Downloads

28,050 INFERENCES · $834.94 SPENT
FileContents
data.csvAggregated by (variant, model, jd). One row per cell with n, mean score, and recommend rate.
summary.mdSame table as data.csv plus per-model cost & token breakdown.
matrix.jsonPer-(axis, variant, model) mean Δ aggregated across JDs.
results.ndjson.gzFull run-level corpus. One JSON object per inference run.
raw/audit-verdicts.csvOne row per audited cell. Both first-run and median-run verdicts plus the verdicts_agree flag, so you can recompute the disagreement rate yourself.
resumes.jsonFull text of all 30 résumé variants.