Analysis only: no engine change, nothing adopted, no every-army run. Read from the every-army raw table at BSData 374f505 (build/tier-raw-all/021a6f986ec8.json, T-583) composed at B0 (round 11, best legal size) and B1 (minimum size) with tools/tierbench.js. The scratch scripts sit in the git-ignored build/ folder (build/sizeodd.js, sizeodd2.js and sizeodd3.js). No panel judges anything here. Derived numbers only.
How DPS (Hammer) is built, and what size can touch
Hammer raw = 0.75 × the peak target + 0.25 × the meta-weighted mean of points destroyed per 100 pts, delivered before death, for each target. Delivered = per-turn damage per 100 pts × the alive-weighted turns in range (tools/tiertest.js go4: each turn counts clamp(T − turns already exposed, 0, 1)). T is the survival term, Σ over the six attacker classes of the fire weight × clamp(x, 1, 5), where x is the absolute hits to destroy the unit over half the median unit's (round 5's "survival absolute"). DPS is the pool percentile of that raw.
What a size change does in the code (sizeVariant): per-turn damage per 100 pts × (m₁/p₁)/(m₀/p₀), which is 1 when points go per model; hits to destroy per 100 pts likewise, so the absolute hits to destroy, and T, scale with size. For a pair, the leader's damage and points stay fixed (pairVariant). Lends, auras and Banner don't enter Hammer. So at a new size, Hammer can move through five parts: turns alive (delivery), per-turn damage per point, the context picked at that size (detachment, leader pair), the way in picked, and the pool shift (every other unit moved too, so the same raw lands on a different percentile).
The parts, unit by unit (B0 → B1)
Per-turn: the bare per-turn points destroyed per 100 pts (rankings.json Offense × points per wound, no context), GEQ / MEQ / TEQ / LV / HV. Turns: x per attacker class (small arms, special, plasma, elite, anti-tank, mortal) and T as used. The last column splits the DPS move exactly (Shapley over the five parts; they sum to the move).
| Unit | Size, pts | Context, way | Per-turn per 100 pts | Delivered per 100 pts | Turns x by class; T used | Hammer raw | DPS | Move: pool / context / per-turn / turns / way |
|---|---|---|---|---|---|---|---|---|
| Hive Guard | 6, 160 → 3, 80 | Subterranean Assault; walk → reserves | 56/37/41/87/59, unchanged | 223/153/204/475/329 → 174/114/131/285/196 | 6.0, 7.1, 4.2, 9.0, 4.8, 4.8 → 3.0, 3.6, 2.1, 4.5, 2.4, 2.4; 4.62 → 2.70 | 429 → 259 | 86 → 65 | +4.9 / 0 / 0 / −26.3 / 0 |
| Zoanthropes (with the Neurotyrant, both) | 6, 320 → 3, 220 | Subterranean Assault; walk | 44/35/32/52/37 → 41/36/29/42/29 | 277/268/272/443/328 → 232/238/221/344/253 | all ≥ 6.0 → 4.4 to 5.9; 5.00 → 4.78 | 413 → 322 | 85 → 79 | +4.7 / 0 / −8.2 / −2.2 / 0 |
| Tyrannofex | 1, 170 (one size) | Subterranean Assault; walk | unchanged | unchanged | 16.0, 14.2, 7.4, 12.0, 4.2, 3.2; 4.71 | 399 | 84 → 88 | +4.8 (pool only) |
| Termagants | 20, 110 → 10, 60 | Subterranean Assault; reserves | 99/45/34/43/30 → 91/41/32/40/27 | 458/208/158/199/134 → 233/106/81/101/69 | 1.3 … 8.0 → 0.6 … 4.0; 3.52 → 2.09 | 223 → 114 | 52 → 30 | +5.9 / 0 / −3.8 / −24.0 / 0 |
| Hellblaster Squad (Space Marines) | 10, 220 → 5, 155 | Vanguard Spearhead alone, walk → Saga of the Beastslayer with a Lieutenant, transport | 73/81/48/63/50 → 68/81/50/60/47 | 226/312/268/425/354 → 130/142/127/490/367 | 4.11 → 2.33 | 399 → 434 | 84 → 91 | +3.9 / +29.4 / −0.7 / −25.6 / 0 |
| Bladeguard Veteran Squad (Space Marines) | 6, 160 → 3, 120 | Liberator Assault Group alone, transport → Saga of the Beastslayer with the Bladeguard Ancient, reserves | 58/81/48/34/26 → 52/59/35/26/19 | 467/379/332/536/393 → 123/132/96/642/284 | 4.02 → 2.86 | 504 → 547 | 92 → 97 | +2.6 / +27.1 / −3.2 / −21.4 / 0 |
Hive Guard is the clean case: same detachment, points per model unchanged (80 for 3, 160 for 6), so per-turn damage per point is identical and the whole fall is turns alive (4.62 → 2.70: half the models, half the absolute hits to destroy). Zoanthropes are not: B1 composes them at 79, not 65, because the Neurotyrant's wounds keep the pair's turns near the cap of 5. Their fall is mostly per-turn damage: the leader's fixed 130 pts weigh more beside three Zoanthropes than six. (Alone, Zoanthropes would fall 86 → 55 at three, all turns: 4.41 → 2.35. The analysis's "85 → 65" isn't reproduced by the table.) The Tyrannofex's rise is the pool shift alone: the pool's median Hammer raw falls 218 → 195 as squads shrink, so an unchanged raw ranks higher. The two Space Marine squads lose about as much to turns as Hive Guard, but at minimum size a leader pair becomes their best context, and that more than makes up the loss.
The regression (every multi-size unit, best size → minimum size)
298 fieldable units with more than one legal size (two spore units with no damage left out); 230 change size between B0 and B1, 68 already sat at minimum. Mean DPS move −13.1 (−17.8 among the 230). ΔDPS regressed on the pool shift (in percentile points) and the log change of each other part: R² 0.82 (0.80 among the 230). Each part's share of the R² (averaged over every order the parts can enter), and its share of the absolute moves (the exact Shapley split):
| Part | Share of R² (all 298) | Among the 230 | Share of the absolute moves | Mean contribution (230) |
|---|---|---|---|---|
| Turns alive (delivery) | 58% | 50% | 61% | −20.4 |
| Context re-picked at the new size (detachment, leader pair) | 21% | 27% | 13% | +0.4 (large both ways; nets out) |
| Per-turn damage per point (a joined leader's fixed points; uneven points per model) | 10% | 9% | 8% | −1.7 |
| Way in picked | 6% | 7% | 2% | −0.7 |
| Pool shift | 4% | 7% | 16% | +4.6 (everyone, the same way) |
Turns alive is the largest part in 135 of the 146 drops of 10 or more. Among the 80 units whose best job at best size is DPS and whose size changes, the mean move is −16.2, of which −21.3 is turns. Mean T over the 230 falls 4.21 → 3.05. (The turns part is the remainder after per-turn damage, context and way; on the 65 squads alone in the same context and the same way it tracks the log change in T with slope 1.3, R² 0.49. The rest is delivery's shape: turns out of range don't count, and a turn alive counts in full only up to the cap.)
The answer
(a): the drop is mostly turns alive. Hammer counts damage per point but survival in absolute hits, so halving a squad halves its turns and its damage delivered per point. Hive Guard and Termagants are pure cases. The Zoanthropes' drop is smaller and is (b), per-turn dilution by the joined Neurotyrant's fixed points; the Space Marine pairs are rescued by a context re-pick. The Tyrannofex's rise is the pool shift.
What it means for the minimum-size convention: under today's survival term, minimum size isn't neutral. It penalises every multi-model damage unit by about 20 percentile points through turns alive, and best legal size did the reverse (round 2 learnt the same thing at the other end: "per-point damage × whole-unit survival rewards size"). Keep minimum size only with a survival term that doesn't scale with squad size. Otherwise the convention ranks squads by how many models they bring, not by what a point buys.
The field's mirror (design/matrix.md 7.5) makes survival a hazard rate per wound, the same at any size, so it would remove the turns part, about 60% of these moves, and most of the pool shift with it. That has nothing to do with targets: it holds whatever the field shoots at. Hive Guard and Termagants would keep their best-size DPS at minimum size. The leader dilution (Zoanthropes) and the context re-picks would remain, because they're per-turn damage and picking, not survival. Matrix.md's question 7 (a fixed enemy fire budget) would bring back a small size effect, by design.
Paste-ready line for design/decisions.md
## 3 Oct: the size oddity explained (T-586, analysis only)
- **Finding:** B0 → B1 (best size → minimum size) moves DPS by −17.8 on average over 230 units that change size; ΔDPS on its parts, R² 0.82: turns alive 58% (61% of the absolute moves; Hive Guard 86 → 65 and Termagants 52 → 30 are all turns), context re-picked 21%, per-turn damage 10% (a joined leader's fixed points: Zoanthropes with the Neurotyrant 85 → 79, not 65), way 6%, pool shift 4% (+4.6 for everyone: the Tyrannofex 84 → 88). Cause: Hammer is damage per point × survival in absolute hits (round 5), so size enters through turns alive. Round 2 learnt the same at the other end ("whole-unit survival rewards size").
- **Kept / reversed:** nothing changed. Minimum size is neutral only with a size-free survival term; the matrix's mirror (survival as a hazard per wound) is one, and removes the turns part for a reason unrelated to targets. Report: reports/tiers/2026-10-03-size-oddity.md.