What: Track B's switches read per-army data (Burst, preferred targets, measured support, points over a life). Before this card the 14 other armies had no preferred-target or life data, so on them preferredTargets and attachChars fell back to plain Burst, and a Tyranid-fitted setting couldn't be examined there. Now every army has all four (built by tools/burst.js --field, --prefer, --life and tools/supportfx.js, merged by tools/ledger-merge.js; tools/ledger-all.js's chain builds them). The Space Marines get their life file. Their lack of scarcity and Soak-effects files is unchanged, so Track B's scarcity and soakfx still fall back there. Building those would have moved their Track B numbers. No Tyranid or Space Marine number moved: Track A and Track B reproduce exactly.
How: node tools/exam.js <settings…> (19 s for three settings). Each setting goes through the page's own compute() at wCarry 0 (as tools/ledgercarry.js). It is scored by pairwise accuracy over tiers S to D: of the pairs a list puts in different tiers, the share our Value orders the same way. An army scores the mean over its lists. Pooled is the mean over the 15 armies, each counted once; the Tyranids are the training set and aren't pooled. The intervals are 95% paired bootstraps against the first setting, 2,000 draws: each army's graded units are resampled, the same draw for every setting. Pooled has two intervals: "units" (the armies fixed) and "armies" (the 15 armies' differences resampled).
Settings: A is Track A (step 34b). B is Track B. B+4 is Track B with the four dials from reports/tiers/2026-10-05-dials.md (arrivalKill on at 1.5, secondaryVP on at 0.5, supportMeasuredW 3, shockPressureW 2), saved as the candidate tools/ledger-benchmarks/2026-10-05-trackB-dials4.json.
| Army (lists, graded units) | A | B | B+4 | B − A [95%] | B+4 − A [95%] |
|---|---|---|---|---|---|
| Space Marines (2, 84) | 64.5 | 62.0 | 61.0 | −2.5 [−7.5, +3.1] | −3.4 [−9.9, +2.9] |
| Adeptus Mechanicus (1, 29) | 73.0 | 59.6 | 62.1 | −13.4 [−25.8, −3.6] | −10.9 [−23.5, +0.7] |
| Agents of the Imperium (1, 26) | 58.3 | 62.8 | 61.0 | +4.5 [−5.1, +14.8] | +2.7 [−6.1, +12.0] |
| Astra Militarum (1, 61) | 66.9 | 77.2 | 74.9 | +10.2 [+1.2, +20.7] | +8.0 [−0.5, +17.0] |
| Blood Angels (1, 19) | 61.3 | 57.3 | 47.6 | −4.0 [−16.7, +7.1] | −13.7 [−33.1, +1.8] |
| Dark Angels (1, 77) | 54.1 | 51.0 | 51.3 | −3.1 [−7.9, +2.0] | −2.8 [−7.2, +2.4] |
| Death Guard (1, 36) | 53.4 | 50.6 | 51.3 | −2.7 [−15.1, +9.2] | −2.1 [−16.4, +11.6] |
| Drukhari (1, 23) | 57.5 | 58.9 | 64.7 | +1.4 [−11.7, +13.9] | +7.2 [−5.2, +20.5] |
| Emperor's Children (1, 23) | 66.0 | 69.1 | 66.5 | +3.1 [−6.3, +11.7] | +0.5 [−10.2, +10.6] |
| Genestealer Cults (2, 24) | 56.4 | 51.2 | 49.4 | −5.2 [−14.1, +3.3] | −7.0 [−19.1, +5.4] |
| Leagues of Votann (2, 20) | 59.2 | 69.2 | 66.9 | +10.0 [−3.6, +24.7] | +7.8 [−6.1, +23.7] |
| Orks (1, 45) | 63.7 | 66.7 | 64.0 | +3.0 [−2.1, +9.2] | +0.3 [−7.1, +8.6] |
| Space Wolves (1, 20) | 51.4 | 48.6 | 60.7 | −2.9 [−14.3, +7.6] | +9.3 [−1.9, +22.1] |
| Thousand Sons (1, 28) | 37.9 | 47.6 | 49.5 | +9.7 [+0.7, +20.1] | +11.7 [+1.7, +23.8] |
| World Eaters (3, 30) | 67.8 | 69.4 | 68.9 | +1.6 [−10.3, +12.4] | +1.1 [−11.8, +13.2] |
| Pooled, 15 armies (units) | 59.4 | 60.1 | 60.0 | +0.7 [−2.0, +3.2] | +0.6 [−2.4, +3.6] |
| Pooled (armies resampled) | +0.7 [−2.7, +3.8] | +0.6 [−3.1, +4.3] | |||
| Armies better than A | 8 / 15 | 9 / 15 | |||
| Tyranids (5, 52; fitted on) | 79.3 | 77.7 | 81.4 | −1.6 | +2.2 |
B+4 against B directly: pooled −0.1 [−2.4, +2.4] (armies resampled −0.1 [−2.3, +2.3]), better in 6 of 15 armies. On the Tyranids the four dials add +3.7 in sample.
Read: neither Track B nor the four dials transfer. Track B ties Track A off the Tyranids (+0.7, interval across 0). Its gains (Astra Militarum, Thousand Sons, Leagues of Votann) and losses (Adeptus Mechanicus, Genestealer Cults) cancel. The four dials gain the Tyranids +3.7, and on the 15 armies they add nothing to Track B (−0.1 ± 2.4). They are Tyranid-fitted, as the nested check in the dials report already warned. Most single armies grade one list of 20 to 30 units, so a per-army interval is around ±12 points. Only the pooled row can tell settings apart, and it resolves about ±2.5 points. The armies' spread (Thousand Sons 38 to 50, Adeptus Mechanicus 73 to 60) is bigger than any setting's effect.
Next: a setting joins Track B only if it holds on the exam's pooled row as well as on the Tyranids. Run node tools/exam.js <A> <B> <candidate> on each new candidate. Nothing is adopted here.