Replay and capacity intelligence
Inference Replay records the timing and token-count shape of production traffic without storing prompt
or output bodies. Session, parent-session, and shared-prefix identities are tenant-secret HMACs, not raw
identifiers. A trace records its window, sample count, schema, canonical digest, and evidence class.
This command only captures persisted observations. It sends no workload and creates no infrastructure.
To approximate the captured mean token sizes and peak concurrency with AIPerf, opt in explicitly:
Execution is an AIPerf approximation, not a reproduction of prompt semantics or exact arrival timing.
The distinction is preserved in output and evidence.
Capacity intelligence
Capacity Intelligence groups this tenant’s observed lifecycle attempts by provider adapter, runtime,
compute mode, region, and GPU. It reports sample counts, success rate, normalized capacity/runtime
failures, and p50/p95 operation duration. It does not claim global availability or combine customer data.
Artifact placement
The artifact cache contract records short-lived provider-native observations (present, prefetching,
missing, or unknown). Prefetch requests are durable and idempotent, but execution belongs to the
selected infrastructure adapter. InferCrane does not implement a model CDN or download protocol.
Cache observations expire after at most 24 hours. Missing or stale evidence must never be treated as a
cache hit or a readiness guarantee.
Current boundary
Local qualification proves capture privacy, deterministic digests, tenancy, aggregation, and delegated
prefetch semantics. Real cache behavior and capacity distributions require the consolidated provider
qualification run.