpointPolicy: policy}); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	value := decodeCapturedEvidence(t, encoder)
+	if len(value.Contexts) != 1 || value.Contexts[0].EvidenceRefs[0].LaneID != "notes" {
+		t.Fatalf("evidence context = %#v, want checkpointed normalized output", value)
+	}
+}
+
+func TestRunnerSkipsEvidenceContextWhenOutputDoesNotOptIn(t *testing.T) {
+	prepared := preparedConcurrentPipeline(t, 1)
+	encoder := &capturingEvidenceOutput{}
+	prepared.output = encoder
+	projected := 0
+	prepared.evidencePlan = nil
+	if _, err := New().Run(context.Background(), RunInput{Prepared: prepared, RawInput: []byte("input")}); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	if projected != 0 || len(encoder.requests) != 1 || encoder.requests[0].EvidenceContext != nil {
+		t.Fatalf("projected = %d requests = %#v, want no evidence work", projected, encoder.requests)
+	}
+}

internal/framework/pipeline/evidence_policy.go

@@ -0,0 +1,20 @@
+package pipeline
+
+// EvidenceContextPolicy controls optional source-context publication by an
+// output encoder.
+type EvidenceContextPolicy struct {
+	Enabled     bool
+	WindowUnits int
+	LaneIDs     []string
+}
+
+// EvidenceContextPolicyProvider is implemented by output encoders that opt in
+// to evidence-context publication.
+type EvidenceContextPolicyProvider interface {
+	EvidenceContextPolicy() EvidenceContextPolicy
+}
+
+func cloneEvidenceContextPolicy(policy EvidenceContextPolicy) EvidenceContextPolicy {
+	policy.LaneIDs = append([]string(nil), policy.LaneIDs...)
+	return policy
+}

internal/framework/pipeline/evidence_preparation_test.go

@@ -0,0 +1,119 @@
+package pipeline
+
+import (
+	"context"
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/core/source"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+)
+
+type testEvidenceOutput struct {
+	policy EvidenceContextPolicy
+}
+
+func (output testEvidenceOutput) Key() string { return "output" }
+func (output testEvidenceOutput) Encode(context.Context, contracts.OutputRequest) (contracts.OutputResult, error) {
+	return contracts.OutputResult{}, nil
+}
+func (output testEvidenceOutput) EvidenceContextPolicy() EvidenceContextPolicy {
+	return cloneEvidenceContextPolicy(output.policy)
+}
+
+func TestEvidencePolicyResolutionAndPreparationSurviveLaneFiltering(t *testing.T) {
+	registries, _ := constructionRegistries(t, nil, nil)
+	registries.ArtifactEvidence = NewArtifactEvidenceRegistry()
+	policy := EvidenceContextPolicy{Enabled: true, WindowUnits: 2, LaneIDs: []string{"artifact", "inactive"}}
+	profileValidated := false
+	registerTestEvidenceOutputWithProfileValidation(t, &registries, policy, func(context OutputProfileOptionContext, _ map[string]any) error {
+		profileValidated = true
+		if want := []string{"artifact", "inactive"}; !reflect.DeepEqual(context.LaneIDs, want) {
+			return fmt.Errorf("configured lane ids = %#v, want %#v", context.LaneIDs, want)
+		}
+		return nil
+	})
+	if err := RegisterArtifactEvidence(registries.ArtifactEvidence, "test/notes", func(codecNotes) []source.SourceRef { return nil }); err != nil {
+		t.Fatal(err)
+	}
+	profile := constructionProfile()
+	profile.Artifacts["inactive"] = profile.Artifacts["artifact"]
+	profile.Output.Options = map[string]any{"known": true}
+	resolved, err := ResolvePipeline(profile, ResolveOptions{Only: []string{"artifact"}}, registries.catalog())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !profileValidated {
+		t.Fatal("output profile options were not validated")
+	}
+	if lanes := resolved.AllArtifactLanes(); len(lanes) != 1 || lanes[0].ID != "artifact" {
+		t.Fatalf("resolved lanes = %#v, want only the invocation-selected lane", lanes)
+	}
+	prepared, err := Prepare(resolved, registries, ModuleDependencies{})
+	if err != nil {
+		t.Fatalf("Prepare() error = %v, want nil", err)
+	}
+	if prepared.evidencePlan == nil || !reflect.DeepEqual(prepared.evidencePlan.policy.LaneIDs, []string{"artifact", "inactive"}) {
+		t.Fatalf("evidence plan = %#v, want complete configured policy", prepared.evidencePlan)
+	}
+	if got := prepared.evidencePlan.lanes; len(got) != 1 || got[0].laneID != "artifact" {
+		t.Fatalf("active evidence lanes = %#v, want artifact only", got)
+	}
+	prepared.evidencePlan.policy.LaneIDs[0] = "mutated"
+	if policy := prepared.output.(EvidenceContextPolicyProvider).EvidenceContextPolicy(); policy.LaneIDs[0] != "artifact" {
+		t.Fatalf("prepared plan mutated provider policy: %#v", policy)
+	}
+}
+
+func TestPrepareEvidencePlanRejectsMissingAndMismatchedCapabilities(t *testing.T) {
+	for _, test := range []struct {
+		name      string
+		configure func(*Registries)
+		want      string
+	}{
+		{name: "missing registry", configure: func(registries *Registries) { registries.ArtifactEvidence = nil }, want: "artifact evidence registry"},
+		{name: "unsupported kind", configure: func(registries *Registries) {}, want: "artifact evidence \"test/notes\" is not registered"},
+		{name: "mismatched type", configure: func(registries *Registries) {
+			if err := RegisterArtifactEvidence(registries.ArtifactEvidence, "test/notes", func(codecScore) []source.SourceRef { return nil }); err != nil {
+				t.Fatal(err)
+			}
+		}, want: "requires Go type"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			registries, _ := constructionRegistries(t, nil, nil)
+			registries.ArtifactEvidence = NewArtifactEvidenceRegistry()
+			registerTestEvidenceOutput(t, &registries, EvidenceContextPolicy{Enabled: true, LaneIDs: []string{"artifact"}})
+			test.configure(&registries)
+			profile := constructionProfile()
+			profile.Output.Options = map[string]any{"known": true}
+			resolved, err := ResolvePipeline(profile, ResolveOptions{}, registries.catalog())
+			if err != nil {
+				t.Fatal(err)
+			}
+			_, err = Prepare(resolved, registries, ModuleDependencies{})
+			if err == nil || !strings.Contains(err.Error(), test.want) {
+				t.Fatalf("Prepare() error = %v, want %q", err, test.want)
+			}
+		})
+	}
+}
+
+func registerTestEvidenceOutput(t *testing.T, registries *Registries, policy EvidenceContextPolicy) {
+	t.Helper()
+	registerTestEvidenceOutputWithProfileValidation(t, registries, policy, nil)
+}
+
+func registerTestEvidenceOutputWithProfileValidation(t *testing.T, registries *Registries, policy EvidenceContextPolicy, validateProfile OutputProfileOptionValidator) {
+	t.Helper()
+	registry := NewOutputEncoderRegistry()
+	if err := registry.RegisterBuilderWithProfileValidation(testModuleSpec("output", StageOutput), func(options map[string]any) error {
+		return RejectUnknownOptions(options, "known")
+	}, validateProfile, func(BuildRequest) (contracts.OutputEncoder, error) {
+		return testEvidenceOutput{policy: cloneEvidenceContextPolicy(policy)}, nil
+	}); err != nil {
+		t.Fatal(err)
+	}
+	registries.Outputs = registry
+}

internal/framework/pipeline/input_registry.go

@@ -24,10 +24,6 @@ func NewInputAdapterRegistry() *InputAdapterRegistry {
 	}
 }
 
-func (r *InputAdapterRegistry) Register(key string, constructor InputAdapterConstructor) error {
-	return r.RegisterWithSpec(defaultModuleSpec(key, StageInput), constructor)
-}
-
 func (r *InputAdapterRegistry) RegisterWithSpec(spec ModuleSpec, constructor InputAdapterConstructor) error {
 	if constructor == nil {
 		return fmt.Errorf("input adapter constructor for %q must not be nil", strings.TrimSpace(spec.Key))

internal/framework/pipeline/input_registry_test.go

@@ -11,10 +11,14 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 )
 
+func registerTestInput(registry *InputAdapterRegistry, key string, constructor InputAdapterConstructor) error {
+	return registry.RegisterWithSpec(testModuleSpec(key, StageInput), constructor)
+}
+
 func TestInputAdapterRegistryRegisterAndBuild(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	if err := registry.Register("generic-input", fakeInputAdapterConstructor("generic-input")); err != nil {
+	if err := registerTestInput(registry, "generic-input", fakeInputAdapterConstructor("generic-input")); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
 	}
 
@@ -30,7 +34,7 @@ func TestInputAdapterRegistryRegisterAndBuild(t *testing.T) {
 func TestInputAdapterRegistryRegisterAndBuildTrimKeys(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	if err := registry.Register(" generic-input ", fakeInputAdapterConstructor("generic-input")); err != nil {
+	if err := registerTestInput(registry, " generic-input ", fakeInputAdapterConstructor("generic-input")); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
 	}
 
@@ -46,10 +50,11 @@ func TestInputAdapterRegistryRegisterAndBuildTrimKeys(t *testing.T) {
 func TestInputAdapterRegistryRegisterWithSpecStoresMetadata(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 	spec := ModuleSpec{
-		Key:      " generic-input ",
-		Stage:    StageInput,
-		Provides: []string{" parsed-source ", "source-document", "parsed-source", ""},
-		Requires: []string{" raw-bytes ", "raw-bytes", ""},
+		Key:            " generic-input ",
+		Stage:          StageInput,
+		ExecutionClass: contracts.ExecutionClassDeterministic,
+		Provides:       []string{" parsed-source ", "source-document", "parsed-source", ""},
+		Requires:       []string{" raw-bytes ", "raw-bytes", ""},
 	}
 
 	if err := registry.RegisterWithSpec(spec, fakeInputAdapterConstructor("generic-input")); err != nil {
@@ -61,10 +66,11 @@ func TestInputAdapterRegistryRegisterWithSpecStoresMetadata(t *testing.T) {
 		t.Fatal("Spec() ok = false, want true")
 	}
 	want := ModuleSpec{
-		Key:      "generic-input",
-		Stage:    StageInput,
-		Provides: []string{"parsed-source", "source-document"},
-		Requires: []string{"raw-bytes"},
+		Key:            "generic-input",
+		Stage:          StageInput,
+		ExecutionClass: contracts.ExecutionClassDeterministic,
+		Provides:       []string{"parsed-source", "source-document"},
+		Requires:       []string{"raw-bytes"},
 	}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("Spec() = %#v, want %#v", got, want)
@@ -80,10 +86,10 @@ func TestInputAdapterRegistryRegisterWithSpecStoresMetadata(t *testing.T) {
 	}
 }
 
-func TestInputAdapterRegistryRegisterStoresDefaultSpec(t *testing.T) {
+func TestInputAdapterRegistryRegisterWithSpecStoresMinimalMetadata(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	if err := registry.Register(" generic-input ", fakeInputAdapterConstructor("generic-input")); err != nil {
+	if err := registerTestInput(registry, " generic-input ", fakeInputAdapterConstructor("generic-input")); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
 	}
 
@@ -91,7 +97,7 @@ func TestInputAdapterRegistryRegisterStoresDefaultSpec(t *testing.T) {
 	if !ok {
 		t.Fatal("Spec() ok = false, want true")
 	}
-	want := ModuleSpec{Key: "generic-input", Stage: StageInput}
+	want := ModuleSpec{Key: "generic-input", Stage: StageInput, ExecutionClass: contracts.ExecutionClassDeterministic}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("Spec() = %#v, want %#v", got, want)
 	}
@@ -100,7 +106,7 @@ func TestInputAdapterRegistryRegisterStoresDefaultSpec(t *testing.T) {
 func TestInputAdapterRegistryRegisterWithSpecRejectsWrongStage(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	err := registry.RegisterWithSpec(ModuleSpec{Key: "generic-input", Stage: StageExtract}, fakeInputAdapterConstructor("generic-input"))
+	err := registry.RegisterWithSpec(ModuleSpec{Key: "generic-input", Stage: StageExtract, ExecutionClass: contracts.ExecutionClassDeterministic}, fakeInputAdapterConstructor("generic-input"))
 
 	if err == nil {
 		t.Fatal("RegisterWithSpec() error = nil, want error")
@@ -121,7 +127,7 @@ func TestInputAdapterRegistrySpecRejectsUnknownKey(t *testing.T) {
 func TestInputAdapterRegistryRegisterRejectsEmptyKey(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	err := registry.Register(" \t", fakeInputAdapterConstructor("generic-input"))
+	err := registerTestInput(registry, " \t", fakeInputAdapterConstructor("generic-input"))
 
 	if err == nil {
 		t.Fatal("Register() error = nil, want error")
@@ -133,11 +139,11 @@ func TestInputAdapterRegistryRegisterRejectsEmptyKey(t *testing.T) {
 
 func TestInputAdapterRegistryRegisterRejectsDuplicateKey(t *testing.T) {
 	registry := NewInputAdapterRegistry()
-	if err := registry.Register("generic-input", fakeInputAdapterConstructor("generic-input")); err != nil {
+	if err := registerTestInput(registry, "generic-input", fakeInputAdapterConstructor("generic-input")); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
 	}
 
-	err := registry.Register(" generic-input ", fakeInputAdapterConstructor("generic-input"))
+	err := registerTestInput(registry, " generic-input ", fakeInputAdapterConstructor("generic-input"))
 
 	if err == nil {
 		t.Fatal("Register() error = nil, want error")
@@ -150,7 +156,7 @@ func TestInputAdapterRegistryRegisterRejectsDuplicateKey(t *testing.T) {
 func TestInputAdapterRegistryRegisterRejectsNilConstructor(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 
-	err := registry.Register("generic-input", nil)
+	err := registerTestInput(registry, "generic-input", nil)
 
 	if err == nil {
 		t.Fatal("Register() error = nil, want error")
@@ -176,7 +182,7 @@ func TestInputAdapterRegistryBuildRejectsUnknownKey(t *testing.T) {
 func TestInputAdapterRegistryBuildWrapsConstructorError(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 	constructorErr := errors.New("constructor failed")
-	if err := registry.Register("generic-input", func() (contracts.InputAdapter, error) {
+	if err := registerTestInput(registry, "generic-input", func() (contracts.InputAdapter, error) {
 		return nil, constructorErr
 	}); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
@@ -197,7 +203,7 @@ func TestInputAdapterRegistryBuildWrapsConstructorError(t *testing.T) {
 
 func TestInputAdapterRegistryBuildRejectsNilAdapter(t *testing.T) {
 	registry := NewInputAdapterRegistry()
-	if err := registry.Register("generic-input", func() (contracts.InputAdapter, error) {
+	if err := registerTestInput(registry, "generic-input", func() (contracts.InputAdapter, error) {
 		return nil, nil
 	}); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
@@ -215,7 +221,7 @@ func TestInputAdapterRegistryBuildRejectsNilAdapter(t *testing.T) {
 
 func TestInputAdapterRegistryBuildRejectsAdapterKeyMismatch(t *testing.T) {
 	registry := NewInputAdapterRegistry()
-	if err := registry.Register("generic-input", fakeInputAdapterConstructor("other-input")); err != nil {
+	if err := registerTestInput(registry, "generic-input", fakeInputAdapterConstructor("other-input")); err != nil {
 		t.Fatalf("Register() error = %v, want nil", err)
 	}
 
@@ -232,7 +238,7 @@ func TestInputAdapterRegistryBuildRejectsAdapterKeyMismatch(t *testing.T) {
 func TestInputAdapterRegistryRegisteredKeysReturnsSortedCopy(t *testing.T) {
 	registry := NewInputAdapterRegistry()
 	for _, key := range []string{"zeta", "alpha", "middle"} {
-		if err := registry.Register(key, fakeInputAdapterConstructor(key)); err != nil {
+		if err := registerTestInput(registry, key, fakeInputAdapterConstructor(key)); err != nil {
 			t.Fatalf("Register(%q) error = %v, want nil", key, err)
 		}
 	}
@@ -253,7 +259,7 @@ func TestInputAdapterRegistryRegisteredKeysReturnsSortedCopy(t *testing.T) {
 func TestInputAdapterRegistryNilRegistryBehavior(t *testing.T) {
 	var registry *InputAdapterRegistry
 
-	if err := registry.Register("generic-input", fakeInputAdapterConstructor("generic-input")); err == nil {
+	if err := registerTestInput(registry, "generic-input", fakeInputAdapterConstructor("generic-input")); err == nil {
 		t.Fatal("Register() error = nil, want error")
 	}
 	if _, err := registry.Build("generic-input"); err == nil {

internal/framework/pipeline/module.go

@@ -23,23 +23,19 @@ const (
 type ModuleSpec struct {
 	Key            string
 	Stage          ModuleStage
+	ExecutionClass contracts.ExecutionClass
 	ArtifactKind   contracts.ArtifactKind
 	Provides       []string
 	Requires       []string
 	ReferenceSlots []contracts.ReferenceSlot
 }
 
-func defaultModuleSpec(key string, stage ModuleStage) ModuleSpec {
-	return ModuleSpec{
-		Key:   key,
-		Stage: stage,
-	}
-}
-
 func normalizeModuleSpec(spec ModuleSpec) ModuleSpec {
+	executionClass := contracts.ExecutionClass(strings.TrimSpace(string(spec.ExecutionClass)))
 	return ModuleSpec{
 		Key:            strings.TrimSpace(spec.Key),
 		Stage:          spec.Stage,
+		ExecutionClass: executionClass,
 		ArtifactKind:   normalizeArtifactKind(spec.ArtifactKind),
 		Provides:       normalizeCapabilities(spec.Provides),
 		Requires:       normalizeCapabilities(spec.Requires),
@@ -76,6 +72,7 @@ func cloneModuleSpec(spec ModuleSpec) ModuleSpec {
 	return ModuleSpec{
 		Key:            spec.Key,
 		Stage:          spec.Stage,
+		ExecutionClass: spec.ExecutionClass,
 		ArtifactKind:   spec.ArtifactKind,
 		Provides:       append([]string(nil), spec.Provides...),
 		Requires:       append([]string(nil), spec.Requires...),
@@ -90,6 +87,12 @@ func validateModuleSpec(kind string, expectedStage ModuleStage, spec ModuleSpec)
 	if spec.Stage != expectedStage {
 		return fmt.Errorf("%s %q must use %q stage, got %q", kind, spec.Key, expectedStage, spec.Stage)
 	}
+	if spec.ExecutionClass == "" {
+		return fmt.Errorf("%s %q execution class must not be empty", kind, spec.Key)
+	}
+	if spec.ExecutionClass != contracts.ExecutionClassDeterministic && spec.ExecutionClass != contracts.ExecutionClassLLMBacked {
+		return fmt.Errorf("%s %q has unsupported execution class %q", kind, spec.Key, spec.ExecutionClass)
+	}
 	if spec.ArtifactKind != "" && spec.Stage != StageExtract && spec.Stage != StageMerge && spec.Stage != StageNormalize {
 		return fmt.Errorf("%s %q must not declare an artifact kind", kind, spec.Key)
 	}

internal/framework/pipeline/module_test.go

@@ -1,12 +1,45 @@
 package pipeline
 
 import (
+	"reflect"
 	"strings"
 	"testing"
 
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 )
 
+func TestValidateModuleSpecRequiresSupportedExecutionClass(t *testing.T) {
+	for _, test := range []struct {
+		name  string
+		class contracts.ExecutionClass
+		want  string
+	}{
+		{name: "missing", want: "execution class"},
+		{name: "unsupported", class: "remote", want: "unsupported"},
+		{name: "deterministic", class: contracts.ExecutionClassDeterministic},
+		{name: "llm backed", class: contracts.ExecutionClassLLMBacked},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			spec := normalizeModuleSpec(ModuleSpec{Key: "module", Stage: StageChunk, ExecutionClass: test.class})
+			err := validateModuleSpec("chunker", StageChunk, spec)
+			if test.want == "" && err != nil {
+				t.Fatalf("validateModuleSpec() error = %v, want nil", err)
+			}
+			if test.want != "" && (err == nil || !strings.Contains(err.Error(), test.want)) {
+				t.Fatalf("validateModuleSpec() error = %v, want %q", err, test.want)
+			}
+		})
+	}
+}
+
+func TestCloneModuleSpecPreservesExecutionClass(t *testing.T) {
+	spec := normalizeModuleSpec(ModuleSpec{Key: " module ", Stage: StageChunk, ExecutionClass: contracts.ExecutionClassLLMBacked})
+	cloned := cloneModuleSpec(spec)
+	if !reflect.DeepEqual(cloned, spec) {
+		t.Fatalf("cloneModuleSpec() = %#v, want %#v", cloned, spec)
+	}
+}
+
 func TestValidateModuleSpecAllowsReferenceSlotsForEligibleStages(t *testing.T) {
 	tests := []struct {
 		name  string
@@ -22,8 +55,9 @@ func TestValidateModuleSpecAllowsReferenceSlotsForEligibleStages(t *testing.T) {
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
 			spec := normalizeModuleSpec(ModuleSpec{
-				Key:   "module",
-				Stage: test.stage,
+				Key:            "module",
+				Stage:          test.stage,
+				ExecutionClass: contracts.ExecutionClassDeterministic,
 				ReferenceSlots: []contracts.ReferenceSlot{
 					{Name: "roster", Description: "Character roster", MaxBytes: 1024},
 				},
@@ -50,8 +84,9 @@ func TestValidateModuleSpecRejectsReferenceSlotsForIneligibleStages(t *testing.T
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
 			spec := normalizeModuleSpec(ModuleSpec{
-				Key:   "module",
-				Stage: test.stage,
+				Key:            "module",
+				Stage:          test.stage,
+				ExecutionClass: contracts.ExecutionClassDeterministic,
 				ReferenceSlots: []contracts.ReferenceSlot{
 					{Name: "roster"},
 				},
@@ -109,6 +144,7 @@ func TestValidateModuleSpecRejectsInvalidReferenceSlotsForEligibleStages(t *test
 				spec := normalizeModuleSpec(ModuleSpec{
 					Key:            "module",
 					Stage:          stage.stage,
+					ExecutionClass: contracts.ExecutionClassDeterministic,
 					ReferenceSlots: invalid.slots,
 				})
 				err := validateModuleSpec(stage.kind, stage.stage, spec)

internal/framework/pipeline/normalizer_registry_test.go

@@ -26,7 +26,7 @@ func TestNormalizerRegistryErasureClonesRetryDirective(t *testing.T) {
 		FallbackWarnings: []contracts.Warning{{Scope: "fallback", ReasonCode: "omitted", Message: "fallback warning"}},
 	}
 	registry := NewNormalizerRegistry()
-	if err := RegisterNormalizer(registry, ModuleSpec{Key: "test/retry-normalize", Stage: StageNormalize, ArtifactKind: "test/notes"}, func() (contracts.Normalizer[codecNotes], error) {
+	if err := RegisterNormalizer(registry, ModuleSpec{Key: "test/retry-normalize", Stage: StageNormalize, ExecutionClass: contracts.ExecutionClassDeterministic, ArtifactKind: "test/notes"}, func() (contracts.Normalizer[codecNotes], error) {
 		return retryingNotesNormalizer{warnings: warnings, retry: retry}, nil
 	}); err != nil {
 		t.Fatalf("RegisterNormalizer() error = %v", err)

internal/framework/pipeline/options.go

@@ -2,7 +2,7 @@ package pipeline
 
 import "fmt"
 
-func validateResolvedOptions(resolved ResolvedPipeline, catalog ModuleCatalog) error {
+func validateResolvedOptions(resolved ResolvedPipeline, catalog ModuleCatalog, configuredLaneIDs []string) error {
 	if err := catalog.Inputs.ValidateOptions(resolved.Input.Module, resolved.Input.Options); err != nil {
 		return moduleOptionsError(resolved.ID, "", StageInput, resolved.Input.Module, err)
 	}
@@ -42,6 +42,9 @@ func validateResolvedOptions(resolved ResolvedPipeline, catalog ModuleCatalog) e
 	if err := catalog.Outputs.ValidateOptions(resolved.Output.Module, resolved.Output.Options); err != nil {
 		return moduleOptionsError(resolved.ID, "", StageOutput, resolved.Output.Module, err)
 	}
+	if err := catalog.Outputs.ValidateProfileOptions(resolved.Output.Module, OutputProfileOptionContext{LaneIDs: configuredLaneIDs}, resolved.Output.Options); err != nil {
+		return moduleOptionsError(resolved.ID, "", StageOutput, resolved.Output.Module, err)
+	}
 	return nil
 }
 

internal/framework/pipeline/output_registry.go

@@ -10,24 +10,28 @@ import (
 type OutputEncoderConstructor func() (contracts.OutputEncoder, error)
 type OutputEncoderBuilder func(BuildRequest) (contracts.OutputEncoder, error)
 
+type OutputProfileOptionContext struct {
+	LaneIDs []string
+}
+
+type OutputProfileOptionValidator func(OutputProfileOptionContext, map[string]any) error
+
 type OutputEncoderRegistry struct {
-	builders         map[string]OutputEncoderBuilder
-	optionValidators map[string]OptionValidator
-	specs            map[string]ModuleSpec
+	builders          map[string]OutputEncoderBuilder
+	optionValidators  map[string]OptionValidator
+	profileValidators map[string]OutputProfileOptionValidator
+	specs             map[string]ModuleSpec
 }
 
 func NewOutputEncoderRegistry() *OutputEncoderRegistry {
 	return &OutputEncoderRegistry{
-		builders:         make(map[string]OutputEncoderBuilder),
-		optionValidators: make(map[string]OptionValidator),
-		specs:            make(map[string]ModuleSpec),
+		builders:          make(map[string]OutputEncoderBuilder),
+		optionValidators:  make(map[string]OptionValidator),
+		profileValidators: make(map[string]OutputProfileOptionValidator),
+		specs:             make(map[string]ModuleSpec),
 	}
 }
 
-func (r *OutputEncoderRegistry) Register(key string, constructor OutputEncoderConstructor) error {
-	return r.RegisterWithSpec(defaultModuleSpec(key, StageOutput), constructor)
-}
-
 func (r *OutputEncoderRegistry) RegisterWithSpec(spec ModuleSpec, constructor OutputEncoderConstructor) error {
 	if constructor == nil {
 		return fmt.Errorf("output encoder constructor for %q must not be nil", strings.TrimSpace(spec.Key))
@@ -38,6 +42,12 @@ func (r *OutputEncoderRegistry) RegisterWithSpec(spec ModuleSpec, constructor Ou
 }
 
 func (r *OutputEncoderRegistry) RegisterBuilderWithSpec(spec ModuleSpec, validateOptions OptionValidator, builder OutputEncoderBuilder) error {
+	return r.RegisterBuilderWithProfileValidation(spec, validateOptions, nil, builder)
+}
+
+// RegisterBuilderWithProfileValidation registers an output builder with an
+// optional validator that can inspect all configured lane identities.
+func (r *OutputEncoderRegistry) RegisterBuilderWithProfileValidation(spec ModuleSpec, validateOptions OptionValidator, validateProfile OutputProfileOptionValidator, builder OutputEncoderBuilder) error {
 	if r == nil {
 		return fmt.Errorf("output encoder registry must not be nil")
 	}
@@ -62,11 +72,15 @@ func (r *OutputEncoderRegistry) RegisterBuilderWithSpec(spec ModuleSpec, validat
 	if r.optionValidators == nil {
 		r.optionValidators = make(map[string]OptionValidator)
 	}
+	if r.profileValidators == nil {
+		r.profileValidators = make(map[string]OutputProfileOptionValidator)
+	}
 	if r.specs == nil {
 		r.specs = make(map[string]ModuleSpec)
 	}
 	r.builders[normalizedSpec.Key] = builder
 	r.optionValidators[normalizedSpec.Key] = validateOptions
+	r.profileValidators[normalizedSpec.Key] = validateProfile
 	r.specs[normalizedSpec.Key] = cloneModuleSpec(normalizedSpec)
 	return nil
 }
@@ -116,6 +130,21 @@ func (r *OutputEncoderRegistry) ValidateOptions(key string, options map[string]a
 	return validateRegisteredOptions(validator, options)
 }
 
+func (r *OutputEncoderRegistry) ValidateProfileOptions(key string, context OutputProfileOptionContext, options map[string]any) error {
+	if r == nil {
+		return fmt.Errorf("output encoder registry must not be nil")
+	}
+	normalizedKey := strings.TrimSpace(key)
+	validator, ok := r.profileValidators[normalizedKey]
+	if !ok {
+		return fmt.Errorf("output encoder %q is not registered", normalizedKey)
+	}
+	if validator == nil {
+		return nil
+	}
+	return validator(OutputProfileOptionContext{LaneIDs: append([]string(nil), context.LaneIDs...)}, cloneOptions(options))
+}
+
 func (r *OutputEncoderRegistry) Spec(key string) (ModuleSpec, bool) {
 	if r == nil {
 		return ModuleSpec{}, false