Evidence stress test
Module registry
Eleven computation-first checks, with executed, partial and registered states kept separate.
| Module | Question | Independent unit | Current state | Claim impact |
|---|---|---|---|---|
| NWLE-00 | Release and provenance integrity | release object or rendered route | RELEASED EXECUTED | Supports reproducibility of the released resource; it does not support a biological claim. |
| NWLE-01 | Same-context predictive audit | held-out perturbation target within repeated balanced folds | EXECUTED BOUNDED STRICT GATE 0 OF 16 | Supports a small same-context RMSE improvement only; strict direct-head support is not established. |
| NWLE-02 | External aggregate transfer diagnostic | external perturbation target | EXECUTED NO DIRECTIONAL TRANSFER SUPPORT | Falsifies a broad directional-transfer claim on the current 55-target panel; it is not DMD or muscle truth. |
| NWLE-03 | Public myogenic and DMD context triangulation | public study, target or reference trajectory | EXECUTED REFERENCE ONLY | Can downgrade or contextualize hypotheses; cannot create independent DMD candidate-perturbation validation. |
| NWLE-04 | Source-balanced DMD prior multiverse | source-aware gene-context estimate | REGISTERED NOT EXECUTED | Will improve disease-prior robustness, but still will not constitute perturbation efficacy. |
| NWLE-05 | Leakage-safe G2 backtest | held-out gene family or pathway | REGISTERED NOT EXECUTED | Required before an unseen-family generalization claim. |
| NWLE-06 | In-silico perturbation-mode boundary | perturbation × context × modality | PARTIAL BOUNDED SIMULATION ONLY | Observed same-HepG2 CRISPRi and additive null previews are usable; KO, overexpression, fate and synergy claims remain locked. |
| NWLE-07 | Uncertainty, OOD and abstention audit | target or context-level prediction | PARTIAL BOOTSTRAP AVAILABLE OOD ABSTENTION NOT TRAINED | Sampling intervals are available for selected executed analyses; trained OOD-aware abstention is not yet supported. |
| NWLE-08 | Retrospective selection-enrichment backtest | outcome-blinded candidate selection set | REGISTERED NOT EXECUTED OUTCOME LEAKAGE GUARD | Only valid when a pre-outcome model snapshot and eligible outcome universe can be proven. |
| NWLE-09 | Independent computational reproduction | independent analyst execution | REGISTERED EXTERNAL REPRODUCTION NOT EXECUTED | Independent reproduction can strengthen computational reliability, but not biological validity. |
| NWLE-10 | UniPert cross-domain representation feasibility | candidate gene, held-out target family or compound scaffold | PARTIAL SOURCE AND IDENTIFIER AUDIT EXECUTED MODEL EVALUATION PENDING | Supports operational feasibility only; no embedding benefit, cross-domain transfer or DMD therapeutic value has been established. |
Boundary: these modules can stress-test computational reliability and evidence compatibility. They cannot establish DMD efficacy, function, safety or treatment utility.