diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/CHUNK_MAP.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/CHUNK_MAP.md
new file mode 100644
index 00000000..e30144fb
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/CHUNK_MAP.md
@@ -0,0 +1,17 @@
+# Chunk Map: WS-POL-003 - Unified Project Guide Compilation
+
+All chunks are L1, one PR each, proposed, and inactive.
+
+| Chunk | Purpose | Hard dependency |
+|---|---|---|
+| WS-POL-003-01 | Strict unified contracts, safe evidence references, merged ART-04B1 pre-submit projection, and durable CHECKER/POL post-submit projection | AUTH-12B2 merged; consume ART-04B1 PR #276 exact contract |
+| WS-POL-003-02 | One OpenAI Agents SDK unified adapter method and fake-runtime tests | 01 |
+| WS-POL-003-03 | Immutable compilation persistence, trusted validator, and action-specific service provenance | 02; AUTH-12F/12G and exact XINT/AUTH compilation request+execute activation merged |
+| WS-POL-003-04 | Initial setup cutover from sufficiency + artifact-policy calls to one compilation | 03 |
+| WS-POL-003-05 | Project Manager approval and trusted project pre-submit policy compilation | 04 |
+| WS-POL-003-06 | Deterministic post-submit projection compilation with zero second model call | 05; AUTH-12G merged |
+| WS-POL-003-07 | One typed checker service port with one complete pre and one complete post command | 06; ART-04B1-04B3 pre-submit executor/evidence-writer contract merged and callable |
+| WS-POL-003-08 | Visibility, generation-safe correction, activation compatibility, checker-route clean cut, and legacy inference cleanup | 07; AUTH-12H/14 and ART-05B merged |
+
+No chunk starts automatically. Each contract must be reconciled against the
+then-current main and dependency merge before human start.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DECISIONS.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DECISIONS.md
new file mode 100644
index 00000000..a15a0fbc
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DECISIONS.md
@@ -0,0 +1,63 @@
+# Decisions: WS-POL-003 - Unified Project Guide Compilation
+
+1. One logical structured model attempt is used per immutable source snapshot,
+   catalogue snapshot, and setup generation. A durable attempt row and provider
+   idempotency key serialize dispatch. Retry/recovery uses that exact key and
+   retrieves or reuses its accepted result; it never creates another call.
+   Invalid or unsafe output consumes the attempt and blocks that generation.
+2. `ProjectGuideCompilation` is immutable provenance/proposal evidence, not a
+   canonical policy replacement.
+3. Existing policy objects and Project Manager approval gates remain separate.
+4. Agent-derived projections are immutable. Corrections create a new
+   generation. A manual replacement, if retained, has separate provenance and
+   invalidates every dependent unified proposal.
+5. The fixed `workstream.project.setup` service performs compilation and
+   service-owned projection mutations using fresh action-specific PREP. Human
+   authority only requests/recoveries, acknowledges, corrects through an
+   approved replacement path, and approves.
+6. Platform coverage and selectable project capabilities remain separate
+   projections even when supplied by the same phase owner.
+7. ART-04B1 owns the complete pre-submit catalogue and effective-plan compiler,
+   including mandatory platform entries and its closed selectable project-rule
+   namespace. CHECKER/POL owns durable post-submit capability truth. WS-POL-003
+   consumes these exact owners and creates no parallel dispatch registry.
+8. Unsupported required capabilities block activation. Optional/advisory gaps
+   require explicit Project Manager acknowledgement.
+9. Evidence references are closed structured identifiers; raw excerpts,
+   provider responses, hidden reasoning, URLs, paths, credentials, and
+   executable content are not persisted.
+10. Representative task context is optional and bounded; its absence cannot
+    block project guide compilation.
+11. Setup failures, capability gaps, timeouts, and retries create no
+    ContributionRecord, payment, award, or negative reputation evidence.
+12. No backward-compatibility aliases or dual model-inference paths survive
+    final cleanup.
+13. Pre-submit has no standalone feedback/execution API. One canonical
+    submission preparation/admission request executes one effective plan that
+    contains mandatory platform checks plus exact task-locked project rules.
+14. Post-submit normal execution is automatically dispatched once from the
+    successful Submission creation/finalization boundary. Callers cannot
+    select or separately invoke platform, project, or individual checkers.
+15. An authorized checker repair/requeue command may exist only when it accepts
+    a Submission/run identity and atomically claims the canonical phase-attempt
+    row under the phase owner's repository transaction. The idempotency key is
+    the phase, exact locked material/plan lineage, and attempt ID. Concurrent
+    repair/requeue calls either observe the existing terminal result or one
+    caller resumes the same non-terminal run; they cannot rerun completed
+    members or create a second business effect. A genuinely new evaluation
+    requires a new attempt identity. This is not an alternative checker API.
+16. Setup approval/correction-request APIs configure policy, and read APIs
+    expose bounded evidence; neither is a checker execution path.
+17. AUTH must activate two narrow compilation actions before runtime cutover:
+    a Project Manager dispatch/recovery request and a fixed
+    `workstream.project.setup` execution action. Execution owns only the model
+    call and immutable compilation parent/supersession; 12E/12F/12G retain
+    custody of their separate canonical projections.
+18. CHECKER exposes one internal typed service port with exactly two phase
+    commands: one complete pre-submit evaluation and one complete post-submit
+    evaluation. Artifact-flow orchestration invokes each command once at the
+    ART material boundary and never calls an individual checker.
+19. The pre command is a facade over ART-04B1-04B3's sole compiler, executor,
+    attempt, result, and evidence writer. The post command uses CHECKER's sole
+    durable executor/repository. The facade creates no duplicate member rows or
+    evidence and returns only the canonical phase result/reference.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DISCOVERY.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DISCOVERY.md
new file mode 100644
index 00000000..4991d24c
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/DISCOVERY.md
@@ -0,0 +1,102 @@
+# Discovery: WS-POL-003 - Unified Project Guide Compilation
+
+Baseline inspected: `origin/main` `bb77ff4a0ab61120b94d6d4763934b444c39207d`
+on 2026-08-05.
+
+## Current behavior
+
+- `backend/app/interfaces/project_agents.py` exposes separate
+  `analyze_guide_sufficiency`, `derive_submission_artifact_policy`, and
+  `derive_post_submit_checker_policy` contracts.
+- `backend/app/adapters/project_agents/openai_agent_sdk.py` implements three
+  prompts/model calls behind `ProjectGuideAgentRuntime`.
+- `backend/app/modules/projects/sufficiency_mutation_service.py` now performs
+  sufficiency execution only as the fixed setup service over ART-verified
+  material. The Project Manager route is an asynchronous dispatch/recovery
+  request.
+- `project_setup.py` still sequences sufficiency,
+  submission-policy derivation, approval/compilation continuation, and later
+  post-submit derivation.
+- `backend/app/modules/checkers/compiler.py` is the current legacy pre-submit
+  policy compiler now integrated with ART-04B1 catalogue definitions.
+- `backend/app/modules/checkers/catalogue.py` implements the immutable
+  `PreSubmissionCheckerCatalogue`, exact `v0.1` definition manifest,
+  startup-fixed disabled state, and canonical manifest hash.
+- `backend/app/modules/checkers/effective_plan.py` implements the pure
+  `EffectivePreSubmissionExecutionPlan` compiler and exact locked lineage,
+  rule-instance, configuration, catalogue, and plan hashes. It performs no
+  checker execution or durable write.
+- `backend/app/modules/checkers/runner.py` registers the current durable
+  checker implementations. `check_acceptance_criteria_present` is the only
+  current non-default project-selectable post-submit checker.
+- The obsolete `POST /tasks/{task_id}/submission-precheck` route and direct
+  `POST /submissions/{submission_id}/checker-runs` trigger remain reachable.
+  ART PLAN5 superseded 04A4 and moved the standalone precheck clean cut to
+  ART-05B. Later AUTH-14/cleanup must constrain ordinary post execution to the
+  single typed command and preserve only bounded same-attempt repair.
+- `backend/app/modules/projects/service.py` deliberately prevents mutation of
+  agent-derived policy bodies; that immutability must be preserved.
+
+## Canonical dependencies
+
+- AUTH-12F: submission-artifact policy mutation/provenance authority.
+- AUTH-12G: post-submit policy mutation/provenance authority.
+- AUTH-12B2: fixed setup-service worker call-graph cutover.
+- AUTH-12H: terminal guide activation authority.
+- ART-04B1: merged PR #276. The immutable catalogue is exactly
+  `workstream.pre_submission_checkers` `v0.1` with schema
+  `pre_submission_checker_catalogue.v1`; the pure effective plan is
+  `effective_pre_submission_plan.v1` and binds its manifest hash plus locked
+  source/effective/pre-submit policy lineage.
+- ART-04B2/04B3: sealed scratch/default execution facts consumed through a
+  typed boundary; WS-POL-003 does not change those ART behaviors.
+- CHECKER/POL: canonical durable post-submit defaults/selectable rules and one
+  typed evaluation service with a complete pre and complete post command.
+- Artifact-flow orchestration invokes the pre command once while material is
+  sealed in scratch and the post command once after verified storage/binding.
+  It does not call individual checkers.
+
+## Existing tests to preserve
+
+- `backend/tests/test_projects.py`: setup generation, agent failure,
+  idempotency, policy derivation/approval, correction, Celery, and provenance.
+- `backend/tests/test_checkers.py`: pre/post compiler and checker registry.
+- `backend/tests/test_checker_catalogue.py`: exact 26-entry ART-04B1 catalogue,
+  availability, immutable manifest, effective-plan lineage, policy coverage,
+  default weakening, and stale/invalid plan proof.
+- `backend/tests/test_authorization.py`: action/catalogue/PREP/fixed-service
+  isolation.
+- `backend/tests/test_tasks.py`: task-locked guide and policy context.
+- `backend/tests/test_alembic.py`: migration topology and round trip.
+- `backend/tests/test_guide_bindings.py`: ART-verified guide material custody.
+
+## Confirmed risks and gaps
+
+- A manually edited agent projection would invalidate unified result
+  provenance. Agent projections must be immutable.
+- Unified fixed-service execution needs explicit fresh PREP custody for
+  each protected durable boundary; no synthetic human context is acceptable.
+- Free-text model output can echo secrets, raw guide excerpts, paths, URLs, or
+  prompt injection. Evidence references require a closed structured grammar
+  and all persisted text requires bounded sanitization.
+- ART-04B1 is merged but intentionally performs no checker execution or durable
+  write. WS-POL-003 must consume its exact immutable manifest/effective-plan
+  contracts rather than creating an interim registry or assuming 04B2/04B3.
+- A post-submit proposal produced early becomes stale if its compilation,
+  artifact-policy projection, pre-submit proposal, catalogue snapshot, or
+  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.
+- 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
+
+- Final names and limits for evidence-reference and safe-text schemas.
+- 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.
+- Whether separately manual policies remain supported after clean cut. If so,
+  they require independent provenance and cannot reuse unified proposals.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/INTENT.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/INTENT.md
new file mode 100644
index 00000000..ea92bf72
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/INTENT.md
@@ -0,0 +1,67 @@
+# Intent: WS-POL-003 - Unified Project Guide Compilation
+
+## Human goal
+
+Compile one immutable Project Guide source snapshot into one coherent,
+reviewable setup proposal with one logical, idempotency-keyed model attempt per
+setup generation. The result must cover guide sufficiency, submission-artifact
+policy, atomic guide requirements, supported project-specific pre-submit and
+post-submit checker bindings, human-review/lifecycle dispositions, and visible
+capability gaps.
+
+The purpose is to remove repeated guide reads and inconsistent independent
+agent conclusions without giving the model policy, authorization, checker, or
+approval authority.
+
+## Success state
+
+- One verified guide snapshot and setup generation owns one durable model-attempt
+  identity and one provider idempotency key. Retries recover that same attempt;
+  they cannot issue a second provider request under a different key.
+- A structurally invalid or unsafe provider result consumes and terminally
+  blocks that generation. Correction or another genuine evaluation requires a
+  new setup generation; transport uncertainty is reconciled under the original
+  key and can only reuse an already accepted result.
+- Trusted server validation projects the immutable result into the existing
+  canonical policy objects; `ProjectGuideCompilation` does not replace them.
+- Platform checks remain mandatory and non-selectable.
+- ART-04B1 owns the complete pre-submit catalogue: mandatory platform entries
+  plus its closed selectable project-rule namespace. CHECKER/POL owns the
+  durable post-submit capability registry/compiler. WS-POL-003 consumes both
+  read-only and creates neither a duplicate catalogue nor new ART behavior.
+- Unsupported required requirements visibly block activation; optional gaps
+  require explicit acknowledgement.
+- Agent-derived projections are immutable. Correction creates a new setup
+  generation, or a separately proven manual policy that cannot claim unified
+  agent provenance.
+- Project Managers request/recover and approve bounded results; the fixed
+  `workstream.project.setup` service alone performs compilation and service
+  projection mutations with fresh transaction-bound authorization.
+- No prepared handle, guide bytes, extracted content, credentials, or scratch
+  path enters a Celery payload.
+
+## Non-goals
+
+- Dynamic checker discovery, generated code, project-provided plugins, network
+  checks before submission, or a second checker registry.
+- Replacing human review or changing task, review, revision, contribution,
+  compensation, payment, or reputation semantics.
+- Replacing `ProjectSetupRun`, `GuideSufficiencyReport`,
+  `SubmissionArtifactPolicy`, effective policy, `PreSubmitCheckerPolicy`, or
+  `PostSubmitCheckerPolicy`.
+- Implementing ART-04B1 through 04B3 inside this initiative.
+- Changing ART scratch, storage, provider, binding, or lifecycle behavior. This
+  initiative provides one typed checker-service call per phase for later
+  artifact-flow integration at ART material boundaries.
+
+## Human decisions already captured
+
+- Prefer one unified inference over three complete guide-reading inferences.
+- Keep durable policy lifecycles and approval gates separate.
+- Treat the model as an untrusted proposal generator.
+- Preserve async-first execution and fixed-service authorization custody.
+- Do not preserve compatibility aliases or dual inference paths in v0.1.
+- Keep one authoritative execution entry per checker phase: pre-submit runs
+  only inside canonical submission preparation/admission, and post-submit runs
+  automatically from successful Submission creation/finalization. Platform,
+  project, and individual checkers are never separate caller-selected APIs.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/PLAN.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/PLAN.md
new file mode 100644
index 00000000..82585620
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/PLAN.md
@@ -0,0 +1,333 @@
+# Plan: WS-POL-003 - Unified Project Guide Compilation
+
+## Objective
+
+Replace the current three complete project-guide inference passes with one
+bounded logical `ProjectGuideCompilationAgent` attempt for each exact immutable
+guide source, capability-catalogue snapshot, and setup generation. A durable
+attempt identity and provider idempotency key enforce that cardinality across
+dispatch, timeout, reconciliation, and retry.
+
+The invocation proposes:
+
+1. guide sufficiency and findings;
+2. a submission-artifact policy draft;
+3. an atomic, traceable requirement inventory;
+4. supported project-specific pre-submit bindings;
+5. supported project-specific post-submit bindings;
+6. platform-covered requirements;
+7. human-review and lifecycle-policy requirements; and
+8. non-executable capability suggestions for unsupported requirements.
+
+The model is untrusted. Trusted server code validates, canonicalizes, hashes,
+persists, compiles, and submits separate canonical policies for approval.
+
+## Prerequisite sequence
+
+AUTH-12F, 12G, 12B2, and 12H complete the existing policy mutation,
+fixed-service worker, and activation authorization boundaries. WS-POL-003 must
+reuse those action-specific boundaries rather than create a broad
+`project.guide.compile_everything` permission.
+
+ART-04B1 supplies the complete typed pre-submit catalogue and effective-plan
+compiler: mandatory/non-selectable platform coverage plus its closed selectable
+project-rule namespace. ART-04B2/04B3 retain ART scratch/default execution.
+WS-POL-003 consumes those typed projections and does not change ART ownership.
+
+The merged pre-submit projection is consumed exactly as implemented:
+
+- catalogue ID `workstream.pre_submission_checkers`, version `v0.1`, schema
+  `pre_submission_checker_catalogue.v1`;
+- immutable canonical manifest and `manifest_sha256`;
+- stable definition ID/version, owner, phase/order/dependencies,
+  classification, typed inputs, result schema, failure code, resource budget,
+  state/disabled behavior, policy trace, and dispatch identity;
+- pure `effective_pre_submission_plan.v1` bound to exact project, guide/source,
+  effective policy, pre-submit policy, catalogue manifest, ordered entries,
+  rule-instance identities/configuration hashes, and `plan_sha256`.
+
+WS-POL-003 does not infer missing catalogue fields, mutate availability, or
+reconstruct plan identity independently.
+
+Durable CHECKER/POL ownership supplies post-submit defaults and registered
+selectable project rules. WS-POL-003 creates no pre-submit or post-submit
+dispatch registry; the unified agent sees read-only projections from both
+canonical phase owners.
+
+Before persistence/runtime cutover, an XINT/AUTH amendment must activate:
+
+- `project.guide_compilation.request`: Project Manager dispatch/recovery bound
+  to exact actor/link/grant, project, draft guide, snapshot, setup
+  run/generation, operation, request digest, and idempotency identity;
+- `project.guide_compilation.execute`: fixed `workstream.project.setup`
+  authority bound to exact canonical input hash, source and phase-owned
+  capability snapshot hashes, setup run/generation, instruction/agent version,
+  prior compilation when superseding, session/root transaction, and result.
+
+Execute authorizes only the model call plus immutable compilation parent
+creation/supersession. It does not authorize canonical projections: 12E owns
+sufficiency, 12F owns submission/effective/pre-submit policy mutations, and
+12G owns post-submit policy mutations. Each projection consumes fresh
+action-specific PREP at its protected transaction.
+
+## Single checker execution surfaces
+
+There are two lifecycle phases exposed through one internal typed checker
+service port, with exactly one complete command per phase:
+
+```text
+ART sealed scratch material
+-> checker_service.evaluate_pre_submission(...) exactly once
+   -> mandatory ART platform plan + exact locked project pre-submit plan
+ART verified stored/bound Submission material
+-> checker_service.evaluate_post_submission(...) exactly once
+   -> durable platform defaults + exact locked project post-submit plan
+```
+
+Artifact-flow orchestration supplies exact ART material facts and invokes the
+phase command at the corresponding material boundary. The checker service
+facade invokes the canonical phase executor once and returns one typed bounded
+result; no caller invokes an individual checker. For pre-submit, ART-04B1-04B3
+remain the sole plan compiler/executor/evidence writer behind the facade. For
+post-submit, the durable CHECKER executor/repository is the sole writer. The
+facade never reruns members or persists a competing evidence set.
+
+These commands are internal typed service APIs, not contributor-facing HTTP
+checker routes. Callers cannot provide checker names or invoke platform and
+project rules separately. Automatic orchestration and any bounded repair use
+the same command and deterministic attempt identity.
+
+Setup proposal, approval, correction-request, and visibility APIs remain
+separate because they configure or observe policy rather than execute a
+submission. Read-only checker-run visibility also remains bounded and separate.
+
+This initiative owns the checker service contract and composition. Later
+artifact-flow integration consumes it at ART's scratch and verified-storage
+boundaries without WS-POL-003 modifying ART code or forcing ART lifecycle changes.
+
+## Input contract
+
+`ProjectGuideCompilationContext` is strict (`extra="forbid"`) and contains:
+
+- exact existing ART-verified `GuideSourceMaterial`;
+- 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.
+
+Representative task context is tenant-local, server-redacted, and limited to
+policy shape. It excludes actor/user IDs, emails, submission artifacts, review
+decisions, payment/compensation data, credentials, secrets, URLs, and raw task
+bodies not already part of approved guide/setup source material.
+
+The canonical input binds source snapshot ID/hash, catalogue hashes, setup
+run/generation, instruction version, configured agent identity, SHA-256, and
+byte count. Provider responses, reasoning traces, credentials, scratch paths,
+and raw duplicate source text are not persisted.
+
+## Project capability and stage contract
+
+Both canonical phase catalogues are closed, typed, versioned, and registered
+explicitly at their composition roots. The pre-submit projection uses only the
+fields in merged ART-04B1; it does not invent per-entry timeout or safety
+metadata absent from `v0.1`. Catalogue mutation requires code, tests,
+deployment, and startup parity validation; projects and model output cannot
+register it.
+
+A project capability is eligible for pre-submit only when ART-04B1 registers
+it as an enabled `policy_primitive` in `project_policy`, its exact trusted
+implementation satisfies the closed primitive contract, and its compiled
+configuration matches the definition's policy fields. The later executor owns
+the server-side 60-second aggregate project-rule ceiling; the model cannot
+change stage, order, resource budget, disabled behavior, or timeout. Sixty
+seconds is a ceiling, not permission. Mandatory ART work retains its separate
+platform budgets.
+
+Deterministic work outside that contract is post-submit. Expert judgment is
+human review. Claiming, assignment, deadlines, routing, revision, acceptance,
+payment, and other transitions are lifecycle policy. Timeout, cancellation,
+or infrastructure failure is retryable platform state, never contributor
+failure.
+
+Post-submit is not an escape hatch: a selectable post-submit capability must
+also be registered for that stage, closed-schema, deterministic,
+side-effect-free, bounded, credential-safe, and implemented by trusted code.
+It cannot execute arbitrary project code, shell, network calls, or model
+judgment. Its timeout and resource budget are catalogue-owned and enforced.
+
+Initial project pre-submit capability identities, enabled only when backed by
+registered implementations, are:
+
+- `policy.submission_packet.validate`
+- `policy.storage_scheme.enforce`
+- `policy.manifest_field.require` (Workstream manifest fields only)
+- `policy.hash.verify`
+- `policy.file.require`
+- `policy.evidence.minimum`
+- `policy.artifact.forbid`
+- `policy.attestation.require`
+- `policy.file_size.limit`
+- `policy.package_size.limit`
+- `policy.packaging.require`
+- `policy.generated_quality.warn`
+
+`policy.manifest_field.require` does not validate arbitrary domain files. Such
+a requirement remains a capability gap until a dedicated typed structured-
+document capability exists.
+
+The initial project-selectable post-submit truth is only
+`check_acceptance_criteria_present` unless the running build registers more.
+Durable platform defaults are non-selectable and cannot be repeated in project
+bindings. Unknown support becomes a capability suggestion; it is never
+manufactured to make a guide appear complete.
+
+## Output contract
+
+`ProjectGuideCompilationResult` is strict, size/count bounded, and contains:
+
+- `guide_blocked`, `draft_ready`, or `draft_ready_with_warnings`;
+- sufficiency findings;
+- nullable submission-artifact policy projection;
+- atomic requirements with exactly one disposition;
+- pre-submit and post-submit binding proposals;
+- bounded capability suggestions;
+- bounded safe setup notes;
+- server-verified agent/schema identity.
+
+Allowed dispositions are `platform_covered`, `supported_pre_submit`,
+`pre_submit_capability_gap`, `supported_post_submit`,
+`post_submit_capability_gap`, `human_review`,
+`project_lifecycle_policy`, `guide_blocker`, and `informational`.
+
+Bindings may select only the exact ID/version/stage exposed through
+`project_capabilities`. Suggestions are engineering work items and can contain
+no capability ID, source code, command, URL, import, dependency, or executable
+expression.
+
+## Evidence and text safety
+
+Evidence uses a closed `GuideEvidenceRef` structure minted/validated by trusted
+server code from the immutable source-item and extraction lineage. It never
+contains raw excerpts, URLs, paths, credentials, signed references, or caller
+text.
+
+Every persisted operator-readable model field passes centralized bounded safe
+text validation/redaction. Rejection is atomic: unsafe or structurally invalid
+output produces retryable/blocked setup evidence and no policy projection.
+
+## Trusted validation
+
+Validation must, in order:
+
+1. enforce strict shape and count/size limits;
+2. lock and revalidate service identity, source lineage, setup run/generation,
+   and canonical input identity;
+3. validate sufficiency/finding consistency;
+4. validate artifact policy with existing strict/default-merge rules;
+5. validate requirement/binding/suggestion referential integrity;
+6. reject platform/default selection or repetition;
+7. reject unknown, disabled, stale-version, or wrong-stage capabilities;
+8. validate typed configuration, deterministic/side-effect-free constraints,
+   and phase-owned resource controls exactly as exposed: ART `resource_budget`
+   plus executor gates and the aggregate pre-submit ceiling, or post-submit
+   timeout/budget fields only when their canonical source defines them;
+9. sanitize every persisted text field;
+10. canonicalize and hash the result and each projection; and
+11. prepare every required action-specific fixed-service PREP, consume all of
+    them inside the one root database transaction owning compilation and
+    projection persistence, then commit mutations and authorization evidence
+    together; any validation, consumption, or write failure rolls back the
+    entire unit without borrowing authority between actions.
+
+The agent cannot order platform execution. Catalogue phases, dependencies, and
+the trusted compiler determine order.
+
+## Persistence and provenance
+
+Add immutable `ProjectGuideCompilation` provenance with exact project, guide,
+source snapshot, catalogue snapshots, setup run/generation, agent/instruction
+identity, canonical input/result hashes, component hashes, created service
+identity, and append-only supersession.
+
+Existing `GuideSufficiencyReport`, `SubmissionArtifactPolicy`,
+`PreSubmitCheckerPolicy`, and `PostSubmitCheckerPolicy` link to the exact
+compilation ID/result/component hashes. They remain canonical business objects.
+
+Agent-derived projections cannot be edited. Correction creates a new setup
+generation and compilation. If separately manual policies remain supported,
+they carry manual provenance, invalidate unified downstream proposals, and
+cannot claim or reuse agent compilation approval.
+
+## Lifecycle
+
+```text
+ART verified extraction
+-> automatic fixed-service setup continuation
+-> canonical platform/capability projections
+-> one unified model invocation
+-> trusted validation and immutable compilation
+-> separate sufficiency/policy proposals
+-> Project Manager review/approval
+-> trusted effective + pre-submit compilation
+-> deterministic post-submit proposal compilation (zero model calls)
+-> separate PostSubmitCheckerPolicy approval
+-> existing authorized guide activation
+```
+
+Blocked compilation creates no policy projections. A required capability gap
+blocks activation with an exact operator-visible setup status/error. The setup
+generation has one durable model-attempt row and provider idempotency key. A
+timeout, cancellation, or infrastructure failure before known acceptance may
+retry or reconcile only that key; an unknown outcome must be retrieved or
+replayed idempotently, never redispatched with a new key. Once an accepted
+result is persisted, retry reuses it. Invalid or unsafe output terminally
+consumes the attempt and requires a new setup generation for correction. None
+of these states is contributor failure or negative contribution evidence.
+
+## Alternatives rejected
+
+- Keep three guide-reading inference calls: repeated cost and inconsistent
+  conclusions.
+- Store one combined canonical policy: collapses separate ownership and
+  approval lifecycles.
+- Allow model-created checker code or identifiers: unsafe and non-auditable.
+- Build POL-local pre/post primitive maps or duplicate ART platform/project
+  entries: creates competing dispatch authority instead of consuming the two
+  canonical phase owners.
+- Permit in-place edits to agent projections: destroys result-hash provenance.
+- Grant one broad compilation permission: bypasses action-specific AUTH
+  custody and atomic evidence.
+- Keep standalone precheck or caller-selected checker triggers: creates a
+  parallel execution path that can drift from the locked effective plan.
+
+## Verification strategy
+
+- Strict contract and prompt-injection tests.
+- Registry/version/stage/configuration/default-isolation tests.
+- Fixed-service identity, PREP replay/session/transaction/resource tests.
+- Postgres migration, append-only supersession, concurrent idempotency, and
+  rollback tests.
+- Single-logical-attempt lifecycle tests covering concurrent dispatch,
+  timeout-after-provider-acceptance recovery under the same idempotency key,
+  accepted-result reuse, terminal invalid/unsafe output, and zero-call
+  post-submit continuation.
+- Stale source/catalogue/setup/policy invalidation tests.
+- Task-lock and activation-chain regression tests.
+- OpenAPI/import/reachability tests proving standalone precheck is absent,
+  no caller can select checkers, artifact-facing composition has one command per
+  phase, and repair converges on the same attempt identity.
+- Hosted CI full suite/coverage; changed backend subsystems remain at least 90
+  percent and repository floor remains at least 78 percent.
+
+## Completion boundary
+
+Completion requires the old three runtime methods/prompts and second
+post-submit model invocation to be deleted after all callers cut over. No
+compatibility alias, dual inference path, second registry, or independently
+invocable legacy precheck survives. The two canonical phase catalogues and one
+typed checker service port remain; the port has exactly one complete command
+per pre/post phase and no individual-checker product entry.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/RISKS.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/RISKS.md
new file mode 100644
index 00000000..1d9c97d7
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/RISKS.md
@@ -0,0 +1,19 @@
+# Risks: WS-POL-003 - Unified Project Guide Compilation
+
+| Risk | Severity | Mitigation |
+|---|---|---|
+| Unified model result is trusted as policy | Critical | Strict schemas, canonical registry lookup, trusted validation, separate approvals, and final locked revalidation. |
+| Setup service borrows PM/admin authority | Critical | Finish AUTH-12F/12G/12B2/12H; use fresh fixed-service PREP per protected transaction. |
+| Pre-submit capability dispatch is duplicated | Critical | Consume ART-04B1's complete platform-plus-project catalogue read-only; add no POL pre-submit registry or primitive map. |
+| Manual correction breaks compilation provenance | Critical | Immutable agent projections; new generation or independent manual replacement provenance. |
+| Model echoes secrets or injected guide instructions | High | Closed evidence references, safe-text validation/redaction, strict limits, no tools/network, and prompt-injection tests. |
+| Early post-submit proposal becomes stale | High | Bind all projections to compilation/result hashes; invalidate on any source/catalogue/policy/generation change. |
+| Required unsupported capability silently passes | High | Required gaps block activation and have explicit operator-visible status/code. |
+| Platform defaults are repeated or weakened | High | Non-selectable platform projection plus trusted compiler rejection of repetition/reordering/downgrade. |
+| One result/record becomes an oversized god object | Medium | Keep canonical policies separate; bounded immutable result and projection hashes only. |
+| Setup starts requiring representative tasks | Medium | Task context remains optional and bounded. |
+| Migration races concurrent initiatives | High | Allocate the then-current migration only after dependencies merge; prove one Alembic head. |
+| ART is forced to own post-submit policy | Critical | ART retains only its current pre-submit catalogue/material custody; durable CHECKER/POL ownership remains post-submit. |
+| Individual checker calls bypass the effective plan | Critical | One checker-service port exposes one pre and one post command; artifact-flow orchestration invokes once per material boundary and cannot select checkers. |
+| Facade duplicates ART pre-submit execution/evidence | Critical | Pre command delegates once to ART-04B1-04B3 and returns its canonical result/reference; ART remains the sole pre attempt/evidence writer. |
+| Compilation parent lacks exact AUTH custody | Critical | Merge narrow XINT/AUTH request+execute activation before chunk 03; projections retain separate 12E/12F/12G PREP. |
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/STATUS.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/STATUS.md
new file mode 100644
index 00000000..4775ae34
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/STATUS.md
@@ -0,0 +1,27 @@
+# Status: WS-POL-003 - Unified Project Guide Compilation
+
+Status: planning candidate under review; no implementation is active.
+
+Baseline: `origin/main` `bb77ff4a0ab61120b94d6d4763934b444c39207d`.
+
+## Dependency gates
+
+| Dependency | Required before | Current status at baseline |
+|---|---|---|
+| AUTH-12F | service submission-policy projection writes | Proposed after merged 12E |
+| AUTH-12G | service post-submit projection writes | Proposed after 12F |
+| AUTH-12B2 | unified Celery call-graph cutover | Proposed after 12F/12G |
+| AUTH-12H | terminal activation integration | Proposed after 12B2 |
+| ART PLAN5 / 04A4 | 04A4 superseded; standalone precheck clean cut moved to ART-05B | PLAN5 merged through PR #273 |
+| ART-04B1 complete pre-submit catalogue/effective-plan contract | exact immutable platform plus closed project-rule projection consumed read-only | Merged through PR #276 |
+| ART-04B2/04B3 | sealed execution and immutable evidence writer used behind the pre facade | Proposed after merged ART-04B1 |
+| ART-05B | standalone precheck and legacy Submission path clean cut | Proposed after ART/XINT admission sequence |
+| CHECKER/POL post-submit catalogue | durable defaults plus registered selectable project rules | Consumed read-only/hardened by POL work |
+| CHECKER typed evaluation port | one pre call in scratch and one post call after verified storage/binding | Owned by WS-POL-003-07 |
+| XINT/AUTH compilation activation | exact PM request and fixed-service execute actions for immutable compilation custody | Not yet planned/merged |
+
+## Chunk state
+
+All WS-POL-003 chunks are proposed and inactive. Planning does not authorize
+implementation. The first implementation chunk requires explicit human start
+after this plan and its applicable dependencies are merged.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-01-contract-catalogue-projection.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-01-contract-catalogue-projection.md
new file mode 100644
index 00000000..8a69df4b
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-01-contract-catalogue-projection.md
@@ -0,0 +1,41 @@
+# Chunk Contract: WS-POL-003-01 - Unified Contract and Catalogue Projection
+
+Status: Proposed, inactive. 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.
+
+## Allowed files
+
+`backend/app/interfaces/project_agents.py`, canonical ART and CHECKER/POL
+projection interfaces/composition only, focused tests, and WS-POL-003 docs.
+
+## 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.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-02-unified-agent-adapter.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-02-unified-agent-adapter.md
new file mode 100644
index 00000000..6ba78131
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-02-unified-agent-adapter.md
@@ -0,0 +1,33 @@
+# Chunk Contract: WS-POL-003-02 - Unified Agent Adapter
+
+Status: Proposed after 01. Risk: L1.
+
+## Goal
+
+Implement one bounded `compile_project_guide` adapter call with untrusted-data
+instructions, strict structured output, cancellation, timeout, and sanitized
+failures. Do not rewire production Celery orchestration.
+
+## Allowed files
+
+`backend/app/adapters/project_agents/**`, project-agent interfaces/configuration,
+fake adapter tests, and WS-POL-003 docs.
+
+## Not allowed
+
+Database, authorization, Celery orchestration, policy approval, registry, or checker runtime
+changes; no provider trace persistence or tool/network capability.
+
+## Acceptance
+
+- One method consumes the canonical context and returns the strict result.
+- Guide/task contents remain untrusted data and cannot alter instructions.
+- Prompt/input limits, timeout, cancellation, and sanitized error behavior are
+  preserved.
+- Agent cannot emit code, commands, URLs, capabilities outside the projection,
+  or approval decisions.
+
+## Verification and review
+
+Fake-runtime, injection, timeout/cancellation, and serialization tests. Required
+reviewers: security, architecture, QA, product, test delta, CI integrity.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-03-compilation-persistence-validator.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-03-compilation-persistence-validator.md
new file mode 100644
index 00000000..4da1b47d
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-03-compilation-persistence-validator.md
@@ -0,0 +1,51 @@
+# Chunk Contract: WS-POL-003-03 - Compilation Persistence and Validator
+
+Status: Proposed after 02 and required AUTH dependencies. Risk: L1.
+
+Hard gate: the exact XINT/AUTH registration and activation for
+`project.guide_compilation.request` and
+`project.guide_compilation.execute` is merged. Generic fixed-service authority
+or 12E/12F/12G projection authority cannot substitute for it.
+
+## Goal
+
+Add immutable compilation provenance, trusted validation, component hashes,
+append-only supersession, and action-specific fixed-service mutation custody.
+
+## Allowed files
+
+Project models/schemas/repository/validator/composition, AUTH prepared-resource
+composition only where exact existing actions require it, one then-current
+Alembic migration, focused tests, and WS-POL-003/AUTH specification docs.
+
+## Not allowed
+
+Celery call-graph cutover, approval behavior, checker execution, broad compilation
+permission, synthetic human authority, compatibility path, or ART semantics.
+
+## Acceptance
+
+- A database `UNIQUE` constraint covers exact `project_id`, `guide_id`, source
+  snapshot ID, pre- and post-catalogue snapshot hashes, setup run ID, and setup
+  generation. The setup-run current-compilation pointer advances only by
+  compare-and-swap against its locked expected generation/current ID. Concurrent
+  requests therefore converge on one immutable compilation rather than two
+  current rows.
+- Strict validation and sanitization precede atomic persistence.
+- Existing policy projections bind exact compilation/component hashes.
+- Fresh action-specific fixed-service PREPs are prepared separately, but all
+  required handles are consumed in the single root database transaction that
+  owns compilation, projection links, and authorization evidence. Replay,
+  copied/wrong handle, stale context, or any partial failure rolls back the
+  whole unit and creates no durable effect.
+- Agent-derived projections cannot be updated in place.
+- Compilation creation/supersession consumes the exact fixed-service execute
+  action; PM recovery consumes the exact request action. Projection writes each
+  consume their separate 12E/12F/12G PREP in that same root transaction and
+  cannot borrow compilation authority.
+
+## Verification and review
+
+Postgres unique-key, compare-and-swap, concurrent-insert, PREP atomicity, and
+rollback tests plus AUTH all-pairs denials and 90% changed-subsystem coverage.
+Required reviewers: all L1 tracks.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-04-initial-setup-cutover.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-04-initial-setup-cutover.md
new file mode 100644
index 00000000..66243dcb
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-04-initial-setup-cutover.md
@@ -0,0 +1,41 @@
+# Chunk Contract: WS-POL-003-04 - Initial Setup Pipeline Cutover
+
+Status: Proposed after 03. Risk: L1.
+
+## Goal
+
+Replace the separate sufficiency and submission-policy inference calls with one
+unified compilation, while producing the existing canonical report and draft
+policy projections.
+
+## Allowed files
+
+Project setup worker/service/queue/composition, unified compilation services,
+focused project/authorization/ART-boundary tests, and WS-POL-003 docs.
+
+## Not allowed
+
+Post-submit continuation cutover, approval semantics, checker execution,
+serialized handles/material, or fallback to legacy guide excerpts.
+
+## Acceptance
+
+- Automatic and manual recovery converge on one deterministic setup run/task.
+- Each generation persists one model-attempt row and provider idempotency key
+  derived from the exact setup run/generation and compilation input identity.
+  Dispatch and recovery atomically claim that row; retries use the same key.
+- Provider acceptance must be recoverable by idempotent replay/result lookup,
+  and the accepted result is persisted before downstream continuation. A
+  timeout-after-acceptance retry reuses that result without another invocation.
+- Invalid or unsafe output terminally consumes the generation's attempt;
+  another evaluation requires a new setup generation.
+- Blocked output creates no policy projection.
+- Ready output atomically links sufficiency and draft artifact policy to one
+  compilation.
+- Stale/revoked/wrong-service/output failures occur before protected mutation.
+
+## Verification and review
+
+Celery concurrent-claim, retry/replay, timeout-after-acceptance, accepted-result
+reuse, terminal-invalid-output, stale-generation, provider-failure, and rollback
+tests plus hosted full coverage. Required reviewers: all L1 tracks.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-05-approval-pre-submit-integration.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-05-approval-pre-submit-integration.md
new file mode 100644
index 00000000..248653ae
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-05-approval-pre-submit-integration.md
@@ -0,0 +1,40 @@
+# Chunk Contract: WS-POL-003-05 - Approval and Pre-Submit Integration
+
+Status: Proposed after 04. Risk: L1.
+
+## Goal
+
+Bind Project Manager approval to the exact immutable compilation and compile
+approved project pre-submit bindings through ART-04B1's effective-plan
+compiler while preserving its platform entries as mandatory and non-selectable.
+
+## Allowed files
+
+Project policy approval/service/repository/router/schema surfaces,
+ART-04B1 catalogue/compiler integration,
+authorization resource composition, focused tests, and specifications.
+
+## Not allowed
+
+Second registry/compiler, platform-default selection, checker execution,
+post-submit compilation, unrelated approval semantics, task/review/payment
+behavior, or in-place agent edits. This chunk does own Project Manager approval
+binding to the exact compilation, including hash binding, stale invalidation,
+and activation blocking.
+
+## Acceptance
+
+- Approval locks exact compilation/result/artifact/pre-submit hashes.
+- Platform defaults are composed only by ART and cannot be selected, repeated,
+  weakened, reordered, or downgraded.
+- Required capability gaps block approval/activation with exact operator code.
+- Catalogue/source/generation/projection changes stale prior approval.
+- Effective and pre-submit outputs commit atomically with authorization evidence.
+- The approved project plan and mandatory ART platform entries compose only
+  through the later single checker-service pre-submit command; this chunk does
+  not change ART or expose an execution route.
+
+## Verification and review
+
+Postgres approval races, stale hashes, default isolation, compiler parity,
+AUTH denial, and task-lock regression tests. Required reviewers: all L1 tracks.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-06-post-submit-deterministic-cutover.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-06-post-submit-deterministic-cutover.md
new file mode 100644
index 00000000..bc2a229a
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-06-post-submit-deterministic-cutover.md
@@ -0,0 +1,42 @@
+# Chunk Contract: WS-POL-003-06 - Deterministic Post-Submit Cutover
+
+Status: Proposed after 05. Risk: L1.
+
+## Goal
+
+Compile the stored unified post-submit proposal after effective/pre-submit
+approval without rereading the guide or invoking a second model.
+
+## Allowed files
+
+Project post-submit policy/compiler/Celery/repository surfaces, exact AUTH-12G
+composition, the canonical durable CHECKER/POL post-submit capability source,
+focused tests, and WS-POL-003 docs.
+
+## Not allowed
+
+Checker execution, review/revision behavior, new checker registration, default
+checker repetition, manual reuse of agent provenance, or ART changes.
+
+## Acceptance
+
+- Continuation performs zero model invocations.
+- Exact compilation, effective policy, pre-submit plan, catalogue, setup
+  generation, approval record ID, approval actor identity, and approval hash are
+  locked and revalidated.
+- Any correction, replacement, catalogue change, or change to the stored
+  approval record/actor/hash invalidates the proposal.
+- Platform-default repetition and unknown/wrong-stage checkers fail closed.
+- Only registered project entries from the canonical durable CHECKER/POL
+  post-submit source are selectable; ART pre-submit entries cannot be selected.
+- Tests explicitly reject current legacy behavior that repeats any durable
+  default in either required or warning project bindings.
+- Compiled post-submit policy retains separate PM approval; that later approver
+  may differ, but cannot substitute for or alter the exact approval identity
+  authorizing selection of the stored proposal.
+
+## Verification and review
+
+Zero-call, invalidation, replay, approval-race, default-isolation, and atomic
+evidence tests. Runtime dispatch is owned by later consumers of the chunk-07
+port. Required reviewers: all L1 tracks.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-07-single-checker-service-port.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-07-single-checker-service-port.md
new file mode 100644
index 00000000..3665cba3
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-07-single-checker-service-port.md
@@ -0,0 +1,68 @@
+# Chunk Contract: WS-POL-003-07 - Single Checker Service Port
+
+Status: Proposed after 06. Risk: L1.
+
+## Goal
+
+Provide one internal typed checker service port with exactly two complete
+phase commands:
+
+- `evaluate_pre_submission(...)`, one logical attempt initiated by artifact-flow
+  orchestration while exact material is sealed in `ArtifactScratchManager`
+  custody;
+- `evaluate_post_submission(...)`, one logical attempt initiated by
+  artifact-flow orchestration after exact verified content is durably stored,
+  bound, and attached to the Submission lineage.
+
+Each command invokes the complete canonical phase executor and returns one
+typed result. The pre command is a facade over ART-04B1-04B3's existing
+effective-plan execution; it does not reimplement or rerun ART entries. The
+post command invokes the durable CHECKER executor. Artifact-flow callers never
+select or call individual platform/project checkers.
+
+## Allowed files
+
+Checker interfaces/service/composition, project policy plan adapters, typed ART
+material/result interfaces only, focused checker/project tests, and WS-POL-003
+docs.
+
+## Not allowed
+
+ART scratch/storage/provider/binding/lifecycle changes, public contributor
+checker routes, caller-selected checker names, per-checker endpoints, dynamic
+plugins, arbitrary code/network execution, or prepared handles in payloads.
+
+## Acceptance
+
+- The service exposes exactly one pre and one post command and no generic
+  `run_checker(name, ...)` product boundary.
+- Pre composes mandatory ART platform entries with exact task-locked project
+  pre-submit entries through ART-04B1-04B3 and evaluates them once against one
+  sealed scratch generation.
+- Post composes durable defaults with exact task-locked project post-submit
+  entries and evaluates them against one verified stored/bound content lineage.
+- Both commands bind exact project/task/assignment, guide/policy, artifact,
+  manifest, generation, attempt, action, service identity, and transaction
+  facts; stale/replay/cross-phase/cross-resource calls fail closed.
+- The port requires a deterministic attempt identity. Bounded retry/repair may
+  call the command again for that same logical attempt, but replay returns the
+  existing canonical result without rerunning completed members. A genuinely
+  new evaluation requires a new attempt identity; this chunk does not alter or
+  claim the external call sites.
+- ART's pre-submit attempt/result/evidence repository is the only pre writer;
+  CHECKER's durable repository is the only post writer. The facade returns the
+  canonical result/reference and cannot persist partial or duplicate member
+  results. Each phase-owner repository transaction atomically claims its
+  attempt and commits the canonical member set, phase result, and authorization
+  evidence; concurrent callers observe or resume that same row.
+- One bounded phase result contains every platform/project member with exact
+  definition/version/policy trace; infrastructure failure is never contributor
+  blame or a review decision.
+
+## Verification and review
+
+All-pairs phase/identity/resource denial, exact-once composition, no-individual-
+dispatch reachability, timeout/cancellation, deterministic attempt replay,
+scratch and stored-content contract parity, and 90% changed-subsystem coverage.
+Required reviewers: all L1 tracks. Human focus: the port permits one complete
+call per phase and cannot select individual checkers.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-08-visibility-correction-cleanup.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-08-visibility-correction-cleanup.md
new file mode 100644
index 00000000..a1af6bd0
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/chunks/WS-POL-003-08-visibility-correction-cleanup.md
@@ -0,0 +1,48 @@
+# Chunk Contract: WS-POL-003-08 - Visibility, Correction, and Cleanup
+
+Status: Proposed after 07 and AUTH-12H. Risk: L1.
+
+## Goal
+
+Expose a bounded Project Manager review package, implement generation-safe
+correction, prove activation compatibility, and delete retired multi-inference
+paths and parallel project-checker maps.
+
+## Allowed files
+
+Project setup visibility/correction/API/service/worker surfaces, agent
+interfaces/adapters cleanup, checker project-catalogue and ordinary trigger
+route cleanup, task/activation regression tests, docs, and WS-POL-003 memory.
+
+## Not allowed
+
+ART code, raw guide/model output exposure, in-place agent projection edits,
+contributor compilation routes, public checker selection APIs, or compatibility
+aliases.
+
+## Acceptance
+
+- Project Manager sees bounded sufficiency, policy proposals, requirements,
+  bindings, gaps, and safe notes without raw material.
+- Correction creates a new setup generation and compilation; immutable history
+  remains linked and superseded.
+- Existing authorized activation consumes only the complete current approved
+  chain.
+- Three retired runtime methods/prompts and the second post-submit model call
+  are deleted; the canonical ART-04B1 pre-submit catalogue, canonical
+  CHECKER/POL post-submit catalogue, and one checker service port remain.
+- OpenAPI/import/reachability tests prove ART-05B has removed the standalone
+  `/submission-precheck` route and remove the ordinary
+  caller-triggered `POST /submissions/{id}/checker-runs` execution path. A
+  bounded repair command may remain only by Submission/run identity and must
+  use the same deterministic attempt contract.
+- Static/import tests find no per-checker product entry, caller-selected
+  checker names, dual project registry, stale terminology, or compatibility
+  inference path.
+- Failures/gaps create no contribution, payment, or reputation evidence.
+
+## Verification and review
+
+API privacy, correction, activation, task-lock, stale wording/link,
+registry/reachability, full hosted CI, and 90% changed-subsystem coverage.
+Required reviewers: all L1 tracks.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-internal-review-evidence.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-internal-review-evidence.md
new file mode 100644
index 00000000..d03dd68f
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-internal-review-evidence.md
@@ -0,0 +1,51 @@
+# WS-POL-003 Planning Internal Review Evidence
+
+Reviewed baseline was reconciled after ART-04B1 merged to `origin/main`
+`bb77ff4a0ab61120b94d6d4763934b444c39207d`.
+
+## Scope reviewed
+
+- Intent, discovery, plan, decisions, risks, status, chunk map, and all eight
+  proposed chunk contracts.
+- Current project-agent, project setup, checker compiler/catalogue/effective
+  plan/runner/router, AUTH prepared-capability, ART PLAN5/04B1-04B3, and POL-002
+  boundaries.
+
+## Reviewer results
+
+| Track | Final result | Resolution |
+|---|---|---|
+| Architecture | PASS WITH LOW RISKS | Clarified ART-04B1 as the unchanged complete pre-submit catalogue; collapsed duplicate status wording; preserved external call-site ownership. |
+| Security | PASS WITH LOW RISKS | Added exact XINT/AUTH request+execute gate, immutable projection rules, bounded safe model input/output, and separate phase-owned evidence writers. |
+| Product/operations | PASS WITH LOW RISKS | Replaced ambiguous ART-dispatch wording with artifact-flow integration at ART material boundaries; kept setup APIs distinct from execution. |
+| QA/test | PASS WITH LOW RISKS | Made pre command a facade over ART-04B1-04B3, named sole pre/post evidence writers, added default-repetition tests, and strengthened chunk-07 dependency. |
+
+After PR #276 merged, the plan was reconciled again against the implemented
+`PreSubmissionCheckerCatalogue`, exact v0.1 manifest, pure effective-plan
+compiler, and PLAN5 route-removal sequence before the PR head was refreshed.
+The final architecture review passed. Final QA and security reviews identified
+the same non-blocking precision issue: preserve disabled advisory entries in
+the full manifest and consume phase-owned resource controls without inventing
+a pre-submit per-entry timeout. Both corrections are incorporated.
+
+CodeRabbit then identified eight concurrency, idempotency, approval-binding,
+and transaction-boundary gaps. The plan now defines database uniqueness and
+compare-and-swap, one durable provider attempt/key per setup generation,
+timeout-after-acceptance recovery, exact approval identity, phase-owned checker
+attempt claims, and one root transaction consuming all separate PREPs. Fresh
+architecture, security, QA, and product/operations rereviews all passed with no
+remaining findings.
+
+No blocking finding remains. All reviewer sessions completed.
+
+## Checks
+
+- `python3 scripts/check_markdown_links.py` — passed.
+- `python3 scripts/check_stale_workstream_wording.py` — passed.
+- Targeted `scripts/check_stale_authorization_docs.py` scan of all 17 planning
+  files — passed.
+- `git diff --check` — passed.
+- Temporary root draft `latest_new_addtion_plan.md` — deleted.
+
+No application code, migration, workflow, coverage threshold, or runtime
+behavior changed in this planning-only chunk.
diff --git a/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-pr-trust-bundle.md b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-pr-trust-bundle.md
new file mode 100644
index 00000000..20fc7b73
--- /dev/null
+++ b/.agent-loop/initiatives/WS-POL-003-unified-project-guide-compilation/reviews/WS-POL-003-PLAN-pr-trust-bundle.md
@@ -0,0 +1,80 @@
+# WS-POL-003 Planning PR Trust Bundle
+
+## Chunk
+
+`WS-POL-003-PLAN` — Unified Project Guide Compilation planning.
+
+## Goal and human-approved intent
+
+Replace three complete Project Guide inference passes with one bounded unified
+compilation per immutable setup generation, while retaining separate canonical
+policy objects, approvals, authorization boundaries, and checker ownership.
+
+Provide one internal typed checker-service facade with one complete pre-submit
+command at sealed-scratch custody and one complete post-submit command after
+verified storage/binding. No caller selects or invokes individual checkers.
+
+## What changed and why
+
+- Added the complete WS-POL-003 intent, discovery, plan, decisions, risks,
+  status, and eight bounded implementation contracts.
+- Consumed merged ART-04B1's exact catalogue/effective-plan contract and
+  preserved future ART-04B2/04B3 as the sole pre-submit executor/evidence
+  writer, with durable CHECKER/POL as the post-submit owner.
+- Added exact future XINT/AUTH compilation request/execute custody before
+  persistence can start.
+- Made unified agent projections immutable, safe-text bounded, capability
+  closed-world, and generation/hash bound.
+- Deleted the temporary root planning draft after incorporating its valid intent.
+
+The original draft had the right product direction but stale baseline,
+ambiguous correction provenance, under-specified fixed-service authority, and
+unclear phase execution/evidence ownership.
+
+## Design and scope control
+
+- One model invocation; no model policy authority.
+- ART's complete pre-submit catalogue is consumed unchanged.
+- No new checker registry, dynamic plugin, generated code, or compatibility path.
+- Pre facade delegates once to ART and persists no duplicate evidence.
+- Post facade uses CHECKER's sole durable writer.
+- No application code, API, database, migration, workflow, or CI change.
+
+Rejected alternatives include three repeated inference calls, one combined
+canonical policy object, in-place edits to agent projections, broad compilation
+authority, POL-local pre-submit maps, and per-checker product APIs.
+
+## Acceptance evidence and test delta
+
+This is documentation/planning only. Markdown links, stale wording, and diff
+integrity pass. There is no runtime test delta and no CI or coverage weakening.
+
+## Reviewer results and external review
+
+Architecture, security, product/operations, and QA all pass with low residual
+dependency-discipline risks after repairs. CodeRabbit and hosted CI are pending
+on the PR head.
+
+## Remaining risks and follow-up
+
+- Implementation must honor every dependency gate; planning does not activate
+  any action or chunk.
+- Exact XINT/AUTH compilation request/execute activation must merge before
+  WS-POL-003-03.
+- ART-04B2/04B3 must merge and be callable behind merged ART-04B1 before the
+  checker facade chunk.
+- Each implementation chunk requires separate human start, evidence, review,
+  PR, and human merge.
+
+## Human review focus
+
+- Does one unified inference preserve separate policy and approval truth?
+- Is ART pre-submit ownership unchanged and free of duplicate evidence?
+- Is post-submit authority confined to durable CHECKER/POL ownership?
+- Are capability gaps, correction, replay, and stale generation fail-closed?
+- Are the eight chunks genuinely bounded and correctly dependency-gated?
+
+## Human merge ownership
+
+This PR must be merged only after explicit human approval. Merge authorizes the
+plan record only; it does not start an implementation chunk.