Status registered 2026-10-31; amended 2026-11-01 (calibration); rerun in full under emphasis b21083f and projected submitted 8 November (arrays v9, 2,400 jobs, queued behind V7; amendments of 2026-11-07 and 2026-11-08), a fit on the 1000-iteration cap read as the estimate; first pass on 182055b (622 of 2,400 fits; 9 of 15 calibrated cells sampler-limited, no tree read) archived under results/V9-182055b
Series the validation series (replaces E11, E17 (E47's alternative-generator arm))
Package build first pass: emphasis 182055b (1.0.0.9000), R 4.5.3; relaunch: b21083f (1.1.0.9000)
Driver sim/V9.R

Question

The ladder chooses nk on 54 of the 170 real trees, with a speciation factor above 1 on all 54. Processes with no \(K\) effect can still make some clades keep splitting: a heritable latent rate (ClaDS), protracted speciation, a fast region colonised rarely. If they make the ladder choose nk with factors like the real ones, nk on real trees reads as a heritable rate, not \(K\) in particular.

Design

generators latent rate (\(\sigma\) 0.10, 0.25, 0.40; trend-free), protracted speciation (duration 0.5, 2, 5), two regions (extirpation equal or none from A)
turnover 0.10 and 0.40
cells 16, each scale calibrated to a median of 100 tips (V9cal)
trees 30 a cell
fits the ladder's five models, as in V7; controls are V7's cr and nk cells
jobs 2400 (after 16 calibration jobs)

Pre-set rules

  1. A cell reproduces the K result when nk is chosen on more than 10% of read trees with a median factor per sd of \(K\) of at least 1.48.
  2. A cell with more than 20% of trees unread is sampler-limited.

Amendment 2026-11-01

The first calibration left 10 of 16 cells off the 100-tip target. The protracted-speciation generator had passed its parameters to PBD::pbd_sim in the wrong order (speciation completed at the extinction rate, incipient species split at the inverse of the mean duration); it now runs pbd_sim's own event loop with the order fixed and an explicit lineage cap, and returns the same trees as pbd_sim at the same seed (150 of 150). The search widens to scales 0.005-30 with 400 clades a step; a cell is calibrated when its median lies within 70-140 tips, and one that cannot be is reported with its size distribution and enters no rule. The latent rate at \(\sigma = 0.40\), whose surviving clades either die or pass the cap, stays and is read under that rule. The six calibrated cells (latent rate \(\sigma\) 0.10, the four two-region cells) keep their scales.

Recalibrated: 15 of 16 cells lie within 70-140 tips (all six protracted-speciation cells at 94-104; the latent rate at \(\sigma\) 0.25 at 84 and 84; at \(\sigma\) 0.40 and turnover 0.10 at 105). The latent rate at \(\sigma = 0.40\) and turnover 0.40 reaches a median of 35 tips with a third of its surviving clades over the cap: it is not calibrated and enters no rule.

Results

Interim, on the 622 fits of 5 November (sim/V9-analysis.R, tables in results/V9/). A tree is read when its five ladder fits converged with an effective sample of at least 10.

Rule 2. 9 of the 15 calibrated cells already have more than 20% of their 30 trees unread and enter no rule: the latent rate at \(\sigma\) 0.10 and 0.25 (both turnovers) and 0.40 (turnover 0.10), protracted speciation at durations 0.5 and 5 (turnover 0.10), and two regions with no extirpation from A (both turnovers). The other six (protracted speciation at duration 2 (0.10) and 0.5, 2, 5 (0.40); two regions with equal extirpation, both turnovers) are 1–4 unread trees from the same state.

Rule 1. No tree is read in any cell, so no cell reproduces the K result and none is ruled out. The controls are in the same state: V7's cr and nk +0.35 cells have no read tree, and both cr cells are budget-limited under V7's rule 3. The factor per sd of \(K\) needs the trees regenerated (--trees), which needs the calibration file and PBD/diversitree.