# Worker integration acceptance — 2026-10-28 Source baseline: develop `df5584d21461f68eb05ffb59ae79ddde4acc622e`, including backend commits 6102556 and 39f5b70. Existing graph-cache timestamp, .specify/ and evidence/ changes were not owned and are excluded. Sibling consumer work was excluded. Graph navigation: copied graph.json to `/opt/data/cache/scratch/seo-http-impact`, queried `SeoTrackingController SeoTrackingService authentication snapshots worker`. Original SHA256 6a73473cee9e207543fdeeb668bea544f766200a1b16b8a3a9ce18d4e42e51af; source binding unknown. Graph is navigation only; directly inspected module/controller/service, PrismaService, tests and worker contract at baseline. ## Executed - Python `python3 -m unittest discover -v`: 17 tests pass, exit 0. - Backend `SEO_TEST_DATABASE_URL=postgresql://seo_test@127.0.0.1:55439/postgres npm test -- --runInBand seo-tracking`: 4 suites, 16 tests pass, exit 0. Receipt in evidence/real-http-receipt.json and focused output in evidence/backend-focused.txt. - `npm run build`: exit 0. - Full backend `npm test -- --runInBand` without integration env: 94 suites pass, 1 fails, 4 skip; 815 tests pass, 1 fails, 8 skip. Failure is existing WikiService pagination assertion, previously reproduced on clean baseline by backend owner; see backend/src/seo-tracking/VERIFICATION.md. Full output in evidence/backend-full.txt. This is NOT full-suite green. - Spec Kit 1.2.3 prerequisite script with explicit feature directory, --require-spec --require-tasks --include-tasks: exit 0. Artifact presence only, not full acceptance. New integration test imports the actual SeoTrackingModule, replaces only PrismaService with a real PrismaClient pointed at disposable PostgreSQL, uses a local Passport JWT strategy with actual shared JwtAuthGuard/RolesGuard, and binds loopback ephemeral TCP. Python calls real requests, creates finding, patches feedback (version 1), rejects stale feedback action (409), links correct action (version 2), reads worker findings, makes actual read-only Google requests, persists SQLite state, closes/reopens state, delivers snapshot through real HTTP, GETs database history, retries identical POST (same ID), rejects changed payload (409), and verifies one history row. Missing/invalid JWT and customer role are denied. Fixture blog/finding and cascading snapshots are cleaned up. This is test-auth, NOT production user lookup/Redis/API-key acceptance. Actual Google result for existing published blackseed article, 2026-10-12..25: HTTP 200 GSC 5 clicks / 32 impressions / position 3.4375; query rows empty (limited coverage). HTTP 200 GA4 2 active users / 33 engagement seconds / 16.5 seconds per user / bounce 0.5; timezone America/Los_Angeles. Scroll, commerce and Clarity remain null/unavailable. No causal sales or automated scaling claim. First integration run failed because pg_ctl defaulted to port 5432, not the allowed 55439. Restarted this known disposable cluster with explicit loopback port 55439 and scratch Unix socket; rerun passed. This was test setup failure, NOT behavioral RED or evidence of a contract bug. No worker/backend contract mismatch was found, so no production implementation fix or claimed new RED/GREEN cycle. New integration assertions are acceptance coverage of existing code. ## Reproduce Provision a disposable PostgreSQL 16 schema from backend/prisma/schema.prisma (never production) at the exact test URL. Then from backend: `SEO_TEST_DATABASE_URL=postgresql://seo_test@127.0.0.1:55439/postgres SEO_HTTP_RECEIPT=/absolute/scratch/receipt.json npm test -- --runInBand seo-tracking.worker-http.spec.ts` Requires Python requests/google-auth and existing read-only Google credential path used by worker; never copy credentials into Git. This opt-in test makes Google reads and only local test-database writes. No cron or deployment exists. ## Still open Actual publisher consumers/finalization enforcement, many-actions/goals/history requirements, production identity/API-key, browser/full frontend matrix, real department alert receipt, Clarity reporting, event instrumentation and causal commerce attribution. Full release acceptance remains open; no push/CI/Stage/deployment/production writes authorized or performed. Broader spec tasks intentionally remain unchecked rather than conflating partial evidence with closure.