Instability Prediction on 28-system external pool pooled transfer
0.69Spearman's RhoModular screened score
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Modular screened scorePredictor / protocol=Modular screened score (ours), Notes=best overall2026.05 | 0.69 | 71 | 0.04 | 1 | |
| Large-scale human-only gold (1,200 cells)Predictor / protocol=Large-scale human-only gold (1,200 cells), Notes=700 annot.-hrs2026.05 | 0.68 | 70 | 0.04 | 5.12 | |
| Evaluator-reconfig instab. (6 judges at t0)Predictor / protocol=Evaluator-reconfig instab. (6 judges at t0), Notes=strong, no screen2026.05 | 0.64 | 62 | 0.06 | 0.72 | |
| Rule-based blocked-channel scorePredictor / protocol=Rule-based blocked-channel score, Notes=our rule impl. alone2026.05 | 0.62 | 63 | 0.06 | 0.41 | |
| Trace-only learned instability predictorPredictor / protocol=Trace-only learned instability predictor, Notes=end-to-end trace2026.05 | 0.61 | 58 | 0.07 | 0.55 |