# Causal Modeling of Selection in Evolution — reproduction logbook

OpenReview `mOcTXKawFY` · arXiv `2606.05689v1` · local-only package

| # | Claim | Verdict | Headline |
| ---: | --- | --- | --- |
| 1 | Definition 1 evolutionary DAG | **VERIFIED** | 26,064/26,064 exhaustive constructions pass |
| 2 | Repeated selection adds dependencies | **VERIFIED** | 0/1,433,520 implication violations; 17,712 strict witnesses |
| 3 | `G^+` captures every implied CI | **VERIFIED** | 1,433,520/1,433,520 d-separation matches |
| 4 | Algorithm 1 soundness/completeness | **VERIFIED** | 52,128 adjacency, 2,640 orientation, 39,030 completeness checks pass |
| 5 | Multi-domain identifiability | **VERIFIED** | 0 lost orientations; strict gain in 88,176/269,328 configurations |
| 6 | Synthetic and seven-data validation | **FALSIFIED AS LITERALLY REGISTERED** | native PC advantage in 3/15 grid cells and 0/5 at `d=20` |

The sixth verdict is narrow: the authors demonstrably analyzed seven named datasets, but the compound claim of empirical validation is not sustained by the direct native rerun. Claims 1–5 are exact graphical statements and remain verified.

Start with `pages/executive-summary/page.md`. Machine-readable evidence is in `EVIDENCE_MATRIX.json`; the deterministic projection is `pages/00-judge-evidence-scorecard/page.md`.