Experimental benchmark

Repeat the run,
verify the evidence.

The controlled text-rag Benchmark Pack fixes high-precision inputs and records its Pack, runtime environment, and build revision.

Read the guideExplore the API
Reproducibility contractP0 → P8
Profilehigh_precisionTop-K 8
P0 pack5 / 6000evidence / text units
P8 pack3 / 6000only changed setting

Server-owned

Three durable facts.

Manifest SHA-256, environment SHA-256 and build revision are persisted on every run.

Comparable only when proven

Direct P0 lineage.

Different Packs, inputs, environments or builds never produce an inferred comparison.

No-key walkthrough

Run the real stack.

make console-real-tutorial-benchmark-e2e