textmachine/eval/dovodka/blind-read
2026-09-02 22:23:34 +03:00
..
KEY-2-do-not-show-judges.json Freeze the second blind key before judging: thirty battle pairs on manifest-stitched drafts plus ten low-against-low active nulls 2026-09-02 21:48:28 +03:00
KEY-do-not-show-judges.json
PREREG-2-BLIND-READ.md Fix my own timing claim: 21:49 was the task file's creation, not the result, so the addendum was written blind mid-panel rather than after the numbers 2026-09-02 22:16:12 +03:00
PREREG-BLIND-READ.md
RESULT-2-BLIND-READ.md Record the negative result: judges named a winner on 29 of 30 pure-noise pairs, so the earlier cheap-arm advantage is indistinguishable from generation randomness 2026-09-02 22:23:34 +03:00
RESULT-BLIND-READ.md Record the blind-read result after adversarial attack, with seven errata naming the truncated draft that left the accuracy lens without a reference on a third of the text 2026-09-02 20:32:11 +03:00
verdicts-99.json Record the blind-read result after adversarial attack, with seven errata naming the truncated draft that left the accuracy lens without a reference on a third of the text 2026-09-02 20:32:11 +03:00
verdicts-120.json Record the negative result: judges named a winner on 29 of 30 pure-noise pairs, so the earlier cheap-arm advantage is indistinguishable from generation randomness 2026-09-02 22:23:34 +03:00