feat(dogfood): in-process runtime regression gate

[os-zhuang]

Why

Static gates — build, unit tests, spec-liveness, CodeQL — verify each layer in isolation, usually against mocks. They cannot catch a break that only appears when the real engine + strategies + services + HTTP context run together.

#2018 ("org timezone drives analytics date bucketing") is the canonical case: it was green on every static gate — 900+ unit tests included — because each of its three broken seams (analytics strategy routing, in-memory count, REST execution-context) was individually correct and individually mocked. The break was only visible by booting the app and comparing a date bucket under UTC vs a non-UTC org timezone. Unit tests that mock the protocol/server (e.g. rest.test.ts) structurally cannot catch it.

What

@objectstack/dogfood (private, never published): a harness that boots real example apps in-process against in-memory SQLite — wired with the same service plugins objectstack dev loads — and exercises them through the real HTTP surface via Hono request-injection (no ports, no sockets, CI-stable). Tests act as a browser client: sign in, hit /api/v1/..., assert on real responses.

• src/harness.ts — bootDogfoodStack(config) → { kernel, api, raw, signIn, apiAs, stop }. Registers ObjectQL + SQLite + HonoServer + App + Settings + Analytics + Auth + Security + REST + Dispatcher, runs kernel.bootstrap() (seed + dev-admin + route registration), and injects requests through the live Hono app.
• test/analytics-timezone.dogfood.test.ts — golden regression for ADR-0053 Phase 2 · Slice 5: timezone-aware analytics date bucketing #1982/fix(analytics): make organization timezone drive date-dimension bucketing (#1982) #2018. Inserts leads at a tz-boundary instant (2024-03-01T03:00:00Z), sets the org timezone through the real settings route, and asserts the date bucket shifts (UTC → 2024-03-01, America/Los_Angeles → 2024-02-29) with the correct count.
• CI — new Dogfood Regression Gate job (gated on the core paths filter); also runs under turbo run test.

Proof it's a real gate

Reverting just one of the three #2018 seams (NativeSQLStrategy routing) and rebuilding turns the LA assertion red (expected undefined to be 3 — the bucket didn't shift). Restoring it → green. A gate that stays green on the bug is not a gate; this one fails on it.

Boot-to-assert ≈ 2s. Private package — no changeset.

Notes

• Pre-existing (out of scope): example apps' standalone tsc --noEmit is already red (they import Theme/Webhook/etc. from @objectstack/spec root, which the built dist re-exports only via subpaths). CI only type-checks @objectstack/spec, so this gate is unaffected.
• Next dogfood golden tests (follow-ups): array-field writes (fix: array field types persist + Field.time accepts time-of-day #2004), form-section crash, member-edit RLS (fix(security)[P0]: enforce RLS on by-id writes — close member-edits-others'-records hole (#1985) #1994).

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
spec	Ready	Preview, Comment	Jun 18, 2026 5:31am

[github-actions]

📓 Docs Drift Check

No hand-written docs reference the 1 changed package(s). ✅