EvolMoves: modern-versus-archived comparison limits

keep chain-specific points next to pooled results. starts, decoder, version, model, budget all differ here, so don't squint at a scatterplot and blame everything on one thing.

diagnostic panel: 24 families, two 30-minute chains per family. The deck compares archived BAli-Phy outputs, modern BAli-Phy 4.2 and EvolMoves. It explicitly does not call the modern computation an exact rerun of archived BAli-Phy 2.3.x.

confounds to retain: starting conditions, indel model, posterior decoder, chain duration and software version. A scatterplot difference cannot isolate a version effect when those other choices differ.

Alignment recall and precision across 24 families.

BAli-Phy and EvolMoves versions, chains, and comparison budgets.

metric interpretation: SP recall and Modeler precision are plotted as scores, with point size representing expansion; one minus each score gives the corresponding error but is not the plotted axis. Filled points are pooled estimates and open points are chain-specific.

files: benchmarks/output/t0-baliphy-reproduction/REPORT.md, benchmarks/experiments/t0-baliphy-reproduction/ANALYSIS.md, benchmarks/output/t0-baliphy-reproduction/analysis/summary.json.

review question: does a pooled result hide disagreement between chains? The deck says these short chains are not scientifically converged; treat them as diagnostics rather than a definitive ranking.