In plain English: the exam scores a Tyranid-fitted setting on the other armies. It used to average 15 armies, Leagues of Votann among them. Votann's two reviewer panels are 10th-edition reference lists, and the 3 Oct decision said they would never be scored. Jordan decided on 5 Oct to take Votann out of the pooled number. It now averages 14 armies. Votann still gets its own row in the table, marked apart. No ledger value moves, only the yardstick.
What changes: Track B's small edge over Track A goes. Votann was its third-best army (+10.0), so without it Track B ties Track A exactly (−0.0). The cross-army trio was the one setting whose gain cleared zero. It loses half a point and its interval now touches zero: +2.0 [−0.3, +4.3], against +2.5 [+0.3, +4.7] before. Nothing passes or fails because of the change. The trio is still the best cross-army result and still not adopted (it waits for T-673).
How: node tools/exam.js <settings…> now pools 14 armies. --with-votann pools 15 again and reproduces T-668's table and the trio's ingredients numbers to the bit. The bootstrap still draws over all 15 armies in the same order, so no single army's interval moves with the switch. Same method as the exam: wCarry 0, pairwise over tiers S to D, each army the mean of its lists, 2,000 paired draws against Track A (seed 1).
Settings: A is Track A (step 34b). B is Track B. B+4 is Track B with the four dials. Trio is A with aircraftBoard 1, smallPiece 0.4 and attachChars (tools/ledger-benchmarks/2026-10-05-cross-army-trio.json).
| Pooled | A | B | B+4 | Trio | B − A [95%] | B+4 − A [95%] | Trio − A [95%] |
|---|---|---|---|---|---|---|---|
| 14 armies, new (units) | 59.4 | 59.4 | 59.5 | 61.4 | −0.0 [−2.7, +2.6] | +0.1 [−3.0, +3.0] | +2.0 [−0.3, +4.3] |
| 14 armies (armies resampled) | −0.0 [−3.2, +3.0] | +0.1 [−3.8, +3.9] | +2.0 [−0.5, +4.3] | ||||
| Armies better than A | 7 / 14 | 8 / 14 | 9 / 14 | ||||
| 15 armies, old (units) | 59.4 | 60.1 | 60.0 | 61.9 | +0.7 [−2.0, +3.2] | +0.6 [−2.4, +3.6] | +2.5 [+0.3, +4.7] |
| 15 armies (armies resampled) | +0.7 [−2.7, +3.8] | +0.6 [−3.1, +4.3] | +2.5 [−0.0, +4.9] | ||||
| Armies better than A | 8 / 15 | 9 / 15 | 10 / 15 | ||||
| Leagues of Votann (2 lists, 20 units; apart) | 59.2 | 69.2 | 66.9 | 69.2 | +10.0 [−3.6, +24.7] | +7.8 [−6.1, +23.7] | +10.0 [−0.1, +24.1] |
| Tyranids (5, 52; fitted on) | 79.3 | 77.7 | 81.4 | 76.5 | −1.6 | +2.2 | −2.7 |
B+4 against B directly: 14 armies +0.1 [−2.4, +2.6] (armies resampled +0.1 [−2.4, +2.7]), better in 6 of 14. It was −0.1 [−2.4, +2.4] on 15. The other 14 armies' rows are unchanged; they are in the exam, and the trio's are in ingredients.
Read: Track A stays at 59.4 because Votann scored about the pool's mean under it (59.2). Every setting that lifted Votann loses a little: B −0.7, B+4 −0.5, the trio −0.5. That was the risk with a 20-unit army graded by out-of-date lists. No conclusion changes. Track B and the four dials don't transfer. The trio holds in 9 of 14 armies, but its interval now crosses zero, so it is a lean, not a result.
Next: T-681 (Space Marines' scarcity and Soak-effects inputs, a new snapshot) is measured on this 14-army pool. T-673's character dials and the trio after it are measured on it too.