WS-POL-003-01: add unified guide compilation contracts

.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DISCOVERY.md

@@ -95,14 +95,15 @@ on 2026-08-08.
   setup generation changes.
 - Representative task material is optional bounded context. Guide setup must
   not depend on tasks already existing.
-- Current post-submit compilation must gain a trusted hard rejection for
-  platform-default repetition; prompt instructions are insufficient.
+- The new unified proposal validator must reject platform-default repetition;
+  POL-01 does not change historical post-submit compiler behavior.
 - A catalogue is not an execution API. The checker service must expose exactly
   one typed call per phase and accept no caller-selected checker names.
 
-## Unknowns to resolve in the planning PR
+## Remaining unknowns after POL-01 contract repair
 
-- Final names and limits for evidence-reference and safe-text schemas.
+- POL-01 freezes evidence-reference and safe-text names/limits in its active
+  executable contract; implementation must prove them before later chunks.
 - Exact AUTH action/resource binding for creation of the compilation record;
   use narrow XINT/AUTH compilation request+execute actions for the parent while
   preserving separate 12E/12F/12G projection actions.

.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/PLAN.md

@@ -133,15 +133,19 @@ boundaries without WS-POL-003 modifying ART code or forcing ART lifecycle change
 
 `ProjectGuideCompilationContext` is strict (`extra="forbid"`) and contains:
 
-- exact existing ART-verified `GuideSourceMaterial`;
+- a deeply immutable canonical snapshot of exact ART-verified
+  `GuideSourceMaterial`, its payload hash, and source evidence lineage;
 - optional bounded representative task context;
-- non-selectable `platform_coverage` generated from ART-04B1 platform entries
-  plus CHECKER-owned durable post-submit defaults;
-- selectable `project_capabilities` generated from ART-04B1's project-rule
-  namespace for pre-submit and CHECKER/POL's registered rules for post-submit;
-- server-owned classification policy and schema versions;
-- optional bounded correction feedback tied to an exact superseded
-  compilation.
+- one exact `pre_submission_capabilities` projection generated from ART-04B1,
+  carrying both non-selectable platform entries and selectable project-policy
+  primitives;
+- one exact `post_submission_capabilities` projection generated from
+  CHECKER-owned registration plus the frozen durable default snapshot; and
+- server-owned setup run/generation, instruction version, and agent identity.
+
+Later persistence/correction chunks bind classification policy and bounded
+correction feedback to an exact superseded compilation; POL-01 does not expose
+those fields prematurely.
 
 Representative task context is tenant-local, server-redacted, and limited to
 policy shape. It excludes actor/user IDs, emails, submission artifacts, review

.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-01-contract-catalogue-projection.md

@@ -1,41 +1,181 @@
 # Chunk Contract: WS-POL-003-01 - Unified Contract and Catalogue Projection
 
-Status: Proposed, inactive. Risk: L1.
+Status: Active after explicit human start on 2026-08-08. Risk: L1.
 
 ## Goal
 
-Add strict unified input/result/evidence models and read-only projections from
-ART-04B1's complete pre-submit catalogue and CHECKER/POL's durable post-submit
-capability truth. No model call, persistence, registry, or lifecycle change.
+Add strict bounded unified input/result/evidence contracts and two read-only
+capability projections from existing phase-owner truth. No model call,
+persistence, registry, compiler behavior, composition-root, or lifecycle
+change is permitted.
+
+## Why this chunk exists
+
+The later unified adapter needs one closed model-facing contract without
+copying ART's 26-entry catalogue or treating CHECKER's runtime registry as a
+second POL-owned registry.
 
 ## Allowed files
 
-`backend/app/interfaces/project_agents.py`, canonical ART and CHECKER/POL
-projection interfaces/composition only, focused tests, and WS-POL-003 docs.
+```text
+backend/app/interfaces/project_agents.py
+backend/app/modules/checkers/catalogue.py
+backend/app/modules/projects/post_submit_policy.py
+backend/scripts/run_test_lanes.py
+backend/tests/test_ci_test_lanes.py
+backend/tests/test_project_guide_compilation_contracts.py
+.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/**
+```
+
+`catalogue.py` and `post_submit_policy.py` may add pure read-only projection
+functions only. Their existing registry, definitions, defaults, compiler,
+validation, hashing, and execution semantics are frozen.
+
+The CI lane files may change only to assign the new test module exactly once
+to the existing task-lifecycle lane and update its exact-inventory assertion.
 
 ## Not allowed
 
-ART/CHECKER catalogue changes, duplicate project registry, database/model/Celery
-changes, action activation, or checker execution.
-
-## Acceptance
-
-- Strict bounded schemas reject extra/executable/unsafe fields.
-- ART platform coverage is non-selectable. Pre-submit project capabilities come
-  only from ART-04B1; post-submit project capabilities come only from the
-  canonical durable CHECKER/POL source.
-- Pre-submit input preserves the exact full 26-entry ART-04B1 manifest,
-  including enabled and disabled advisory entries, catalogue ID/version/schema,
-  `manifest_sha256`, definition identities and states, dispatch kinds,
-  classifications, phases, policy fields, resource budgets, and disabled
-  behavior. Selection remains limited to enabled policy primitives. POL invents
-  no missing timeout/safety fields and cannot mutate catalogue state.
-- `GuideEvidenceRef` is closed and raw excerpts/URLs/paths cannot enter it.
-- Optional representative task context does not gate compilation.
-- Unknown/default/wrong-stage bindings fail closed.
-
-## Verification and review
-
-Focused schema/catalogue tests, Ruff, type checks, stale-registry scan. Required
-reviewers: architecture, security, QA, product, reuse, test delta, CI integrity.
-Human focus: no second registry and no executable model fields.
+```text
+new or changed ART/CHECKER catalogue definitions/defaults/registrations
+new registry, service locator, composition-root state, or startup wiring
+database/model/migration/repository/Celery/API/authorization changes
+agent adapter, prompt, provider/model call, checker dispatch, or execution
+post-submit compiler behavior changes, including default-checker handling
+effective-plan reconstruction or independent canonical/hash algorithms
+open Any/dict model-visible configuration or executable suggestion fields
+```
+
+Default-checker repetition is rejected only by the new unified proposal
+validator. This chunk does not change the historical post-submit compiler.
+
+## Frozen contracts and bounds
+
+- New Pydantic contracts use `extra="forbid"` and strict scalar validation.
+- At most: 100 findings, 200 atomic requirements, 100 pre bindings, 100 post
+  bindings, 50 capability suggestions, 20 setup notes, and 20 evidence refs
+  per item. Safe operator text is at most 1,000 characters per field and
+  rejects control characters, URL/scheme/path/credential/command/import/
+  dependency-shaped content where the field is model-produced.
+- `GuideEvidenceRef` contains only server-minted immutable lineage:
+  `source_item_id`, `extraction_usage_id`, canonical output SHA-256, and bounded
+  numeric start/end ordinals. It contains no raw excerpt, URL, path,
+  credential, signed reference, caller text, or provider/scratch handle.
+- Representative task context is optional and contains only bounded
+  server-redacted policy-shape identifiers. It contains no actor/user ID,
+  email, raw task body, submission/review/payment data, secret, URL, or path.
+- The ART guide material must be marked verified, contain no legacy raw
+  representative-task items, and expose only the canonical `content_markdown`
+  guide field plus ART-verified source items. The unified context receives a
+  deeply immutable canonical payload snapshot, payload hash, and immutable
+  source lineage so post-validation mutation cannot change evidence truth.
+  Extracted guide content remains explicitly untrusted model input; it is not
+  accepted as output evidence.
+- Capability configuration is a tuple of closed key/value parameters. Keys
+  must be present in the selected canonical definition's policy fields; values
+  are bounded JSON scalars or bounded scalar tuples, never nested objects,
+  source code, commands, imports, dependencies, URLs, or paths.
+
+## Canonical pre-submit projection
+
+- The projection consumes one exact startup-composed
+  `PreSubmissionCheckerCatalogue`; it never calls a second builder in product
+  execution.
+- It preserves the complete immutable envelope, `manifest_sha256`, and all 26
+  exact definition projections in canonical order, including disabled advisory
+  rows and all 19 existing definition fields.
+- The first 14 `platform_capability` definitions are non-selectable platform
+  coverage. Only enabled `project_policy`/`policy_primitive` definitions are
+  selectable project capabilities.
+- Disabled mandatory catalogue state makes the projection unavailable;
+  disabled advisory rows remain visible but non-selectable.
+- The projection does not invent timeout, safety, implementation-version, or
+  other metadata absent from ART-04B1.
+
+## Canonical post-submit projection
+
+- The projection remains in `post_submit_policy.py`, adjacent to the existing
+  compiler-version default snapshot, and consumes
+  `default_checker_registry().names()` plus
+  `POST_SUBMIT_DEFAULT_CHECKERS_BY_COMPILER_VERSION`.
+- Envelope identity is
+  `workstream.post_submission_checkers`, schema
+  `post_submission_checker_capability_projection.v1`, and source version
+  `POST_SUBMIT_COMPILER_VERSION`; its canonical SHA-256 commits the sorted
+  registered names, frozen default names, stage, and selectability.
+- Capability identity is the registered checker name; capability version is
+  the frozen source/compiler version; stage is exactly `post_submit`.
+- The eight frozen v0.1 defaults are non-selectable platform coverage. A
+  separate frozen compiler-version snapshot explicitly owns project-selectable
+  capabilities; it is disjoint from the defaults and currently contains only
+  `check_acceptance_criteria_present`. Registered but unlisted capabilities are
+  neither default nor selectable.
+- Default/unknown/wrong-stage/stale-snapshot bindings fail in unified proposal
+  validation. No registry or compiler mutation occurs.
+
+## Acceptance criteria
+
+- Strict input/result/evidence schemas reject extra, unsafe, executable,
+  over-limit, PII-bearing, raw-source, and nested-open configuration fields.
+- One atomic requirement has exactly one canonical disposition. Binding and
+  evidence references resolve to existing requirement/source lineage.
+- `platform_covered` requires an exact stage/ID/version reference to an
+  enabled mandatory pre-submit platform capability or a canonical post-submit
+  default. Advisory pre-submit rows remain visible but cannot satisfy required
+  coverage. Ready/blocked status must agree with warning, blocking-gap, and
+  capability-gap evidence.
+- ART projection equality covers the exact 26-entry manifest and hash; no field
+  is dropped, retyped, inferred, or reordered.
+- Platform/default and disabled definitions cannot be selected. Enabled
+  project capabilities can be selected only at their exact stage/version and
+  with catalogue-owned policy fields.
+- Post-submit projection is derived solely from the canonical registry and
+  frozen default snapshot; parity drift fails closed.
+- Optional representative task context may be omitted without invalidating the
+  compilation context.
+- Static proof finds no new registry, model/provider call, persistence, route,
+  Celery, authorization, or catalogue mutation.
+
+## Verification commands
+
+```bash
+(cd backend && .venv/bin/python -m ruff check app/interfaces/project_agents.py app/modules/checkers/catalogue.py app/modules/projects/post_submit_policy.py tests/test_project_guide_compilation_contracts.py)
+(cd backend && .venv/bin/python -m pytest -q tests/test_project_guide_compilation_contracts.py tests/test_checker_catalogue.py)
+# Hosted Backend lane (supplies WORKSTREAM_TEST_DATABASE_URL/Postgres):
+(cd backend && .venv/bin/python -m pytest -q tests/test_project_guide_compilation_contracts.py tests/test_checker_catalogue.py tests/test_checkers.py --cov=app.interfaces.project_agents --cov=app.modules.checkers.catalogue --cov=app.modules.projects.post_submit_policy --cov-report=term-missing --cov-fail-under=90)
+# Hosted full Backend matrix and repository coverage gate:
+(cd backend && .venv/bin/python -m pytest -q)
+python3 scripts/check_stale_workstream_wording.py
+python3 scripts/check_stale_authorization_docs.py
+python3 scripts/check_markdown_links.py
+! git diff origin/main -- backend/app/interfaces/project_agents.py backend/app/modules/checkers/catalogue.py backend/app/modules/projects/post_submit_policy.py | rg '^\+.*(class .*Registry|\.register\(|responses\.create|Runner\.run|Celery|Mapped\[|APIRouter)'
+git diff --check
+```
+
+No configured repository type-check command exists on current main; strict
+Pydantic construction, Ruff, focused tests, and the full hosted Backend matrix
+are the type/runtime gates for this chunk.
+
+## Required reviewers
+
+- architecture
+- security/auth and data safety
+- QA/test
+- product/operations
+- senior engineering
+- reuse/dedup
+- test delta
+- CI integrity
+- docs
+
+## Human review focus
+
+Confirm exact phase-owner reuse, no second registry, no executable or leaking
+model fields, and no change to catalogue/compiler/runtime behavior.
+
+## Stop conditions
+
+Stop if current canonical sources cannot provide a deterministic read-only
+projection without inventing authority metadata, if a projection requires
+registry/compiler mutation, or if safe strict configuration requires an open
+model-visible object.

.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-01-internal-review-evidence.md

@@ -0,0 +1,54 @@
+# WS-POL-003-01 Internal Review Evidence
+
+Date: 2026-08-08. Risk: L1.
+
+## Deterministic evidence
+
+- Scoped Ruff: passed.
+- Focused non-database tests: 91 passed after external-review corrections.
+- Changed-subsystem coverage with the neighboring checker tests reached above
+  90 percent; database-backed completion remains assigned to the hosted
+  Backend lane because it supplies Postgres and `WORKSTREAM_TEST_DATABASE_URL`.
+- Stale Workstream wording, stale authorization docs, Markdown links, static
+  boundary scan, and `git diff --check`: passed.
+- No workflow, coverage threshold, skip, xfail, or bypass change exists.
+
+## Review results
+
+- Architecture: pass after replacing mutable legacy material with an immutable
+  canonical payload/hash/lineage snapshot and excluding advisory/disabled
+  capabilities from required platform coverage.
+- Security: pass after closing evidence lineage, non-finite scalar, unsafe
+  text/path/PII, service-owned version, status, platform-coverage, and mutable
+  context gaps.
+- Product/operations: pass after ready/blocked status consistency and exact
+  platform-coverage proof were enforced.
+- QA: pass; strict scalar, catalogue parity, immutable projection, status,
+  binding, evidence, and stage cases are covered.
+- Senior engineering: pass after deep immutability and canonical snapshot hash
+  validation were added.
+- Reuse/dedup: pass; the pre-submit projection consumes
+  `manifest_entry()` and the post-submit projection consumes the existing
+  registry/default snapshot without another registry.
+- Test delta: pass after invalid parameter ownership and all platform-coverage
+  branches received regression tests.
+- CI integrity: pass; local non-database and hosted database/full-suite duties
+  are explicit and no gate is weakened.
+- Docs: pass; the active plan now distinguishes POL-01 context fields from
+  later correction/persistence fields.
+
+## External-review correction re-review
+
+- Architecture: pass after explicit post-submit selectability, default/selectable
+  disjointness, and matching chunk-contract wording were added.
+- Security: pass after bare value-shaped credential forms, complete non-empty
+  ART lineage, unavailable mandatory pre-submit coverage, and closed
+  post-submit selectability were enforced.
+- QA: pass; all six CodeRabbit findings have regression proof.
+- Senior engineering: pass after credential detection was narrowed to preserve
+  ordinary security-policy language.
+- Test delta: pass after missing selectable-registration and snapshot-overlap
+  regressions were added; no tests were removed or skipped.
+
+All blocking findings were corrected and re-reviewed. No reviewer session
+remains open.

.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-01-pr-trust-bundle.md

@@ -0,0 +1,53 @@
+# WS-POL-003-01 PR Trust Bundle
+
+## Intent
+
+Create the strict unified project-guide compilation contract and read-only
+pre/post capability projections without activating a model call, persistence,
+API, worker, authorization action, or checker execution path.
+
+## Design and scope
+
+- Snapshot exact ART-verified guide material as immutable canonical bytes,
+  payload hash, and immutable source lineage.
+- Treat model output as an untrusted bounded proposal with closed findings,
+  atomic requirements, artifact policy, capability bindings, evidence refs,
+  suggestions, and setup notes.
+- Project all 26 ART pre-submit definitions from the supplied canonical
+  catalogue and preserve its manifest/hash/state.
+- Project post-submit capabilities from the existing checker registry and
+  frozen compiler-version defaults.
+- Validate exact stage/version/selectability, catalogue-owned parameters,
+  status consistency, evidence lineage, and mandatory/default platform
+  coverage fail closed.
+
+## Explicitly unchanged
+
+No registry/default mutation, compiler behavior change, model/provider call,
+database model or migration, route, Celery task, authorization action,
+composition wiring, or live lifecycle change.
+
+## Evidence
+
+- Ruff passed.
+- Focused tests: 66 passed.
+- Changed subsystem coverage remains above 90 percent in the combined focused
+  run; hosted Backend lanes own the Postgres-backed completion and repository
+  coverage floor.
+- Stale wording, stale authorization docs, Markdown links, static boundary
+  scan, and whitespace validation passed.
+- Architecture, security, product/operations, QA, senior engineering,
+  reuse/dedup, test-delta, CI-integrity, and docs reviewers passed after all
+  valid findings were fixed.
+
+## Human review focus
+
+Confirm that the snapshot is deeply immutable, phase-owner catalogue truth is
+reused without duplication, advisory/default rows cannot masquerade as
+required project coverage, and this chunk does not activate runtime behavior.
+
+## Remaining external proof
+
+GitHub must run the complete Backend matrix, database-backed focused coverage,
+repository coverage floor, and external CodeRabbit review on the exact PR
+head before merge.