ci: retire blocking mutation workflow

.agent-loop/initiatives/WS-QUAL-001-backend-coverage-floor/CHUNK_MAP.md

@@ -24,12 +24,12 @@ remain stopped historical experiments. Do not resume them.
 | `WS-QUAL-001-PLAN3R1` | Resolve five valid late CodeRabbit findings from PR #272 | L1 | Merged PR #278 |
 | `WS-QUAL-001-04P` | Establish protected hash-verified mutation dependency authority | L1 | Merged PR #281 |
 | `WS-QUAL-001-04M` | Pilot pinned changed-scope mutation evidence without a score gate | L1 | Merged PR #285 as `7f395d47`; hosted calibration accepted |
-| `WS-QUAL-001-05M` | Add calibrated blocking behavior-mutation policy | L1 | Active by explicit human instruction |
+| `WS-QUAL-001-05M` | Add calibrated blocking behavior-mutation policy | L1 | Retired after callable-wide enforcement proved unsuitable; requires a fresh changed-line-aware plan |
 
 ## Dependency rule
 
 `PLAN3 -> PLAN3R1 -> 04P -> 04M -> human calibration checkpoint -> 05M`.
 
-Each chunk maps to one PR. `04M` may prove that the candidate engine or target
-strategy is unsuitable and stop without `05M`. Planning does not pre-authorize
-either implementation chunk.
+The completed pilot evidence remains historical input. Do not restart `05M` or
+another blocking mutation workflow without a fresh bounded plan proving that
+unchanged executable lines cannot block a declaration-only change.

.agent-loop/initiatives/WS-QUAL-001-backend-coverage-floor/STATUS.md

@@ -10,7 +10,7 @@ wall time, and a 464.471-second slowest lane.
 The global blocking floor remains 78 percent by explicit human decision. Named
 new or materially changed subsystem checks remain blocking at 90 percent.
 
-## Current gate
+## Mutation-gate disposition
 
 `WS-QUAL-001-PLAN3` merged through PR #272. Its planning-only correction
 `WS-QUAL-001-PLAN3R1` merged through PR #278 after resolving all late CodeRabbit
@@ -27,19 +27,18 @@ error, timeout, or suspicious outcomes. Strong calibration killed two
 representative mutants and the deliberately weak calibration left two alive.
 The human accepted this evidence and explicitly started `WS-QUAL-001-05M`.
 
-The corrected proposal remains two-stage:
+The subsequent blocking rollout proved unsuitable for ordinary work: its
+callable-wide selection treated unchanged executable lines as part of every
+small changed declaration and produced unresolvable survivor sets. The hosted
+workflow is therefore retired pending a separately reviewed changed-line-aware
+design. Existing policy and evidence files remain historical input, not an
+active PR requirement.
 
-1. `04M` — merged bounded, pinned, changed-scope mutation pilot with complete
-   evidence and no blocking score.
-2. Human calibration checkpoint — accepted.
-3. `05M` — implemented and internally reviewed bounded blocking survivor
-   policy for eligible changed logic and explicit test-only behavior claims;
-   exact-head hosted CI and external review remain before human merge.
-
-The mutation score remains observational. Existing Backend semantic lanes,
-global 78-percent coverage, and protected 90-percent subsystem floors remain
-unchanged and blocking on their existing terms.
+Existing Backend semantic lanes, global 78-percent coverage, protected
+90-percent subsystem floors, lint, and review gates remain unchanged and
+blocking on their existing terms.
 
 ## Stop condition
 
-Stop after the 05M PR is merge-ready. Do not start another QUAL chunk.
+Do not restart mutation enforcement without a fresh bounded plan and proof that
+unchanged executable lines cannot block a declaration-only change.

.agent-loop/initiatives/WS-QUAL-001-backend-coverage-floor/reviews/WS-QUAL-001-05M-external-review-response.md

@@ -24,3 +24,41 @@ behavior. The textual TOML rewrite remains fail closed; typed failure is the
 required safety property for unsupported legacy shapes.
 
 Exact-head hosted CI and CodeRabbit rereview remain required after publication.
+
+## PR #289 retirement review on `54cd358a`
+
+Comments addressed:
+
+- Rewrote the PR description using the complete repository trust-bundle
+  template so its title, intent, scope, evidence, reviewer results, deliberate
+  workflow retirement, remaining risk, and follow-up boundary match the final
+  diff.
+- Added the same reactivation guard to `CONTRIBUTING.md`, the behavior-claim
+  guide, and the Backend operations guide: enforcement cannot resume without an
+  approved fresh changed-line-aware plan proving unchanged executable lines do
+  not block declaration-only changes.
+- Reworded the remaining claim-discovery and fail-closed guidance as historical
+  behavior rather than an active contribution requirement.
+
+Comments deferred:
+
+- Two mutation-policy inline threads are outdated because the referenced
+  implementation was fully reverted and is absent from the final PR diff.
+
+Human decisions needed:
+
+- A repository administrator must remove the retired check from external
+  branch-protection settings if it was configured there.
+
+Commands rerun:
+
+```text
+python3 scripts/check_markdown_links.py
+python3 scripts/check_stale_workstream_wording.py
+git diff --check
+```
+
+Remaining risks:
+
+- Behavior mutation is no longer enforced in hosted CI. Reintroduction requires
+  the separately approved changed-line-aware design recorded in current status.

.ci/behavior-claims/README.md

@@ -1,37 +1,49 @@
 # Behavior mutation claims
 
-Schema-v1 claim files provide bounded owning pytest nodes for mutation targets.
-They are additive: every eligible changed production or CI-runtime Python target
-is selected independently, and a claim cannot remove or replace one.
-
-The filename and `chunk_id` must match. Targets are repository-relative Python
-files under `backend/app/` or `backend/scripts/`; each target also names its
+The hosted behavior-mutation workflow is temporarily retired. These files are
+retained as historical design input and are not currently required for pull
+requests. The claim-discovery and fail-closed rules below describe the retired
+design only; do not infer a blocking check from them. Behavior-mutation
+enforcement must not resume until a fresh changed-line-aware plan is approved
+and proves that unchanged executable lines cannot block a declaration-only
+change.
+
+Historically, schema-v1 claim files provided bounded owning pytest nodes for
+mutation targets. They were additive: every eligible changed production or
+CI-runtime Python target was selected independently, and a claim could not
+remove or replace one.
+
+Under that retired design, the filename and `chunk_id` had to match. Targets
+were repository-relative Python files under `backend/app/` or
+`backend/scripts/`; each target also named its
 qualified callables, exact owning pytest nodes, typed observable outcomes, and
 any essential real boundaries. Unknown fields, unsafe paths, missing files,
-duplicate entries, unowned changed targets, or stale chunk identifiers fail
+duplicate entries, unowned changed targets, or stale chunk identifiers failed
 closed.
 
-The required behavior-mutation check discovers the one claim changed by the
+The retired behavior-mutation check discovered the one claim changed by the
 pull request; labels, workflow inputs, environment variables, and PR prose
-cannot select it. Copy `example.behavior-claim.json`, rename it to the bounded
-chunk identifier, and replace every example target, callable, test, outcome,
-and boundary. Eligible production changes without exactly one changed claim
-fail closed. A test-only behavior claim is additive and cannot remove an
+could not select it. Contributors copied `example.behavior-claim.json`, renamed
+it to the bounded chunk identifier, and replaced every example target,
+callable, test, outcome, and boundary. Eligible production changes without exactly one changed claim
+failed closed. A test-only behavior claim was additive and could not remove an
 eligible changed target.
 
-The check has no mutation percentage. Killed mutants pass. A meaningful
+The retired check had no mutation percentage. Killed mutants passed. A meaningful
 survivor, timeout, suspicious result, engine error, malformed or stale evidence,
-target escape, or excluded mutant inside the selected callable scope blocks.
-The only surviving control allowed by policy is Workstream's exact deliberately
-weak calibration callable; contributors cannot add survivor allowlists,
+target escape, or excluded mutant inside the selected callable scope blocked.
+The only surviving control allowed by policy was Workstream's exact deliberately
+weak calibration callable; contributors could not add survivor allowlists,
 free-form exemptions, or source mutation pragmas.
 
-Changes with no eligible target and no claim produce typed `not_applicable`
-evidence before the mutation toolchain is installed. Ordinary PR verdicts are
+Under the retired design, changes with no eligible target and no claim produced
+typed `not_applicable`
+evidence before the mutation toolchain was installed. Ordinary PR verdicts were
 calculated by the evaluator and Git-delta helper archived from protected base,
 not by PR-head policy code.
 
-Validate claim discovery locally from the repository root:
+For historical diagnostics only, claim discovery can still be inspected locally
+from the repository root:
 
 ```bash
 backend/.venv/bin/python backend/scripts/mutation_policy.py \
@@ -42,7 +54,7 @@ backend/.venv/bin/python backend/scripts/mutation_policy.py \
   --selection-output /tmp/workstream-mutation-selection.json
 ```
 
-An unrelated delta reports `applicability: not_applicable`. An applicable
-delta must report the exact changed targets, callable ownership, and owning
-tests expected by the contributor. Discovery errors are policy failures; fix
-the claim or delta rather than editing generated evidence.
+Under the retired design, an unrelated delta reported
+`applicability: not_applicable`. An applicable delta reported the exact changed
+targets, callable ownership, and owning tests expected by the contributor.
+This command does not produce active PR evidence or authorize reactivation.

CONTRIBUTING.md

@@ -72,34 +72,17 @@ rerun affected checks; unchanged evidence does not need ceremonial repetition.
 
 ## Behavior Mutation Claims
 
-Eligible Python changes under `backend/app/` or `backend/scripts/` require one
-schema-v1 claim at `.ci/behavior-claims/<chunk-id>.json`. Start from
-[the copyable example](.ci/behavior-claims/example.behavior-claim.json) and name
-the exact changed callable, its owning pytest node, the observable outcome, and
-any essential PostgreSQL, MinIO, HTTP, lock, trigger, or concurrency boundary.
-The [claim guide](.ci/behavior-claims/README.md) contains the closed rules.
-
-Run `cd backend && .venv/bin/python -m pytest -q <owning-test-node>` before
-opening the PR. From the repository root, validate discovery against the PR
-delta with:
-
-```bash
-backend/.venv/bin/python backend/scripts/mutation_policy.py \
-  --repository-root . \
-  --base-sha "$(git merge-base origin/main HEAD)" \
-  --head-sha "$(git rev-parse HEAD)" \
-  --discover \
-  --selection-output /tmp/workstream-mutation-selection.json
-```
-
-Inspect `applicability`, `changed_targets`, `changed_callables`, `tests`, and
-`target_owners` in that output. The required mutation check derives targets from the exact git
-delta and uses the claim only for bounded callable/test ownership. It does not
-use a global score: meaningful survivors and incomplete or unsafe evidence
-block. Repair a survivor by strengthening the owning behavior assertion or by
-correcting the production behavior; do not add skips, exclusions, allowlists,
-or mutation pragmas. Unrelated changes return `not_applicable` automatically
-and do not install the mutation engine.
+The hosted behavior-mutation check is temporarily retired because its
+callable-wide survivor policy blocked declaration-only changes by mutating
+unchanged executable lines. Do not treat a behavior claim as a required PR
+gate while the replacement is being designed.
+
+Existing claim, schema, policy, dependency, and evidence files remain as
+historical design input. They do not replace focused tests, hosted Backend
+lanes, coverage floors, internal review, CodeRabbit, or human merge approval.
+Behavior-mutation enforcement must not resume until a fresh changed-line-aware
+plan is approved and proves that unchanged executable lines cannot block a
+declaration-only change.
 
 ## Durable Records