Jev Evidence Kit
Free · Open source · Runs in your browser

Check the answer
before the savings.

Compare a baseline and a local result against an expected answer. Only credit observed ChatGPT tokens when both answers pass the same exact-match check.

Synthetic example · Edit any field

Try one paired result

Observed token records

Download the offline tool ↗No account or API key needed.
150 ChatGPT tokens avoided credited

Both outputs match the expected answer. This is synthetic sample data, not a benchmark or cost claim.

Exact match ignores leading and trailing whitespace and normalizes Windows line endings. It does not judge meaning, validate provider billing, or prove general task quality. The offline Python tool is the reproducible version; this browser demo is a quick illustration.

Need help evaluating this with your team? Ask about a pilot ↗Share a use case, not private prompts, outputs, or API keys.

Install the CLI in an isolated pipx environment ↗

Run the synthetic exercise in Google Colab ↗ Colab uses a hosted runtime; use synthetic data only.