EvolMoves: alignment-space convergence diagnostics

flat alignment length trace can still hide completely different homology. ugh, need diagnostics that actually notice that. Save idea: animate samples w/ residue identities fixed, watch the pairings move instead.

why scalar traces are insufficient: the deck gives schematic alignments with identical ungapped sequences and width three but different residue-pair homologies. A constant width trace can coexist with complete disagreement over which residues align.

Convergence diagnostic matrix: PASS, FAIL, and NA.

Chain traces for alignment length, tree length, split frequency, and log likelihood.

recorded diagnostic rules: distinguish PASS, FAIL and NA; scalar R-hat, bulk/tail ESS and ASDSF answer different questions. Distance-based alignment/topology ESS uses at most 200 actual draws per chain; constant or insufficient traces remain NA. Within-chain efficiency cannot rule out separate modes across chains.

• timing limitation: BAli-Phy prefixes use measured CPU fractions, whereas EvolMoves has only a final point because per-draw wall timestamps are zero. Do not fabricate an intermediate EvolMoves trajectory. Draw index is not common elapsed time, and tool-specific likelihood values are not directly comparable.

• pointers: the t0 reproduction report/analysis; benchmarks/prospective/other-experiments/independent-contributions-plan.md for the alignment-disagreement illustration.

• review question: how can an alignment-space diagnostic reveal disagreement missed by likelihood/width while retaining enough actual samples for a meaningful estimate?

Same alignment width, different residue-homology pairs.