Registered 2026-10-31 from the report/experiment-history.pdf: of 56 experiments, 11 stand, 24 stand in part, 20 no longer stand and E46 is running. Each method the two papers use is validated by one experiment below, on one build (emphasis 182055b), under one protocol (00-design.md, The validation series): E43's calibrated generators with no size window, the conditioned likelihood, the exponential link, fits run to the stopping rule.

Validates Replaces Status
V1 the complete-data likelihood E44 complete: per-tree rule met in 29 of 64 cells; pooled, 138 of 150 coefficients within [0.9, 1.1]
V2 proposal densities and efficiency E3, E6b, E7 complete: surrogate unbiased (30 of 31 reference cells); proposals agree where both are well sampled; ED models sampler-limited on both
V3 the stopping rule and the box E9, E13, E41 complete for the reference and two rules under b21083f (projected, and mc_error at fixed draws; 320 trees each): projected stops cr, 11 of 20 dd and 28 of 280 lineage-level fits; speciation within 2% of the box span of the reference on 8 of 20 model–turnover pairs (7 of 10 at turnover 0.10); the box holds the reference on 15 of 20; the rel and mc arms are to be rerun
V4 agreement with the exact estimate E1, E42, E45 dd complete: rule holds for cr and dd on the real trees under both stopping rules; Picidae dd stops 14.8 below
V5 recovery and tree size E2, E12, E16, E18, E19, E21, E24, E29, E30 running since 10 November on two local hosts (1,710 fits at 50 and 100 tips, b21083f, projected): 423 written on 12 November, no cell read; covariate slope within a factor of 2 on 74% of 414 finished fits
V6 the sampling fraction E27's \(\rho\) arms, E35(b) rerun under b21083f and projected submitted 8 November (3,200 jobs, behind V7); first pass archived (993 of 3,200 fits; no cell–condition read; 225 fits with no conditioned iteration)
V7 model choice: pairwise and the ladder E8, E23, E27, E31 complete (3,580 fits, b21083f, projected): 22 of 32 cells read, 10 budget-limited; false choice holds on dd at turnover 0.10 and no lineage-level model is chosen on 54 null trees; power fails in 18 of 19 lineage-level cells (highest nk 12 of 20, nd 10 of 19)
V8 the covariate path E20, E21b, E22, E33 registered; submitted once V7's dd fits are written
V9 misspecification E11, E17 rerun under b21083f and projected submitted 8 November (2,400 jobs, behind V7); first pass archived (622 of 2,400 fits; 15 of 16 cells calibrated, 9 sampler-limited, no tree read)
V10 extinction E29, E32, E35(a), E38 registered (analysis)
V11 real phylogenies E26, E28, E40, E48 submitted 11 November (875 fits, b21083f, projected), ahead of the readings of V5–V10 (amendment of 2026-11-11)
V12 cost E4, E15 registered (analysis)

State on 7 November. The lineage-level models have no reading in V3, V5, V6, V7 or V9: their fits end on the iteration cap or the 22-hour clock, many of them with the clock spent in the unconditioned start that precedes the conditioned segments (see Open issues). Under the amendments of 7 and 8 November their fits rerun on emphasis b21083f under the package's default rule projected: V3 (its ref, proj and mcfix arms; rel and mc later), V7, V6 and V9 are all submitted. Since the evening amendment of 8 November a fit on the 1000-iteration cap is read as the estimate, on V3's reading (the capped fits' speciation coefficients within 2% of the box span of the reference; extinction not identified); a fit on the clock stays budget-limited. cr and dd fits converge in minutes and are read wherever the rules allow.