Benchmark v0.1: three reproducible Receipt API scenarios

[PactVerityHQ]

PactVerity has published AI Agent Evidence Benchmark v0.1, a small known-answer suite for the operational Receipt API beta.

Published scenarios

• Healthy agent response — expected pass
• Latency threshold breach — expected fail
• Output-integrity mismatch — expected fail

The first-party live run produced the expected 3/3 results. The verified external-run register remains at 0. This benchmark tests deterministic evaluation and receipt verification on published inputs; it does not prove the truth of caller-supplied measurements, customer adoption, revenue, custody, settlement or audit assurance.

Reproduce it

After extracting the developer kit:

node benchmark/run.mjs
node benchmark/run.mjs --execute

Please publish the commit, environment, timestamp, complete command output and three receipt digests, and disclose any modifications or failures.

• Benchmark and developer kit: https://pactverity.com/benchmark
• Developer guide: https://pactverity.com/developers
• Evidence questions: https://github.com/PactVerityHQ/pactverity/issues

Independent reproduction attempts and technical criticism are welcome.