The question. T-675's screen dials lift World Eaters by about 5 held out, but they lift every cheap squad, the ones reviewers rate low (Wracks D, Hearthkyn C, Daemonettes C) as well as the ones they rate high. Hypothesis: an army needs only a couple of screens, and reviewers credit the best ones for that job. Answer: no, not across armies. Slots cut the cost to the other armies by about two thirds. But the screen each army ranks best by our measures is often the squad reviewers rate low, so slots add nothing held out.
What was built (ledger.js, code only, every dial 0 = today exactly):
screenSlotsk (1 to 3): each army's units are ranked by screen worth, Σ w × Q over T-675's area, block, score and trade at the dial strengths, taken at each unit's best solo form. Only the top k get the screen credit, still w × Q × P ÷ points.screenNextx: the next unit gets x of the credit and the one after gets x².screenRanksets the ranking. 0 ranks per point among the units that can do the job (they can do actions, have OC above 0 and have 3 or more models). 1 ranks by the worth itself, over every unit.- One fix: Battle-shock's reference Value now leaves the screen credit out, as the screens' own reference already did. Before, screen credit raised shock's P and inflated shock units: Culexus's shock went from 328 to 731 at area 2. T-675's figures barely move: trade 0.01 is still +0.15 / Tyranids +0.19, area 2 −3.3 / Tyranids −9.0 (was −8.3).
Held out across armies. Strengths come from coordinate ascent over area, block, score and trade (grid 0 to 3) on 10 armies and are scored on the other 5. That's 3 seeded splits, the same as T-675's (Track A: pooled 59.4, Tyranids 79.3, World Eaters 67.8). The in-sample columns are the strengths picked on all 15.
| k | x | Per point among screens: held-out Δ | in sample: pooled / Tyr / WE | Worth overall: held-out Δ | in sample: pooled / Tyr / WE |
|---|---|---|---|---|---|
| no slots | −0.06 ± 0.03 | +0.10 / −0.17 / +0.21 | |||
| 1 | 0 | −0.35 ± 0.19 | +0.40 / −2.01 / +0.50 | −0.34 ± 0.14 | +0.36 / −0.75 / −2.13 |
| 2 | 0 | −0.21 ± 0.18 | +0.28 / −1.13 / +3.50 | −0.14 ± 0.33 | +0.46 / +0.42 / +0.23 |
| 3 | 0 | +0.03 ± 0.41 | +0.57 / −2.57 / +4.19 | −0.56 ± 0.17 | +0.62 / −0.14 / −0.39 |
| 2 | 0.5 | −0.14 ± 0.22 | +0.18 / −0.65 / +1.17 | −0.03 ± 0.49 | +0.76 / +0.07 / −3.22 |
| 3 | 0.25 | −0.60 ± 0.28 | +0.63 / −1.71 / +3.88 | −0.03 ± 0.37 | +0.57 / +0.57 / −1.70 |
The other four structures (x 0.25 or 0.5 at k 1 and 2, and k 3 x 0.5) held out between −0.63 and −0.20. Letting the split's 10 armies choose the structure too gave −0.23, −0.04 and −0.71, a mean of −0.33. Ranking without trade (trade can push Terminators into a slot) tells the same story: every fixed structure ≤ −0.02, structure chosen too +0.15. Every in-sample gain (+0.2 to +0.8) disappears held out.
The exam (node tools/exam.js, 2,000 draws, against Track A):
| Candidate | Pooled Δ [95%, units] | Armies better | Tyranids | World Eaters | Worst armies |
|---|---|---|---|---|---|
| Cross-army pick: k 3 per point, area 0.5, block 0.01, trade 0.05 (in sample) | +0.6 [−0.4, +1.6] | 7/15 | −2.6 | +4.2 | Drukhari −1.4, DG −1.1 |
| World Eaters' slotted pick: k 1, x 0.25 per point, area 1.5, block 1.5, score 0.1, trade 0.35 | −0.7 [−2.3, +0.9] | 3/15 | −3.0 | +7.4 | Drukhari −7.7, LoV −5.4, EC −2.6 |
| World Eaters' unslotted pick: area 1.5, block 0.1, score 0.2, trade 0.1 | −2.1 [−4.5, +0.3] | 4/15 | −5.2 | +8.6 | AoI −8.5, LoV −6.7, Orks −6.3 |
World Eaters' lists, one held out (the setting chosen on the other two lists):
| Setting chosen | Red Path | Tactical Sugar | Exalted | Mean held out | Pooled 15 | Tyranids |
|---|---|---|---|---|---|---|
| Unslotted (T-675's dials) | +13.4 | +1.5 | +6.6 | +7.2 | −1.8 to −4.9 | −4.5 to −12.9 |
| Slotted (structure chosen too; k 2 each time) | +9.5 | +6.3 | +4.3 | +6.7 | −0.5 to −0.9 | −3.1 to −5.7 |
Cheap units (percentile by Value among the units the army's lists grade, as T-675's table; letters in list order):
| Unit (army, Track A's points) | Letters | Track A | Cross-army pick | WE slotted | WE unslotted |
|---|---|---|---|---|---|
| Jakhals (World Eaters, 130) | S S S | 34 | 69 | 100 | 93 |
| Flesh Hounds (World Eaters, 150) | A A A | 28 | 28 | 83 | 69 |
| Khorne Berzerkers (World Eaters, 160) | A A S | 48 | 41 | 38 | 59 |
| Gargoyles (Tyranids, 155) | B B B A S | 12 | 12 | 12 | 39 |
| Termagants (Tyranids, 110) | B B C B A | 27 | 51 | 67 | 67 |
| Hormagaunts (Tyranids, 120) | S A B A S | 67 | 63 | 63 | 71 |
| Cadian Shock Troops (Astra Militarum, 145) | A | 80 | 93 | 80 | 95 |
| Kasrkin (Astra Militarum, 105) | A | 92 | 88 | 92 | 90 |
| Gretchin (Orks, 80) | B | 11 | 52 | 100 | 91 |
| Wracks (Drukhari, 120) | D | 57 | 74 | 100 | 78 |
| Hearthkyn Warriors (Leagues of Votann, 90) | C C | 79 | 84 | 100 | 84 |
| Hernkyn Yaegirs (Leagues of Votann, 90) | A – | 16 | 63 | 16 | 63 |
| Nurglings (Death Guard, 90) | A | 11 | 11 | 11 | 37 |
| Fenrisian Wolves (Space Wolves, 75) | B | 10 | 10 | 10 | 70 |
| Daemonettes (Emperor's Children, 100) | C | 100 | 100 | 95 | 100 |
Read. Slots do what they were built to do. At the World Eaters' strengths, the pooled cost falls from about 2 to about 0.7, and the World Eaters keep almost all of their gain (+6.7 held out against +7.2). Jakhals become the army's top unit, and Flesh Hounds rise from the 28th percentile to the 83rd.
The hypothesis fails because of which unit each army ranks first. With one slot, the Drukhari's best screen per point is the Wracks (D), Votann's is the Hearthkyn (C) rather than the Yaegirs (A), and Death Guard's is the Poxwalkers rather than the Nurglings (A). The slot then lifts the squad reviewers rate low to the top of its army. At three slots, tough units also take slots through trade (Terminator Assault Squads, Meganobz, Blightlords). Reviewers rate the screens our quantities can't tell apart, so ranking by those quantities picks the wrong one about as often as the right one.
Decision: kept at 0, not on the page. Nothing held out positive across armies (best +0.03 ± 0.41; the structure chosen too gave −0.33). For the World Eaters alone, slotted screens are the better army-level setting: one or two slots ranked per point, area about 1.5, block about 1.5, trade about 0.35 (the -we.json candidate). They give +6.7 held out, at a third of the unslotted cost elsewhere.
How it's done: node tools/screenslots.js [--write] (about 70 s). Each army goes through compute() once at 0, and each setting is then a sum per solo form. That fast path is checked against compute() on all 16 armies (largest gap 2e−13). Candidates are in tools/ledger-benchmarks/2026-10-05-screen-slots.json, -we.json and -we0.json.