Skip to content

Metrics Reference

Metrics endpoint

GOVERN Probe exposes Prometheus-format metrics at /metrics (configurable via HEALTH_METRICS_PATH).

Terminal window
curl http://localhost:4020/metrics

Inference metrics

MetricTypeDescription
govern_inferences_totalCounterTotal inference requests proxied
govern_inferences_scored_totalCounterInferences that completed scoring
govern_inferences_flagged_totalCounterInferences that exceeded a threshold (flag/block mode)
govern_inferences_blocked_totalCounterInferences blocked (block mode only)
govern_inferences_unscored_totalCounterInferences where scoring failed

Labels: model, provider, scorer (for flagged/blocked)

Latency metrics

MetricTypeDescription
govern_proxy_latency_msHistogramEnd-to-end proxy latency (p50, p95, p99)
govern_upstream_latency_msHistogramTime waiting for upstream model response
govern_scoring_latency_msHistogramTime to score the inference (async, not in path)

Score distribution metrics

MetricTypeDescription
govern_score_securityHistogramDistribution of security scores
govern_score_biasHistogramDistribution of bias scores
govern_score_accuracyHistogramDistribution of accuracy scores
govern_score_driftHistogramDistribution of drift scores
govern_score_costHistogramDistribution of cost utilization scores

Telemetry metrics

MetricTypeDescription
govern_telemetry_batches_flushed_totalCounterTotal batches successfully flushed
govern_telemetry_events_flushed_totalCounterTotal events successfully transmitted
govern_telemetry_dropped_totalCounterEvents dropped (ring buffer overflow or max retries)
govern_telemetry_retry_totalCounterFlush retry attempts
govern_ring_buffer_sizeGaugeCurrent events in the ring buffer
govern_ring_buffer_utilizationGaugeRing buffer fill percentage (0.0–1.0)

Token and cost metrics

MetricTypeDescription
govern_tokens_input_totalCounterTotal input tokens proxied
govern_tokens_output_totalCounterTotal output tokens proxied
govern_cost_usd_totalCounterEstimated total spend in USD
govern_tokens_per_hourGaugeRolling hourly token rate
govern_cost_usd_per_hourGaugeRolling hourly spend rate
govern_budget_utilizationGaugeBudget burn percentage (0.0–1.0)

Health metrics

MetricTypeDescription
govern_probe_upGauge1 if probe is running, 0 if not
govern_upstream_reachableGauge1 if upstream is reachable, 0 if not
govern_telemetry_connectedGauge1 if GOVERN platform is reachable
govern_build_infoGaugeProbe version (via labels)

Sample Prometheus scrape output

# HELP govern_inferences_total Total inference requests proxied
# TYPE govern_inferences_total counter
govern_inferences_total{model="claude-sonnet-4",provider="anthropic"} 1247
# HELP govern_proxy_latency_ms End-to-end proxy latency in milliseconds
# TYPE govern_proxy_latency_ms histogram
govern_proxy_latency_ms_bucket{le="1"} 0
govern_proxy_latency_ms_bucket{le="5"} 987
govern_proxy_latency_ms_bucket{le="10"} 1241
govern_proxy_latency_ms_bucket{le="+Inf"} 1247
govern_proxy_latency_ms_sum 3982.4
govern_proxy_latency_ms_count 1247
# HELP govern_ring_buffer_utilization Ring buffer fill percentage
# TYPE govern_ring_buffer_utilization gauge
govern_ring_buffer_utilization 0.12

Grafana dashboard

A pre-built Grafana dashboard is available for import:

Terminal window
# Dashboard ID for Grafana.com
GOVERN Probe Overview: 21847

Or import from the GOVERN dashboard: Settings → Integrations → Grafana → Export Dashboard JSON.