Jordan asked (3 Oct): do settings with the intangible lines on (Hold, Score, Actions, Presence) find closer parity across armies than the damage-and-soak search, and do reviewers grade leaders alone or leading? Scratch scripts kept in tools/scratch/ (ledgerjoint.js, ledgerleaders.js), through docs/lab/ledger/ledger.js's own compute(). Floors: Score, Actions, Presence ≥ 0.5, Hold ≥ 1. Objective: the mean ρ over the five panels. Nothing on the site changes.
The joint search and its checks
| Run | Tyr Auspex | Tyr Hivemind | Tyr Second | SM Tista Minis | SM Auspex | Mean |
|---|---|---|---|---|---|---|
| Joint, leaders alone (3 random starts) | 0.59–0.61 | 0.53 | 0.44–0.47 | 0.14–0.18 | 0.37–0.40 | 0.42–0.43 |
| Joint, leaders leading their best squad (3 starts) | 0.58–0.60 | 0.52–0.56 | 0.44–0.47 | 0.19–0.26 | 0.25–0.26 | 0.40–0.42 |
| Scrambled letters, same search (8 runs) | 0.02–0.25 (median 0.16) |
Fit on four, test on the fifth (leaders alone): Tyr Auspex 0.57, Tyr Hivemind 0.46, Tyr Second 0.42, SM Tista Minis 0.07, SM Auspex 0.25. The Tyranid panels hold up when held out (each within 0.05 of its fitted value); the Marine panels don't (Auspex 0.40 → 0.25, Tista Minis 0.15 → 0.07).
Reading: the joint search clears the scrambled letters by about 0.18 (0.43 against at most 0.25), far wider than the Tyranid-only search's +0.02 to +0.07. Settings found on several reviewers and two armies are much less a fit to noise. The Tyranid side generalises across reviewers; Space Marines are the weak side, and the Marine reviewers disagree most (Tista Minis especially).
Leaders: alone or leading? (at the joint benchmark)
Per reviewer, which of our two bands (leader alone; leader with its best squad) is closer to the reviewer's letter:
| Reviewer | Closer alone | Closer leading | Tie |
|---|---|---|---|
| Tyranids, Auspex | 3 | 1 | 2 |
| Tyranids, Hivemind | 4 | 0 | 3 |
| Tyranids, Second | 3 | 1 | 2 |
| Space Marines, Tista Minis | 0 | 2 | 1 |
| Space Marines, Auspex | 5 | 3 | 1 |
Over whole panels: leading raises Tista Minis (0.14 → 0.24) and lowers Marine Auspex (0.45 → 0.29). Few leaders are graded (8 Tyranid, 12 Marine with any letter), so this is suggestive, not settled: Auspex and the Tyranid lists look like they grade leaders alone; Tista Minis may grade them leading. Notable: the Swarmlord and the Hive Tyrant are S alone (matching S/S/A and S/A/S) and B with Tyrant Guard; the Neurotyrant is C alone and A leading Zoanthropes (Auspex: A); the Librarian is C alone and A leading (Tista Minis A, Auspex S).