ExperimentDeterministic eval mode

LLM Evaluation Playground

Choose a test suite, edit two candidate responses, and compare them with reproducible graders. Every score comes from visible rules rather than another model's opinion.

01 / test-suite

Choose an evaluation suite

Deterministic eval mode

Task prompt

Return JSON with status and priority. Do not include uncertain language.

Candidate A

39/3,000

Candidate B

61/3,000

Evaluated locally by this site. No response is sent to an external model or stored.

02 / evaluation-report

Regression report

Run the suite to compare pass rates, weighted scores, and individual test failures.

Define expectations

Turn product requirements into explicit pass/fail checks and weights.

Compare candidates

Run the same graders against a baseline and a proposed response.

Catch regressions

Inspect failed checks instead of trusting one aggregate score.