v0.4.0 — Production-readiness pack

v0.4.0 — Production-readiness pack

This release focuses on correctness and operability: end-of-cycle parity, safe replays for late/backfilled events, first-class metrics/dashboards/alerts, and a concrete operator runbook.

Highlights

• Test Clocks parity demo — prove end-of-cycle invoice parity (drift ≈ 0 within ε).
• Replay API — POST /v1/replay with dry-run/apply and watermark/cursor semantics (safe, idempotent reprocessing).
• Shadow Mode (per meter) — push to Stripe test with deterministic idempotency keys; live invoices unaffected.
• Observability pack — /metrics labels, Prometheus scrape example, Grafana dashboard JSON.
• ALERTS.md — PromQL recipes (drift, p95 latency, queue lag, 5xx) + Alertmanager sample.
• RECONCILIATION.md — triage & repair runbook; copy-paste commands and drift report.

Try in 5 minutes → Verify in 30 seconds

git clone https://github.com/geminimir/stripemeter && cd stripemeter
cp .env.example .env && docker compose up -d && pnpm -r build && pnpm dev
curl -fsS http://localhost:3000/health/ready | jq . || true
TENANT_ID=$(uuidgen 2>/dev/null || cat /proc/sys/kernel/random/uuid) bash examples/api-calls/send.sh
curl -fsS http://localhost:3000/metrics | head -n 30  # duplicate counted once

Baseline (laptop) numbers

• p95 ingest: ~10–25 ms
• Re-aggregate 10k late events: ≤ 2 s
• Idempotency window 24 h: 0 double-counts

Upgrade notes

• Replay via API: POST /v1/replay (dry-run → apply). See README and docs/api/replay.md.
• Grafana dashboard JSON provided under ops/grafana/.
• See ALERTS.md and RECONCILIATION.md for production guardrails.

Thanks

Thanks to our contributors @coryli and @louispy!
If you can share attribution: “We caught a ~3% drift before invoice close.” — <team/company>

Links:

• README banner and quick demo at the top of the repo
• Full notes: docs/RELEASE_NOTES_v0.4.0.md