Add SLI/SLO definitions and burn-rate alerts for Tier-1 services

[bdevierno1]

Summary

• docs/slo-policy.md — In-repo SLO policy: service tier classification (all three services are Tier 1 per the org policy; no Tier-0 payments/auth exists in this app), SLI definitions sourced from the existing /api/metrics Prometheus endpoint (vecna_operations_total), SLO targets (vecna-api 99.9 %, vecna-agent 99.0 %, vecna-frontend 99.9 %), error-budget table, Green/Yellow/Red band thresholds, operational constraints, and review triggers.
• monitoring/alerts/slo-burn-rate.yml — Prometheus multi-window burn-rate rules for vecna-api and vecna-agent. Six recording rules (5 m, 30 m, 1 h, 6 h, 3 d error-rate windows) plus 6 alert rules covering all three budget states: Red (≥ 14.4 × / 1 h + 5 m guard), Yellow (≥ 6 × / 6 h + 30 m guard), Green-elevated (≥ 3 × / 3 d + 6 h guard). Every alert links to the runbook.
• docs/runbooks/slo-burn-rate-runbook.md — Response runbook: alert reference table, diagnosis commands against SQLite and /api/health/deep, per-severity actions (Red = page + freeze deploys, Yellow = business-hours page + ticket, Green = sprint ticket).

What is not in scope

• Latency SLI (vecna_operation_duration_seconds histogram): the backend does not currently emit duration histograms. Noted in slo-policy.md §2.1; tracked for follow-up.
• vecna-frontend + vecna-desktop alert rules: both require a Blackbox Exporter probe configured in the Prometheus scrape config, which is out of scope for this PR. SLO targets for both are defined in the policy doc.
• Prometheus scrape config / prometheus.yml: not present in this repo. The monitoring/alerts/slo-burn-rate.yml is a drop-in rule_files: entry or Kubernetes PrometheusRule CR.

Test plan

• grep -c '^+' <(git diff main...) — confirm ≤ 400 added lines (actual: 385).
• Paste monitoring/alerts/slo-burn-rate.yml into Prometheus rule linter or run promtool check rules monitoring/alerts/slo-burn-rate.yml — should pass with no errors.
• Confirm every alert annotation contains runbook_url pointing to docs/runbooks/slo-burn-rate-runbook.md.
• Review docs/slo-policy.md against org policy (file_id a86fb303) — targets and budget minutes match pinned memory.

Rollback

Delete the three new files and the monitoring/ directory. No existing code changed.

---

🤖 Generated with Vecna

Co-Authored-By: Vecna noreply@vecna.ai