Issues

• Tetris real-time results correction: premature information and limited corrected rerun

  potto007/unridden#39 · opened on Sep 27, 2026

• Possible future work: evaluate concurrent inference requests

  potto007/unridden#38 · opened on Sep 27, 2026

• Investigate calibrated E4B-to-26B quality escalation in the gateway

  potto007/unridden#37 · opened on Sep 27, 2026

• Skip, not break, on a failed label alphabet candidate

  potto007/unridden#11 · opened on Sep 22, 2026

• Rescore near-tie questions alone in batched mode

  potto007/unridden#10 · opened on Sep 22, 2026

• Add a padded-shape control for batch-shape movement

  potto007/unridden#9 · opened on Sep 22, 2026

• Cluster bootstrap intervals for paired run comparisons

  potto007/unridden#8 · opened on Sep 22, 2026

• Report top-two margin and normalised entropy per answer

  potto007/unridden#7 · opened on Sep 22, 2026

• Measure option-order permutation flip rate on the suites

  potto007/unridden#6 · opened on Sep 22, 2026

• Add ECE, AURC and risk-coverage to the suite scorer

  potto007/unridden#5 · opened on Sep 22, 2026

• Fit an offline temperature and report calibrated confidence

  potto007/unridden#4 · opened on Sep 22, 2026