Phase 1A score recomputation

Cards recomputed

14
Phase 1A source cards and repeat-control cards.

lDDT delta

0.0
Maximum absolute difference versus stored scores.

TM-score delta

0.0
Maximum absolute difference versus stored scores.

Command

USALIGN_BIN=/absolute/path/to/USalign \
  pilot/m3max-server/.venv/bin/python pilot/scripts/recompute_phase1a_scores.py

Output summary

{
  "cards_recomputed": 14,
  "score_rows": 14,
  "max_lddt_abs_delta": 0.0,
  "max_tm_score_abs_delta": 0.0,
  "all_lddt_ok": true,
  "all_tm_ok": true
}

Artifacts

phase1a_recomputed_scores.json
phase1a_recomputed_scores.csv
phase1a_release_sha256.txt

This verifies the scoring layer from saved predictions and public RCSB references. It does not rerun SimpleFold-3B inference.