V5 — from how many tips is each lineage-level coefficient recovered?
| Status | running since 10 November 2026 on two local hosts (24 and 8 workers); 423 of 1,710 fits written on 12 November; no cell read yet |
| Series | the validation series (E46 run as an experiment of the series; E46's first pass on 182055b is E46) |
| Package build | emphasis b21083f (1.1.0.9000); rule projected, a fit on the 1000-iteration cap read as the estimate |
| Driver | sim/V5.R |
Question
From how many tips is each coefficient of the eight lineage-level models recovered under the conditioned likelihood, by the rules E46 fixed in advance. E46's first pass read 1 of 65 model–cell pairs; the fits that ended on the clock spent it in the unconditioned start (Open issues), which emphasis b21083f caps.
Design
| cells | every reachable E43 cell of d, nd, ed, edc, ned, nedc, k, nk at 50 and 100 tips (turnover 0.10 and 0.40, the coefficient at ±0.35): 29 at 50 tips, 28 at 100 |
| trees | 30 a cell, the series' trees (v_tree) |
| fit | the generating model by estimate_rates(): defaults, conditioned on survival, the box of auto_bounds(), projected, ρ = 1, 22-hour clock |
| jobs | 1,710 |
| next | the 200- and 400-tip cells for the parameters not reliable at 100 tips |
Pre-set rules
E46's. A cell is read when at least 80% of its fits finished (converged or on the cap) and the median final effective sample is at least 10. A coefficient is reliable in a cell when the median ratio to its generating value lies in [0.8, 1.25] and the central 80% of ratios within [0.5, 2]; identified when reliable with the right sign on at least 95% of trees. The answer is the smallest size from which each coefficient is reliable.
Results
Interim, 12 November: 423 of 1,710 fits written, about a quarter of every cell. 414 finished (67 met the stopping rule, 347 on the iteration cap, read as the estimate) and 9 ended on the 22-hour clock. No cell has its 30 trees, so none is read by the rule; over the finished fits:
- Covariate slope. Within a factor of 2 of its generating value on 74% of fits, right sign on 92% (fits that met the rule: 79% of 67; capped fits: 73% of 347).
- By model, effect and turnover (50 and 100 tips together; share within a factor of 2, median ratio):
| turnover 0.10, − | turnover 0.10, + | turnover 0.40, − | turnover 0.40, + | |
|---|---|---|---|---|
d |
94%, 1.12 | 71%, 0.92 | 88%, 0.97 | 59%, 0.79 |
nd |
88%, 1.38 | 50%, 0.64 | 54%, 0.61 | 40%, 0.42 |
ed |
93%, 1.01 | 88%, 1.18 | 69%, 0.74 | 87%, 1.20 |
edc |
93%, 1.11 | 56%, 0.67 | 92%, 1.02 | 82%, 1.04 |
ned |
46%, 0.99 | 47%, 1.86 | no trees | 73%, 1.11 |
nedc |
67%, 1.02 | 79%, 0.73 | 83%, 0.95 | 83%, 0.98 |
k |
100%, 1.13 | 60%, 0.77 | no trees | 38%, 0.43 |
nk |
71%, 1.20 | 87%, 0.85 | no trees | 93%, 1.21 |
- Intercepts. Speciation within a factor of 2 on 81% of fits; extinction on 11%.
- Box edges. The extinction intercept is on an edge of the box on 67% of
edfits, 64% ofnedc, 56% ofned, 36% ofnk, 27% ofndand ofk, 11% ofd, 2% ofedc. - Effective sample at the estimate. Median 29 to 53 of 200 draws for
d,edc,k,nd,nkand 184 to 190 fored,ned,nedc; fits with the extinction intercept on an edge keep a median 188, the others 35. 46 of 414 fits end below 10.
Tables: results/V5/cells.csv, coef.csv, fits.csv, slope.csv (sim/V5-analysis.R).