Evidence capture tool

AI Voice Benchmark Run Recorder

Capture a reproducible test record in your browser, then export it as JSON or CSV. Nothing is uploaded by this tool.

Run metadata

Identify exactly what was tested

Scores

Use the public 1–5 methodology

Revision evidence

Record attempts and production time

Evidence rules

A benchmark record is only useful if someone else can reproduce it.

Always record the exact model or voice, settings, test date and any workaround. Do not replace difficult phrases for one provider unless the modified input is clearly disclosed. Scores should follow the public VoicePilot methodology.