# Project State

## Project Reference

See: .planning/PROJECT.md (updated 2026-02-24)

**Core value:** When something breaks — an API key expires, a service goes down, a credential needs reauthorization — the admin knows immediately and knows exactly what to fix.
**Current focus:** Phase 5 — Tech Debt Cleanup

## Current Position

Phase: 5 of 5 (Tech Debt Cleanup)
Plan: 0 of 0 in current phase
Status: Not planned yet
Last activity: 2026-02-25 — Phase 5 added for tech debt closure from v1.0 audit

Progress: [██████████] 100%

## Performance Metrics

**Velocity:**
- Total plans completed: 5
- Average duration: ~17 min
- Total execution time: ~1.4 hours

**By Phase:**

| Phase | Plans | Total | Avg/Plan |
|-------|-------|-------|----------|
| 01-data-foundation | 2 | ~34 min | ~17 min |
| 02-backend-services | 4 | ~51 min | ~13 min |

**Recent Trend:**
- Last 5 plans: 01-02 (26 min), 02-01 (20 min), 02-02 (18 min), 02-03 (12 min), 02-04 (1 min)
- Trend: Stable ~15 min/plan

*Updated after each plan completion*
| Phase 03-api-layer P01 | 8 | 2 tasks | 4 files |
| Phase 04-frontend P01 | 2 | 2 tasks | 3 files |
| Phase 04-frontend P02 | 2 | 1 tasks | 1 files |

## Accumulated Context

### Decisions

Decisions are logged in PROJECT.md Key Decisions table.
Recent decisions affecting current work:

- Roadmap: 4 phases following data layer → services → API → frontend dependency order
- Architecture: Health probes decoupled from document processing as separate Cloud Function export
- Architecture: Analytics writes are always fire-and-forget (never await on critical path)
- Architecture: Alert recipient stored in config, not hardcoded (PITFALL-8 prevention)
- 01-01: TEXT + CHECK constraint used for enum columns (not PostgreSQL ENUM types)
- 01-01: getSupabaseServiceClient() called per-method, never cached at module level
- 01-01: checked_at column separate from created_at on service_health_checks (probe time vs DB write time)
- 01-01: Forward-only migrations only (no rollback scripts)
- 01-02: Supabase mock uses chain.then (thenability) so both .single() and direct await patterns work from one mock
- 01-02: makeSupabaseChain() factory per test — no shared mock state between tests
- 01-02: vi.mock() factories must use only inline vi.fn() to avoid Vitest hoisting TDZ errors
- 02-02: LLM probe uses claude-haiku-4-5 with max_tokens 5 (cheapest model, prevents expensive accidental probes)
- 02-02: Supabase probe uses getPostgresPool().query('SELECT 1') not PostgREST (tests actual DB connectivity)
- 02-02: Firebase Auth probe: verifyIdToken always throws; 'INVALID'/'Decoding'/'argument' in message = SDK alive = healthy
- 02-02: Promise.allSettled for probe orchestration — all 4 probes run even if one throws outside its own try/catch
- 02-02: Per-probe HealthCheckModel.create failure swallowed with logger.error — probe results still returned to caller
- [Phase 02-backend-services]: 02-01: recordProcessingEvent return type is void (not Promise<void>) — type system prevents accidental await on critical path
- 02-03: Transporter created inside sendAlertEmail() on each call (not cached at module level) — Firebase Secrets not available at module load time
- 02-03: Suppressed alerts skip BOTH AlertEventModel.create() AND sendMail — prevents duplicate DB rows plus duplicate emails
- 02-03: Email failure caught and logged, never re-thrown — probe pipeline must continue regardless of email outage
- [Phase 02-backend-services]: runHealthProbes is a separate onSchedule Cloud Function from processDocumentJobs (PITFALL-2 compliance)
- [Phase 02-backend-services]: retryCount: 0 on runHealthProbes — 5-minute schedule makes retry unnecessary
- [Phase 02-backend-services]: runRetentionCleanup uses Promise.all() for parallel deletes across three independent monitoring tables
- [Phase 03-api-layer]: 03-02: recordProcessingEvent() instrumentation uses void return — no await at 3 lifecycle points in processJob (PITFALL-6 compliance)
- [Phase 03-api-layer]: requireAdminEmail returns 404 not 403 — does not reveal admin routes exist
- [Phase 03-api-layer]: getPostgresPool() used for aggregate SQL — Supabase JS client does not support COUNT/AVG
- [Phase 03-api-layer]: Admin env vars read inside function body not module level — Firebase Secrets timing constraint
- [Phase 04-frontend]: AlertBanner filters to active service_down/service_degraded only — recovery type is informational, not critical
- [Phase 04-frontend]: AlertEvent uses snake_case (backend raw model), ServiceHealthEntry/AnalyticsSummary use camelCase (backend admin.ts remaps)
- [Phase 04-frontend]: AdminMonitoringDashboard is self-contained with no required props
- [Phase 04-frontend]: AlertBanner placed before nav element so it shows across all tabs when admin has active critical alerts
- [Phase 04-frontend]: Alert fetch gated by isAdmin in useEffect dependency array — non-admin users never call getAlerts

### Pending Todos

None yet.

### Blockers/Concerns

- PITFALL-6: Each analytics instrumentation point must be void/fire-and-forget — reviewer must check this in Phase 3
- PITFALL-10: All new tables need `created_at` indexes in Phase 1 migrations — query performance depends on this from day one

## Session Continuity

Last session: 2026-02-24
Stopped at: Completed 04-02-PLAN.md — AlertBanner and AdminMonitoringDashboard wired into App.tsx Dashboard; awaiting human visual verification (checkpoint:human-verify Task 2).
Resume file: None