Prepare version 0.8.0 release

.woodpecker/release.yml

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

AGENTS.md

@@ -1,7 +1,5 @@
 Please review `docs/development.md` for initial orientation in this repository
 and follow its task-specific reading guide.
 
-For every planning and implementation task, follow the
-[`Implementation Discipline`](docs/policy/architecture.md#implementation-discipline):
-keep machinery proportionate, do not invent restrictions, and do not elevate
-incidental implementation details into repository policy.
+Review and follow `docs/policy/agents.md` before recommending, planning, or
+implementing any changes to this repository.

docs/cli.md

@@ -12,8 +12,8 @@ defined in [Operations](operations.md).
 notarius help
 notarius --version
 notarius run <pipeline-id> --input path/to/source.json [--json] [flags]
-notarius config validate [--config path/to/config.yml] [--config-profile name] [--pipeline pipeline-id] [--only lane-a,lane-b]
-notarius pipelines list [--config path/to/config.yml] [--config-profile name] [--json]
+notarius config validate [--config path/to/config.yml] [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
+notarius pipelines list [--config path/to/config.yml] [--config-overlay name] [--json]
 ~~~
 
 Running Notarius without arguments, or with **help**, **--help**, or **-h**,
@@ -42,7 +42,7 @@ pipeline ID and **--input** are required.
 | Flag | Meaning |
 | --- | --- |
 | **--config path** | Use this configuration file. When omitted, configuration discovery applies; see [Configuration](config.md). |
-| **--config-profile name** | Select a ConfigKit configuration overlay. Empty and **default** select the composed base configuration. |
+| **--config-overlay name** | Select a ConfigKit configuration overlay. Empty and **default** select the composed base configuration. |
 | **--input path** | Source input file to process. Required. |
 | **--output-dir path** | Override the configured output root for this run. |
 | **--json** | Write the successful run-result receipt as JSON to standard output. |
@@ -155,7 +155,7 @@ OPENROUTER_API_KEY=your-api-key \
 ## config validate
 
 ~~~
-notarius config validate [--config path/to/config.yml] [--config-profile name] [--pipeline pipeline-id] [--only lane-a,lane-b]
+notarius config validate [--config path/to/config.yml] [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
 ~~~
 
 This command loads and validates the selected composed configuration. With **--pipeline**, it also
@@ -184,7 +184,7 @@ validation; it cannot run a provider-backed pipeline.
 ## config effective
 
 ~~~
-notarius config effective [--config path/to/config.yml] [--config-profile name]
+notarius config effective [--config path/to/config.yml] [--config-overlay name]
 ~~~
 
 This command writes the selected composed configuration as deterministic,
@@ -194,19 +194,19 @@ pipeline, prepare modules, materialize references, or contact a provider.
 ## config explain
 
 ~~~
-notarius config explain [--config path/to/config.yml] [--config-profile name] <json-pointer>
+notarius config explain [--config path/to/config.yml] [--config-overlay name] <json-pointer>
 ~~~
 
 This command writes ConfigKit's redacted, human-readable explanation for one
 canonical JSON Pointer. Supply `""` explicitly to explain the root. A malformed
 pointer is a syntax error; a well-formed unknown path is a command failure.
-The explanation can describe a value displaced by a selected profile. Standard
+The explanation can describe a value displaced by a selected overlay. Standard
 output contains only the explanation.
 
 ## pipelines list
 
 ~~~
-notarius pipelines list [--config path/to/config.yml] [--config-profile name] [--json]
+notarius pipelines list [--config path/to/config.yml] [--config-overlay name] [--json]
 ~~~
 
 This command lists configured pipeline IDs in sorted order. By default, it
@@ -225,7 +225,7 @@ errors are written to standard error.
 
 Configuration load errors include ConfigKit's field diagnostics and source
 coordinates, followed by a source-ID legend identifying the acquired files or
-source labels. This includes imported files and the selected configuration profile.
+source labels. This includes imported files and the selected configuration overlay.
 
 For **run --json**, actionable process warnings remain on standard error and standard output is a
 machine-readable success result only. Syntax and runtime diagnostics remain on

docs/config.md

@@ -20,12 +20,13 @@ For configuration values, precedence is:
 
 1. built-in defaults;
 2. composed YAML sources, including imports;
-3. the selected ConfigKit profile; and
+3. the selected ConfigKit overlay; and
 4. explicitly supplied configuration-valued run flags.
 
-`--config-profile` selects a ConfigKit overlay from `profiles/<name>.yml`
+`--config-overlay` selects a ConfigKit overlay from `overlays/<name>.yml`
 beneath the selected configuration root. Empty and `default` select no overlay.
-Configuration profiles are distinct from PromptKit LLM profiles selected by
+`profiles/<name>.yml` is not a fallback location, and `--config-profile` is no
+longer accepted. Configuration overlays are distinct from PromptKit LLM profiles selected by
 `llm_profile` or `--llm-profile`. The retired operational environment variables
 have no effect; `NOTARIUS_CONFIG` remains a source-discovery input.
 
@@ -79,14 +80,14 @@ umask, setgid inheritance, and default ACLs; see [Operations](operations.md#cach
 
 Every nonempty relative configured destination or reference path resolves from
 the canonical directory of the file that supplies its surviving declaration.
-This includes imported fragments and selected configuration profiles. Built-in
+This includes imported fragments and selected configuration overlays. Built-in
 relative destinations use the selected main configuration file's directory;
 explicit `--output-dir` and `--debug-dir` values use the invocation working
 directory. Absolute values remain unchanged.
 
 The root file can use ConfigKit imports to compose disjoint definitions. Imports
-and selected profiles remain under the configuration root; they do not merge or
-append a pipeline's `steps` sequence. A profile replaces sequences as defined
+and selected overlays remain under the configuration root; they do not merge or
+append a pipeline's `steps` sequence. An overlay replaces sequences as defined
 by ConfigKit, so changing an individual ordered binding requires supplying the
 whole replacement sequence.
 
@@ -101,6 +102,7 @@ version: 5
 promptkit:
   profile_dir: ./profiles
   # profile_file: ./profiles.yml
+  transient_retries: 2
   local_backend:
     endpoint: http://localhost:8000/v1
     concurrency_limit: 2
@@ -110,6 +112,7 @@ promptkit:
 | --- | --- | --- |
 | **profile_dir** | string | Non-empty directory containing profile files. |
 | **profile_file** | string | Non-empty profile file. |
+| **transient_retries** | integer | Non-negative additional built-in HTTP attempts for eligible transient provider failures; defaults to 2. Set 0 to disable these retries. |
 | **local_backend** | object | Optional registration for the conventional PromptKit backend ID **local**. |
 | **local_backend.endpoint** | string | Required when **local_backend** is present; absolute HTTP or HTTPS URL with a host. |
 | **local_backend.concurrency_limit** | integer | Optional non-negative limit; defaults to 0. |
@@ -201,7 +204,7 @@ debug:
 **concurrency.total_llm** must be greater than zero. The only supported
 **concurrency.stage_workers** key is **extract**; its value must be from 1
 through **total_llm**. When omitted, it is recalculated from the effective
-**total_llm** after imports, profile selection, and explicit ConfigKit
+**total_llm** after imports, overlay selection, and explicit ConfigKit
 overrides.
 
 **cache.chunk_plans.mode** accepts **auto**, **bypass**, or **refresh**.
@@ -440,7 +443,7 @@ pipeline, lane, and binding scopes and precedence are defined by the
 replace a configured generated artifact handoff. Required slots must be bound
 after configuration and CLI reference actions are applied. A YAML reference
 path resolves from the file that supplies its surviving declaration, including
-an imported fragment or selected profile. CLI reference paths resolve relative
+an imported fragment or selected overlay. CLI reference paths resolve relative
 to the process working directory.
 
 ### D&D Reference Slots
@@ -601,5 +604,5 @@ Use [config effective](cli.md#config-effective) to write the selected composed
 configuration as deterministic, redacted YAML. Use
 [config explain](cli.md#config-explain) with a canonical JSON Pointer to inspect
 the effective or historical provenance for one path, including a value displaced
-by a selected profile. Neither command resolves a pipeline, prepares modules,
+by a selected overlay. Neither command resolves a pipeline, prepares modules,
 materializes references, or requires provider credentials.

docs/consumers/subprocess.md

@@ -15,10 +15,10 @@ configured lane inventory, and downstream acceptance checklist, see
 Optionally preflight a selected configuration and pipeline before work starts:
 
 ```sh
-notarius config validate --config /path/to/notarius.yml --config-profile production --pipeline pipeline-id
+notarius config validate --config /path/to/notarius.yml --config-overlay production --pipeline pipeline-id
 ```
 
-Omit `--config-profile` when the composed base configuration is the intended
+Omit `--config-overlay` when the composed base configuration is the intended
 selection; it is independent from any `--llm-profile` run override.
 
 Invoke the run with explicit paths and machine-readable output. Capture
@@ -28,7 +28,7 @@ processing the result.
 ```sh
 notarius run pipeline-id \
   --config /path/to/notarius.yml \
-  --config-profile production \
+  --config-overlay production \
   --input /path/to/source.json \
   --output-dir /path/to/output-root \
   --json

docs/development.md

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

docs/integrations/pkg-configkit.md

@@ -2,7 +2,7 @@
 
 Notarius uses the public root package
 [`gitea.maximumdirect.net/eric/ConfigKit`](https://gitea.maximumdirect.net/eric/ConfigKit)
-at v0.3.0 for version 5 configuration acquisition, imports, profiles, explicit
+at v0.4.0 for version 5 configuration acquisition, imports, overlays, explicit
 overrides, diagnostics, and safe inspection rendering.
 
 Notarius owns the typed document, domain defaults and validation, runtime

docs/integrations/pkg-promptkit.md

@@ -1,11 +1,11 @@
 # PromptKit Integration
 
 Notarius pins
-[`gitea.maximumdirect.net/eric/promptkit` v0.9.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0)
+[`gitea.maximumdirect.net/eric/promptkit` v0.11.0](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0)
 as its in-process prompt engine. The upstream
-[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/consumers/pkg-promptkit.md)
+[Go package consumer guide](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/consumers/pkg-promptkit.md)
 owns the public engine API, and the upstream
-[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md)
+[format reference](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/formats.md)
 owns prompt, profile, and schema file contracts.
 
 ## Supported Boundary
@@ -27,7 +27,7 @@ Notarius relies on the root `promptkit` package to:
   admission exhaustion through `ErrCapacityExceeded`.
 
 The pinned
-[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/backends.go)
+[`BackendLocal`, `LocalBackend`, and `WithBackend` API](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/backends.go)
 owns the registration and backend-capacity contract.
 
 For one completion, the adapter calls `PrepareExecution`, takes a
@@ -53,7 +53,7 @@ Notarius sends one stable effective session through PromptKit's direct session
 field, which is authoritative for provider session behavior. It also retains
 the same value as the `session_id` prompt variable for maintained prompt
 compatibility. The generated identifier is 76 ASCII characters, within
-PromptKit v0.9.0's 256-code-point session limit. Session IDs are non-secret
+PromptKit's 256-code-point session limit. Session IDs are non-secret
 correlation identifiers and may be exposed to providers and provider
 observability. The CLI contract owns generation and override behavior.
 
@@ -71,7 +71,15 @@ endpoint-only profiles have no such backend limit. The adapter translates
 PromptKit capacity rejection into the provider-neutral Notarius
 `ErrLLMCapacityExceeded` contract. It may include the normalized selected
 backend ID in safe diagnostic context, without exposing PromptKit's capacity
-error type, and leaves retries to the calling pipeline stage.
+error type. PromptKit's built-in client retries eligible transient HTTP failures
+with its v0.11.0 default of two additional attempts per generation; Notarius
+keeps pipeline-stage retries outside that completion boundary.
+
+The adapter projects built-in generation failures as the application-owned
+`LLMGenerationError`, retaining a safe failure kind, HTTP status when present,
+and retry-exhaustion attempt count. Transport, malformed successful envelopes,
+provider statuses, and generic generation failures remain distinct without
+exposing PromptKit error values to Notarius callers.
 
 ## Profile Sources And Compatibility
 
@@ -86,16 +94,16 @@ configuration and deployment workflow are defined in
 [Operations](../operations.md#promptkit-profile-deployment).
 
 PromptKit owns `base_profile` resolution under its
-[pinned format rules](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.9.0/docs/formats.md).
+[pinned format rules](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/formats.md).
 Notarius records the selected leaf identity and resolved target without parsing
 or merging inheritance. An unset filesystem `api_key_env` is optional and may
 reach the provider without authorization, which can result in a 401 or 403.
 
-PromptKit v0.9.0 accepts only the `developer`, `system`, `user`, and
+PromptKit accepts only the `developer`, `system`, `user`, and
 `assistant` text-chat roles after normalizing case and surrounding whitespace.
 Maintained Notarius prompt definitions use only `system` and `user`.
 
-For application-owned semantic correction, Notarius uses PromptKit v0.9.0's
+For application-owned semantic correction, Notarius uses PromptKit's
 `RunRequest.AppendedMessages` after the ordinary rendered prompt. It supplies
 exactly two messages in order: the latest validated producer response with
 role `assistant`, then deterministic validation guidance with role `user`.
@@ -112,12 +120,28 @@ and terminal summaries retain only safe counts, digests, identities, and usage;
 complete appended messages remain limited to the explicitly requested detailed
 debug trace.
 
-PromptKit now obtains its maintained OpenRouter and Rakestrawhome backend and
+PromptKit obtains its maintained OpenRouter and Rakestrawhome backend and
 profile catalogs from independently versioned transitive modules. Notarius
 does not import or register either catalog; PromptKit retains catalog source,
 identity, precedence, credential, and capacity ownership.
 
-Notarius supports this boundary against PromptKit v0.9.0. Its fallback source,
+External catalog profile names and settings can change between dependency
+releases, including removal without aliases. Review directly selected IDs and
+inherited `base_profile` values when upgrading. An application-owned alias that
+inherits an external profile does not freeze that profile's settings.
+PromptKit's opt-in `promptkit-openrouter-standard` and
+`promptkit-openrouter-auto` names provide name continuity within its
+selected-and-tested dependency scope, not a fixed model or provider availability
+guarantee. See the upstream
+[catalog compatibility policy](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/docs/policy/release.md#catalog-profile-compatibility).
+
+Use a complete operator-owned profile with explicit backend or endpoint, model,
+and settings when those choices must remain operator-controlled. The embedded
+`dnd-extraction` fallback already follows this model and does not inherit an
+external catalog profile. An upgrade updates Notarius's compiled-in catalog
+checkpoint identity so prior profile-source checkpoints are not reused.
+
+Notarius supports this boundary against PromptKit v0.11.0. Its fallback source,
 prepared-execution, inspection, and typed capacity APIs are used as public
 upstream contracts; other PromptKit APIs or file-format behavior are not
 implicitly supported. A dependency upgrade requires reviewing the adapter,
@@ -146,3 +170,9 @@ profile/configuration contracts against the upstream documentation. Maintained
 production prompts use PromptKit's bounded structural-repair contract; their
 current declaration is one additional repair attempt. Notarius retains the
 transport-neutral boundary and does not expose PromptKit types to modules.
+
+PromptKit v0.11.0 is licensed under
+[LGPL-3.0-or-later](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/LICENSE).
+Its accompanying [GPLv3 notice](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/COPYING)
+is incorporated by that license. This dependency change does not alter
+Notarius's source license or distribution workflow.

docs/internal/cli.md

@@ -35,7 +35,7 @@ options before dispatch so that a missing production dependency fails as a
 command error rather than reaching execution.
 
 Commands that need configuration use one shared ConfigKit-backed loader. The
-CLI discovers the file, passes the requested profile and only explicit
+CLI discovers the file, passes the requested overlay and only explicit
 configuration-valued run overrides, and receives a **LoadedConfig**. The
 configured discovery and precedence contract is in
 [Configuration](../config.md), while the loading and resolution mechanics are

docs/internal/configuration.md

@@ -28,7 +28,7 @@ at their respective [CLI](cli.md), [pipeline](pipeline.md), and
 
 ## Loading And Validation
 
-ConfigKit acquires the selected root, imports, optional selected profile, and
+ConfigKit acquires the selected root, imports, optional selected overlay, and
 explicit CLI overrides in precedence order. Its document definition supplies
 typed defaults, source-aware path normalization, and structural validation
 before conversion to the independently owned runtime **Config**. Only
@@ -44,7 +44,9 @@ concurrency settings, malformed bindings and references, invalid retries, and
 invalid pipeline, step, or lane structure. PromptKit local-backend validation
 accepts only an absolute HTTP or HTTPS endpoint with a host and no user
 information, query, or fragment, and rejects a negative local concurrency
-limit. Its errors retain the closest known pipeline, lane, and binding context.
+limit. It also rejects a negative `promptkit.transient_retries` value; decoded
+omission resolves to two, while explicit zero remains a distinct opt-out. Its
+errors retain the closest known pipeline, lane, and binding context.
 It deliberately does not require modules to be registered: that requires a
 catalog and belongs to resolution.
 
@@ -112,13 +114,15 @@ recording arbitrary environment values or provider requests.
 
 ## Invariants To Preserve
 
-- ConfigKit composition applies defaults, imports, an optional profile, and
+- ConfigKit composition applies defaults, imports, an optional overlay, and
   explicit overrides in one direction with retained origin metadata.
 - A configuration is structurally valid before it is resolved, and a resolved
   pipeline is compatible with the supplied catalog before preparation begins.
 - Whitespace-normalized identifiers are unique wherever they identify a
   pipeline, step, lane, worker, or reference slot.
 - Runtime conversion and resolved-pipeline publication return detached data.
+- The resolved PromptKit transient-retry budget is always passed to the shared
+  built-in client; document presence preserves an explicit zero.
 - The resolved digest changes when resolved composition changes and never
   includes itself.
 
@@ -126,7 +130,7 @@ recording arbitrary environment values or provider requests.
 
 - **internal/core/config/document_test.go** and
   **operational_document_test.go** cover ConfigKit document loading,
-  normalization, imports, profiles, and overrides.
+  normalization, imports, overlays, and overrides.
 - **internal/core/config/validation_contract_test.go** covers configuration
   invariants and contextual failures.
 - **internal/core/config/effective_config_contract_test.go** covers defaults,

docs/internal/llm.md

@@ -216,6 +216,19 @@ than output-validation failures. Apart from documented context, capacity, and
 invalid-output categories, provider error values and types do not cross the
 adapter error chain; callers receive only a credential-redacted diagnostic.
 
+`LLMGenerationError` preserves the provider-neutral failure kind (`transport`,
+`provider_status`, `malformed_response`, or generic `generation`) and any HTTP
+status. When PromptKit exhausts an eligible transient-status retry budget, it
+also records that exhaustion and the attempts in that generation. Those fields
+do not include repair generations, and zero attempts means unavailable rather
+than proof that no request was sent.
+
+Every execution that reaches PromptKit returns an optional request-local
+`execution` summary. It contains only terminal aggregate attempts, retry and
+repair counts, durations, outcome and failure classification, status, and
+presence-aware token usage. Preparation failures have no summary; later local
+decode failures retain the upstream terminal summary without changing it.
+
 When PromptKit rejects backend admission before generation, the adapter maps
 `promptkit.ErrCapacityExceeded` to
 `contracts.ErrLLMCapacityExceeded`, retaining prompt context and a redacted
@@ -258,14 +271,18 @@ diagnostics; see
 ## Timeout Ownership
 
 The caller context remains the outer cancellation authority. PromptKit applies
-a positive effective generation timeout as an inner request deadline; an
-explicit zero disables only that generation deadline. The HTTP client timeout
-is a separate transport-wide cap. Notarius forwards the caller context and
-does not install another timeout wrapper around PromptKit.
+a positive effective generation timeout as an inner request deadline for each
+built-in generation attempt; an explicit zero disables only that generation
+deadline. The HTTP client timeout is a separate transport-wide cap. Notarius
+forwards the caller context and does not install another timeout wrapper around
+PromptKit.
 
-The selected PromptKit profile owns generation settings. Notarius binding
-retries remain outside the adapter and repeat the complete module operation
-and validation chain. PromptKit does not add a provider retry loop.
+The selected PromptKit profile owns generation settings. The resolved global
+`promptkit.transient_retries` configuration is passed to the shared PromptKit
+engine and controls additional built-in attempts for eligible transient HTTP
+failures in one generation. It defaults to two; explicit zero disables those
+attempts. Notarius binding retries remain outside the adapter and repeat the
+complete module operation and validation chain.
 Operator-facing behavior is summarized in
 [Operations](../operations.md#operational-limits), and the pinned upstream
 contract is identified in

docs/internal/pipeline.md

@@ -52,7 +52,7 @@ consumers do not repeat profile inheritance or configuration resolution.
 Each LLM request receives its own copy of that resolved value. PromptKit spends
 it only for structural correction inside one completion; the runner's binding
 retry policy remains the separate outer budget for complete stage attempts.
-Configuration resolution supplies the selected profile and catalog; see
+Configuration resolution supplies the selected overlay and catalog; see
 [Configuration Internals](configuration.md).
 
 External reference materialization happens before preparation. The materializer

docs/internal/state.md

@@ -1,5 +1,9 @@
 # Run State Internals
 
+LLM debug response artifacts and their safe per-call references retain an
+optional terminal execution summary when one is available. This metadata is
+diagnostic only and does not change receipts, checkpoints, or aggregate usage.
+
 This document describes the implementation collaborators behind output, cache,
 and debug state. User-visible fields belong in [Configuration](../config.md),
 and physical layout, retention, recovery, reason codes, and cleanup belong in

docs/operations.md

@@ -147,13 +147,32 @@ their exit classes are defined in the [CLI reference](cli.md#output-streams-and-
 
 ## Validation Retries And Terminal Outcomes
 
+When debug recording is enabled, per-call response artifacts include the
+terminal execution summary for retry, repair, timing, and presence-aware usage
+diagnosis. Normal output, receipts, and warnings do not aggregate or publish
+this diagnostic metadata.
+
 Each producer binding has one outer **retries** budget. It covers complete
 producer attempts for operational failures, invalid structured output,
 normalizer fallback retry, and semantic correction. It is independent from
-PromptKit's structural-repair calls inside one completion and from an
-LLM-backed validator's own retry budget. A semantic correction rebuilds the
-ordinary producer request and supplies only the latest rejected model response
-plus aggregated validator guidance; it is not a conversation replay.
+PromptKit's structural-repair calls inside one completion, its configured
+`promptkit.transient_retries` built-in HTTP attempts, and an LLM-backed
+validator's own retry budget. The transient setting defaults to two additional
+attempts for eligible provider statuses; set it explicitly to zero when an
+operator needs one HTTP attempt per generation. A semantic correction rebuilds
+the ordinary producer request and supplies only the latest rejected model
+response plus aggregated validator guidance; it is not a conversation replay.
+
+These budgets multiply only by their declared scopes. With two outer producer
+retries, two transient retries, and one structural repair, a single producer
+can issue up to 18 HTTP attempts: three producer attempts, each with an initial
+generation and one repair generation, each allowing three HTTP attempts. This
+is an illustration, not a total-deadline or exactly-once guarantee. Every
+built-in HTTP attempt has its own generation timeout; caller cancellation
+remains the overall authority and interrupts Retry-After waiting. The shared
+Notarius scheduler keeps its permit through those waits, so retry budgets can
+increase latency and provider cost. Observed usage can remain incomplete when
+providers omit accounting fields.
 
 After the applicable budgets are exhausted, the resolved
 [`validation_policy`](config.md#pipelines) determines the result. Structural
@@ -319,7 +338,7 @@ content. Neither surface is a cache input. Do not treat a debug bundle as safe
 to share merely because its configuration summary is redacted. Invocation
 metadata omits reasoning effort when it is inherited, records the replacement
 value when one is supplied, and records an empty value when inherited reasoning
-was explicitly cleared. It records `config_profile` only when a non-base
+was explicitly cleared. It records `config_overlay` only when a non-base
 configuration overlay was selected.
 
 Notarius never creates debug state without an explicit request and never
@@ -352,7 +371,9 @@ Provider execution settings and the generation timeout come from the selected
 PromptKit profile. The invocation-only **--reasoning-effort** and
 **--clear-reasoning-effort** controls may replace or clear that profile setting
 for all LLM-backed calls in one run without changing the profile. PromptKit
-structural output repair happens within one structured-completion call. Its
+v0.11.0 also retries eligible transient HTTP failures with its built-in default
+of two additional attempts per generation. This retry work is separate from
+structural output repair, which happens within one structured-completion call. Its
 effective `structured_output_repair_attempts` limit is resolved from the
 selected binding, then the pipeline, then the prompt declaration; see
 [module bindings](config.md#module-bindings-and-validators). This is distinct
@@ -360,19 +381,20 @@ from Notarius binding **retries**, which rerun the complete module operation
 and validation chain and do not consume or replenish the structural-repair
 limit. The maintained production prompts declare one repair attempt, paid only
 after a structural failure. One structured completion with repair budget **R**
-makes at most **R + 1** serial provider calls. If one stage attempt performs
-**C** structured completions, a binding with **retries: N** has a maximum of
-**(N + 1) * C * (R + 1)** provider calls; LLM-backed validators have their own
-corresponding invocation counts and budgets. This is an upper bound, not a
-promise that every call reaches a provider.
+contains at most **R + 1** serial generations. Each generation can now contain
+more than one provider call when PromptKit retries an eligible transient
+failure. LLM-backed validators have their own corresponding invocation and
+repair budgets. These are bounds on attempted work, not a promise that every
+call reaches a provider.
 
 Timeouts are layered. Caller cancellation is the outer authority. A positive
-effective generation timeout adds an inner request deadline, while zero
-disables only that generation deadline. The HTTP client timeout remains a
-transport-wide cap. Notarius does not add another timeout around PromptKit.
-Repairs are serial within the same caller context, so their worst-case latency
-and cost follow the provider-call bound above; provision run deadlines and
-provider budgets accordingly. Credentials remain optional unless the selected
+effective generation timeout adds an inner request deadline to each built-in
+generation attempt, while zero disables only that generation deadline. The HTTP
+client timeout remains a transport-wide cap. Notarius does not add another
+timeout around PromptKit. Retries and repairs are serial within the same caller
+context, so their worst-case latency and cost can exceed the structural-repair
+bound above; provision run deadlines and provider budgets accordingly.
+Credentials remain optional unless the selected
 PromptKit profile requires one, in which case preparation fails before a
 provider call when its configured credential is unavailable.
 The pinned upstream boundary and profile-format links are in

docs/policy/agents.md

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

docs/policy/architecture.md

@@ -1,11 +1,25 @@
 # Architecture
 
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius adapts the base template to its fixed pipeline, artifact families,
+external adapters, state boundaries, and source-only distribution.
+
 This document defines the intended high-level architecture of Notarius and the
 invariants that changes must preserve. Implemented component details belong in
 [Internal Overview](../internal/overview.md) and its linked documents. The
 reasoning behind significant architectural choices belongs in
 [ADRs](../adr/).
 
+An adopted constraint can govern future changes without implying that every
+component it could govern exists. Descriptions of implemented structure must
+remain accurate; proposals and transition status belong in roadmaps and ADRs.
+Constraints reflect selected requirements, not incidental implementation choices.
+
 ## System Shape
 
 Notarius is a small, dependency-light Go application for extracting structured
@@ -38,35 +52,7 @@ Notarius is contract-first without being abstraction-heavy. Interfaces and
 extension points should protect demonstrated boundaries. New abstraction is not
 itself an architectural goal.
 
-## Implementation Discipline
-
-Simplify the mechanism, not the user's choices.
-
-**Keep implementation machinery proportionate.** Use straightforward,
-idiomatic code and introduce only the machinery needed to support the requested
-behavior and flexibility. Optimize for conceptual simplicity and
-maintainability, not for the fewest lines, fewest options, or narrowest product
-surface.
-
-**Do not invent restrictions.** Do not reject, prohibit, constrain, cap,
-harden, or otherwise narrow behavior unless required by the user, an explicit
-repository contract, or an unavoidable correctness constraint. Preserve
-operator and caller choice where policy has not been established. If an
-additional restriction appears advisable, present it as a decision rather than
-silently implementing it.
-
-**An implementation detail is not repository policy.** Tests and documentation
-must not elevate incidental behavior into a durable guarantee or restriction
-unless that behavior was expressly selected as part of the contract. Existing
-explicit contracts remain authoritative; changing or weakening one requires a
-deliberate decision rather than being inferred from this principle.
-
-[ADR-0017](../adr/0017-defer-cache-access-policy-to-deployment.md) illustrates
-these principles. Notarius previously elevated private cache creation modes
-into application policy even though Unix deployments already express access
-policy through directory permissions, process umask, setgid inheritance, and
-default ACLs. The revised design retains application-owned correctness
-boundaries while leaving deployment security policy to the operator.
+Implementation discipline is owned by [Agent Policy](agents.md).
 
 ## Layers And Dependency Direction
 
@@ -309,6 +295,13 @@ model, correction, or large payload content.
 
 ## State, Output, And Safety
 
+Source material and model responses cross application validation boundaries.
+Host access policy belongs to the deployment; hypothetical privileged actors
+concurrently changing the filesystem do not justify new application defenses
+without an explicit decision. This scope statement does not remove the adopted
+correctness boundaries below or claim excluded scenarios cannot occur. Follow
+[Agent Policy](agents.md) when a new threat or guarantee needs a decision.
+
 Notarius exposes three filesystem surfaces with independent roots and
 lifecycle:
 
@@ -364,8 +357,9 @@ best-effort development and compilation environment, while Windows is
 unsupported. Notarius distributes source releases only: an immutable source
 tag and its checked-in release note identify a release. The project does not
 publish executable binaries, archives, installers, container images,
-checksums, signatures, or package-manager entries. Maintainer release commands
-and tag guards belong to [Source Releases](../release.md).
+checksums, signatures, or package-manager entries. [Release Policy](release.md)
+owns publication requirements and completion; maintainer commands and guards
+belong to [Source Releases](../release.md).
 
 ## Architectural Non-Goals
 

docs/policy/documentation.md

@@ -1,5 +1,14 @@
 # Documentation Policy
 
+**Base Policy Version:** 1.3
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Notarius retains subprocess-consumer ownership, the distinction between
+descriptive behavior and policy, and no placeholders for absent interfaces.
+
 ## Purpose
 
 This policy assigns each documentation topic to one canonical owner. Its goal is
@@ -24,8 +33,12 @@ contract, maintain it only in the owning document.
 Outside `docs/roadmap/`, documentation describes implemented behavior only.
 Partial features may be described only to their implemented boundary.
 
-ADRs are the narrow exception: an ADR may record an accepted architectural
-decision before implementation, but acceptance must not be presented as proof
+Architectural constraints may govern future changes without implying that the
+components or capabilities they would govern already exist. This does not permit
+describing a proposed system structure as implemented.
+
+ADRs may record an accepted architectural decision before implementation, but
+acceptance must not be presented as proof
 that the behavior exists. The roadmap owns implementation status and sequencing
 until the decision is implemented. Current architecture, user, operator,
 integration, and internal documentation are updated when the behavior lands.
@@ -38,7 +51,7 @@ policy, default, or limit as authoritative merely because the current code
 implements it. Documentation may accurately describe current behavior without
 making it a permanent requirement; distinguish descriptive behavior from
 normative policy. The
-[implementation-discipline rules](architecture.md#implementation-discipline)
+[implementation-discipline rules](agents.md#implementation-discipline)
 own the criteria for adopting product restrictions and for elevating behavior
 into repository policy.
 
@@ -71,16 +84,18 @@ secret values.
 | --- | --- | --- | --- |
 | Product orientation and minimal end-to-end quickstart | `README.md` | What Notarius is, why it is useful, one shortest successful invocation, and links onward. | Complete command reference, configuration reference, operational procedures, implementation detail. |
 | Contributor entry point | `docs/development.md` | Task-oriented reading guide, minimal contributor orientation, baseline validation commands, and links to canonical docs. | Package inventory, architecture rules, subsystem behavior, detailed change recipes. |
-| Current application architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Application architecture policy | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Agent policy | `docs/policy/agents.md` | Implementation discipline, rules against invented restrictions or guarantees, and local commit authorization. | Application contracts, architecture, testing policy, and contributor procedures. |
 | Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
 | Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression-test policy, and criteria for adding, rewriting, or deleting tests. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
 | CLI contract | `docs/cli.md` | Commands, arguments, flags, invocation semantics, and exit codes. | End-to-end operating procedures, configuration field definitions, runtime filesystem layout, module implementation details. |
 | Configuration contract | `docs/config.md` | Discovery and precedence, file schema, fields, defaults, environment overrides, validation rules, and user-selectable module or validator keys. | Complete example files, CLI syntax, runtime state lifecycle, module implementation details. |
 | Operations | `docs/operations.md` | Runtime workflows, physical filesystem and state layout, output, cache, and debug handling, resume, cleanup, permissions, recovery, and operational limits. | CLI flag syntax, configuration field definitions, logical output schemas, implementation mechanics. |
-| Source release procedure | `docs/release.md` | Maintainer release selection, candidate validation, tagging, publication guards, verification, and immutable-tag recovery. | Product installation summary, CLI version semantics, historical release summaries, CI implementation detail. |
+| Release policy | `docs/policy/release.md` | Versioning, compatibility, integrity, immutability, proportionate gate selection, publication controls, and release completion. | Concrete commands, product installation, historical notes, and CI implementation detail. |
+| Repository-specific release procedure | `docs/release.md` | Prerequisites, candidate preparation, publication command, automated guards, optional diagnostics, and failure recovery. | General release requirements, product installation guidance, historical summaries, and current compatibility contracts. |
 | Release-note history | `docs/releases/` | One checked-in historical summary for each source release made under the procedure. The note at the immutable tag is that release's record. | Current commands, behavior, contracts, and compatibility definitions. |
 | Public HTTP contract, if introduced | `docs/api.md` | Routes, authentication, media types, request and response schemas, status codes, pagination, caching, idempotency, rate limits, and HTTP retry semantics. | Client walkthroughs, upstream or downstream integration internals, implementation detail. |
-| Consumer guidance, if a public package or API is introduced | `docs/consumers/` | Task-oriented use of the public interface, minimal client examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| Consumer guidance | `docs/consumers/` | Task-oriented subprocess use, minimal client examples, consumer responsibilities, and guidance for any future public package or API. | Wire semantics, external protocol contracts, internal implementation detail. |
 | External and durable integration contracts | `docs/integrations/` | External file formats and protocols, upstream and downstream contracts, logical output bundle paths and schemas, media types, and compatibility behavior. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
 | Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
 | Internal component behavior | Other files under `docs/internal/` | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
@@ -111,11 +126,13 @@ other contracts.
 
 ### Releases
 
-`docs/release.md` owns the source-release procedure. Release notes are
+The [release policy](release.md) owns requirements and invariants; the
+[repository procedure](../release.md) owns the commands that satisfy them.
+Release notes are
 historical summaries, not current-state contract owners: the checked-in note at
 an immutable tag records that release, while current canonical documentation
-must change with the behavior it describes. Do not use a release note to defer
-or replace current documentation updates.
+must change with the behavior it describes. Do not use a release procedure or
+note to weaken policy, redefine current contracts, or defer documentation updates.
 
 ### Contracts And Implementation
 

docs/policy/release.md

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

docs/policy/testing.md

@@ -1,5 +1,14 @@
 # Testing Policy
 
+**Base Policy Version:** 1.1
+
+**Last Updated**: 2026-09-25
+
+**Upstream Source:** `policy` commit `45591a79004c6811950f14dca554f6ff69b7f58e`.
+
+Adopted without substantive deviations; Notarius's established contracts remain
+subject to [Agent Policy](agents.md).
+
 ## Purpose
 
 Our tests exist to make **incorrect changes expensive and correct changes cheap**.
@@ -8,29 +17,16 @@ We do not optimize for test count, line coverage, exhaustive isolation, or the f
 
 ## Every test has a cost
 
-Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
-
-A test must be:
-
-- written and reviewed;
-- understood by future maintainers and coding agents;
-- executed in local and CI workflows;
-- diagnosed when it fails;
-- updated when legitimate behavior changes;
-- maintained as fixtures, APIs, and dependencies evolve; and
-- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
-
-Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+Every test has an immediate and continuing cost: writing, review, execution,
+diagnosis, fixture maintenance, and updates when legitimate behavior changes.
+Tests can also constrain refactoring, duplicate policy, and add failure noise.
 
 A test is warranted only when the confidence it provides justifies these costs.
-
-Apply this cost-benefit analysis at two levels:
+Evaluate that tradeoff at two levels:
 
 1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
 2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
 
-The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
-
 Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
 
 The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
@@ -55,7 +51,7 @@ Control clocks, randomness, environment variables, and other process-global or m
 
 ## What deserves tests
 
-Prioritize tests for:
+To the extent the following features are present in the application, prioritize tests for:
 
 1. Public and package-level contracts.
 2. Domain rules and important invariants.
@@ -67,7 +63,18 @@ Prioritize tests for:
 
 A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
 
-For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+Choose tests for requested behavior and established contracts. Tests must not
+introduce new restrictions, guarantees, or threat scenarios merely because they
+could be asserted. Follow the [agent policy](agents.md) when deciding whether a
+behavior belongs to the contract; current implementation behavior alone is not
+sufficient authority.
+
+For established behavior involving **data integrity, destructive operations,
+compatibility, security, concurrency, idempotency, or recovery**, favor durable
+tests for realistic failures whose consequences warrant protection. Identify
+the failure and the existing contract it would violate. These categories call
+for attention, not an automatic test inventory or permission to strengthen the
+application's guarantees. Account for protection already provided at other layers.
 
 Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
 
@@ -85,18 +92,7 @@ Do not force all behavior through oversized end-to-end tests. Do not test every
 
 ## Test behavior, not implementation
 
-A test should protect a decision, contract, or invariant—not memorialize the current implementation.
-
-Tests must not create a stronger product contract than the requirement they
-protect. In particular, do not turn an incidental limit, restriction,
-permission choice, validation rule, or implementation default into policy by
-asserting it without an expressly selected contract. Follow the
-[implementation-discipline rules](architecture.md#implementation-discipline)
-when deciding what behavior deserves durable protection.
-
-Before adding or retaining a test, ask:
-
-> What realistic defect would this test catch?
+A test should protect a decision, contract, or invariant. A test must not memorialize the current implementation.
 
 A test is suspect when its main purpose is to detect that someone:
 
@@ -109,8 +105,6 @@ A test is suspect when its main purpose is to detect that someone:
 
 Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
 
-A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
-
 ## Expected effects of different changes
 
 Use the following expectations when evaluating test failures and test maintenance:
@@ -128,10 +122,11 @@ A test failing is not the same as a test needing to be edited. Many tests may co
 
 Configurable thresholds and defaults must not be duplicated throughout the test suite.
 
-For example, do not encode an internal concurrency limit indirectly:
+For this example, assume the selected contract rejects a request to start more
+processes than a configured limit. Do not encode its production default indirectly:
 
 ```go
-// Production policy:
+// Production default:
 const maxConcurrency = 4
 
 // Brittle test:
@@ -149,25 +144,30 @@ require.NoError(t, runner.Start(limit))
 require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
 ```
 
-The test should prove:
+For this selected contract, the test should prove:
 
 - the configured limit is accepted; and
 - one beyond the configured limit is rejected.
 
+Rejection is not a general requirement for concurrency control. If the selected
+contract queues excess work or limits active workers while accepting larger
+batches, test that behavior instead. Do not add rejection or a configurable
+limit solely to reproduce this example.
+
 The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
 
 Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
 
-For concurrency limits, test both kinds of behavior when relevant:
+For established concurrency limits, distinguish these behaviors where the contract requires them:
 
-1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+1. **Configuration handling:** values are accepted or rejected according to the selected contract.
 2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
 
 Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
 
 ## Avoid semantic duplication across layers
 
-Each behavior should have a clear test owner.
+Each behavior should have a clear test owner. Where these layers exist:
 
 - Parser tests own parsing cases.
 - Validator tests own validation rules.
@@ -179,8 +179,6 @@ Each behavior should have a clear test owner.
 
 Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
 
-Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
-
 ## Use test doubles deliberately
 
 Choose the least elaborate test double that provides the required control or observation.
@@ -204,15 +202,16 @@ Do not use mocks merely to isolate every object or reproduce the implementation'
 
 ## Go-specific guidance
 
-Use:
+Select techniques according to the behavior and risk being tested. The following
+are options, not a checklist that every package or application must satisfy:
 
 - table-driven tests for meaningful behavioral categories and boundaries;
 - `t.TempDir()` for real filesystem behavior;
 - `httptest.Server` for realistic HTTP interactions;
-- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- fuzz tests when exploring broad input spaces, such as parsers or path handling, adds useful protection beyond representative cases;
 - golden files only when the complete output is intentionally stable;
 - integration tests where correctness depends on component interaction; and
-- a small number of representative end-to-end tests.
+- a small number of representative end-to-end tests when assembled workflows expose meaningful risks not adequately covered at narrower boundaries.
 
 Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
 
@@ -230,8 +229,6 @@ Use it to find untested critical branches and unexpectedly weak packages. Do not
 
 Pure domain logic will often warrant higher coverage than CLI wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
 
-Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
-
 ## Regression tests
 
 A bug fix should normally include a regression test that fails before the fix and passes afterward.
@@ -260,44 +257,26 @@ Strong candidates include tests that:
 
 Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
 
-Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
-
 ## Reviewing a proposed test
 
 Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
 
-1. What realistic defect would it catch?
-2. How likely is that defect?
-3. How consequential would it be?
-4. Is the behavior already protected elsewhere?
-5. At which layer should this behavior be owned?
-6. Does the test assert a durable contract or an incidental implementation detail?
-7. Could the implementation be refactored without changing the behavior and without editing this test?
-8. What should cause this test to fail?
-9. What legitimate changes should not cause this test to fail?
-10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
-11. Is there a smaller or more direct test that protects the same risk?
-
-Do not add the test when its expected lifetime cost exceeds its expected protective value.
-
-When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+1. Which requested behavior or established contract does it protect, and what realistic defect would violate it?
+2. How likely and consequential is that defect, and is it already protected elsewhere?
+3. What is the narrowest stable boundary that can test it clearly?
+4. Would a correct internal refactor or a change to an incidental default require test edits?
+5. Does its confidence justify its maintenance, execution, and diagnostic cost, or would a smaller test suffice?
 
 ## Definition of sufficient
 
 A test suite is sufficient when:
 
-- important contracts and invariants are protected;
-- meaningful boundaries and failure modes are exercised;
-- realistic and consequential regressions are credibly protected against silent recurrence;
-- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
-- important external boundaries have realistic integration coverage;
-- representative complete workflows are tested;
+- important requested behavior and established contracts are protected against realistic, consequential failures;
+- boundaries and component interactions have coverage where they introduce meaningful risks not protected elsewhere;
 - failures provide useful signal rather than redundant noise;
 - legitimate internal changes usually do not require test edits; and
 - additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
 
-Sufficiency is a risk judgment, not a coverage percentage or test count. Reassess it as the application, its users, and the consequences of failure evolve.
-
-The governing rule is:
-
-> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.
+Sufficiency is a risk judgment, not a coverage percentage, test count, or required
+mix of testing techniques. Reassess it as the application, its users, and the
+consequences of failure evolve.

docs/release.md

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

docs/releases/v0.7.1.md

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

docs/releases/v0.8.0.md

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

docs/roadmap/implementation.md

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

docs/roadmap/remediation.md

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

docs/roadmap/upgrade-kit.md

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

examples/dnd-complete.config.yml

@@ -4,6 +4,7 @@ configkit:
     - path: pipelines/dnd-session.yml
 promptkit:
   profile_file: ./llm-profiles/dnd-extraction.yml
+  transient_retries: 2
 concurrency:
   total_llm: 2
   stage_workers:

go.mod

@@ -3,8 +3,8 @@ module gitea.maximumdirect.net/eric/notarius
 go 1.25.5
 
 require (
-	gitea.maximumdirect.net/eric/ConfigKit v0.3.0
-	gitea.maximumdirect.net/eric/promptkit v0.9.0
+	gitea.maximumdirect.net/eric/ConfigKit v0.4.0
+	gitea.maximumdirect.net/eric/promptkit v0.11.0
 	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 )
@@ -12,7 +12,7 @@ require (
 require golang.org/x/text v0.40.0
 
 require (
-	gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0 // indirect
-	gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0 // indirect
+	gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1 // indirect
+	gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0 // indirect
 	go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect
 )

go.sum

@@ -1,11 +1,11 @@
-gitea.maximumdirect.net/eric/ConfigKit v0.3.0 h1:LgXEUjR0KsgwGBznOWa8FJTQ+iElctHtI1via5vKdu4=
-gitea.maximumdirect.net/eric/ConfigKit v0.3.0/go.mod h1:tYWpsrbLyu05r8+1MapWp5fN6ZrX9I8hU2xEPedYmYk=
-gitea.maximumdirect.net/eric/promptkit v0.9.0 h1:IpvDRC8L6xRxQ9hpuyKOmMc5b6MeLTKYyx+h1YAjy08=
-gitea.maximumdirect.net/eric/promptkit v0.9.0/go.mod h1:oMJ/WUJImUtwJ5e+6MAGECPYAErAkOaKel0G+3T/b4E=
-gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0 h1:lc062euk2qseO//D762i3JaFyulDNML3eQQX7DkYTho=
-gitea.maximumdirect.net/eric/promptkit-backend-openrouter v1.0.0/go.mod h1:AIa7kAu2mfrRQgcspe4L+DW51WqgnALQT60lqkEywJI=
-gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0 h1:j9YY7wsTVjzke2kHH4YAzpU0oUpM+x+nXwl1IeS+2eg=
-gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome v1.0.0/go.mod h1:4RNS+LILDg4JbS4Ts9Lwy1C92wauXJIbeQaalps4Koo=
+gitea.maximumdirect.net/eric/ConfigKit v0.4.0 h1:fo+vKff3zTGeWqJIA38NtT84b6OUP//9ZmBH3lFuKgk=
+gitea.maximumdirect.net/eric/ConfigKit v0.4.0/go.mod h1:tYWpsrbLyu05r8+1MapWp5fN6ZrX9I8hU2xEPedYmYk=
+gitea.maximumdirect.net/eric/promptkit v0.11.0 h1:x7JguOUiInPVXK0GEitjyoRFqqvcuSTjYUvqJ+whtik=
+gitea.maximumdirect.net/eric/promptkit v0.11.0/go.mod h1:0zEBVlJJC+ldDdXazAgybYBXNSe74vwFx5dyOuSos/w=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1 h1:S2/Q/JDE+m8Ca3pDhAjVbfYrdlEhs9blj0STgNopHo0=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v2 v2.20260922.1/go.mod h1:a3VBbh9oaGXD6ruxhtwHj+AWW6DY/SSyby1QtMmYSUo=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0 h1:Y0EXc3eh12eKW2NrOviVqdk3oBo6UUjopt1YAa1B3Oc=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v2 v2.20260922.0/go.mod h1:mtgD1E0SGm0e/I8Ffhz+AYX7GN7RKUWJXehNFIKHs4Y=
 github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI=
 github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 h1:KRzFb2m7YtdldCEkzs6KqmJw4nqEVZGK7IN2kJkjTuQ=

internal/cli/catalog.go

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

internal/cli/configkit_commands.go

@@ -16,7 +16,7 @@ func runConfigEffective(args []string, stdout, stderr io.Writer, opts Options) i
 	fs := flag.NewFlagSet("config effective", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
-	profile := fs.String("config-profile", "", "configuration profile")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
 	if err := fs.Parse(args); err != nil {
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
 		return 2
@@ -25,7 +25,7 @@ func runConfigEffective(args []string, stdout, stderr io.Writer, opts Options) i
 		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
 		return 2
 	}
-	loaded, _, err := loadConfigKit(*configPath, *profile, nil, opts)
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
 	if err != nil {
 		writeConfigLoadError(stderr, err)
 		return 1
@@ -41,7 +41,7 @@ func runConfigExplain(args []string, stdout, stderr io.Writer, opts Options) int
 	fs := flag.NewFlagSet("config explain", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
-	profile := fs.String("config-profile", "", "configuration profile")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
 	if err := fs.Parse(args); err != nil {
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
 		return 2
@@ -56,7 +56,7 @@ func runConfigExplain(args []string, stdout, stderr io.Writer, opts Options) int
 		fmt.Fprintf(stderr, "notarius: %q is not a canonical JSON Pointer\n", pointer)
 		return 2
 	}
-	loaded, _, err := loadConfigKit(*configPath, *profile, nil, opts)
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
 	if err != nil {
 		writeConfigLoadError(stderr, err)
 		return 1
@@ -72,7 +72,7 @@ func runConfigValidate(args []string, stdout, stderr io.Writer, opts Options) in
 	fs := flag.NewFlagSet("config validate", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
-	profile := fs.String("config-profile", "", "configuration profile")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
 	pipelineID := fs.String("pipeline", "", "pipeline ID")
 	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
 	if err := fs.Parse(args); err != nil {
@@ -92,7 +92,7 @@ func runConfigValidate(args []string, stdout, stderr io.Writer, opts Options) in
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
 		return 2
 	}
-	loaded, path, err := loadConfigKit(*configPath, *profile, nil, opts)
+	loaded, path, err := loadConfigKit(*configPath, *overlay, nil, opts)
 	if err != nil {
 		writeConfigLoadError(stderr, err)
 		return 1
@@ -123,7 +123,7 @@ func runPipelinesList(args []string, stdout, stderr io.Writer, opts Options) int
 	fs := flag.NewFlagSet("pipelines list", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
-	profile := fs.String("config-profile", "", "configuration profile")
+	overlay := fs.String("config-overlay", "", "configuration overlay")
 	jsonOutput := fs.Bool("json", false, "write JSON output")
 	if err := fs.Parse(args); err != nil {
 		fmt.Fprintf(stderr, "notarius: %v\n", err)
@@ -133,7 +133,7 @@ func runPipelinesList(args []string, stdout, stderr io.Writer, opts Options) int
 		fmt.Fprintf(stderr, "notarius: unexpected argument %q\n", fs.Arg(0))
 		return 2
 	}
-	loaded, _, err := loadConfigKit(*configPath, *profile, nil, opts)
+	loaded, _, err := loadConfigKit(*configPath, *overlay, nil, opts)
 	if err != nil {
 		writeConfigLoadError(stderr, err)
 		return 1

internal/cli/configkit_commands_test.go

@@ -14,22 +14,49 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
 )
 
-func TestConfigKitReadOnlyCommandsLoadImportsAndProfiles(t *testing.T) {
+func TestConfigKitReadOnlyCommandsLoadImportsAndOverlays(t *testing.T) {
 	t.Parallel()
 	root := t.TempDir()
 	path := filepath.Join(root, "notarius.yml")
 	writeConfigKitFile(t, filepath.Join(root, "fragment.yml"), "pipelines:\n  imported:\n    input: {module: input}\n")
-	writeConfigKitFile(t, filepath.Join(root, "profiles", "development.yml"), "pipelines:\n  profiled:\n    input: {module: input}\n")
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "development.yml"), "pipelines:\n  overlaid:\n    input: {module: input}\n")
 	writeConfigKitFile(t, path, "version: 5\nconfigkit:\n  imports:\n    - path: fragment.yml\npipelines:\n  base:\n    input: {module: input}\n")
 	var stdout, stderr bytes.Buffer
-	if code := RunWithOptions([]string{"pipelines", "list", "--config", path, "--config-profile", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || stdout.String() != "base\nimported\nprofiled\n" {
+	if code := RunWithOptions([]string{"pipelines", "list", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || stdout.String() != "base\nimported\noverlaid\n" {
 		t.Fatalf("list = %d, %q, %q", code, stdout.String(), stderr.String())
 	}
 	stdout.Reset()
 	stderr.Reset()
-	if code := RunWithOptions([]string{"config", "validate", "--config", path, "--config-profile", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || !strings.Contains(stdout.String(), "is valid") {
+	if code := RunWithOptions([]string{"config", "validate", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, commandContractOptions(t)); code != 0 || !strings.Contains(stdout.String(), "is valid") {
 		t.Fatalf("validate = %d, %q, %q", code, stdout.String(), stderr.String())
 	}
+	for _, selection := range [][]string{nil, {"--config-overlay", ""}, {"--config-overlay", "default"}} {
+		stdout.Reset()
+		stderr.Reset()
+		args := append([]string{"pipelines", "list", "--config", path}, selection...)
+		if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != 0 || stdout.String() != "base\nimported\n" || stderr.Len() != 0 {
+			t.Fatalf("base list %q = %d, %q, %q", selection, code, stdout.String(), stderr.String())
+		}
+	}
+}
+
+func TestRetiredConfigProfileFlagUsesNormalUnknownFlagBehavior(t *testing.T) {
+	t.Parallel()
+
+	for _, args := range [][]string{
+		{"run", "sample", "--config-profile=development"},
+		{"config", "validate", "--config-profile=development"},
+		{"config", "effective", "--config-profile=development"},
+		{"config", "explain", "--config-profile=development", ""},
+		{"pipelines", "list", "--config-profile=development"},
+	} {
+		t.Run(strings.Join(args[:2], " "), func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			if code := RunWithOptions(args, &stdout, &stderr, commandContractOptions(t)); code != 2 || stdout.Len() != 0 || !strings.Contains(stderr.String(), "not defined") {
+				t.Fatalf("args=%q code=%d stdout=%q stderr=%q", args, code, stdout.String(), stderr.String())
+			}
+		})
+	}
 }
 
 func TestConfigKitReadOnlyCommandsRejectVersionFour(t *testing.T) {
@@ -53,8 +80,8 @@ pipelines:
   "0":
     input: {module: input}
 `)
-	writeConfigKitFile(t, filepath.Join(root, "profiles", "development.yml"), "concurrency:\n  total_llm: 3\n")
-	writeConfigKitFile(t, filepath.Join(root, "profiles", "retired.yml"), "pipelines: null\n")
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "development.yml"), "concurrency:\n  total_llm: 3\n")
+	writeConfigKitFile(t, filepath.Join(root, "overlays", "retired.yml"), "pipelines: null\n")
 
 	providerCalled := false
 	opts := commandContractOptions(t)
@@ -64,7 +91,7 @@ pipelines:
 	}
 
 	var stdout, stderr bytes.Buffer
-	if code := RunWithOptions([]string{"config", "effective", "--config", path, "--config-profile", "development"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), "total_llm: 3") || stderr.Len() != 0 {
+	if code := RunWithOptions([]string{"config", "effective", "--config", path, "--config-overlay", "development"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), "total_llm: 3") || stderr.Len() != 0 {
 		t.Fatalf("effective = %d, %q, %q", code, stdout.String(), stderr.String())
 	}
 	if providerCalled {
@@ -73,7 +100,7 @@ pipelines:
 
 	stdout.Reset()
 	stderr.Reset()
-	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-profile", "development", "/concurrency/total_llm"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/concurrency/total_llm"`) || stderr.Len() != 0 {
+	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-overlay", "development", "/concurrency/total_llm"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/concurrency/total_llm"`) || stderr.Len() != 0 {
 		t.Fatalf("explain effective = %d, %q, %q", code, stdout.String(), stderr.String())
 	}
 
@@ -85,7 +112,7 @@ pipelines:
 
 	stdout.Reset()
 	stderr.Reset()
-	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-profile", "retired", "/pipelines/0"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/pipelines/0"`) || stderr.Len() != 0 {
+	if code := RunWithOptions([]string{"config", "explain", "--config", path, "--config-overlay", "retired", "/pipelines/0"}, &stdout, &stderr, opts); code != 0 || !strings.Contains(stdout.String(), `path: "/pipelines/0"`) || stderr.Len() != 0 {
 		t.Fatalf("explain removed mapping key = %d, %q, %q", code, stdout.String(), stderr.String())
 	}
 }

internal/cli/configkit_load.go

@@ -10,7 +10,7 @@ import (
 	"gitea.maximumdirect.net/eric/notarius/internal/core/config"
 )
 
-func loadConfigKit(configPath, profile string, overrides []configkit.Override, opts Options) (*config.LoadedConfig, string, error) {
+func loadConfigKit(configPath, overlay string, overrides []configkit.Override, opts Options) (*config.LoadedConfig, string, error) {
 	path, err := discoverConfigPath(configPath, opts)
 	if err != nil {
 		return nil, "", err
@@ -19,7 +19,7 @@ func loadConfigKit(configPath, profile string, overrides []configkit.Override, o
 	if err != nil {
 		return nil, "", fmt.Errorf("determine invocation directory: %w", err)
 	}
-	loaded, err := config.Load(path, config.LoadOptions{Profile: profile, Overrides: overrides, InvocationDirectory: workingDirectory})
+	loaded, err := config.Load(path, config.LoadOptions{Overlay: overlay, Overrides: overrides, InvocationDirectory: workingDirectory})
 	if err != nil {
 		return nil, "", err
 	}
@@ -31,7 +31,7 @@ func writeConfigLoadError(output io.Writer, err error) {
 	if errors.As(err, &loadError) {
 		if writeErr := configkit.WriteDiagnostics(output, loadError.Diagnostics()); writeErr == nil {
 			// ConfigKit's locations contain source IDs. Keep its renderer and
-			// supply the acquired-source legend so imported/profile errors can
+			// supply the acquired-source legend so imported/overlay errors can
 			// be traced to a file. Quoting also escapes terminal controls.
 			for _, source := range loadError.Sources() {
 				label := source.Path

internal/cli/configkit_load_test.go

@@ -13,16 +13,16 @@ import (
 
 func TestConfigLoadDiagnosticsIdentifyComposedSources(t *testing.T) {
 	t.Parallel()
-	for _, profile := range []bool{false, true} {
-		t.Run(strconv.FormatBool(profile), func(t *testing.T) {
+	for _, overlay := range []bool{false, true} {
+		t.Run(strconv.FormatBool(overlay), func(t *testing.T) {
 			root := t.TempDir()
 			main := filepath.Join(root, "config.yml")
 			fragment := filepath.Join(root, "pipeline\n\x1b.yml")
 			args := []string{"config", "validate", "--config", main}
-			if profile {
-				fragment = filepath.Join(root, "profiles", "development.yml")
+			if overlay {
+				fragment = filepath.Join(root, "overlays", "development.yml")
 				writeConfigKitFile(t, main, "version: 5\n")
-				args = append(args, "--config-profile", "development")
+				args = append(args, "--config-overlay", "development")
 			} else {
 				writeConfigKitFile(t, main, "version: 5\nconfigkit:\n  imports:\n    - path: "+strconv.Quote(filepath.Base(fragment))+"\n")
 			}
@@ -43,18 +43,18 @@ func TestConfigLoadDiagnosticsIdentifyComposedSources(t *testing.T) {
 	}
 }
 
-func TestLoadConfigKitUsesDiscoveryProfileAndOverrides(t *testing.T) {
+func TestLoadConfigKitUsesDiscoveryOverlayAndOverrides(t *testing.T) {
 	t.Parallel()
 	root := t.TempDir()
 	path := filepath.Join(root, "notarius.yml")
 	writeConfigKitFile(t, path, "version: 5\npipelines:\n  base:\n    input: {module: input}\n")
-	profiles := filepath.Join(root, "profiles")
-	writeConfigKitFile(t, filepath.Join(profiles, "development.yml"), "pipelines:\n  profiled:\n    input: {module: input}\n")
+	overlays := filepath.Join(root, "overlays")
+	writeConfigKitFile(t, filepath.Join(overlays, "development.yml"), "pipelines:\n  overlaid:\n    input: {module: input}\n")
 	loaded, gotPath, err := loadConfigKit(path, "development", []configkit.Override{{Path: configkit.Keys("output", "directory"), Value: "generated", Flag: "--output"}}, commandContractOptions(t))
 	if err != nil {
 		t.Fatalf("loadConfigKit() error = %v", err)
 	}
-	if gotPath != path || loaded.Config.Pipelines["profiled"].Input.Module != "input" || loaded.Config.Output.Directory == "generated" {
+	if gotPath != path || loaded.Config.Pipelines["overlaid"].Input.Module != "input" || loaded.Config.Output.Directory == "generated" {
 		t.Fatalf("loaded = %#v, path=%q", loaded.Config, gotPath)
 	}
 }

internal/cli/example_contract_test.go

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

internal/cli/production_contract_test.go

@@ -557,11 +557,16 @@ func TestNormalizeOptionsSharesProductionProfileAssetsWithDefaultRuntime(t *test
 func TestProductionLLMClientFactoryUsesConfiguredLocalBackend(t *testing.T) {
 	var providerCalls atomic.Int32
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		providerCalls.Add(1)
+		call := providerCalls.Add(1)
 		if r.URL.Path != "/v1/chat/completions" {
 			t.Errorf("provider path = %q, want /v1/chat/completions", r.URL.Path)
 		}
 		w.Header().Set("Content-Type", "application/json")
+		if call == 1 {
+			w.WriteHeader(http.StatusServiceUnavailable)
+			_, _ = w.Write([]byte(`{"error":{"message":"transient"}}`))
+			return
+		}
 		_, _ = w.Write([]byte(`{
 			"choices": [{"message": {"role": "assistant", "content": "{\"ok\":true}"}}],
 			"usage": {"prompt_tokens": 3, "completion_tokens": 4, "total_tokens": 7}
@@ -596,6 +601,7 @@ output:
 
 	cfg := config.Default()
 	cfg.PromptKit.ProfileFile = profilePath
+	cfg.PromptKit.TransientRetries = 1
 	cfg.PromptKit.LocalBackend = &config.PromptKitLocalBackendConfig{
 		Endpoint:         server.URL + "/v1",
 		ConcurrencyLimit: 2,
@@ -624,8 +630,8 @@ output:
 	if err != nil {
 		t.Fatalf("CompleteStructured() error = %v, want nil", err)
 	}
-	if providerCalls.Load() != 1 {
-		t.Fatalf("provider calls = %d, want 1", providerCalls.Load())
+	if providerCalls.Load() != 2 {
+		t.Fatalf("provider calls = %d, want 2", providerCalls.Load())
 	}
 	provider, ok := client.(contracts.LLMProfileManifestProvider)
 	if !ok {

internal/cli/run.go

@@ -34,10 +34,10 @@ const usage = `Usage:
   notarius help
   notarius --version
   notarius run <pipeline-id> --input path/to/source.json [--json] [flags]
-  notarius config validate --config path/to/config.yml [--config-profile name] [--pipeline pipeline-id] [--only lane-a,lane-b]
-  notarius config effective --config path/to/config.yml [--config-profile name]
-  notarius config explain --config path/to/config.yml [--config-profile name] <json-pointer>
-  notarius pipelines list --config path/to/config.yml [--config-profile name] [--json]
+  notarius config validate --config path/to/config.yml [--config-overlay name] [--pipeline pipeline-id] [--only lane-a,lane-b]
+  notarius config effective --config path/to/config.yml [--config-overlay name]
+  notarius config explain --config path/to/config.yml [--config-overlay name] <json-pointer>
+  notarius pipelines list --config path/to/config.yml [--config-overlay name] [--json]
 `
 
 type Options struct {
@@ -160,7 +160,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 	fs := flag.NewFlagSet("run", flag.ContinueOnError)
 	fs.SetOutput(io.Discard)
 	configPath := fs.String("config", "", "config file path")
-	configProfile := fs.String("config-profile", "", "configuration profile")
+	configOverlay := fs.String("config-overlay", "", "configuration overlay")
 	inputPath := fs.String("input", "", "source input file path")
 	onlyRaw := fs.String("only", "", "comma-separated artifact lanes")
 	outputDir := fs.String("output-dir", "", "output directory")
@@ -268,7 +268,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		return 2
 	}
 
-	loaded, loadedConfigPath, err := loadConfigKit(*configPath, *configProfile, runConfigurationOverrides(chunkCache, *outputDir, *debugDir), opts)
+	loaded, loadedConfigPath, err := loadConfigKit(*configPath, *configOverlay, runConfigurationOverrides(chunkCache, *outputDir, *debugDir), opts)
 	if err != nil {
 		writeConfigLoadError(stderr, err)
 		return 1
@@ -331,7 +331,7 @@ func runPipelineCommand(args []string, stdout, stderr io.Writer, opts Options) i
 		InputPath:               strings.TrimSpace(*inputPath),
 		ConfigPath:              loadedConfigPath,
 		ConfigSource:            configSource(*configPath),
-		ConfigProfile:           effectiveConfigProfile(*configProfile),
+		ConfigOverlay:           effectiveConfigOverlay(*configOverlay),
 		OnlyLanes:               append([]string(nil), only...),
 		ChunkCacheOverride:      chunkCache.explicitValue(),
 		ReasoningEffortOverride: runtimeOverrides.ReasoningEffort,
@@ -877,7 +877,7 @@ func reorderRunArgs(args []string) []string {
 
 func runFlagTakesValue(arg string) bool {
 	switch arg {
-	case "--config", "--config-profile", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--reasoning-effort", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
+	case "--config", "--config-overlay", "--input", "--only", "--output-dir", "--debug-dir", "--llm-profile", "--session-id", "--reasoning-effort", "--chunk_cache", "--reference", "--without-reference", "--recompute-step":
 		return true
 	default:
 		return false
@@ -963,7 +963,7 @@ func chunkPlanStoreForRun(cfg config.ChunkPlanCacheConfig, opts Options) (pipeli
 
 func validateRunFlagValues(args []string) error {
 	for i, arg := range args {
-		if arg != "--config-profile" && arg != "--session-id" && arg != "--reasoning-effort" {
+		if arg != "--config-overlay" && arg != "--session-id" && arg != "--reasoning-effort" {
 			continue
 		}
 		if i+1 >= len(args) || strings.HasPrefix(args[i+1], "-") {
@@ -973,11 +973,11 @@ func validateRunFlagValues(args []string) error {
 	return nil
 }
 
-func effectiveConfigProfile(profile string) string {
-	if profile == "" || profile == "default" {
+func effectiveConfigOverlay(overlay string) string {
+	if overlay == "" || overlay == "default" {
 		return ""
 	}
-	return profile
+	return overlay
 }
 
 func flagWasProvided(args []string, name string) bool {

internal/cli/run_contract_test.go

@@ -38,8 +38,8 @@ func TestRunControlsRejectSyntaxWithoutAllocatingState(t *testing.T) {
 		{name: "blank debug directory", args: func(roots stateTestRoots) []string {
 			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--debug-dir", ""}
 		}},
-		{name: "missing configuration profile", args: func(roots stateTestRoots) []string {
-			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--config-profile"}
+		{name: "missing configuration overlay", args: func(roots stateTestRoots) []string {
+			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--config-overlay"}
 		}},
 		{name: "debug directory without debug", args: func(roots stateTestRoots) []string {
 			return []string{"run", "sample", "--config", roots.config, "--input", roots.input, "--debug-dir", filepath.Join(filepath.Dir(roots.debug), "requested-debug")}
@@ -247,38 +247,38 @@ func TestRunStateRootsHonorEnvironmentFlagsAndDefaults(t *testing.T) {
 	})
 }
 
-func TestRunAppliesConfigurationProfileAndRecordsSelection(t *testing.T) {
+func TestRunAppliesConfigurationOverlayAndRecordsSelection(t *testing.T) {
 	roots := newStateTestRoots(t)
-	profileDirectory := filepath.Join(filepath.Dir(roots.config), "profiles")
-	profileOutput := filepath.Join(profileDirectory, "profile-output")
-	profileDebug := filepath.Join(profileDirectory, "profile-debug")
-	writeConfigKitFile(t, filepath.Join(profileDirectory, "development.yml"), "output:\n  directory: ./profile-output\ndebug:\n  directory: ./profile-debug\n")
+	overlayDirectory := filepath.Join(filepath.Dir(roots.config), "overlays")
+	overlayOutput := filepath.Join(overlayDirectory, "overlay-output")
+	overlayDebug := filepath.Join(overlayDirectory, "overlay-debug")
+	writeConfigKitFile(t, filepath.Join(overlayDirectory, "development.yml"), "output:\n  directory: ./overlay-output\ndebug:\n  directory: ./overlay-debug\n")
 
 	var stdout, stderr bytes.Buffer
 	code := RunWithOptions([]string{
-		"run", "sample", "--config", roots.config, "--config-profile", "development",
+		"run", "sample", "--config", roots.config, "--config-overlay", "development",
 		"--input", roots.input, "--chunk_cache", "bypass", "--debug",
 	}, &stdout, &stderr, newStateTestHarness().options())
 	if code != 0 || stderr.Len() != 0 {
 		t.Fatalf("code=%d stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 	}
-	assertFile(t, filepath.Join(profileOutput, filepath.Base(onlyChildDir(t, profileOutput)), "result.json"))
-	bundle := onlyChildDir(t, profileDebug)
+	assertFile(t, filepath.Join(overlayOutput, filepath.Base(onlyChildDir(t, overlayOutput)), "result.json"))
+	bundle := onlyChildDir(t, overlayDebug)
 	var invocation debugbundle.Invocation
 	readStateTestSummaryJSON(t, bundle, "invocation.json", &invocation)
-	if invocation.ConfigProfile != "development" {
-		t.Fatalf("config profile = %q, want development", invocation.ConfigProfile)
+	if invocation.ConfigOverlay != "development" {
+		t.Fatalf("config overlay = %q, want development", invocation.ConfigOverlay)
 	}
 }
 
-func TestEffectiveConfigProfileOmitsBaseSelection(t *testing.T) {
-	for _, profile := range []string{"", "default"} {
-		if got := effectiveConfigProfile(profile); got != "" {
-			t.Fatalf("effectiveConfigProfile(%q) = %q, want empty", profile, got)
+func TestEffectiveConfigOverlayOmitsBaseSelection(t *testing.T) {
+	for _, overlay := range []string{"", "default"} {
+		if got := effectiveConfigOverlay(overlay); got != "" {
+			t.Fatalf("effectiveConfigOverlay(%q) = %q, want empty", overlay, got)
 		}
 	}
-	if got := effectiveConfigProfile("development"); got != "development" {
-		t.Fatalf("effectiveConfigProfile(development) = %q", got)
+	if got := effectiveConfigOverlay("development"); got != "development" {
+		t.Fatalf("effectiveConfigOverlay(development) = %q", got)
 	}
 }
 

internal/core/config/config.go

@@ -14,9 +14,10 @@ type Config struct {
 }
 
 type PromptKitConfig struct {
-	ProfileDir   string                       `json:"profile_dir,omitempty"`
-	ProfileFile  string                       `json:"profile_file,omitempty"`
-	LocalBackend *PromptKitLocalBackendConfig `json:"local_backend,omitempty"`
+	ProfileDir       string                       `json:"profile_dir,omitempty"`
+	ProfileFile      string                       `json:"profile_file,omitempty"`
+	TransientRetries int                          `json:"transient_retries"`
+	LocalBackend     *PromptKitLocalBackendConfig `json:"local_backend,omitempty"`
 }
 
 type PromptKitLocalBackendConfig struct {
@@ -29,7 +30,10 @@ type ConcurrencyConfig struct {
 	StageWorkers map[string]int `json:"stage_workers"`
 }
 
-const defaultLLMConcurrency = 16
+const (
+	defaultLLMConcurrency   = 16
+	defaultTransientRetries = 2
+)
 
 type OutputConfig struct {
 	Directory string `json:"directory"`
@@ -55,6 +59,7 @@ type DebugConfig struct {
 
 func Default() Config {
 	return Config{
+		PromptKit: PromptKitConfig{TransientRetries: defaultTransientRetries},
 		Pipelines: map[string]pipeline.PipelineProfile{},
 		Concurrency: ConcurrencyConfig{
 			TotalLLM:     defaultLLMConcurrency,

internal/core/config/document.go

@@ -16,9 +16,10 @@ type Document struct {
 }
 
 type PromptKitDocument struct {
-	ProfileDir   *string                        `yaml:"profile_dir,omitempty"`
-	ProfileFile  *string                        `yaml:"profile_file,omitempty"`
-	LocalBackend *PromptKitLocalBackendDocument `yaml:"local_backend,omitempty"`
+	ProfileDir       *string                        `yaml:"profile_dir,omitempty"`
+	ProfileFile      *string                        `yaml:"profile_file,omitempty"`
+	TransientRetries *int                           `yaml:"transient_retries,omitempty"`
+	LocalBackend     *PromptKitLocalBackendDocument `yaml:"local_backend,omitempty"`
 }
 
 type PromptKitLocalBackendDocument struct {

internal/core/config/document_paths_test.go

@@ -77,27 +77,48 @@ func TestDocumentPathsUseInvocationDirectoryForOverrides(t *testing.T) {
 	}
 }
 
-func TestDocumentPathsUseProfileDeclarationDirectory(t *testing.T) {
+func TestDocumentPathsUseOverlayDeclarationDirectory(t *testing.T) {
 	t.Parallel()
 
 	root := t.TempDir()
-	profiles := filepath.Join(root, "profiles")
-	if err := os.MkdirAll(profiles, 0o700); err != nil {
+	overlays := filepath.Join(root, "overlays")
+	if err := os.MkdirAll(overlays, 0o700); err != nil {
 		t.Fatal(err)
 	}
 	main := filepath.Join(root, "notarius.yml")
 	if err := os.WriteFile(main, []byte("version: 5\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
-	if err := os.WriteFile(filepath.Join(profiles, "development.yml"), []byte("output: {directory: ./profile-output}\n"), 0o600); err != nil {
+	if err := os.WriteFile(filepath.Join(overlays, "development.yml"), []byte("output: {directory: ./overlay-output}\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
-	loaded, err := loadConfig(main, LoadOptions{Profile: "development"})
+	loaded, err := loadConfig(main, LoadOptions{Overlay: "development"})
 	if err != nil {
 		t.Fatalf("loadConfig() error = %v", err)
 	}
-	if got := loaded.Config.Output.Directory; got != filepath.Join(profiles, "profile-output") {
-		t.Fatalf("profile directory = %q", got)
+	if got := loaded.Config.Output.Directory; got != filepath.Join(overlays, "overlay-output") {
+		t.Fatalf("overlay directory = %q", got)
+	}
+}
+
+func TestDocumentLoadDoesNotFallBackToLegacyProfileDirectory(t *testing.T) {
+	t.Parallel()
+
+	root := t.TempDir()
+	main := filepath.Join(root, "notarius.yml")
+	if err := os.WriteFile(main, []byte("version: 5\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	legacy := filepath.Join(root, "profiles", "development.yml")
+	if err := os.MkdirAll(filepath.Dir(legacy), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(legacy, []byte("output: {directory: ./legacy-output}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	if _, err := loadConfig(main, LoadOptions{Overlay: "development"}); err == nil {
+		t.Fatal("loadConfig() error = nil, want missing overlay")
 	}
 }
 

internal/core/config/load.go

@@ -10,7 +10,7 @@ import (
 // boundary. It is intentionally narrow: configuration discovery and CLI flag
 // parsing remain outside this package.
 type LoadOptions struct {
-	Profile             string
+	Overlay             string
 	Overrides           []configkit.Override
 	InvocationDirectory string
 }
@@ -30,7 +30,7 @@ func loadDocumentWithSource(source configkit.Source, options LoadOptions, import
 	return configkit.Load(source, configkit.Options[Document]{
 		Definition: documentDefinition(state),
 		Imports:    imports,
-		Profile:    options.Profile,
+		Overlay:    options.Overlay,
 		Overrides:  options.Overrides,
 	})
 }
@@ -63,6 +63,9 @@ func documentDefinition(state *documentLoadState) configkit.Definition[Document]
 
 func documentDefaults() Document {
 	return Document{
+		PromptKit: &PromptKitDocument{
+			TransientRetries: documentValue(defaultTransientRetries),
+		},
 		Pipelines: map[string]PipelineDocument{},
 		Concurrency: &ConcurrencyDocument{
 			TotalLLM:     documentValue(defaultLLMConcurrency),

internal/core/config/loaded_config.go

@@ -94,6 +94,9 @@ func documentToConfig(document Document) (Config, error) {
 		if document.PromptKit.ProfileFile != nil {
 			config.PromptKit.ProfileFile = *document.PromptKit.ProfileFile
 		}
+		if document.PromptKit.TransientRetries != nil {
+			config.PromptKit.TransientRetries = *document.PromptKit.TransientRetries
+		}
 		if document.PromptKit.LocalBackend != nil {
 			backend := PromptKitLocalBackendConfig{}
 			if document.PromptKit.LocalBackend.Endpoint != nil {

internal/core/config/loaded_config_test.go

@@ -2,6 +2,7 @@ package config
 
 import (
 	"bytes"
+	"strconv"
 	"strings"
 	"testing"
 
@@ -58,6 +59,44 @@ pipelines:
 	}
 }
 
+func TestLoadedConfigResolvesPromptKitTransientRetriesAndRetainsInspection(t *testing.T) {
+	t.Parallel()
+
+	for _, test := range []struct {
+		name     string
+		document string
+		want     int
+	}{
+		{name: "omitted", document: "version: 5\n", want: defaultTransientRetries},
+		{name: "disabled", document: "version: 5\npromptkit:\n  transient_retries: 0\n", want: 0},
+		{name: "configured", document: "version: 5\npromptkit:\n  transient_retries: 5\n", want: 5},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			loaded, err := loadConfigSource(configkit.Bytes("document.yml", []byte(test.document)), LoadOptions{})
+			if err != nil {
+				t.Fatal(err)
+			}
+			if got := loaded.Config.PromptKit.TransientRetries; got != test.want {
+				t.Fatalf("transient retries = %d, want %d", got, test.want)
+			}
+			var effective bytes.Buffer
+			if err := loaded.WriteEffective(&effective); err != nil {
+				t.Fatal(err)
+			}
+			if !strings.Contains(effective.String(), "transient_retries: "+strconv.Itoa(test.want)) {
+				t.Fatalf("effective configuration = %q", effective.String())
+			}
+			var explanation bytes.Buffer
+			if err := loaded.WriteExplanation(&explanation, "/promptkit/transient_retries"); err != nil {
+				t.Fatal(err)
+			}
+			if !strings.Contains(explanation.String(), strconv.Itoa(test.want)) {
+				t.Fatalf("explanation = %q", explanation.String())
+			}
+		})
+	}
+}
+
 func TestLoadedConfigOwnsRuntimeValuesAndKeepsFrozenInspection(t *testing.T) {
 	t.Parallel()
 

internal/core/config/operational_document.go

@@ -167,6 +167,11 @@ func validateDocumentPromptKit(promptKit *PromptKitDocument, validation *configk
 	if err := validatePromptKitProfileSources(profileDir, profileFile); err != nil {
 		validation.Error(configkit.Keys("promptkit"), err.Error())
 	}
+	if promptKit.TransientRetries == nil {
+		validation.Error(configkit.Keys("promptkit", "transient_retries"), "must be an integer")
+	} else if err := validatePromptKitTransientRetries(*promptKit.TransientRetries); err != nil {
+		validation.Error(configkit.Keys("promptkit", "transient_retries"), err.Error())
+	}
 	if promptKit.LocalBackend == nil {
 		return
 	}
@@ -283,6 +288,13 @@ func validatePromptKitLocalBackendConcurrencyLimit(concurrencyLimit int) error {
 	return nil
 }
 
+func validatePromptKitTransientRetries(transientRetries int) error {
+	if transientRetries < 0 {
+		return fmt.Errorf("promptkit.transient_retries must not be negative")
+	}
+	return nil
+}
+
 func validateRequiredPath(name, value string) error {
 	if strings.TrimSpace(value) == "" {
 		return fmt.Errorf("%s must not be empty", name)

internal/core/config/operational_document_test.go

@@ -105,28 +105,28 @@ debug:
 	}
 }
 
-func TestDocumentLoadDerivesWorkersAfterProfileAndCLIOverrides(t *testing.T) {
+func TestDocumentLoadDerivesWorkersAfterOverlayAndCLIOverrides(t *testing.T) {
 	t.Parallel()
 
 	directory := t.TempDir()
 	main := filepath.Join(directory, "config.yml")
-	profiles := filepath.Join(directory, "profiles")
-	if err := os.Mkdir(profiles, 0o755); err != nil {
+	overlays := filepath.Join(directory, "overlays")
+	if err := os.Mkdir(overlays, 0o755); err != nil {
 		t.Fatal(err)
 	}
 	if err := os.WriteFile(main, []byte("version: 5\nconcurrency:\n  total_llm: 4\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
-	if err := os.WriteFile(filepath.Join(profiles, "development.yml"), []byte("concurrency:\n  total_llm: 3\n"), 0o600); err != nil {
+	if err := os.WriteFile(filepath.Join(overlays, "development.yml"), []byte("concurrency:\n  total_llm: 3\n"), 0o600); err != nil {
 		t.Fatal(err)
 	}
 
-	profiled, err := loadDocument(main, LoadOptions{Profile: "development"})
+	overlaid, err := loadDocument(main, LoadOptions{Overlay: "development"})
 	if err != nil {
 		t.Fatal(err)
 	}
-	if profiled.Config.Concurrency.TotalLLM == nil || *profiled.Config.Concurrency.TotalLLM != 3 || profiled.Config.Concurrency.StageWorkers["extract"] != 3 {
-		t.Fatalf("profiled concurrency = %#v", profiled.Config.Concurrency)
+	if overlaid.Config.Concurrency.TotalLLM == nil || *overlaid.Config.Concurrency.TotalLLM != 3 || overlaid.Config.Concurrency.StageWorkers["extract"] != 3 {
+		t.Fatalf("overlaid concurrency = %#v", overlaid.Config.Concurrency)
 	}
 
 	overridden, err := loadDocument(main, LoadOptions{Overrides: []configkit.Override{{
@@ -138,6 +138,16 @@ func TestDocumentLoadDerivesWorkersAfterProfileAndCLIOverrides(t *testing.T) {
 	if overridden.Config.Concurrency.TotalLLM == nil || *overridden.Config.Concurrency.TotalLLM != 2 || overridden.Config.Concurrency.StageWorkers["extract"] != 2 {
 		t.Fatalf("overridden concurrency = %#v", overridden.Config.Concurrency)
 	}
+
+	overlaidAndOverridden, err := loadDocument(main, LoadOptions{Overlay: "development", Overrides: []configkit.Override{{
+		Path: configkit.Keys("concurrency", "total_llm"), Value: 2, Flag: "--total-llm",
+	}}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if overlaidAndOverridden.Config.Concurrency.TotalLLM == nil || *overlaidAndOverridden.Config.Concurrency.TotalLLM != 2 || overlaidAndOverridden.Config.Concurrency.StageWorkers["extract"] != 2 {
+		t.Fatalf("overlaid and overridden concurrency = %#v", overlaidAndOverridden.Config.Concurrency)
+	}
 }
 
 func TestDocumentLoadRetainsExplicitWorkerAndRejectsInvalidOperationalValues(t *testing.T) {
@@ -213,6 +223,13 @@ cache:
 	assertDocumentDiagnostic(t, err, configkit.CodeValidateError, "/promptkit/local_backend/endpoint")
 }
 
+func TestDocumentLoadRejectsNegativePromptKitTransientRetries(t *testing.T) {
+	t.Parallel()
+
+	_, err := loadDocumentSource(configkit.Bytes("promptkit.yml", []byte("version: 5\npromptkit:\n  transient_retries: -1\n")), LoadOptions{})
+	assertDocumentDiagnostic(t, err, configkit.CodeValidateError, "/promptkit/transient_retries")
+}
+
 func assertDocumentDiagnostic(t *testing.T, err error, code configkit.Code, path string) {
 	t.Helper()
 	var loadError *configkit.LoadError

internal/core/config/validation.go

@@ -46,6 +46,9 @@ func validatePromptKit(cfg PromptKitConfig) error {
 	if err := validatePromptKitProfileSources(cfg.ProfileDir, cfg.ProfileFile); err != nil {
 		return err
 	}
+	if err := validatePromptKitTransientRetries(cfg.TransientRetries); err != nil {
+		return err
+	}
 	if cfg.LocalBackend == nil {
 		return nil
 	}

internal/core/config/validation_contract_test.go

@@ -89,6 +89,22 @@ func TestValidatePromptKitSourcesAreMutuallyExclusive(t *testing.T) {
 	assertValidationContains(t, cfg, "promptkit profile_dir and profile_file are mutually exclusive")
 }
 
+func TestValidatePromptKitTransientRetries(t *testing.T) {
+	if got := Default().PromptKit.TransientRetries; got != defaultTransientRetries {
+		t.Fatalf("default transient retries = %d, want %d", got, defaultTransientRetries)
+	}
+	for _, retries := range []int{0, 4} {
+		cfg := Default()
+		cfg.PromptKit.TransientRetries = retries
+		if err := cfg.Validate(); err != nil {
+			t.Fatalf("Validate() with transient retries %d error = %v", retries, err)
+		}
+	}
+	negative := Default()
+	negative.PromptKit.TransientRetries = -1
+	assertValidationContains(t, negative, "promptkit.transient_retries must not be negative")
+}
+
 func TestValidatePromptKitLocalBackendEndpoints(t *testing.T) {
 	tests := []struct {
 		name          string

internal/core/debugbundle/summary.go

@@ -39,7 +39,7 @@ type Invocation struct {
 	InputPath               string    `json:"input_path,omitempty"`
 	ConfigPath              string    `json:"config_path,omitempty"`
 	ConfigSource            string    `json:"config_source,omitempty"`
-	ConfigProfile           string    `json:"config_profile,omitempty"`
+	ConfigOverlay           string    `json:"config_overlay,omitempty"`
 	OnlyLanes               []string  `json:"only_lanes,omitempty"`
 	ChunkCacheOverride      string    `json:"chunk_cache_override,omitempty"`
 	ReasoningEffortOverride *string   `json:"reasoning_effort_override,omitempty"`

internal/framework/contracts/contracts.go

@@ -21,15 +21,45 @@ type StructuredCompletionRequest struct {
 }
 
 type StructuredCompletionResponse struct {
-	Content          json.RawMessage   `json:"content"`
-	Provider         string            `json:"provider,omitempty"`
-	Model            string            `json:"model,omitempty"`
-	ProfileID        string            `json:"profile_id,omitempty"`
-	PromptTokens     int               `json:"prompt_tokens,omitempty"`
-	CompletionTokens int               `json:"completion_tokens,omitempty"`
-	TotalTokens      int               `json:"total_tokens,omitempty"`
-	RepairAttempts   int               `json:"repair_attempts,omitempty"`
-	Debug            *LLMDebugMaterial `json:"debug,omitempty"`
+	Content          json.RawMessage      `json:"content"`
+	Provider         string               `json:"provider,omitempty"`
+	Model            string               `json:"model,omitempty"`
+	ProfileID        string               `json:"profile_id,omitempty"`
+	PromptTokens     int                  `json:"prompt_tokens,omitempty"`
+	CompletionTokens int                  `json:"completion_tokens,omitempty"`
+	TotalTokens      int                  `json:"total_tokens,omitempty"`
+	RepairAttempts   int                  `json:"repair_attempts,omitempty"`
+	Debug            *LLMDebugMaterial    `json:"debug,omitempty"`
+	Execution        *LLMExecutionSummary `json:"execution,omitempty"`
+}
+
+type LLMObservedTokenCount struct {
+	Value    int  `json:"value"`
+	Known    bool `json:"known"`
+	Complete bool `json:"complete"`
+}
+
+type LLMObservedTokenUsage struct {
+	PromptTokens     LLMObservedTokenCount `json:"prompt_tokens"`
+	CompletionTokens LLMObservedTokenCount `json:"completion_tokens"`
+	TotalTokens      LLMObservedTokenCount `json:"total_tokens"`
+	CachedTokens     LLMObservedTokenCount `json:"cached_tokens"`
+	CacheWriteTokens LLMObservedTokenCount `json:"cache_write_tokens"`
+}
+
+type LLMExecutionSummary struct {
+	Attempts         uint64                `json:"attempts"`
+	TransientRetries uint64                `json:"transient_retries"`
+	RepairCalls      uint64                `json:"repair_calls"`
+	CapacityWaitMS   int64                 `json:"capacity_wait_ms"`
+	GenerationMS     int64                 `json:"generation_ms"`
+	RetryDelayMS     int64                 `json:"retry_delay_ms"`
+	DurationMS       int64                 `json:"duration_ms"`
+	Outcome          string                `json:"outcome"`
+	FailureKind      string                `json:"failure_kind"`
+	StatusCode       int                   `json:"status_code"`
+	RetriesExhausted bool                  `json:"retries_exhausted"`
+	Usage            LLMObservedTokenUsage `json:"usage"`
 }
 
 type StructuredLLMClient interface {

internal/framework/contracts/errors.go

@@ -13,16 +13,42 @@ var ErrLLMCapacityExceeded = errors.New("LLM capacity exceeded")
 // ErrLLMGeneration identifies a provider generation failure.
 var ErrLLMGeneration = errors.New("LLM generation failed")
 
+type LLMGenerationFailureKind string
+
+const (
+	LLMGenerationFailureGeneration        LLMGenerationFailureKind = "generation"
+	LLMGenerationFailureTransport         LLMGenerationFailureKind = "transport"
+	LLMGenerationFailureProviderStatus    LLMGenerationFailureKind = "provider_status"
+	LLMGenerationFailureMalformedResponse LLMGenerationFailureKind = "malformed_response"
+)
+
+type LLMGenerationMetadata struct {
+	FailureKind      LLMGenerationFailureKind
+	RetriesExhausted bool
+	Attempts         uint64
+}
+
 type LLMGenerationError struct {
-	status     int
-	diagnostic string
+	status      int
+	diagnostic  string
+	failureKind LLMGenerationFailureKind
+	exhausted   bool
+	attempts    uint64
 }
 
 func NewLLMGenerationError(status int, diagnostic string) *LLMGenerationError {
+	return NewLLMGenerationErrorWithMetadata(status, diagnostic, LLMGenerationMetadata{})
+}
+
+func NewLLMGenerationErrorWithMetadata(status int, diagnostic string, metadata LLMGenerationMetadata) *LLMGenerationError {
 	if status < 0 {
 		status = 0
 	}
-	return &LLMGenerationError{status: status, diagnostic: diagnostic}
+	kind := metadata.FailureKind
+	if kind == "" {
+		kind = LLMGenerationFailureGeneration
+	}
+	return &LLMGenerationError{status: status, diagnostic: diagnostic, failureKind: kind, exhausted: metadata.RetriesExhausted, attempts: metadata.Attempts}
 }
 func (e *LLMGenerationError) Error() string {
 	if e == nil || e.diagnostic == "" {
@@ -37,3 +63,21 @@ func (e *LLMGenerationError) StatusCode() int {
 	}
 	return e.status
 }
+
+func (e *LLMGenerationError) FailureKind() LLMGenerationFailureKind {
+	if e == nil || e.failureKind == "" {
+		return LLMGenerationFailureGeneration
+	}
+	return e.failureKind
+}
+
+func (e *LLMGenerationError) RetriesExhausted() bool {
+	return e != nil && e.exhausted
+}
+
+func (e *LLMGenerationError) Attempts() uint64 {
+	if e == nil {
+		return 0
+	}
+	return e.attempts
+}

internal/framework/contracts/errors_test.go

@@ -0,0 +1,22 @@
+package contracts
+
+import (
+	"errors"
+	"testing"
+)
+
+func TestLLMGenerationErrorPreservesApplicationMetadata(t *testing.T) {
+	legacy := NewLLMGenerationError(503, "provider failure")
+	if !errors.Is(legacy, ErrLLMGeneration) || legacy.StatusCode() != 503 || legacy.FailureKind() != LLMGenerationFailureGeneration || legacy.RetriesExhausted() || legacy.Attempts() != 0 {
+		t.Fatalf("legacy generation error = %#v", legacy)
+	}
+
+	err := NewLLMGenerationErrorWithMetadata(429, "provider failure", LLMGenerationMetadata{
+		FailureKind:      LLMGenerationFailureProviderStatus,
+		RetriesExhausted: true,
+		Attempts:         3,
+	})
+	if !errors.Is(err, ErrLLMGeneration) || err.StatusCode() != 429 || err.FailureKind() != LLMGenerationFailureProviderStatus || !err.RetriesExhausted() || err.Attempts() != 3 {
+		t.Fatalf("generation error = %#v", err)
+	}
+}

internal/framework/llm/promptkit_client.go

@@ -26,15 +26,16 @@ type PromptKitLocalBackendConfig struct {
 }
 
 type PromptKitClientConfig struct {
-	ProfileDir      string
-	ProfileFile     string
-	LocalBackend    *PromptKitLocalBackendConfig
-	Assets          *AssetRegistry
-	Timeout         time.Duration
-	HTTPClient      *http.Client
-	EngineOptions   []promptkit.Option
-	Recorder        *LLMProfileRecorder
-	ReasoningEffort *string
+	ProfileDir       string
+	ProfileFile      string
+	TransientRetries *int
+	LocalBackend     *PromptKitLocalBackendConfig
+	Assets           *AssetRegistry
+	Timeout          time.Duration
+	HTTPClient       *http.Client
+	EngineOptions    []promptkit.Option
+	Recorder         *LLMProfileRecorder
+	ReasoningEffort  *string
 }
 
 type PromptKitClient struct {
@@ -79,9 +80,10 @@ func NewPromptKitClient(cfg PromptKitClientConfig) (*PromptKitClient, error) {
 	options = append(options, cfg.EngineOptions...)
 
 	engine, err := promptkit.NewEngine(promptkit.Config{
-		ProfileDir: profileSource.ProfileDir,
-		Timeout:    cfg.Timeout,
-		HTTPClient: cfg.HTTPClient,
+		ProfileDir:       profileSource.ProfileDir,
+		Timeout:          cfg.Timeout,
+		HTTPClient:       cfg.HTTPClient,
+		TransientRetries: cfg.TransientRetries,
 	}, options...)
 	if err != nil {
 		return nil, fmt.Errorf("create PromptKit engine: %w", err)
@@ -128,10 +130,10 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 		return contracts.StructuredCompletionResponse{}, fmt.Errorf("structured completion prompt_id must not be empty")
 	}
 	sessionID := strings.TrimSpace(req.SessionID)
-	var execution *promptkit.ExecutionTargetOverride
+	var targetOverride *promptkit.ExecutionTargetOverride
 	if c.reasoningEffort != nil {
 		reasoningEffort := *c.reasoningEffort
-		execution = &promptkit.ExecutionTargetOverride{
+		targetOverride = &promptkit.ExecutionTargetOverride{
 			ReasoningEffort: &reasoningEffort,
 		}
 	}
@@ -143,9 +145,15 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 		SessionID:        sessionID,
 		Inputs:           promptKitInputs(req.Inputs),
 		Vars:             promptKitVars(req, sessionID),
-		Execution:        execution,
+		Execution:        targetOverride,
 		AppendedMessages: appendedMessages,
 	}
+	var executionSummary *contracts.LLMExecutionSummary
+	runReq.Observer = func(event promptkit.ExecutionEvent) {
+		if event.Kind == promptkit.ExecutionFinished {
+			executionSummary = promptKitExecutionSummary(event)
+		}
+	}
 	if req.StructuredOutputRepairAttempts != nil {
 		inspection, err := c.engine.InspectPrompt(ctx, promptID, strings.TrimSpace(req.PromptVersion))
 		if err != nil {
@@ -170,12 +178,12 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 	result, err := c.engine.RunPrepared(ctx, prepared)
 	if err != nil {
 		if ctxErr := ctx.Err(); ctxErr != nil {
-			return contracts.StructuredCompletionResponse{}, ctxErr
+			return promptKitResponseWithExecution(contracts.StructuredCompletionResponse{}, executionSummary), ctxErr
 		}
 		if errors.Is(err, promptkit.ErrCapacityExceeded) {
 			var capacityErr *promptkit.CapacityError
 			if errors.As(err, &capacityErr) && strings.TrimSpace(capacityErr.BackendID) != "" {
-				return contracts.StructuredCompletionResponse{}, fmt.Errorf(
+				return promptKitResponseWithExecution(contracts.StructuredCompletionResponse{}, executionSummary), fmt.Errorf(
 					"run PromptKit prompt %q on backend %q: %w: %v",
 					promptID,
 					strings.TrimSpace(capacityErr.BackendID),
@@ -183,7 +191,7 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 					redactPromptKitError(err),
 				)
 			}
-			return contracts.StructuredCompletionResponse{}, fmt.Errorf(
+			return promptKitResponseWithExecution(contracts.StructuredCompletionResponse{}, executionSummary), fmt.Errorf(
 				"run PromptKit prompt %q: %w: %v",
 				promptID,
 				contracts.ErrLLMCapacityExceeded,
@@ -192,20 +200,26 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 		}
 		if errors.Is(err, promptkit.ErrLLMGenerate) {
 			var generationErr *promptkit.GenerationError
+			var exhaustedErr *promptkit.RetryExhaustedError
 			status := 0
+			metadata := contracts.LLMGenerationMetadata{FailureKind: promptKitGenerationFailureKind(err)}
+			if errors.As(err, &exhaustedErr) {
+				metadata.RetriesExhausted = true
+				metadata.Attempts = exhaustedErr.Attempts()
+			}
 			response := contracts.StructuredCompletionResponse{Debug: &contracts.LLMDebugMaterial{Prompt: promptKitDebugPrompt(&preparedDetails)}}
 			if errors.As(err, &generationErr) {
 				status = generationErr.StatusCode()
 				response.Debug.Response = promptKitDebugGenerationError(&preparedDetails, generationErr)
 			}
-			return response, contracts.NewLLMGenerationError(status, fmt.Sprintf("run PromptKit prompt %q: %v", promptID, redactPromptKitError(err)))
+			return promptKitResponseWithExecution(response, executionSummary), contracts.NewLLMGenerationErrorWithMetadata(status, fmt.Sprintf("run PromptKit prompt %q: %v", promptID, redactPromptKitError(err)), metadata)
 		}
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run PromptKit prompt %q: %v", promptID, redactPromptKitError(err))
+		return promptKitResponseWithExecution(contracts.StructuredCompletionResponse{}, executionSummary), fmt.Errorf("run PromptKit prompt %q: %v", promptID, redactPromptKitError(err))
 	}
 	if result == nil {
-		return contracts.StructuredCompletionResponse{}, fmt.Errorf("run PromptKit prompt %q: %w: empty result", promptID, contracts.ErrInvalidStructuredOutput)
+		return promptKitResponseWithExecution(contracts.StructuredCompletionResponse{}, executionSummary), fmt.Errorf("run PromptKit prompt %q: %w: empty result", promptID, contracts.ErrInvalidStructuredOutput)
 	}
-	response := c.responseFromResult(result, &preparedDetails)
+	response := promptKitResponseWithExecution(c.responseFromResult(result, &preparedDetails), executionSummary)
 	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, "; "))
 	}
@@ -218,6 +232,61 @@ func (c *PromptKitClient) CompleteStructured(ctx context.Context, req contracts.
 	return response, nil
 }
 
+func promptKitResponseWithExecution(response contracts.StructuredCompletionResponse, execution *contracts.LLMExecutionSummary) contracts.StructuredCompletionResponse {
+	response.Execution = execution
+	return response
+}
+
+func promptKitExecutionSummary(event promptkit.ExecutionEvent) *contracts.LLMExecutionSummary {
+	count := func(value promptkit.ObservedTokenCount) contracts.LLMObservedTokenCount {
+		return contracts.LLMObservedTokenCount{Value: value.Value, Known: value.Known, Complete: value.Complete}
+	}
+	return &contracts.LLMExecutionSummary{
+		Attempts: event.Attempts, TransientRetries: event.TransientRetries, RepairCalls: event.RepairCalls,
+		CapacityWaitMS: event.CapacityWait.Milliseconds(), GenerationMS: event.GenerationDuration.Milliseconds(), RetryDelayMS: event.RetryDelay.Milliseconds(), DurationMS: event.Duration.Milliseconds(),
+		Outcome: promptKitExecutionOutcome(event.Outcome), FailureKind: promptKitExecutionFailure(event.Failure), StatusCode: event.StatusCode, RetriesExhausted: event.RetriesExhausted,
+		Usage: contracts.LLMObservedTokenUsage{PromptTokens: count(event.Usage.PromptTokens), CompletionTokens: count(event.Usage.CompletionTokens), TotalTokens: count(event.Usage.TotalTokens), CachedTokens: count(event.Usage.CachedTokens), CacheWriteTokens: count(event.Usage.CacheWriteTokens)},
+	}
+}
+
+func promptKitExecutionOutcome(value promptkit.ExecutionOutcome) string {
+	switch value {
+	case promptkit.ExecutionOutcomeSuccess:
+		return "success"
+	case promptkit.ExecutionOutcomeInvalidOutput:
+		return "invalid_output"
+	case promptkit.ExecutionOutcomeError:
+		return "error"
+	default:
+		return "other"
+	}
+}
+
+func promptKitExecutionFailure(value promptkit.ExecutionFailureKind) string {
+	switch value {
+	case promptkit.ExecutionFailureNone:
+		return ""
+	case promptkit.ExecutionFailureCanceled, promptkit.ExecutionFailureDeadline, promptkit.ExecutionFailureTransport, promptkit.ExecutionFailureProviderStatus, promptkit.ExecutionFailureMalformedResponse, promptkit.ExecutionFailureCapacity, promptkit.ExecutionFailureInvalidConfig, promptkit.ExecutionFailureInvalidRequest, promptkit.ExecutionFailurePromptNotFound, promptkit.ExecutionFailureProfileNotFound, promptkit.ExecutionFailurePromptLoad, promptkit.ExecutionFailureProfileLoad, promptkit.ExecutionFailureArtifactLoad, promptkit.ExecutionFailurePromptRender, promptkit.ExecutionFailureValidation, promptkit.ExecutionFailureGeneration:
+		return string(value)
+	default:
+		return "other"
+	}
+}
+
+func promptKitGenerationFailureKind(err error) contracts.LLMGenerationFailureKind {
+	switch {
+	case errors.Is(err, promptkit.ErrTransport):
+		return contracts.LLMGenerationFailureTransport
+	case errors.Is(err, promptkit.ErrMalformedResponse):
+		return contracts.LLMGenerationFailureMalformedResponse
+	}
+	var generationErr *promptkit.GenerationError
+	if errors.As(err, &generationErr) && generationErr.StatusCode() != 0 {
+		return contracts.LLMGenerationFailureProviderStatus
+	}
+	return contracts.LLMGenerationFailureGeneration
+}
+
 func promptKitCorrectionMessages(correction *contracts.SemanticCorrection) ([]promptkit.RenderedMessage, error) {
 	owned, err := contracts.CloneSemanticCorrection(correction)
 	if err != nil {

internal/framework/llm/promptkit_client_test.go

@@ -55,6 +55,10 @@ func TestPromptKitClientMapsPromptRequestAndUnmarshalsOutput(t *testing.T) {
 	if resp.PromptTokens != 11 || resp.CompletionTokens != 7 || resp.TotalTokens != 18 {
 		t.Fatalf("usage = %#v, want mapped token counts", resp)
 	}
+	if resp.Execution == nil || resp.Execution.Outcome != "success" || resp.Execution.Attempts != 1 ||
+		!resp.Execution.Usage.PromptTokens.Known || resp.Execution.Usage.PromptTokens.Value != 11 {
+		t.Fatalf("execution summary = %#v", resp.Execution)
+	}
 	if resp.Debug == nil || resp.Debug.Prompt == nil {
 		t.Fatalf("debug prompt = nil, want prepared prompt material")
 	}
@@ -971,6 +975,83 @@ model: local-model
 	}
 }
 
+func TestPromptKitClientAppliesConfiguredBuiltInTransientRetries(t *testing.T) {
+	for _, test := range []struct {
+		name      string
+		retries   int
+		statuses  []int
+		wantCalls int32
+		wantErr   bool
+	}{
+		{name: "eligible failure then success", retries: 1, statuses: []int{http.StatusServiceUnavailable, http.StatusOK}, wantCalls: 2},
+		{name: "disabled", retries: 0, statuses: []int{http.StatusServiceUnavailable}, wantCalls: 1, wantErr: true},
+		{name: "exhausted", retries: 1, statuses: []int{http.StatusServiceUnavailable, http.StatusServiceUnavailable}, wantCalls: 2, wantErr: true},
+		{name: "noneligible client error", retries: 2, statuses: []int{http.StatusBadRequest}, wantCalls: 1, wantErr: true},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			var calls atomic.Int32
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+				index := calls.Add(1) - 1
+				if int(index) >= len(test.statuses) {
+					t.Errorf("unexpected HTTP attempt %d", index+1)
+					w.WriteHeader(http.StatusInternalServerError)
+					return
+				}
+				status := test.statuses[index]
+				w.Header().Set("Content-Type", "application/json")
+				if status != http.StatusOK {
+					w.WriteHeader(status)
+					_, _ = w.Write([]byte(`{"error":{"message":"provider failure"}}`))
+					return
+				}
+				_, _ = w.Write([]byte(`{"choices":[{"message":{"content":"{\"ok\":true}"}}]}`))
+			}))
+			defer server.Close()
+
+			profilePath := filepath.Join(t.TempDir(), "profile.yml")
+			if err := os.WriteFile(profilePath, []byte("id: retry-profile\nendpoint: "+server.URL+"/v1\nmodel: retry-model\n"), 0o600); err != nil {
+				t.Fatal(err)
+			}
+			client, err := NewPromptKitClient(PromptKitClientConfig{
+				Assets:           newTestPromptKitAssets(t),
+				ProfileFile:      profilePath,
+				TransientRetries: &test.retries,
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			var out map[string]any
+			_, err = client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
+				PromptID:  "adapter.test",
+				ProfileID: "retry-profile",
+				SessionID: "retry-test",
+				Inputs: contracts.LLMInputSet{
+					"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", ""),
+				},
+			}, &out)
+			if (err != nil) != test.wantErr {
+				t.Fatalf("CompleteStructured() error = %v, want error=%t", err, test.wantErr)
+			}
+			if test.wantErr {
+				var generation *contracts.LLMGenerationError
+				if !errors.As(err, &generation) || generation.FailureKind() != contracts.LLMGenerationFailureProviderStatus {
+					t.Fatalf("generation error = %#v, want provider-status classification", generation)
+				}
+				wantExhausted := test.name == "exhausted"
+				if generation.RetriesExhausted() != wantExhausted {
+					t.Fatalf("retries exhausted = %t, want %t", generation.RetriesExhausted(), wantExhausted)
+				}
+				if wantExhausted && (generation.Attempts() != 2 || generation.StatusCode() != http.StatusServiceUnavailable) {
+					t.Fatalf("exhausted generation = %#v", generation)
+				}
+			}
+			if got := calls.Load(); got != test.wantCalls {
+				t.Fatalf("HTTP calls = %d, want %d", got, test.wantCalls)
+			}
+		})
+	}
+}
+
 func TestLLMProfileRecorderDistinguishesEffectiveTargets(t *testing.T) {
 	recorder := NewLLMProfileRecorder()
 	for _, profile := range []artifacts.LLMProfileManifest{
@@ -1309,6 +1390,51 @@ func TestPromptKitClientForwardsConfiguredTransportTimeout(t *testing.T) {
 	}
 }
 
+func TestPromptKitClientClassifiesBuiltInTransportAndMalformedResponses(t *testing.T) {
+	for _, test := range []struct {
+		name      string
+		kind      contracts.LLMGenerationFailureKind
+		roundTrip func(*http.Request) (*http.Response, error)
+	}{
+		{
+			name: "transport",
+			kind: contracts.LLMGenerationFailureTransport,
+			roundTrip: func(*http.Request) (*http.Response, error) {
+				return nil, errors.New("transport unavailable")
+			},
+		},
+		{
+			name: "malformed response",
+			kind: contracts.LLMGenerationFailureMalformedResponse,
+			roundTrip: func(request *http.Request) (*http.Response, error) {
+				return &http.Response{StatusCode: http.StatusOK, Header: make(http.Header), Body: io.NopCloser(strings.NewReader("not-json")), Request: request}, nil
+			},
+		},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			client, err := NewPromptKitClient(PromptKitClientConfig{
+				Assets:     newTestPromptKitAssets(t),
+				HTTPClient: &http.Client{Transport: roundTripFunc(test.roundTrip)},
+				EngineOptions: []promptkit.Option{promptkit.WithProfiles(promptkit.OpenAICompatibleProfile(promptkit.OpenAICompatibleProfileConfig{
+					ID: "default-profile", Endpoint: "http://promptkit.test/v1", Model: "default-model",
+				}))},
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			var out map[string]any
+			_, err = client.CompleteStructured(context.Background(), contracts.StructuredCompletionRequest{
+				PromptID: "adapter.test", SessionID: "failure-kind-test",
+				Inputs: contracts.LLMInputSet{"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{"source":true}`), "", "")},
+			}, &out)
+			var generation *contracts.LLMGenerationError
+			if !errors.As(err, &generation) || generation.FailureKind() != test.kind || generation.RetriesExhausted() || generation.Attempts() != 0 {
+				t.Fatalf("generation error = %#v, want kind %q", generation, test.kind)
+			}
+		})
+	}
+}
+
 func TestPromptKitClientClassifiesEmptyStructuredCompletion(t *testing.T) {
 	client := newTestPromptKitClient(t, &fakePromptKitLLM{allowEmpty: true})
 

internal/framework/llm/promptkit_observation_test.go

@@ -0,0 +1,245 @@
+package llm
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func newObservedHTTPClient(t *testing.T, handler http.HandlerFunc, retries int, transport http.RoundTripper) *PromptKitClient {
+	t.Helper()
+	server := httptest.NewServer(handler)
+	t.Cleanup(server.Close)
+	client, err := NewPromptKitClient(PromptKitClientConfig{
+		Assets: newTestPromptKitAssets(t), TransientRetries: &retries,
+		HTTPClient: &http.Client{Transport: transport},
+		EngineOptions: []promptkit.Option{promptkit.WithProfiles(promptkit.OpenAICompatibleProfile(promptkit.OpenAICompatibleProfileConfig{
+			ID: "default-profile", Endpoint: server.URL + "/v1", Model: "test-model",
+		}))},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	return client
+}
+
+func observedRequest() contracts.StructuredCompletionRequest {
+	return contracts.StructuredCompletionRequest{
+		PromptID: "adapter.test", SessionID: "observation-test",
+		Inputs: contracts.LLMInputSet{"transcript": contracts.NewLLMInputMaterial("transcript", "application/json", []byte(`{}`), "", "")},
+	}
+}
+
+func TestPromptKitObservationsComposeTransientRetriesAndRepair(t *testing.T) {
+	var calls atomic.Int32
+	client := newObservedHTTPClient(t, func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		switch calls.Add(1) {
+		case 1, 3:
+			w.WriteHeader(http.StatusServiceUnavailable)
+			_, _ = io.WriteString(w, `{"error":{"message":"try again"}}`)
+		case 2:
+			_, _ = io.WriteString(w, `{"choices":[{"message":{"content":"{\"bad\":true}"}}],"usage":{"prompt_tokens":3,"completion_tokens":2,"total_tokens":5,"prompt_tokens_details":{"cached_tokens":0}}}`)
+		case 4:
+			_, _ = io.WriteString(w, `{"choices":[{"message":{"content":"{\"ok\":true}"}}],"usage":{"prompt_tokens":7,"completion_tokens":0,"total_tokens":7}}`)
+		default:
+			t.Error("unexpected extra HTTP attempt")
+			w.WriteHeader(http.StatusBadRequest)
+		}
+	}, 1, nil)
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	req := observedRequest()
+	repairs := 1
+	req.StructuredOutputRepairAttempts = &repairs
+	var out struct{ OK bool }
+	response, err := client.CompleteStructured(ctx, req, &out)
+	if err != nil || !out.OK {
+		t.Fatalf("completion = (%+v, %v), output = %+v", response, err, out)
+	}
+	summary := response.Execution
+	if calls.Load() != 4 || summary == nil || summary.Attempts != 4 || summary.TransientRetries != 2 || summary.RepairCalls != 1 || summary.Outcome != "success" || summary.RetriesExhausted {
+		t.Fatalf("calls = %d, execution = %+v", calls.Load(), summary)
+	}
+	wantUsage := contracts.LLMObservedTokenUsage{
+		PromptTokens:     contracts.LLMObservedTokenCount{Value: 10, Known: true},
+		CompletionTokens: contracts.LLMObservedTokenCount{Value: 2, Known: true},
+		TotalTokens:      contracts.LLMObservedTokenCount{Value: 12, Known: true},
+		CachedTokens:     contracts.LLMObservedTokenCount{Known: true},
+	}
+	// Failed HTTP attempts contribute unknown usage, not known zero. Neither
+	// terminal totals nor legacy result totals should be added a second time.
+	if summary.Usage != wantUsage {
+		t.Fatalf("observed usage = %+v, want %+v", summary.Usage, wantUsage)
+	}
+	if response.PromptTokens != 10 || response.CompletionTokens != 2 || response.TotalTokens != 12 || response.RepairAttempts != 1 {
+		t.Fatalf("legacy result accounting changed: %+v", response)
+	}
+}
+
+type closeSignalingBody struct {
+	io.ReadCloser
+	closed chan<- struct{}
+}
+
+func (b closeSignalingBody) Close() error {
+	err := b.ReadCloser.Close()
+	b.closed <- struct{}{}
+	return err
+}
+
+func TestPromptKitObservationSurvivesCanceledRetryWait(t *testing.T) {
+	consumed := make(chan struct{}, 2)
+	transport := http.DefaultTransport.(*http.Transport).Clone()
+	t.Cleanup(transport.CloseIdleConnections)
+	var calls atomic.Int32
+	client := newObservedHTTPClient(t, func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		w.Header().Set("Retry-After", "3600")
+		w.WriteHeader(http.StatusServiceUnavailable)
+		_, _ = io.WriteString(w, `{"error":{"message":"busy"}}`)
+	}, 1, roundTripFunc(func(req *http.Request) (*http.Response, error) {
+		response, err := transport.RoundTrip(req)
+		if err == nil {
+			response.Body = closeSignalingBody{ReadCloser: response.Body, closed: consumed}
+		}
+		return response, err
+	}))
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	type result struct {
+		response contracts.StructuredCompletionResponse
+		err      error
+	}
+	done := make(chan result, 1)
+	go func() {
+		var out map[string]any
+		response, err := client.CompleteStructured(ctx, observedRequest(), &out)
+		done <- result{response, err}
+	}()
+	// The failure has been consumed, so cancellation must interrupt the pending
+	// retry wait rather than depend on another provider response or its timeout.
+	select {
+	case <-consumed:
+	case <-time.After(5 * time.Second):
+		t.Fatal("provider response was not consumed")
+	}
+	cancel()
+	select {
+	case got := <-done:
+		if !errors.Is(got.err, context.Canceled) || got.response.Execution == nil {
+			t.Fatalf("canceled completion = (%+v, %v)", got.response, got.err)
+		}
+		summary := got.response.Execution
+		if summary.Outcome != "error" || summary.FailureKind != "canceled" || summary.Attempts != 1 || summary.RetriesExhausted || calls.Load() != 1 {
+			t.Fatalf("canceled execution = %+v, HTTP calls = %d", summary, calls.Load())
+		}
+	case <-time.After(5 * time.Second):
+		t.Fatal("cancellation did not interrupt retry waiting")
+	}
+}
+
+func TestPromptKitObservationsSeparatePreparationAndExecutionFailure(t *testing.T) {
+	var calls atomic.Int32
+	client := newObservedHTTPClient(t, func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		w.WriteHeader(http.StatusServiceUnavailable)
+		_, _ = io.WriteString(w, `{"error":{"message":"busy"}}`)
+	}, 1, nil)
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	req := observedRequest()
+	req.PromptID = "missing-prompt"
+	var out map[string]any
+	response, err := client.CompleteStructured(ctx, req, &out)
+	if err == nil || response.Execution != nil || calls.Load() != 0 {
+		t.Fatalf("preparation failure = (%+v, %v), calls = %d", response, err, calls.Load())
+	}
+	response, err = client.CompleteStructured(ctx, observedRequest(), &out)
+	var generation *contracts.LLMGenerationError
+	if !errors.As(err, &generation) || response.Execution == nil {
+		t.Fatalf("execution failure = (%+v, %v)", response, err)
+	}
+	summary := response.Execution
+	if summary.Outcome != "error" || summary.FailureKind != "provider_status" || summary.StatusCode != http.StatusServiceUnavailable || !summary.RetriesExhausted || summary.Attempts != 2 || summary.TransientRetries != 1 || calls.Load() != 2 {
+		t.Fatalf("failed execution = %+v, calls = %d", summary, calls.Load())
+	}
+	if summary.Usage != (contracts.LLMObservedTokenUsage{}) {
+		t.Fatalf("failed attempts should have unknown usage: %+v", summary.Usage)
+	}
+}
+
+func TestPromptKitObservationsAreIsolatedAcrossConcurrentCalls(t *testing.T) {
+	arrived := make(chan struct{}, 2)
+	release := make(chan struct{})
+	client := newObservedHTTPClient(t, func(w http.ResponseWriter, r *http.Request) {
+		body, err := io.ReadAll(r.Body)
+		if err != nil {
+			t.Error(err)
+			return
+		}
+		tokens := 11
+		if strings.Contains(string(body), "second-call") {
+			tokens = 22
+		}
+		arrived <- struct{}{}
+		select {
+		case <-release:
+		case <-r.Context().Done():
+			return
+		}
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"choices": []any{map[string]any{"message": map[string]string{"content": `{"ok":true}`}}},
+			"usage":   map[string]int{"prompt_tokens": tokens, "completion_tokens": 0, "total_tokens": tokens},
+		})
+	}, 0, nil)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	type result struct {
+		want     int
+		response contracts.StructuredCompletionResponse
+		err      error
+	}
+	done := make(chan result, 2)
+	for name, tokens := range map[string]int{"first-call": 11, "second-call": 22} {
+		go func() {
+			req := observedRequest()
+			req.Vars = map[string]any{"custom": name}
+			var out map[string]any
+			response, err := client.CompleteStructured(ctx, req, &out)
+			done <- result{tokens, response, err}
+		}()
+	}
+	for i := 0; i < 2; i++ {
+		select {
+		case <-arrived:
+		case <-ctx.Done():
+			t.Fatal("requests did not overlap")
+		}
+	}
+	close(release)
+	for i := 0; i < 2; i++ {
+		select {
+		case got := <-done:
+			summary := got.response.Execution
+			if got.err != nil || summary == nil {
+				t.Fatalf("concurrent completion = (%+v, %v)", got.response, got.err)
+			}
+			if summary.Attempts != 1 || summary.Usage.PromptTokens != (contracts.LLMObservedTokenCount{Value: got.want, Known: true, Complete: true}) || summary.Usage.CompletionTokens != (contracts.LLMObservedTokenCount{Known: true, Complete: true}) || summary.Usage.CachedTokens != (contracts.LLMObservedTokenCount{}) {
+				t.Fatalf("execution for %d tokens = %+v", got.want, summary)
+			}
+		case <-ctx.Done():
+			t.Fatal("concurrent completions did not return")
+		}
+	}
+}

internal/framework/llm/promptkit_profile_fingerprint.go

@@ -17,7 +17,7 @@ const (
 	promptKitLocalBackendMarker          = "notarius:promptkit-local-backend:v1"
 	// The built-in profile catalog is compiled into this pinned PromptKit
 	// release. Update this identity when the dependency is upgraded.
-	promptKitBuiltinProfileCatalogID = "promptkit:v0.9.0:builtin-profiles"
+	promptKitBuiltinProfileCatalogID = "promptkit:v0.11.0:builtin-profiles"
 )
 
 func promptKitProfileFingerprint(profileDir, profileFile, fallbackProfileDigest string) (CheckpointFingerprint, error) {

internal/framework/llm/scheduled_client.go

@@ -38,7 +38,7 @@ func (c *scheduledClient) CompleteStructured(ctx context.Context, req contracts.
 		return callErr
 	})
 	if err != nil {
-		return contracts.StructuredCompletionResponse{}, err
+		return response, err
 	}
 	return response, nil
 }

internal/framework/pipeline/debug.go

@@ -113,13 +113,14 @@ type debugSerializedOutput struct {
 type debugStructuredCompletionRequest = contracts.DebugStructuredCompletionRequest
 
 type debugStructuredCompletionResponse struct {
-	Content          string `json:"content,omitempty"`
-	Provider         string `json:"provider,omitempty"`
-	Model            string `json:"model,omitempty"`
-	ProfileID        string `json:"profile_id,omitempty"`
-	PromptTokens     int    `json:"prompt_tokens,omitempty"`
-	CompletionTokens int    `json:"completion_tokens,omitempty"`
-	TotalTokens      int    `json:"total_tokens,omitempty"`
+	Content          string                         `json:"content,omitempty"`
+	Provider         string                         `json:"provider,omitempty"`
+	Model            string                         `json:"model,omitempty"`
+	ProfileID        string                         `json:"profile_id,omitempty"`
+	PromptTokens     int                            `json:"prompt_tokens,omitempty"`
+	CompletionTokens int                            `json:"completion_tokens,omitempty"`
+	TotalTokens      int                            `json:"total_tokens,omitempty"`
+	Execution        *contracts.LLMExecutionSummary `json:"execution,omitempty"`
 }
 
 type debugLLMPromptArtifact struct {
@@ -133,21 +134,23 @@ type debugLLMResponseArtifact struct {
 	Fallback    *debugStructuredCompletionResponse `json:"fallback,omitempty"`
 	ContentPath string                             `json:"content_path,omitempty"`
 	Error       string                             `json:"error,omitempty"`
+	Execution   *contracts.LLMExecutionSummary     `json:"execution,omitempty"`
 }
 
 type debugLLMCallReference struct {
-	CallID              string `json:"call_id"`
-	PromptPath          string `json:"prompt_path,omitempty"`
-	ResponsePath        string `json:"response_path"`
-	ResponseContentPath string `json:"response_content_path,omitempty"`
-	PromptID            string `json:"prompt_id,omitempty"`
-	ProfileID           string `json:"profile_id,omitempty"`
-	Model               string `json:"model,omitempty"`
-	PromptTokens        int    `json:"prompt_tokens,omitempty"`
-	CompletionTokens    int    `json:"completion_tokens,omitempty"`
-	TotalTokens         int    `json:"total_tokens,omitempty"`
-	RepairAttempts      int    `json:"repair_attempts,omitempty"`
-	Error               bool   `json:"error,omitempty"`
+	CallID              string                         `json:"call_id"`
+	PromptPath          string                         `json:"prompt_path,omitempty"`
+	ResponsePath        string                         `json:"response_path"`
+	ResponseContentPath string                         `json:"response_content_path,omitempty"`
+	PromptID            string                         `json:"prompt_id,omitempty"`
+	ProfileID           string                         `json:"profile_id,omitempty"`
+	Model               string                         `json:"model,omitempty"`
+	PromptTokens        int                            `json:"prompt_tokens,omitempty"`
+	CompletionTokens    int                            `json:"completion_tokens,omitempty"`
+	TotalTokens         int                            `json:"total_tokens,omitempty"`
+	RepairAttempts      int                            `json:"repair_attempts,omitempty"`
+	Error               bool                           `json:"error,omitempty"`
+	Execution           *contracts.LLMExecutionSummary `json:"execution,omitempty"`
 }
 
 type debugValidationOutcome struct {
@@ -267,6 +270,7 @@ func (client *debugLLMClient) CompleteStructured(ctx context.Context, req contra
 			Fallback:    fallbackForArtifact,
 			ContentPath: responseContentPath,
 			Error:       errorText,
+			Execution:   cloneLLMExecutionSummary(response.Execution),
 		},
 		Error: errorText,
 	}))
@@ -283,6 +287,7 @@ func (client *debugLLMClient) CompleteStructured(ctx context.Context, req contra
 		TotalTokens:         response.TotalTokens,
 		RepairAttempts:      response.RepairAttempts,
 		Error:               err != nil,
+		Execution:           cloneLLMExecutionSummary(response.Execution),
 	}
 	if scope := debugLLMScopeFromContext(ctx); scope != nil {
 		scope.record(callRef)
@@ -637,6 +642,7 @@ func debugCompletionResponse(response contracts.StructuredCompletionResponse) de
 		PromptTokens:     response.PromptTokens,
 		CompletionTokens: response.CompletionTokens,
 		TotalTokens:      response.TotalTokens,
+		Execution:        cloneLLMExecutionSummary(response.Execution),
 	}
 }
 
@@ -651,7 +657,8 @@ func debugCompletionFallback(response contracts.StructuredCompletionResponse) *d
 		fallback.ProfileID == "" &&
 		fallback.PromptTokens == 0 &&
 		fallback.CompletionTokens == 0 &&
-		fallback.TotalTokens == 0 {
+		fallback.TotalTokens == 0 &&
+		fallback.Execution == nil {
 		return nil
 	}
 	return &fallback
@@ -696,6 +703,15 @@ func cloneDebugFallbackWithoutContent(fallback *debugStructuredCompletionRespons
 	}
 	clone := *fallback
 	clone.Content = ""
+	clone.Execution = cloneLLMExecutionSummary(fallback.Execution)
+	return &clone
+}
+
+func cloneLLMExecutionSummary(summary *contracts.LLMExecutionSummary) *contracts.LLMExecutionSummary {
+	if summary == nil {
+		return nil
+	}
+	clone := *summary
 	return &clone
 }
 

internal/framework/pipeline/debug_execution_test.go

@@ -0,0 +1,122 @@
+package pipeline
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
+	"gitea.maximumdirect.net/eric/notarius/internal/framework/llm"
+)
+
+type executionDebugClient struct {
+	response contracts.StructuredCompletionResponse
+	err      error
+}
+
+func (c executionDebugClient) CompleteStructured(context.Context, contracts.StructuredCompletionRequest, any) (contracts.StructuredCompletionResponse, error) {
+	return c.response, c.err
+}
+
+func TestScheduledDebugClientPreservesExecutionThroughTerminalRecords(t *testing.T) {
+	providerErr := contracts.NewLLMGenerationError(503, "provider unavailable")
+	const content = `{"ok":true}`
+	for _, test := range []struct {
+		name          string
+		err           error
+		content       string
+		debugResponse bool
+	}{
+		{name: "provider failure without body", err: providerErr},
+		{name: "canceled without body", err: context.Canceled},
+		{name: "split debug response", content: content, debugResponse: true},
+		{name: "split fallback response", content: content},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			want := contracts.LLMExecutionSummary{
+				Attempts: 2, TransientRetries: 1, Outcome: "success",
+				CapacityWaitMS: 1, GenerationMS: 3, RetryDelayMS: 5, DurationMS: 9,
+				Usage: contracts.LLMObservedTokenUsage{
+					PromptTokens:     contracts.LLMObservedTokenCount{Value: 7, Known: true},
+					CompletionTokens: contracts.LLMObservedTokenCount{Known: true},
+				},
+			}
+			if test.err != nil {
+				want.Outcome = "error"
+				want.FailureKind = "provider_status"
+				want.StatusCode = 503
+				want.RetriesExhausted = true
+				if errors.Is(test.err, context.Canceled) {
+					want.FailureKind = "canceled"
+					want.StatusCode = 0
+					want.RetriesExhausted = false
+				}
+			}
+			execution := want
+			response := contracts.StructuredCompletionResponse{Execution: &execution}
+			if test.debugResponse {
+				response.Debug = &contracts.LLMDebugMaterial{Response: &contracts.LLMDebugResponse{Content: test.content}}
+			} else {
+				response.Content = json.RawMessage(test.content)
+			}
+			scheduler, err := llm.NewScheduler(1)
+			if err != nil {
+				t.Fatal(err)
+			}
+			recorder := newCapturedDebugRecorder()
+			client := WithDebugLLMRecording(llm.NewScheduledClient(executionDebugClient{response, test.err}, scheduler), recorder)
+			ctx, scope := withDebugLLMScope(context.Background(), "attempt")
+			got, err := client.CompleteStructured(ctx, contracts.StructuredCompletionRequest{StageName: "extract"}, &struct{}{})
+			if !errors.Is(err, test.err) || got.Execution == nil || *got.Execution != want {
+				t.Fatalf("scheduled debug completion = (%+v, %v), want summary and original error", got, err)
+			}
+			terminal := newAttemptTerminalRecorder(recorder, "attempt/terminal", "extract", scope, debugTimedEnvelope{Stage: "extract"})
+			if err := terminal.record(nil, test.err); !errors.Is(err, test.err) {
+				t.Fatalf("terminal recording error = %v", err)
+			}
+			envelope := recorder.envelope(t, "attempt/response-0001.json")
+			encoded, err := json.Marshal(envelope.Payload)
+			if err != nil {
+				t.Fatal(err)
+			}
+			var artifact debugLLMResponseArtifact
+			if err := json.Unmarshal(encoded, &artifact); err != nil {
+				t.Fatal(err)
+			}
+			if artifact.Execution == nil || *artifact.Execution != want {
+				t.Fatalf("persisted response lost execution summary: %+v", artifact)
+			}
+			if !test.debugResponse && (artifact.Fallback == nil || artifact.Fallback.Execution == nil || *artifact.Fallback.Execution != want) {
+				t.Fatalf("persisted fallback lost execution summary: %+v", artifact.Fallback)
+			}
+			if test.content == "" {
+				if artifact.ContentPath != "" {
+					t.Fatalf("bodyless failure has content path %q", artifact.ContentPath)
+				}
+			} else {
+				var compact bytes.Buffer
+				if artifact.ContentPath == "" || json.Compact(&compact, recorder.bytes[artifact.ContentPath]) != nil || compact.String() != test.content {
+					t.Fatalf("split content not preserved at %q", artifact.ContentPath)
+				}
+				if (artifact.Fallback != nil && artifact.Fallback.Content != "") || (artifact.Response != nil && artifact.Response.Content != "") {
+					t.Fatal("split content remains in response metadata")
+				}
+			}
+			refs := recorder.envelope(t, "attempt/terminal.json").LLMCalls
+			if len(refs) != 1 || refs[0].Execution == nil || *refs[0].Execution != want || refs[0].Error != (test.err != nil) {
+				t.Fatalf("terminal references = %+v", refs)
+			}
+			// Decode the persisted JSON, not just Go defaults: omitted false and
+			// zero fields would erase the presence-aware usage contract.
+			usage := envelope.Payload.(map[string]any)["execution"].(map[string]any)["usage"].(map[string]any)
+			unknown := usage["cached_tokens"].(map[string]any)
+			for _, key := range []string{"value", "known", "complete"} {
+				if _, ok := unknown[key]; !ok {
+					t.Fatalf("unknown token usage omits %q: %+v", key, unknown)
+				}
+			}
+		})
+	}
+}

scripts/check-release-source.sh

@@ -2,85 +2,23 @@
 
 set -eu
 
-fail() {
-	printf '%s\n' "check-release-source: $*" >&2
-	exit 1
-}
-
-if [ "$#" -ne 1 ]; then
-	fail "usage: $0 vMAJOR.MINOR.PATCH"
-fi
-
-release_version=$1
-if ! printf '%s\n' "$release_version" | grep -E -x 'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)' >/dev/null; then
-	fail "version must be a stable semantic version tag"
-fi
+fail() { printf '%s\n' "check-release-source: $*" >&2; exit 1; }
 
+[ "$#" -eq 1 ] || fail "usage: $0 vMAJOR.MINOR.PATCH"
+version=$1
+printf '%s\n' "$version" | grep -E -x 'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)' >/dev/null || fail "invalid stable release version"
 script_dir=$(CDPATH= cd "$(dirname "$0")" && pwd)
-repo_root=$(CDPATH= cd "$script_dir/.." && pwd)
-cd "$repo_root"
+cd "$script_dir/.."
+note="docs/releases/$version.md"
+[ -s "$note" ] || fail "missing release note: $note"
+[ "$(head -n 1 "$note")" = "# Notarius $version" ] || fail "release note title does not match version"
 
-temporary_dir=$(mktemp -d "${TMPDIR:-/tmp}/notarius-release-source.XXXXXX") || fail "create temporary directory"
-trap 'rm -rf "$temporary_dir"' 0 HUP INT TERM
+export GOWORK=off
+go test -count=1 ./...
 
-if ! grep -E '^module[[:space:]]+gitea\.maximumdirect\.net/eric/notarius([[:space:]]|$)' go.mod >/dev/null; then
-	fail "go.mod does not declare the expected module path"
-fi
-
-if git ls-files --error-unmatch go.work >/dev/null 2>&1 || git ls-files --error-unmatch go.work.sum >/dev/null 2>&1; then
-	fail "tracked Go workspace files are not allowed"
-fi
-if [ -d vendor ]; then
-	fail "vendor directory is not allowed"
-fi
-if grep -E '^[[:space:]]*replace([[:space:]]|$)' go.mod >/dev/null; then
-	fail "go.mod replace directives are not allowed"
-fi
-
-GOWORK=off go test -count=1 ./...
-GOWORK=off go test -race -count=1 ./...
-GOWORK=off go vet ./...
-GOWORK=off go build ./...
-GOWORK=off go mod tidy -diff
-
-unformatted=$(git ls-files '*.go' | while IFS= read -r go_file; do
-	fmt_output=$(gofmt -l "$go_file")
-	if [ -n "$fmt_output" ]; then
-		printf '%s\n' "$fmt_output"
-	fi
-done)
-if [ -n "$unformatted" ]; then
-	printf '%s\n' "$unformatted" >&2
-	fail "tracked Go files are not gofmt formatted"
-fi
-
-git diff --check
-git diff --cached --check
-
-GOWORK=off go run ./cmd/notarius config validate --config examples/dnd-minimal.config.yml --pipeline dnd-session
-GOWORK=off OPENROUTER_API_KEY=validation-placeholder go run ./cmd/notarius config validate --config examples/dnd-complete.config.yml --pipeline dnd-session
-
-host_os=$(GOWORK=off go env GOOS)
-host_arch=$(GOWORK=off go env GOARCH)
-version_symbol='gitea.maximumdirect.net/eric/notarius/internal/buildinfo.Override'
-
-build_target() {
-	target_os=$1
-	target_arch=$2
-	target_path="$temporary_dir/notarius-$target_os-$target_arch"
-	CGO_ENABLED=0 GOOS="$target_os" GOARCH="$target_arch" GOWORK=off go build -trimpath \
-		-ldflags "-X $version_symbol=$release_version" \
-		-o "$target_path" ./cmd/notarius
-	if [ "$host_os" = "$target_os" ] && [ "$host_arch" = "$target_arch" ]; then
-		"$target_path" --version > "$temporary_dir/version-output"
-		printf 'notarius %s\n' "$release_version" > "$temporary_dir/expected-version-output"
-		if ! cmp -s "$temporary_dir/expected-version-output" "$temporary_dir/version-output"; then
-			fail "version output is incorrect for $target_os/$target_arch"
-		fi
-	fi
-}
-
-build_target linux amd64
-build_target linux arm64
-build_target darwin amd64
-build_target darwin arm64
+# Exercise the actual command and its linker-supplied diagnostic version.
+verification_dir=$(mktemp -d "${TMPDIR:-/tmp}/notarius-release-check.XXXXXX")
+trap 'rm -rf "$verification_dir"' 0 HUP INT TERM
+go build -trimpath -ldflags "-X gitea.maximumdirect.net/eric/notarius/internal/buildinfo.Override=$version" \
+  -o "$verification_dir/notarius" ./cmd/notarius
+[ "$("$verification_dir/notarius" --version)" = "notarius $version" ] || fail "incorrect diagnostic version"

scripts/release.sh

@@ -0,0 +1,56 @@
+#!/bin/sh
+
+set -eu
+
+fail() { printf '%s\n' "release: $*" >&2; exit 1; }
+
+[ "$#" -eq 1 ] || fail "usage: $0 vMAJOR.MINOR.PATCH"
+version=$1
+printf '%s\n' "$version" | grep -E -x 'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)' >/dev/null || fail "invalid stable release version"
+script_dir=$(CDPATH= cd "$(dirname "$0")" && pwd)
+cd "$script_dir/.."
+export GOWORK=off
+candidate=$(git rev-parse 'HEAD^{commit}')
+
+check_candidate() {
+  [ "$(git branch --show-current)" = main ] || fail "release branch must be main"
+  [ -z "$(git status --porcelain)" ] || fail "worktree and index must be clean"
+  [ "$(git rev-parse 'HEAD^{commit}')" = "$candidate" ] || fail "candidate changed during validation"
+  git cat-file -e "$candidate:docs/releases/$version.md" || fail "release note must be committed"
+}
+
+check_publication() {
+  # Query failure is not proof that the tag is unused.
+  remote_tag=$(git ls-remote --tags origin "refs/tags/$version" "refs/tags/$version^{}") || fail "cannot determine remote tag availability"
+  [ -z "$remote_tag" ] || fail "remote tag already exists: $version"
+  if git show-ref --verify --quiet "refs/tags/$version"; then
+    fail "local tag already exists: $version"
+  fi
+  git fetch --no-tags origin refs/heads/main || fail "cannot fetch authoritative main"
+  remote_main=$(git rev-parse 'FETCH_HEAD^{commit}')
+  git merge-base --is-ancestor "$remote_main" "$candidate" || fail "candidate does not contain authoritative main"
+}
+
+check_candidate
+check_publication
+"$script_dir/check-release-source.sh" "$version"
+check_candidate
+check_publication
+
+if ! git -c push.followTags=false push origin "$candidate:refs/heads/main"; then
+  # Do not publish a tag after a failed branch push, even if the branch arrived.
+  git ls-remote origin refs/heads/main >&2 || fail "branch publication status unknown; inspect origin before retrying"
+  fail "branch push failed; inspect the reported remote state before retrying"
+fi
+
+git -c tag.gpgSign=false tag "$version" "$candidate"
+[ "$(git cat-file -t "$version")" = commit ] || fail "tag is not lightweight"
+[ "$(git rev-parse "$version^{commit}")" = "$candidate" ] || fail "tag target changed"
+if ! git -c push.followTags=false push origin "refs/tags/$version:refs/tags/$version"; then
+  # A connection can fail after the server accepted the tag. Preserve it locally
+  # until authoritative state proves whether publication succeeded.
+  remote_tag=$(git ls-remote --tags origin "refs/tags/$version" "refs/tags/$version^{}") || fail "tag publication status unknown; preserve local tag and inspect origin"
+  expected=$(printf '%s\t%s' "$candidate" "refs/tags/$version")
+  [ "$remote_tag" = "$expected" ] || fail "tag push failed; preserve local tag and inspect origin before retrying"
+fi
+printf 'Released %s at %s\n' "$version" "$candidate"

scripts/release_test.go

@@ -0,0 +1,243 @@
+package scripts
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// Keep publication tests offline: real Git repositories exercise ref safety,
+// while a fake Go command represents the expensive candidate validation boundary.
+type releaseFixture struct {
+	t       *testing.T
+	root    string
+	remote  string
+	bin     string
+	log     string
+	initial string
+}
+
+func writeFile(t *testing.T, path, content string, mode os.FileMode) {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(content), mode); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func command(t *testing.T, dir string, args ...string) string {
+	t.Helper()
+	cmd := exec.Command(args[0], args[1:]...)
+	cmd.Dir = dir
+	output, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Fatalf("%v: %v\n%s", args, err, output)
+	}
+	return strings.TrimSpace(string(output))
+}
+
+func newReleaseFixture(t *testing.T) *releaseFixture {
+	t.Helper()
+	dir := t.TempDir()
+	f := &releaseFixture{t: t, root: filepath.Join(dir, "work"), remote: filepath.Join(dir, "origin.git"), bin: filepath.Join(dir, "bin"), log: filepath.Join(dir, "validation.log")}
+	command(t, dir, "git", "init", "--bare", f.remote)
+	command(t, dir, "git", "init", "-b", "main", f.root)
+	f.git("config", "user.name", "Release Test")
+	f.git("config", "user.email", "release@example.invalid")
+	f.git("config", "commit.gpgSign", "false")
+	f.git("config", "tag.gpgSign", "false")
+	for _, name := range []string{"release.sh", "check-release-source.sh"} {
+		content, err := os.ReadFile(name)
+		if err != nil {
+			t.Fatal(err)
+		}
+		writeFile(t, filepath.Join(f.root, "scripts", name), string(content), 0755)
+	}
+	writeFile(t, filepath.Join(f.root, ".gitignore"), "vendor/\n", 0644)
+	f.git("add", ".")
+	f.git("commit", "-m", "Initial source")
+	f.git("remote", "add", "origin", f.remote)
+	f.git("push", "origin", "main")
+	f.initial = f.git("rev-parse", "HEAD")
+	writeFile(t, filepath.Join(f.root, "docs/releases/v0.8.0.md"), "# Notarius v0.8.0\n\nA concise note without fixed sections.\n", 0644)
+	f.git("add", ".")
+	f.git("commit", "-m", "Prepare release")
+	writeFile(t, filepath.Join(f.bin, "go"), `#!/bin/sh
+set -eu
+[ "$GOWORK" = off ]
+printf '%s\n' "$1" >> "$RELEASE_TEST_LOG"
+case "$1" in
+test)
+  [ "${RELEASE_TEST_FAIL:-}" != test ] || exit 1
+  if [ "${RELEASE_TEST_CHANGE:-}" = yes ]; then
+    printf 'changed\n' >> docs/releases/v0.8.0.md
+  fi
+  ;;
+build)
+  while [ "$1" != -o ]; do shift; done
+  shift
+  printf '#!/bin/sh\nprintf "notarius v0.8.0\\n"\n' > "$1"
+  chmod +x "$1"
+  ;;
+*) exit 1 ;;
+esac
+`, 0755)
+	return f
+}
+
+func (f *releaseFixture) git(args ...string) string {
+	f.t.Helper()
+	return command(f.t, f.root, append([]string{"git"}, args...)...)
+}
+
+func (f *releaseFixture) run(version string, env ...string) (string, error) {
+	f.t.Helper()
+	cmd := exec.Command("sh", "scripts/release.sh", version)
+	cmd.Dir = f.root
+	cmd.Env = append(os.Environ(), "PATH="+f.bin+string(os.PathListSeparator)+os.Getenv("PATH"), "RELEASE_TEST_LOG="+f.log)
+	cmd.Env = append(cmd.Env, env...)
+	output, err := cmd.CombinedOutput()
+	return string(output), err
+}
+
+func TestReleasePublishesExactCandidateOnce(t *testing.T) {
+	f := newReleaseFixture(t)
+	// Neither ignored local files nor unrelated local tags are release inputs.
+	writeFile(t, filepath.Join(f.root, "vendor/local.txt"), "ignored", 0644)
+	f.git("tag", "unrelated")
+	f.git("tag", "-a", "unrelated-annotated", "-m", "Do not publish")
+	f.git("config", "push.followTags", "true")
+	candidate := f.git("rev-parse", "HEAD")
+	output, err := f.run("v0.8.0")
+	if err != nil {
+		t.Fatalf("release: %v\n%s", err, output)
+	}
+	for _, ref := range []string{"refs/heads/main", "refs/tags/v0.8.0"} {
+		if got := command(t, f.remote, "git", "rev-parse", ref); got != candidate {
+			t.Fatalf("%s = %s, want %s", ref, got, candidate)
+		}
+	}
+	if got := command(t, f.remote, "git", "cat-file", "-t", "v0.8.0"); got != "commit" {
+		t.Fatalf("tag type = %s", got)
+	}
+	if got := f.git("ls-remote", "--tags", "origin", "refs/tags/unrelated*"); got != "" {
+		t.Fatal("published unrelated tag")
+	}
+	log, err := os.ReadFile(f.log)
+	if err != nil || string(log) != "test\nbuild\n" {
+		t.Fatalf("validation was not performed once: %q, %v", log, err)
+	}
+}
+
+func TestReleaseGuardsPreventPublication(t *testing.T) {
+	for _, name := range []string{"invalid_version", "missing_note", "wrong_title", "dirty", "staged", "wrong_branch", "local_tag", "remote_tag", "remote_ahead", "remote_unavailable", "validation_failure", "candidate_changed"} {
+		t.Run(name, func(t *testing.T) {
+			f := newReleaseFixture(t)
+			version := "v0.8.0"
+			var env []string
+			switch name {
+			case "invalid_version":
+				version = "v00.8.0"
+			case "missing_note":
+				version = "v0.8.1"
+			case "wrong_title":
+				writeFile(t, filepath.Join(f.root, "docs/releases/v0.8.0.md"), "# Wrong version\n", 0644)
+				f.git("add", ".")
+				f.git("commit", "-m", "Wrong title")
+			case "dirty", "staged":
+				writeFile(t, filepath.Join(f.root, "dirty.txt"), "dirty", 0644)
+				if name == "staged" {
+					f.git("add", ".")
+				}
+			case "wrong_branch":
+				f.git("checkout", "-b", "other")
+			case "local_tag":
+				f.git("tag", version)
+			case "remote_tag":
+				f.git("push", "origin", f.initial+":refs/tags/"+version)
+			case "remote_ahead":
+				// Publish an independent descendant of the old remote tip.
+				f.git("checkout", "-b", "other", f.initial)
+				f.git("commit", "--allow-empty", "-m", "Other work")
+				f.git("push", "origin", "HEAD:main")
+				f.initial = f.git("rev-parse", "HEAD")
+				f.git("checkout", "main")
+			case "remote_unavailable":
+				f.git("remote", "set-url", "origin", filepath.Join(t.TempDir(), "missing.git"))
+			case "validation_failure":
+				env = []string{"RELEASE_TEST_FAIL=test"}
+			case "candidate_changed":
+				env = []string{"RELEASE_TEST_CHANGE=yes"}
+			}
+			if output, err := f.run(version, env...); err == nil {
+				t.Fatalf("guard allowed publication: %s", output)
+			}
+			if got := command(t, f.remote, "git", "rev-parse", "main"); got != f.initial {
+				t.Fatal("guard advanced remote branch")
+			}
+			if name != "remote_tag" {
+				if got := command(t, f.remote, "git", "tag", "--list"); got != "" {
+					t.Fatalf("guard published tag: %s", got)
+				}
+			}
+		})
+	}
+}
+
+func TestReleasePushFailureRecovery(t *testing.T) {
+	for _, mode := range []string{"branch_rejected", "tag_rejected", "tag_accepted_disconnect", "tag_unknown"} {
+		t.Run(mode, func(t *testing.T) {
+			f := newReleaseFixture(t)
+			if mode == "branch_rejected" || mode == "tag_rejected" {
+				pattern := "refs/heads/main"
+				if mode == "tag_rejected" {
+					pattern = "refs/tags/*"
+				}
+				writeFile(t, filepath.Join(f.remote, "hooks/update"), "#!/bin/sh\ncase \"$1\" in "+pattern+") exit 1;; esac\n", 0755)
+			} else {
+				realGit, err := exec.LookPath("git")
+				if err != nil {
+					t.Fatal(err)
+				}
+				// Simulate a lost acknowledgement, optionally followed by an outage.
+				writeFile(t, filepath.Join(f.bin, "git"), `#!/bin/sh
+set -eu
+if [ "$1" = -c ] && [ "$3" = push ] && [ "$5" = refs/tags/v0.8.0:refs/tags/v0.8.0 ]; then
+  "$RELEASE_TEST_GIT" "$@"
+  touch "$RELEASE_TEST_MARKER"
+  exit 1
+fi
+if [ "$1" = ls-remote ] && [ -e "$RELEASE_TEST_MARKER" ] && [ "$RELEASE_TEST_MODE" = tag_unknown ]; then
+  exit 1
+fi
+exec "$RELEASE_TEST_GIT" "$@"
+`, 0755)
+				output, err := f.run("v0.8.0", "RELEASE_TEST_GIT="+realGit, "RELEASE_TEST_MARKER="+filepath.Join(f.bin, "marker"), "RELEASE_TEST_MODE="+mode)
+				if (err == nil) != (mode == "tag_accepted_disconnect") {
+					t.Fatalf("unexpected outcome: %v\n%s", err, output)
+				}
+				if got := command(t, f.remote, "git", "rev-parse", "v0.8.0"); got != f.git("rev-parse", "HEAD") {
+					t.Fatal("wrong remote tag")
+				}
+				return
+			}
+			if output, err := f.run("v0.8.0"); err == nil {
+				t.Fatalf("rejected push reported success: %s", output)
+			}
+			if got := command(t, f.remote, "git", "tag", "--list"); got != "" {
+				t.Fatal("rejected push published tag")
+			}
+			if mode == "tag_rejected" {
+				if got := f.git("rev-parse", "v0.8.0"); got != f.git("rev-parse", "HEAD") {
+					t.Fatal("unpublished local tag not preserved")
+				}
+			} else if got := f.git("tag", "--list"); got != "" {
+				t.Fatal("created tag after failed branch push")
+			}
+		})
+	}
+}