Prove the output.
Retest one change.
The method separates evidence collection, deterministic rules and human review so a persuasive explanation can never manufacture a technical pass.
- 01Connect locally
Authenticate to the configured loopback OBS WebSocket and feature-detect the supported requests.
- 02Run controlled cues
Silence, voice, program audio, mixed audio and representative motion create evidence windows.
- 03Inspect the exact finished file
Probe the app-owned recording plus OBS counters and the relevant current-session log segment.
- 04Return a tri-state result
Passed, Blocked or Could Not Verify—each with measurements, thresholds, provenance and limitations.
- 05Retest one controlled change
Compare equivalent evidence. A materially changed configuration cannot masquerade as an equivalent baseline.
An explanation may use AI later; pass/fail may not.
Versioned deterministic rules and fixtures own technical state. Human reviewers can add context without silently rewriting measured evidence.