GenoLeWM-FX Borzoi final outcome report¶
Generated by tools.research.fx_borzoi_final at 2026-06-11T15:00:00Z.
Parent epic: #266. Final issue: #272.
Outcome: no_positive_claim_fragile_lift.
Positive claim allowed: False.
This final report is a no-positive-claim FX rescue outcome. It is not a clinical result, deployment-readiness result, broad VEP superiority claim, broad Carbon comparison, useful-planning claim, ground-truth biology claim, or exact fipip overlap claim.
Reproduce¶
uv run python -m tools.research.fx_borzoi_final \
--overlap-report docs/research/fx-borzoi-overlap-report.json \
--cache-manifest docs/research/fx-borzoi-cache-manifest.json \
--baseline-report docs/research/fx-borzoi-baseline-report.json \
--residual-report docs/research/fx-borzoi-residual-report.json \
--output-json docs/research/fx-borzoi-final-report.json \
--output-md docs/research/fx-borzoi-final-report.md
Final Metrics¶
| Model | AUROC | AUPRC | Balanced accuracy |
|---|---|---|---|
| Strongest simple baseline | 0.6494 | 0.2225 | 0.5080 |
| Residual ensemble | 0.6594 | 0.2300 | 0.5026 |
Paired residual deltas versus the strongest simple baseline:
| Metric | Delta | 95% CI |
|---|---|---|
| AUROC | 0.0099 | [-0.0035, 0.0237] |
| AUPRC | 0.0075 | [-0.0029, 0.0198] |
Negative Findings¶
- The full fipip table exact join was not run; exact fipip overlap is not claimed.
- The residual ensemble's paired AUPRC and AUROC confidence intervals cross zero.
- No trained GenoLeWM-FX positive model-quality claim is supported.
- No FX demo or paper-positive result should ship from this trajectory.
Final Interpretation¶
The rescue path was worth testing because the compact TraitGym-native Borzoi artifact cleared the overlap/cache/baseline gates. The residual model then showed a small mean lift, but the paired confidence intervals cross zero. The correct final outcome is therefore a published no-positive-claim result, not an FX model-quality, demo, clinical, deployment, broad VEP, broad Carbon, or useful-planning claim.