Targets scored
3 / 3
All current provisional slice targets produced predictions and scores.
Median runtime
390 s
Cached SimpleFold-3B CLI runs on M3 Max.
Median peak RSS
38.5 GB
Observed target-level memory is not constant.
Results
| Target | Bin | Length | pLDDT | Wall clock | Peak RSS | sequence-aligned lDDT | TM-score | RMSD | Aligned residues |
|---|---|---|---|---|---|---|---|---|---|
11BC:C | hard | 116 | 81.2384 | 337.455 s | 41.4060 GB | 0.7492 | 0.6569 | 1.97 | 63 |
13EE:A | mid | 161 | 81.6716 | 401.635 s | 38.5429 GB | 0.8582 | 0.9212 | 1.68 | 148 |
13FT:A | mid | 228 | 92.9916 | 390.276 s | 31.2522 GB | 0.8915 | 0.9588 | 1.23 | 212 |
Interpretation
The first cards exposed a residue-numbering scoring artifact: model predictions were numbered from residue 1 while reference PDB chains retained source numbering. The public cards now use single-chain sequence-aligned lDDT. The hard-bin 11BC:C slice adds an alignment-coverage caveat: only 63 reference residues aligned, so lDDT must be interpreted with coverage visible. These corrections are methodological findings, not final cross-model biological conclusions.
Artifacts
Boundary
These runs are scientific benchmark evidence, but they are not the full Phase 1A result by themselves. They are marked as pre-freeze evidence; the full Phase 1A run is now complete, external biological review is still pending, and cross-model claims are out of scope.