Use every proposal exactly as published. No silent exclusions.
EXTERNAL RUNNER / LOCAL BY DEFAULT
RUN.
SCORE.
PROVE.
YOUR OUTPUTS NEVER LEAVE YOUR BROWSER.
Fill the fixed 100-case template, score it locally, download the audit report, publish the complete evidence, and request human verification.
Record the first output, retries, parameters, tools, and prompt.
The browser checks 100 IDs and the exact benchmark SHA-256.
A stable public run file and report are required for review.
LOCAL WORKBENCH
THE SCORE DOES NOT REQUIRE TRUST.
LOCAL RESULT / NOT YET VERIFIED
NO RUN LOADED.
A score appears only after all 100 IDs, predictions, raw outputs, and the exact benchmark hash pass validation.
BENCHMARK CLASS / PUBLIC OPEN-BOOK REGRESSION
THE GOLD LABELS ARE PUBLIC. CONTAMINATION IS POSSIBLE.
CivicProofBench measures reproducible behavior against published rules. It does not establish generalization to hidden cases. A leaderboard score is not proof of legal competence, ethical authority, safety, governance legitimacy, or general intelligence.
Local scoring does not create a leaderboard entry. Only a complete, stable public evidence package can enter human verification.
Submit completed evidence