Structural score
stable
Median sequence-aligned lDDT 0.8613; median TM-score 0.9224.
Runtime
variable
Wall-clock range 285.76-609.38 s.
Peak RSS
31.99 GB
Median over two telemetry-enabled smokes.
Repeat Summary
| Run | pLDDT | Wall clock | Peak RSS | lDDT | TM-score | Execution metadata |
|---|---|---|---|---|---|---|
| first CLI | 81.8822 | 609.38 s | n/a | 0.8627 | 0.9270 | no |
| telemetry | 85.2307 | 378.92 s | 31.4909 GB | 0.8613 | 0.9224 | no |
| execution metadata | 84.1826 | 285.76 s | 32.4891 GB | 0.8569 | 0.9195 | yes |
Aggregate Values
| Metric | n | Min | Median | Max | Mean |
|---|---|---|---|---|---|
| pLDDT mean | 3 | 81.8822 | 84.1826 | 85.2307 | 83.7652 |
| wall clock seconds | 3 | 285.7550 | 378.9160 | 609.3840 | 424.6850 |
| peak RSS GB | 2 | 31.4909 | 31.9900 | 32.4891 | 31.9900 |
| lDDT | 3 | 0.8569 | 0.8613 | 0.8627 | 0.8603 |
| TM-score | 3 | 0.9195 | 0.9224 | 0.9270 | 0.9230 |
Phase 1 Rule
- Final benchmark cards must include
executionmetadata for route, seed, sample count, checkpoint files, auxiliary cache files, progress samples, and peak RSS when available. - One-off wall-clock and pLDDT values should not support strong claims without repeated-run controls or an explicit single-run limitation.
- For model viability tables, report median wall-clock and median peak RSS where repeats exist; otherwise mark values as single-run estimates.
- For confidence interpretation, treat pLDDT as model output under route/cache/runtime conditions until repeated calibration controls exist.