PACTVERITYSystem status

18 / Public benchmark

Three receipts. One reproducible evidence test.

The PactVerity AI Agent Evidence Benchmark is a small, transparent known-answer suite for the operational Receipt API beta. It tests deterministic evaluation and receipt verification, not the truth of caller-supplied measurements.

Reference scenarios3 publicFirst-party live run3 / 3 passVerified external runs0
Evidence boundary:A passing benchmark shows that the published inputs produce the documented receipt checks and that the exported receipt verifies. It does not prove independent monitoring, production adoption, revenue, settlement or audit assurance.

01 / Known answers

Inspect the expected result before running the code.

Scenario data, expected outcomes and the runner are packaged in the first-party developer kit so every input remains inspectable.

PactVerity benchmark scenarios and expected results
ScenarioExpectedWhat it tests
Healthy agent responsePassLatency, uptime and freshness pass; output hash is not evaluated.
Latency threshold breachFailLatency exceeds the agreement while the remaining availability checks pass.
Output integrity mismatchFailAvailability checks pass but expected and observed SHA-256 values differ.

02 / Run it

Dry-run first. Submit labelled test receipts only when ready.

The default command lists the scenarios without calling the API. The explicit execute flag runs them against the configured Receipt API endpoint.

# After extracting pactverity-developer-kit-v0.1.zip
node benchmark/run.mjs
node benchmark/run.mjs --execute
Download the complete developer kit Inspect the 28 August live-run record Inspect the earlier reference run Download the research dataset Read developer guide

03 / Public register

No verified third-party benchmark runs yet.

The register starts at zero. Results are added only after the linked command, environment, commit, outputs and receipt digests can be reproduced.

Verified PactVerity third-party benchmark submissions
SubmissionEvidenceStatus
No entriesNo third-party reproduction evidence has been accepted.Open for review

04 / Participate

Reproduce, disclose, then build.

Step 01Run

Publish the evidence

Share the commit, environment, timestamp, complete command output and three receipt digests. Disclose any modifications or failures.

Step 02Review

Invite reproduction

Use a public GitHub work item for questions and review. Never include credentials, private customer data or vulnerability details.

Open evidence checklist

Step 03Build

Propose an integration

Apply to one of three reference tracks only after reading the public rubric and funding boundary.

Open builder challenge