OSS-first docs
These docs teach the open system first: contracts, generated surfaces, runtimes, governance, and incremental adoption. Studio shows up as the operating layer on top, not as the source of truth.
Auto-Evolution Runbook
Configure sampling, approvals, experiments, and golden tests so your app keeps improving without regressions.
Pipeline overview
- Tracing middleware emits telemetry samples with tenant/user context.
- IntentAggregator batches samples (15m window, configurable).
- IntentDetector raises error/latency/throughput signals.
- SpecGenerator produces proposals with confidence scores.
- ApprovalWorkflow routes low-confidence suggestions to humans.
- SpecExperimentRegistry rolls out approved variants using guardrails.
- TrafficRecorder captures new traffic and GoldenTestGenerator outputs suites.
Key environment variables
# Telemetry sampling AUTO_EVOLUTION_SAMPLE_RATE=0.02 AUTO_EVOLUTION_WINDOW_MS=900000 # Suggestion thresholds AUTO_EVOLUTION_MIN_CONFIDENCE=0.55 AUTO_EVOLUTION_AUTO_APPROVE=0.2 # Guardrails (fallback defaults) AUTO_EVOLUTION_MAX_ERROR_RATE=0.02 AUTO_EVOLUTION_MAX_P99_MS=500
Operations checklist
- Review new `SpecSuggestion` records every morning (UI coming soon; Prisma view today).
- Keep at least one reviewer per domain on-call for approvals.
- Attach `SpecExperimentAdapter.trackOutcome` to every runtime adapter.
- Schedule `contractspec test generate` nightly (or on deploy) to refresh golden suites.
- Feed `runGoldenTests` into CI to gate merges touching evolved specs.
Sampling
Start at 1–2% of traffic. Increase only after sanitization is verified.
Approvals
High-risk ops (money, compliance) should never auto-approve. Use ApprovalWorkflow notes for audit.
On-call alerts
Tie SpecExperimentController rollbacks into PagerDuty/Slack so engineers see guardrail trips immediately.