Evidence
A match without a report is a rumour. Every match ships with one.
What is in the report
For each stage: the probes that were run, the metric, the number, the threshold, and pass or fail. Plus the agent that did the work, how long it took, and the versions of every tool involved. The report is generated by the verification tools, not written by the agent.
Built to catch cheating
The tools were tuned on real matches, including the ways an agent can look good without being good: a constant-gain stage that passes a sweep, a loud render that hides a bad fit, a sparse impulse response that fools a residual metric. Each of those has a probe now. A metric that cannot be trusted reports a question mark instead of a number.