[test-qa-hourly] test: preserve legacy waitlist attribution

[i-xtsu-sixyou-ken-mei]

Intent

Continue continuous QA from current main with one small regression test for legacy waitlist attribution rows.

Inspection

• Previous QA PR [test-qa-hourly] test: pin deterministic metric snapshot tie-break #428 is merged; no open PR matches [test-qa-hourly], label test-qa, or branch prefix test-qa/ongoing-.
• Current main is cf2770f7ec55f7c9241c617f2e48a98ca3542b20 (merge of [test-qa-hourly] test: pin deterministic metric snapshot tie-break #428).
• Previous QA head 23632d7a4ccf606d00f353140a0ae60ab9db3067 CI #2443 (34324571944) completed successfully.
• Reviewed active PRs Clarify reliability priority scores and check counts #429, fix: keep visual fallbacks on topic and simplify scene trace #430, feat(control-center): visualize cross-channel growth journey #431, and fix(podcast): drain renders before Fly rollout #432; none overlaps this test path.
• Reviewed apps/control-center/src/server/services/waitlist-growth.ts: canonical attribution is keyed by social_publish_job_id; legacy jobs may have null language_code and/or social_post_id, and the read model intentionally maps those to unknown language and unavailable views without discarding attribution.
• Reviewed existing waitlist-growth tests: pagination, snapshot consistency, zero views, metric failure, orphaned attribution, batching, latest snapshot, and query contract are covered; the legacy null-language/null-social-post attribution path was not directly pinned.
• .github/workflows/ci.yml still runs quick gates (format repo contracts), code quality (type-check lint deadcode dup), tests (test analytics), E2E, security, and deploy gates.
• Coverage remains available through scripts/coverage.sh summary -> turbo run test:coverage + scripts/coverage-summary.ts.

QA change

Add waitlist-growth-legacy-language.test.ts to assert that a signup referencing a valid legacy social publish job remains socially attributed even when that job has no language_code and no social_post_id. The conversion must remain present with languageCode = unknown, null views/rate, and the signup must not be misclassified as direct/unknown traffic.

This is test-only. No implementation, workflow, docs, ignore, coverage threshold, or agent-instruction change is included.

Validation / commands and results

• GitHub fetch_commit_workflow_runs for previous QA head — PASS: CI #2443 completed successfully.
• Source review of waitlist-growth.ts — PASS: current implementation explicitly preserves canonical job attribution while mapping nullable legacy metadata safely.
• Existing regression test review — PASS: this exact legacy metadata case was not directly asserted.
• .github/workflows/ci.yml inspection — PASS: existing gates unchanged.
• scripts/coverage.sh inspection — PASS: coverage aggregation path remains available.
• Focused local test / broader test, lint, typecheck, coverage — not executable in this GitHub-connector runtime; PR CI is the executable repository validation gate.
• Updated-head PR CI — pending after PR creation.

Scope guard

Changed file only:

• apps/control-center/src/server/services/waitlist-growth-legacy-language.test.ts

Auto-merge

Will only be enabled if repository settings and required checks/reviews permit it and the updated-head diff remains QA-only.

Next QA target

Inspect concurrent mutation during social_post_metrics offset pagination. The smallest useful next invariant is detecting duplicate metric row identities across adjacent pages so a moving snapshot degrades views instead of silently accepting stale pagination.

[i-xtsu-sixyou-ken-mei]

QA run — 2026-09-09 18:34 JST

Inspection

• Reused open QA PR [test-qa-hourly] test: preserve legacy waitlist attribution #433 (test-qa/ongoing-coverage-20260909, label/title contract already satisfied).
• Previous head 0cd0ea42809fa31f0ec139a8709ee5685f3a626b: CI #2446 (34329575886) PASS.
• main remains cf2770f7ec55f7c9241c617f2e48a98ca3542b20; compare shows QA branch ahead 3 / behind 0.
• Reviewed recent commits, PR [test-qa-hourly] test: preserve legacy waitlist attribution #433 diff, waitlist-growth.ts, existing metrics pagination regressions, .github/workflows/ci.yml, and scripts/coverage.sh.
• Reviewed active PR changed paths for Clarify reliability priority scores and check counts #429/fix: keep visual fallbacks on topic and simplify scene trace #430/feat(control-center): visualize cross-channel growth journey #431/fix(podcast): drain renders before Fly rollout #432; none overlaps the waitlist-growth service/test paths changed here.
• CI contract remains quick gates (format repo contracts), quality (type-check lint deadcode dup), tests (test analytics), E2E/security/deploy gates. Coverage remains turbo run test:coverage + scripts/coverage-summary.ts.

Action

CI was green, so this run strengthened one regression-prone behavior: offset pagination can repeat the same social_post_metrics.id across adjacent pages if rows move during collection. Previously that duplicate was silently accepted and could produce a stale-but-plausible 24h view snapshot.

Added a focused regression in waitlist-growth-metrics-pagination.test.ts that makes metric-500 appear at the end of page 1 and again on page 2. Added the minimal implementation guard required by the newly exposed behavior: readViews tracks metric row IDs and throws on a duplicate. The existing caller already catches view-read failures, so persisted signup truth remains available while views24h / signupRate degrade to null with the existing telemetry message.

Commits this run:

• eaa20d94b7ae56e4d6af0b0370808347e2ccd063 — fix(control-center): fail closed on duplicate metric pages
• b5fa9b2ce4d0c661141eaf3f8647b0b2b6b256ab — test(control-center): detect duplicate metric pagination

Commands / results

• GitHub fetch_commit_workflow_runs(0cd0ea4...) — PASS, CI #2446 completed successfully.
• GitHub recent commit inspection — PASS, main unchanged since PR creation.
• GitHub compare_commits(main, test-qa/ongoing-coverage-20260909) — PASS, ahead 3 / behind 0; changed files limited to two tests + minimal service guard.
• GitHub active PR changed-file inspection (Clarify reliability priority scores and check counts #429–fix(podcast): drain renders before Fly rollout #432) — PASS, no path overlap.
• GitHub source review of waitlist-growth.ts and existing metrics pagination tests — PASS, duplicate metric row identity was not previously detected.
• GitHub .github/workflows/ci.yml inspection — PASS, gates unchanged.
• GitHub scripts/coverage.sh inspection — PASS, coverage path unchanged.
• Focused local test / broader test, lint, typecheck, coverage — not executable in this connector runtime; repository CI is the executable validation surface.
• Updated-head CI #2448 (34335664795) — IN PROGRESS at time of this update.

Scope guard

Current PR diff is only:

• apps/control-center/src/server/services/waitlist-growth-legacy-language.test.ts
• apps/control-center/src/server/services/waitlist-growth-metrics-pagination.test.ts
• apps/control-center/src/server/services/waitlist-growth.ts — 5-line minimal fail-closed guard

No workflow/docs/agent/ignore/coverage-threshold/refactor changes.

Auto-merge

Not enabled. Repository setting is currently allow_auto_merge=false; updated-head CI is also still running. No protections will be bypassed.

Next QA target

The duplicate-row guard catches one observable form of a moving offset snapshot. Next inspect the complementary missing-row case (concurrent delete/insert shifts rows without duplication) and whether a bounded count/checksum-style consistency guard is possible without turning telemetry reads into a heavier transactional path.

[i-xtsu-sixyou-ken-mei]

QA run update — 2026-09-09

Inspection

• Previous QA head b5fa9b2ce4d0c661141eaf3f8647b0b2b6b256ab — CI #2448 (34335664795) PASS.
• main remains cf2770f7ec55f7c9241c617f2e48a98ca3542b20; QA branch is ahead 5 / behind 0 after this run.
• PR [test-qa-hourly] test: preserve legacy waitlist attribution #433 is open, non-draft, labeled test-qa, and remains the reusable QA PR.
• Active PR diffs Clarify reliability priority scores and check counts #429–fix(podcast): drain renders before Fly rollout #432 were re-inspected; none changes waitlist-growth.ts or the waitlist-growth QA test paths.
• .github/workflows/ci.yml still runs quick gates (format repo contracts), code quality (type-check lint deadcode dup), tests (test analytics), E2E, security, and deploy gates.
• Coverage remains scripts/coverage.sh summary -> turbo run test:coverage -> scripts/coverage-summary.ts.

Finding

The duplicate-row guard added in the previous run catches the most common offset-pagination shift, but a moving dataset can also return a different unique row that sorts before the last row of the previous page. Without an ordering invariant, that backward page movement is silently accepted and can overwrite a newer 24h view with stale telemetry.

Change

• Added waitlist-growth-metrics-reorder.test.ts to simulate page 1 ending at 14:00:00 and page 2 returning a unique metric from 13:59:59.
• Added the minimal implementation guard in readViews() to require strictly increasing (captured_at, id) order within each 100-post metric batch.
• On backward/reordered pagination, views fail closed through the existing telemetry-only catch: persisted signup counts and attribution remain available while views24h / signupRate degrade to null.
• The sort cursor is reset per JOB_BATCH_SIZE batch so independent batches are not incorrectly compared against each other.

Commits:

• 34444d70aafbfa173e480728abb5d549387c2dac — fix(control-center): reject reordered metric pages
• aa6eabd060edd6289d8289d385802532fa163013 — test(control-center): reject reordered metric pagination

Commands / results

• fetch_commit_workflow_runs b5fa9b2... — PASS, CI #2448 successful.
• compare main...test-qa/ongoing-coverage-20260909 — PASS, ahead 5 / behind 0; four changed files total.
• fetch_file apps/control-center/src/server/services/waitlist-growth.ts — reviewed pagination, batching, ordering, and fail-closed behavior.
• fetch_file apps/control-center/src/server/services/waitlist-growth-metrics-pagination.test.ts — reviewed existing 500-row, tie-break, and duplicate-page regressions.
• list_pr_changed_filenames for Clarify reliability priority scores and check counts #429/fix: keep visual fallbacks on topic and simplify scene trace #430/feat(control-center): visualize cross-channel growth journey #431/fix(podcast): drain renders before Fly rollout #432 — PASS, no target-path overlap.
• fetch_file .github/workflows/ci.yml — PASS, gates unchanged.
• fetch_file scripts/coverage.sh — PASS, coverage path unchanged.
• Focused local Vitest / broader test, lint, typecheck, coverage — not executable in this connector-only runtime; GitHub Actions is the executable validation surface.
• Updated head aa6eabd060edd6289d8289d385802532fa163013 — CI #2450 (34340739395) IN PROGRESS.

Scope guard

Current PR diff remains limited to Control Center waitlist-growth tests plus the minimal readViews() consistency guards. No workflow, docs, agent instructions, ignores, coverage thresholds, or unrelated refactors were changed.

Auto-merge

Not enabled: repository setting allow_auto_merge=false; updated-head CI is also still running. No protection bypass attempted.

Next QA target

Investigate the remaining missing-row form of offset-pagination mutation (delete/insert causing a skipped row without duplicate or backward order). Determine whether a cheap count/cursor invariant can detect it without turning telemetry reads into a heavyweight transactional snapshot.

[i-xtsu-sixyou-ken-mei]

Hourly QA — CI-first repair

Inspection

• Reused this open QA PR ([test-qa-hourly], test-qa/ongoing-coverage-20260909).
• Branch is synced with main: compare before this repair was ahead 5 / behind 0, merge base cf2770f7ec55f7c9241c617f2e48a98ca3542b20.
• Previous head aa6eabd060edd6289d8289d385802532fa163013 ran CI #2450 (34340739395) and failed.
• CI job inspection: coverage, code-quality, e2e, and security passed; the failing job was quick-gates.
• Downloaded and inspected artifact quick-gates-verify-1 (10099650803): repo PASS, contracts PASS, format FAIL.
• Exact formatter diff was only EOF normalization in apps/control-center/src/server/services/waitlist-growth.ts: No newline at end of file -> canonical trailing newline.
• Reviewed active PR changed paths for Clarify reliability priority scores and check counts #429, fix: keep visual fallbacks on topic and simplify scene trace #430, feat(control-center): visualize cross-channel growth journey #431, fix(podcast): drain renders before Fly rollout #432. None overlaps the waitlist-growth QA files in this PR. (fix(podcast): drain renders before Fly rollout #432 changes other Control Center podcast-cost files plus deploy/migration files.)
• Reviewed .github/workflows/ci.yml: quick gates remain format repo contracts; code quality remains type-check lint deadcode dup; tests remain test analytics; E2E/security/deploy gates remain present.
• Reviewed scripts/coverage.sh: coverage remains turbo run test:coverage followed by scripts/coverage-summary.ts.
• Current PR changed paths remain limited to three waitlist-growth regression test files plus the minimal waitlist-growth.ts consistency guard.

Action

Applied only the formatter-required EOF normalization to waitlist-growth.ts.

Commit: fdf9ce26d74bf227485c84e0b06aa93094a5468f — fix(control-center): restore canonical file ending

No behavior, tests, workflow gates, docs, ignores, mocks, or coverage thresholds were changed in this repair.

Commands / results

Repository-authoritative checks used this run:

• fetch_commit_workflow_runs(aa6eabd...) — FAIL, CI #2450.
• CI jobs inspection — PASS for coverage/code-quality/E2E/security; FAIL quick-gates only.
• Download + inspect quick-gates-verify-1 — format FAIL, repo PASS, contracts PASS.
• Formatter artifact diff — only missing final newline in waitlist-growth.ts.
• compare main...test-qa/ongoing-coverage-20260909 — ahead 5 / behind 0 before repair.
• Active PR changed-file inspection Clarify reliability priority scores and check counts #429–fix(podcast): drain renders before Fly rollout #432 — no QA target-path overlap.
• Updated-head CI #2451 (34346015012) — QUEUED for fdf9ce26d74bf227485c84e0b06aa93094a5468f.

Local focused/broader commands are not executable in this connector runtime; GitHub Actions is the executable repository validation surface.

Auto-merge

Not enabled. Updated-head CI is not green yet; do not bypass protections.

Next QA target

After CI is green, inspect a low-cost invariant for missing-row mutation during social_post_metrics offset pagination (row shift without duplicate/backward ordering), rather than adding more pagination guards while CI is red.

[i-xtsu-sixyou-ken-mei]

QA run update

Inspection / source of truth:

• Reused open QA PR [test-qa-hourly] test: preserve legacy waitlist attribution #433 (test-qa/ongoing-coverage-20260909, label/title scope remains appropriate).
• Current base main: cf2770f7ec55f7c9241c617f2e48a98ca3542b20.
• Previous QA head fdf9ce26d74bf227485c84e0b06aa93094a5468f: GitHub Actions CI #2451 (34346015012) completed successfully, so this run did not make a CI repair.
• Reviewed recent/open PRs Clarify reliability priority scores and check counts #429-[test-qa-hourly] test: preserve legacy waitlist attribution #433, current PR diff, waitlist-growth.ts, existing waitlist batching/pagination/legacy attribution tests, CI workflow contract, and coverage script path already documented on this PR.
• Deliberately did not add another speculative missing-row pagination guard: offset deletes/inserts can skip a row without a reliable local invariant, and forcing a heuristic would risk overbuilding telemetry behavior.

Change this run:

• Commit c039afe29ce98d9c0466f447c807ed7395a1ce3b — test(control-center): pin seven-day attribution boundary.
• Strengthened waitlist-growth-batching.test.ts with a boundary regression: with now = 2026-09-09T00:00:00Z, a canonical social signup at exactly 2026-09-02T00:00:00Z must be included in both signups7d and attributedSocial7d; an otherwise identical signup 1 ms earlier remains in the all-time conversion count but outside the 7d attribution window.
• This pins the DB .gte(created_at, since7d) inclusive contract to the in-memory Date.parse(created_at) >= Date.parse(since7d) aggregation, preventing a one-boundary-row split between persisted 7d counts and attribution.

Validation / commands and results:

• GitHub Actions lookup for previous head fdf9ce26... — PASS, CI #2451 conclusion success.
• fetch_pr_patch on [test-qa-hourly] test: preserve legacy waitlist attribution #433 before/after change — PASS, new diff is test-only for this run; pre-existing minimal consistency implementation changes remain unchanged.
• fetch_file apps/control-center/src/server/services/waitlist-growth.ts@branch — PASS, confirmed inclusive .gte query and >= since7d attribution semantics.
• Code search for existing direct 7d-boundary coverage — no existing direct assertion found; current batching coverage covered 100-ID and 500-row boundaries instead.
• Focused local Vitest / broader test, lint, typecheck, coverage — not executable in this connector runtime; updated-head GitHub CI is the repository-authoritative executable validation.

Auto-merge:

• Not enabled. Repository policy previously reports allow_auto_merge=false; never bypassing repository settings/protections. Updated-head CI must also complete first.

Next QA target:

• After CI is green, inspect canonical social_publish_jobs read consistency/error behavior for a small user-visible attribution invariant (especially missing canonical job rows vs direct/unknown classification) before adding more metric pagination machinery.