[test-qa-hourly] test: guard attribution interval endpoint

[i-xtsu-sixyou-ken-mei]

QA scope

Continuous Test QA run from current main (5271fbc07aaafd62a88cb8675ead4171f4e45f96).

Inspection

• Recent commits, including merged QA PR [test-qa-hourly] test: cover impressions reach fallback #440 and integrated Control Center changes.
• Open PRs: none at run start, so a new ongoing QA branch was required.
• Latest main CI run 34491406884: quick-gates, code-quality, tests-related gates, security, coverage and deploy gating completed successfully.
• .github/workflows/ci.yml gate composition.
• scripts/coverage-summary.ts coverage aggregation path.
• social-attribution.ts interval endpoint selection and existing tests.

Change

Added one focused regression test proving computePostActivity() never leaks an observation captured after the snapshot interval into the current interval. With a 100-view baseline, 140-view endpoint at the interval boundary, and a later 1,000-view observation, the expected delta remains 40.

No production code, CI gates, docs, mocks, ignores, or thresholds changed.

Validation

• Focused local validation attempted via DevSpace, but the connector returned the transient account-connection error We couldn't connect your account. Please try again. No local files were modified through that environment.
• Authoritative PR CI will validate the committed test.

Commands / results

• GitHub main commit inspection: PASS
• GitHub open PR inspection: PASS (none open)
• Main CI run 34491406884 jobs inspection: PASS
• CI workflow inspection: PASS
• Coverage script inspection: PASS
• Focused local test: BLOCKED by transient DevSpace connector account error

Next QA target

After this PR is green, validate that standardized 24h growth metrics with collection_status='unavailable' cannot contribute to experiment samples24h, reach, or engagement. Current metrics24hForPostIds() filters only post ID and measurement_window === '24h', while the standardized query intentionally retains collection status.

[i-xtsu-sixyou-ken-mei]

Test QA run — CI-first repair

Inspection

• Reused this open Test QA PR and confirmed branch test-qa/ongoing-coverage-20260911 was already synced with main before the fix (ahead 1 / behind 0).
• Inspected recent main commits, open PR state, active PR patch, CI workflow, failed run #2512 (34496330787), quick-gates artifact, coverage artifact availability, and the new attribution-boundary test.
• CI #2512: code-quality, coverage, e2e, check-dead-env and other non-quick gates passed; quick-gates failed.
• Downloaded quick-gates-verify-1 artifact and inspected .ai-verify/logs/{format,repo,contracts}.log plus result.json.

Root cause

format was the only failed quick gate. Prettier required the helper signature in social-attribution-boundary.test.ts to be multiline. repo and contracts both passed.

Action

Commit 293c362a9a0df88ff8a8cbe8227cb0f000ac6f42 — test(control-center): format attribution boundary helper

Only the test helper formatting changed; no production code, CI gates, ignores, thresholds, docs, or mocks were modified.

Commands / results

• GitHub compare main...test-qa/ongoing-coverage-20260911: PASS (ahead 1 / behind 0 before fix)
• PR patch inspection: PASS (single added test file)
• CI run #2512 jobs inspection: FAIL isolated to quick-gates
• Download/inspect quick-gates-verify-1: PASS
  • format: FAIL — exact Prettier diff identified
  • repo: PASS
  • contracts: PASS
• CI workflow inspection: PASS (format repo contracts / type-check lint deadcode dup / test analytics unchanged)
• Local focused validation: BLOCKED by transient DevSpace connector account error (We couldn't connect your account. Please try again.)
• New-head GitHub CI: pending discovery/run creation at time of this update

Auto-merge

Not enabled while the repaired head is not yet proven green.

Next QA target

Once CI is green, validate that standardized 24h metrics with collection_status='unavailable' cannot contribute to experiment samples24h, reach, or engagement.

[i-xtsu-sixyou-ken-mei]

Hourly QA run — CI-first investigation

CI #2513 (34502732237) is still failing, so this run did not add another regression test.

Inspection

• Reused PR [test-qa-hourly] test: guard attribution interval endpoint #441 (test-qa/ongoing-coverage-20260911); open, non-draft, mergeable.
• Reviewed recent main commits; current main is 5271fbc07aaafd62a88cb8675ead4171f4e45f96.
• Reviewed open PRs: [test-qa-hourly] test: guard attribution interval endpoint #441 is the only open PR.
• Reviewed PR [test-qa-hourly] test: guard attribution interval endpoint #441 diff: still exactly one QA test file, apps/control-center/src/server/services/social-attribution-boundary.test.ts (59 additions, no production/config/docs changes).
• Reviewed CI #2513 jobs: code-quality, e2e, security, and check-dead-env pass. deploy-gates fails at Assert schedules match .github/schedules.json.
• Reviewed .github/workflows/ci.yml, scripts/check-schedules-registry.sh, .github/schedules.json, cron-failure-alert.yml, scheduled workflow inventory, and scripts/coverage-summary.ts.

Root cause

This is repository schedule-registry drift introduced on main, not a failure caused by the QA test. Commit 368116dd7e7655e63bd1078f919dfad30e3e7500 added .github/workflows/ops-operator.yml with workflow name Ops Operator and cron */5 * * * *. GitHub code search shows 9 scheduled workflow files, while .github/schedules.json still has only the previous 8 workflow-sourced cron entries; cron-failure-alert.yml also subscribes only to those previous 8 workflow names and omits Ops Operator.

check-schedules-registry.sh intentionally requires both invariants:

1. every workflow cron must exactly match a workflow-sourced registry row; and
2. every scheduled workflow name must exactly match the cron failure-alert subscriptions.

Therefore fixing #2513 correctly requires updating schedule operational configuration (.github/schedules.json and cron-failure-alert.yml, or otherwise changing the new Ops Operator schedule). Those files are outside this automation's allowed QA-change scope (tests / fixtures / minimal test utilities / minimal implementation fixes), so I am not weakening the gate or smuggling an unrelated config change into this test PR.

Commands / results

• PR metadata / active diff inspection: PASS
• Recent main commits inspection: PASS
• Open PR search: PASS
• CI #2513 workflow-run + job inspection: FAIL isolated to deploy-gates schedule-registry assertion
• Scheduled workflow code search: PASS — 9 scheduled workflows found
• .github/schedules.json inspection: PASS — Ops Operator missing
• cron-failure-alert.yml inspection: PASS — Ops Operator missing from subscriptions
• scripts/check-schedules-registry.sh inspection: PASS — confirms both exact-match checks
• Coverage aggregator inspection: PASS — Istanbul/Vitest + Cobertura/pytest inputs remain supported
• Focused local validation: BLOCKED by transient DevSpace account connection error (We couldn't connect your account. Please try again.)

Action

No code change this run: a safe CI repair is outside the permitted QA diff surface. PR #441 remains unchanged and should not be auto-merged while deploy-gates is red.

Next QA target

Once the repository schedule drift is repaired and CI is green, resume the queued data-quality target: verify that standardized 24h metrics with collection_status='unavailable' cannot contribute to experiment samples24h, reach, or engagement.

[i-xtsu-sixyou-ken-mei]

QA run update

Rechecked the current QA PR, latest main, CI jobs, the active diff, the scheduled workflow registry, and the newly-added Ops Operator workflow.

Result

No new QA code change was made in this run because CI is still blocked by the same repository-level schedule registry drift, not by the QA diff.

• PR head: 293c362a9a0df88ff8a8cbe8227cb0f000ac6f42
• CI run 34502732237 / #2513: tests, coverage, code-quality, quick-gates, e2e, security, and check-dead-env all PASS.
• Only deploy-gates fails at Assert schedules match .github/schedules.json.
• main has not advanced past 5271fbc07aaafd62a88cb8675ead4171f4e45f96, so there is no upstream fix to sync yet.
• .github/workflows/ops-operator.yml defines Ops Operator with cron */5 * * * *.
• .github/schedules.json still has no corresponding ops-operator workflow entry.
• PR diff remains exactly one QA test file (social-attribution-boundary.test.ts, +59/-0).

Commands / results

• Open PR search: PASS — reused [test-qa-hourly] test: guard attribution interval endpoint #441.
• PR metadata/diff inspection: PASS — open, non-draft, mergeable, QA-only diff.
• Recent main commit inspection: PASS — no newer main commit/upstream fix.
• PR CI run/job inspection: FAIL only in deploy schedule consistency gate; all test/quality/coverage/security gates PASS.
• .github/schedules.json inspection: confirms missing Ops Operator registration.
• .github/workflows/ops-operator.yml inspection: confirms scheduled cron exists.
• No code/config mutation performed because fixing the schedule registry/workflow alert configuration is outside this QA task's allowed change scope.

Auto-merge

Not enabled: required CI is red (deploy-gates). No protections were bypassed.

Next QA target

Once the repository-level schedule drift is fixed and CI is green, validate whether standardized 24h metrics with collection_status='unavailable' can incorrectly contribute to experiment samples24h, reach, or engagement; add the smallest regression test/minimal consumer guard only if reproduced.

[i-xtsu-sixyou-ken-mei]

Hourly QA follow-up

Rechecked the repository source of truth before making any new QA change.

• PR [test-qa-hourly] test: guard attribution interval endpoint #441 is still open, non-draft, mergeable, and remains QA-only: one test file (social-attribution-boundary.test.ts), +59/-0.
• Head 293c362a9a0df88ff8a8cbe8227cb0f000ac6f42 CI #2513 is still red only because deploy-gates fails at Assert schedules match .github/schedules.json; code-quality, E2E, security, and dead-env gates pass.
• main is still 5271fbc07aaafd62a88cb8675ead4171f4e45f96; there is no newer upstream commit to sync and no repository-side fix for the schedule-registry drift yet.
• The current PR diff was reviewed again and does not touch workflow/schedule config, so this failure remains unrelated to the QA test.

No code was changed this run. Fixing the current CI failure requires repository workflow/schedule-registry changes outside this QA automation's allowed scope, and weakening/bypassing the gate would be unsafe.

Commands / results

• Recent PR / main inspection — PASS
• PR diff review — PASS, QA-only
• CI #2513 jobs inspection — PASS for diagnosis; blocker unchanged (deploy-gates schedule registry assertion)
• Safe sync with main — no-op; branch base is still current main
• New focused test / broader validation — intentionally not run because CI-first blocker is unresolved

Auto-merge: not enabled; required CI is red.

Next QA target after CI is green: verify that standardized 24h rows with collection_status='unavailable' cannot contribute to experiment samples24h, reach, or engagement; add a regression test/minimal consumer guard only if repository evidence reproduces the issue.

[i-xtsu-sixyou-ken-mei]

Test QA run update

Inspection

• Rechecked PR [test-qa-hourly] test: guard attribution interval endpoint #441 metadata and diff scope: still open, non-draft, mergeable, 1 changed test file only (+59/-0).
• Rechecked head CI run 34502732237 / CI #2513.
• Rechecked latest main: still 5271fbc07aaafd62a88cb8675ead4171f4e45f96; no upstream commit has landed since the prior run.
• Rechecked open PRs: [test-qa-hourly] test: guard attribution interval endpoint #441 is still the only open PR.

CI status

CI is still blocked exclusively by deploy-gates → Assert schedules match .github/schedules.json. code-quality, E2E, security, and check-dead-env are green on this head. This remains the previously identified repository schedule-registry drift introduced by the Ops Operator scheduled workflow, not a failure caused by this QA test diff.

Action

No code change this run. The correct repository fix would require workflow/schedule-registry configuration changes outside this QA automation's allowed test/fixture/minimal-implementation scope. I did not weaken the gate, add an ignore, or create a noise test while CI is red.

Commands / results

• PR metadata inspection: PASS
• Head workflow run inspection: FAIL (deploy-gates schedule registry assertion only)
• CI jobs inspection: PASS for code-quality / E2E / security / check-dead-env; deploy-gates FAIL
• Open PR inspection: PASS ([test-qa-hourly] test: guard attribution interval endpoint #441 only)
• Recent main commits inspection: PASS; no new upstream fix to sync
• Branch sync: NOT NEEDED (main unchanged from PR base)
• Focused/broad validation: NOT RERUN because authoritative CI is blocked before a new QA behavior change is appropriate

Auto-merge

Not enabled: required CI remains red.

Next QA target

Once schedule-registry drift is fixed and CI returns green, validate that standardized 24h metrics with collection_status='unavailable' cannot contribute to experiment samples24h, reach, or engagement.