loop calibrate: recalibration trials over answered relower draws (cut calibrate-harness.r1) #78
2026-11-13
docs: list every figure that came from scratch scripts (Soul s2 scratch-figure-list-incomplete)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
docs: say what loop calibrate produces and what stayed in scratch scripts; restore the mild brief's diversity clause and warn
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
tests: calibrate refusals pinned with answered draws, a partly answered draw, exact class prefixes, unusable counts
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
run: relower and calibrate create their directories after the round is found; paired_floor is NaN under two items instead of underflowing; main: split_text keeps its doc
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
docs: calibration procedure; the map's evidence paragraph points at weksa loop calibrate and states its reproduction
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
cli: loop calibrate exit, refusal and no-leak test; clippy
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
golf_loop: calibrate over the grown bilingual Dao run, exact counts for all and two classes
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
loop calibrate: replay the accept rule over answered relower draws
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
run: relower's scoring becomes read_draws, decide_panel's noise panel_noise, regression doc corrected
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>