Audit & attribution

Per-request audit attribution for billing, compliance, and abuse detection.

When to read this sub-group

You are running an inference service that needs to attribute compute back to a tenant for billing, compliance, or abuse detection. The audit chain already records lease lifecycle and admission decisions; this sub-group covers the per-request inference attribution that closes the loop.

Suggested order

1. Per-Request Audit Attribution for Inference — emits one tamper-evident audit-chain record per completed inference request, carrying tenant, request id, token count, compute time, and the typed completion reason. Downstream billing / compliance / abuse-detection systems join on kind=inference_request_completed.

What’s not here

Per-token observability (token-rate is a telemetry concern, not an audit concern). See observability tagged recipes. General audit chain consumption. See Consuming the Audit Chain under Security, Audit & Cost.