# v1.3 cross-disease leave-one-out audit

## Result

- Diseases: DMD, FSHD, SMA.
- Common module across all three: `Stress_response` only.
- Descriptive leave-one-disease-out stress-direction matches: `3/3`.
- DMD sample-level units: 3 DMD samples versus 2 normal samples.
- FSHD and SMA units: 2 biological lines per genotype.

The stress-response direction is up in all three disease datasets, so each held-out disease is descriptively predicted as up by the other two. This is a reproducible cross-disease direction pattern, not a target-specific perturbation prediction. The other modules are disease- or dataset-specific in the currently frozen tables and are shown as partial context rather than pooled evidence.

## Boundary

No candidate-level DMD perturbation outcome was added (`0/21`). No pooled disease score, therapeutic ranking, website claim, manuscript claim, or released figure was unlocked. Small sample sizes and heterogeneous disease/state designs require the next layer to remain sample-aware and descriptive.

## Artifacts

- `v13_cross_disease_module_effects.tsv`
- `v13_cross_disease_lodo_stress_direction.tsv`
- `v13_cross_disease_module_direction_heatmap.png`
- `v13_cross_disease_module_direction_heatmap.pdf`
- `v13_cross_disease_lodo_summary.json`
