test: publish a privacy-preserving independent benchmark reproduction kit

[AjnasNB]

User problem

Qarinah publishes reproducible local benchmark fixtures, but external developers need a standard way to test their own public or synthetic repositories without contributing private project history or accepting a marketing-only aggregate.

Proposed contract

Create a privacy-preserving reproduction kit that:

• imports only user-selected public or synthetic evidence;
• records fixture identity, command, package version, Node version, platform, estimator, and exact result hashes;
• compares equivalent required current sources on both sides;
• reports retrieval coverage, rank, conflicts, supersession, pack size, and estimated context reduction separately;
• emits a signed or digest-addressed result bundle suitable for an issue attachment or independent archive;
• refuses provider-cost, latency, and answer-quality conclusions unless those measurements are actually supplied.

Acceptance criteria

• At least three redistributable software projects with distinct task shapes.
• One command to reproduce each result from a clean install.
• Machine-readable schema and human-readable report.
• Negative controls for missing evidence, misleading similarity, private paths, and changed fixtures.
• CI verifies arithmetic, source hashes, and result-to-version identity.
• Contribution guide for independent results and responsible claim wording.