Metric Methodology

Decisions traced (24h)

Represents the total number of AI interactions observed by Railguard over the last 24 hours. Each event includes request context, policy evaluation, model metadata, and an immutable receipt.

• Source: Live telemetry streaming from verified production gateways. Events are ingested via append-only queues and deduplicated using receipt hashes. Cross-check trend data anytime on the status dashboard.
• Window: Rolling 24-hour period updated every 60 seconds. Metrics are prorated for accounts with partial onboarding.
• Verification: Sampled receipts are reconciled hourly against /.well-known/build-attestation.json to confirm provenance.

Need raw event access? Email support@railguard.ai for scoped export options.

Attestation status

Reflects whether the active site build is covered by a current attestation and matching SBOM signature.

• Source: Build pipeline emits signed metadata stored at /.well-known/build-attestation.json.
• Validation cadence: Checked on every deployment plus nightly via synthetic monitor.
• Manual verification: Follow the verification guide to validate the signature in under two minutes.
• Artifacts: SBOM library and /.well-known/railguard-gateway-sbom.json (signature at /.well-known/railguard-gateway-sbom.json.sig).

Firewall enforcement time

Displays the p95 decision latency tracked by the AI Firewall simulator and production gateways.

• Source: Aggregated from policy evaluation spans exported via OpenTelemetry.
• Window: Rolling 15-minute window with exponential backoff when traffic is idle.
• Verification: Simulator benchmarks can be reproduced using the AI Firewall sandbox, or by inspecting the latency stream exposed on /status (p95 latency panel).