Roadmap

Ordered by priority — items higher up are nearer-term.

Phase Overview

Phase	Focus	Status
1 — Detection quality & UX	Enrichment, links, readiness	✅ Done
2 — ML maturity	Feature vectors, workload patterns, forecast	🟡 Partial
3 — Operational maturity	Replay mode, dashboards, feedback	🚧 In progress
4 — Observability hardening	Instrumentation fixes, cardinality	🎯 Next
5 — Cluster readiness	Leader election, deploy, real alerts	🔜 Planned
6 — Self-monitoring	PrometheusRules, dashboards, OTel SDK	🔜 Planned

---

✅ Phase 1 — Detection Quality & UX (Done)

• P1.1 — Label-based pivot (anomaly enrichment) with identity extraction, template substitution, bounded concurrency, Redis-backed cache
• P1.2 — Alert payload with Grafana/Tempo/Loki/Runbook deep links anchored at anomaly timestamp
• P1.3 — Complete readiness checks (/readyz probes Prometheus, Loki, Alertmanager, ML)

🟡 Phase 2 — ML Maturity

• P2.1 — Fix ML multivariate (proper feature vector per correlated alert)
• P2.4 — Workload-aware correlation (sibling pod detection, ≥3 pods → workload alert)
• P2.5 — ML feature: replica_anomaly_fraction (blocked on P2.4 prod validation)
• P2.2 — Wire ML Forecast (Prophet) into detection cycle
• P2.3 — Multivariate per-namespace mode

🚧 Phase 3 — Operational Maturity

• P3.1 — Replay mode (16/16 tasks done) — see Replay Mode
• P3.2 — Top noisy workloads dashboard / PrometheusRule — see Monitoring
• P3.3 — Feedback loop (Slack reactions → precision/recall tracking)
• P3.4 — SLO-aware severity adjustment

🎯 Phase 4 — Observability Hardening (Prerequisite for Deploy)

• P4.A.1 — Fix instrumentation bugs — done in 0.7.0 (counter before dry-run, gauge on tick, custom histogram buckets)
• P4.A.2 — Cardinality cleanup — done in 0.7.0 (no identity label on any metric; bounded workload in AM labels)
• P4.A.3 — Multi-cluster constant labels — done in 0.7.0 (WrapRegistererWith{cluster}, eks_cluster at scrape layer); superseded in 0.9.0 — cluster moved to the scrape layer too (ServiceMonitor externalLabels), app emits no cluster identity at all now
• P4.A.4 — Dashboard refresh

🔜 Phase 5 — Cluster Readiness

• P5.1 — K8s Lease leader election (internal/leader/, configurable via controller.leader_election.enabled)
• P5.2 — Deploy to cluster (validate IRSA, ApplicationSet)
• P5.3 — Remove --dry-run, validate real alerts
• P5.4 — Cardinality guard (self-protection)
• P5.5 — Agent API Integration (staffops-chaitops) — invoke Agent API on high-confidence anomalies for automated squad investigation; circuit breaker, bounded concurrency (max 5), Redis dedup. Blocked on P5.3. Spec

🔜 Phase 6 — Self-Monitoring

• P6.1 — PrometheusRules for staffops_ad_* metrics
• P6.2 — Grafana dashboard refresh
• P6.3 — OTel SDK adoption (traceID-correlated logs)