frontend-patterns/integrations/activity-core

Bounded reporting repair for Activity Core

Owner task: FEP-WP-0008 T03, alongside existing ACTIVITY-WP-0041. No new workplan.

The September 27 grant-aware image promotion removed the frontend resolver. Keep that grant-aware base. This overlay changes only the resolver module and its package registration; it does not roll back grants, change schedules, or admit a mutation executor. Diagnosis.

source.json pins the upstream source revision and resolver hashes. Reproduce with an isolated checkout of that exact activity-core revision:

git apply /home/worsch/frontend-patterns/integrations/activity-core/monthly-review.patch
cp /home/worsch/frontend-patterns/integrations/activity-core/Dockerfile Dockerfile.reporting-repair
cp /home/worsch/frontend-patterns/integrations/activity-core/test_frontend_monthly_review.py tests/
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python -m pytest -q -p no:cacheprovider \
  tests/test_frontend_patterns.py tests/test_frontend_monthly_review.py \
  tests/test_repository_grant.py tests/test_resolve_context_binding.py
docker build -f Dockerfile.reporting-repair -t activity-core:fep-reporting-repair-20260928 .

The isolated checkout needs the owner's existing locked development environment. The Docker build itself needs no new dependency installation: it extends the pinned runtime and proves resolver registration plus typed grant support during construction. The review compares all Python source hashes against the base, not only image labels. Run check_runtime.py with the image’s Python (via stdin) to capture those hashes, registered resolver names and a synthetic typed-grant identity; it performs no network requests or application mutations.

The monthly extension reads this repository's feedback/monthly-review-input.json, its cited evidence and learning ledger at the same resolved Git revision. It reports six health dimensions as an owner snapshot, dated and stale after seven days; these fields cannot grant execution authority. Measured research/execution/token/cost sums include unknown-cycle counts and only the reporting month. Costs are not inferred from tokens. At most three bounded decision proposals carry no-response defaults. Missing input/evidence or invalid effort yields a degraded packet while retaining receipt totals. The existing 3,900-character digest bound still applies. Daily and weekly collection remain unchanged. It is not an independent live health observer.

The October fixture uses the built image with network disabled and a future clock. It is not an October production run or completed monthly review. The first actual packet is due October 1.

Deploy only this worker's exact reviewed digest through the existing GitOps source and ArgoCD path; preserve the API and event-router images and all non-image fields. The current deployment is 428f2d7, grant-aware base 713bddad…; use that exact revision as the rollback target, acknowledging that rollback restores broken domain reporting. A rollback must preserve repository grants. Verify resolver and grant hashes, all three definitions, a report-only scheduled smoke and monthly sink after rollout. Do not retry the failed historical run as though it had succeeded naturally.

The owner's release runbook requires 24 healthy hours and authenticated release authority. The domain failure invalidates continuous-health claims; no admitted routine identity exists. The user granted exact attended repair exception abdea390-1039-4da1-b129-33aad00b4396; promotion and all three live scheduled sink smokes subsequently passed. See live proof. That exception does not establish standing release authority. No access/budget/policy expansion is part of the repair.