Local execution + provider evidence
Calculate locally. Compare honestly.
Loading evidence manifest…
Two separate proof modes.The fairness audit recalculates from browser-local rows. The provider tab replays a committed 540-case-per-model benchmark summary. Neither path calls a provider.
Calculated evidence
WAITINGRun the audit to calculate current metrics.
Committed provider benchmark
ARTIFACT REPLAYLoading benchmark evidence…
Adversarial campaigns are maintained separately. Inspect the dedicated 40-probe artifact →