## Problem 025 — rank:18

**Title:** Genome-wide epistasis

**Added binding assumptions (LaTeX):**

```latex
Take $\E[\epsilon\mid g]=0$ and specify the test-genotype distribution. Walsh monomials are orthogonal under the uniform product genotype law, not under an arbitrary factorial subsample. Support recovery needs separation from zero and design identifiability; prediction may remain possible without either. Comparison with latent nonlinear-response models requires anchored response scales and a joint identifiability analysis, not merely two good fits.
```

**Direct edits:** 0. **Fresh external checks:** None; supplied sources retained as provenance.