Prometheus metrics endpoint for production monitoring

[sarvanithin]

Overview

Production deployments need observability. Add a /metrics endpoint exposing Prometheus-compatible metrics so ops teams can monitor guardrail trigger rates, latency, and blocked request counts.

Metrics to expose

medguard_requests_total{guardrail="phi", result="triggered"} 42
medguard_requests_total{guardrail="drug_safety", result="blocked"} 3
medguard_requests_total{guardrail="hallucination", result="flagged"} 7
medguard_pipeline_duration_seconds{quantile="0.99"} 0.45
medguard_pubmed_lookups_total{result="verified"} 15
medguard_llm_calls_total{provider="anthropic", status="success"} 100

What to build

• Add prometheus-client to optional deps: pip install medguard[metrics]
• medguard/api/metrics.py — counter/histogram definitions
• Instrument GuardrailPipeline.run* methods
• Add GET /metrics route to FastAPI app
• Add MEDGUARD_METRICS_ENABLED=true env var toggle

Files to modify

• medguard/api/metrics.py — new file
• medguard/guardrails/pipeline.py — add instrumentation
• medguard/api/routes.py — add /metrics endpoint
• pyproject.toml — add metrics optional dep

Acceptance criteria

• GET /metrics returns valid Prometheus text format
• phi_detection, drug_safety, hallucination, fact_check all have counters
• Pipeline duration histogram with p50/p95/p99 quantiles
• Disabled by default (zero overhead when not enabled)
• Works with standard Prometheus scrape config