The question. The World Eaters have 30 units and three full reviewer lists (Red Path, Tactical Sugar, Exalted). Track A ranks them 67.8% pairwise (Track B 69.4%). How much of the gap can the army's own make-up close, added as dials? The ceiling is how well the reviewers agree with each other on the same pairs. One reviewer's letters scored against another's tiers give 82.4% on average when a tied letter counts half (73.4% when ties count wrong). The mean of two reviewers scored against the third gives 85.2% (80.0%). Our Values almost never tie, so 82 to 85 is the ceiling, and we are about 15 points below it.
Where we miss. Miss is the reviewers' consensus position (the mean letter, as a percentile in the army) minus ours (the Value's), as tools/misses.js computes it. Our characters and melee monsters sit too high. The army's guns, its line squads and its transport sit too low.
| Unit | Letters (RP, TS, Ex) | Theirs | Ours | Miss | Our guess why |
|---|---|---|---|---|---|
| Jakhals | S S S | 97 | 35 | +62 | A cheap screen and objective unit. Objective Ravaged (OC) and Icon of Khorne (more Blessings dice) are noted, not modelled. Track A's Kill weight dwarfs Score and Hold. |
| Maulerfiend | B C C | 15 | 68 | −53 | Big melee Kill per point with every charge landing and no screens. Next to the Forgefiend the reviewers want guns, not another hitter. |
| Chaos Rhino | B B A | 50 | 2 | +48 | A transport for a melee army. It delivers Berzerkers and Jakhals, and the Carry line is off (wCarry 0). |
| Lord on Juggernaut | C B C | 15 | 62 | −47 | A cheap character rated on its solo Kill per point. The reviewers rate it as a leader up against Khârn and the Master of Executions. |
| Angron | B B A | 50 | 95 | −45 | The heal kind (Reborn in Blood, ×1.805) and the aura kind (×1.12) multiply its Value. Both were fitted on the Tyranids. |
| Daemon Prince with wings | C B C | 15 | 58 | −43 | Solo melee Kill per point. It competes with the walking Daemon Prince, which the reviewers rate S S B. |
| Flesh Hounds | A A A | 68 | 28 | +40 | Fast Daemon screens. Hunters from the Warp is a noted arrival, and as Daemons they get no Blessings. |
| Forgefiend | A S S | 90 | 52 | +38 | The army's few guns. Role scarcity weighs the classes a unit kills, not that ranged Kill is rare in this army. |
| Defiler | A A A | 68 | 32 | +37 | The same: guns in a melee army. |
| Khorne Berzerkers | A A S | 80 | 48 | +32 | The core squad. Blood Surge and Icon of Khorne are noted only. Cheap OC bodies. |
What the ledger already uses (step 2). The engine's Kill does include each unit's unconditional unitfx abilities, the detachment's effects (each form takes one detachment, Brazen Engines for most) and a joined leader's lend. It also includes an army rule's plain effects, but not a rule's own choice. That leaves out Blessings of Khorne's dice blessings (Sustained Hits, Lethal Hits and Devastating Wounds in melee), which are partly in armyfx.json. Of the World Eaters' own 71 datasheet rules (Legends aside), 16 are modelled, 9 partly and 46 noted. The noted ones are mostly moves (Blood Surge, Scout-like moves), Battle-shock (Terror of Khorne), mortal wounds outside attacks and heals. They reach Value only through the Tyranid-fitted rule-kind multipliers: heal 1.805, cp 1.88, buff 1.12 and detachment 1.065. One is filed under the wrong kind: Khârn's Berserker Frenzy is fight-on-death, but it is filed as a heal and takes the heal multiplier. Removing the heal or cp multiplier costs the World Eaters 3 to 4 points, and shrinking every kind toward 1 gains at most +0.3. So the multipliers are not what is wrong here.
| Dial (strength 0 = today exactly) | Strength (WE's choice) | WE held out Δ (folds) | Pooled exam Δ, 15 armies [95%] | The other 14, mean Δ ± SE | Tyranids Δ |
|---|---|---|---|---|---|
modeScarcity (new, general: the army's scarcer attack mode is worth more) | 0.2 in every fold | +2.4 (+3.9, +1.8, +1.4) | +0.0 [−1.0, +1.0] | −0.1 ± 0.4 | +0.0 |
armyPick (new, general: an army rule's own choice in Burst; Blessings' Sustained + Lethal Hits) | 0 in every fold (1 gives −2.2) | 0.0 | −0.1 at 1 | 0.0 (no other army has the data) | 0.0 |
| A leader's Value × (1 − d) (a preview, not built: for T-673) | 0.2 in every fold | +2.3 (+4.9, +2.1, 0.0) | – | −0.2 ± 0.3 | −0.1 |
smallPiece (T-670, existing) | 0.15 in every fold | +1.8 (+2.8, +2.4, +0.3) | +0.6 | +0.5 ± 0.4 | −0.6 |
attachChars (existing switch) | on (no strength to choose) | +2.2 in sample | +1.1 | +1.0 ± 0.8 | −0.0 |
modeScarcity + smallPiece, chosen jointly | 0.15 to 0.2 each | +4.5 (+7.0, +5.4, +1.1) | +0.7 | +0.4 ± 0.7 | −0.2 |
aircraftBoard (T-670) | any | 0.0 (the Heldrake can act in the data, so it isn't treated as an aircraft) |
What modeScarcity does. r is the army's ranged share of Kill, summed over each datasheet's best form. Ranged Kill is multiplied by r^−m and melee Kill by (1 − r)^−m, then both are scaled so the army's total Kill stays where it was. The World Eaters' r is 0.11 on Track A's Burst, so at m 0.2 their guns count about 1.5 times as much as their melee. The Forgefiend moves from 15th to 7th and the Defiler from 21st to 10th, while the Maulerfiend falls from 10th to 12th. The Space Wolves (r 0.04), Blood Angels and Emperor's Children (0.08 to 0.09) are just as melee-heavy. The Astra Militarum (0.72) and the Space Marines (0.63) tilt the other way. On the exam it helps 3 armies: the Emperor's Children +4.2 and the Thousand Sons +0.3, beside the World Eaters. It costs the Death Guard 3.4 and 7 more armies 0.7 or less each, and the other 4 score the same.
What armyPick does. tools/burst.js --field --army-pick measures each form's Burst with option 0 of each army rule's choice (tools/ledger.js setArmyPick). The dial blends Burst toward that. Blessings raise the World Eaters' units' Burst by 13 to 45% (the Jakhals by 40%, the Goremongers by 83%). The Daemons (Bloodletters, Bloodcrushers, Hounds, Skarbrand, the Bloodthirster) and the guns get nothing, so the melee monsters rise along with the squads. That moves the order the wrong way.
Read: one dial is worth having. modeScarcity is on /lab/ledger/ as "Army quirks (T-672)", off by default. It held out +2.4 on the World Eaters and costs the other armies nothing beyond noise. With smallPiece it held out +4.5 (72.3%), about a third of the way to the reviewers' ceiling. armyPick stays in code at 0, off the page. Character gaps for T-673: World Eaters leaders are overrated alone (Lord on Juggernaut, Lord Invocatus, the Master of Executions, Khârn). Cutting a leader's Value by 0.2 gains WE +2.3 held out but nothing elsewhere. Their Value comes from solo Kill per point and the Tyranid heal, cp and aura multipliers. Khârn's Berserker Frenzy is filed as a heal. The noted leader rules (Aggressive Advance, Forwards For Blood!, A Worthy Skull's CP) aren't modelled. Nothing is adopted into Track A or Track B: the supervisor decides.
How it's done: Track A is tools/ledger-benchmarks/2026-10-05-step34b-baseline.json at wCarry 0, through compute(), as tools/exam.js scores it (67.8 on WE and 59.4 pooled reproduced exactly). Leave-one-list-out: on each fold the strength with the best mean pairwise on two lists is picked (from a grid: modeScarcity 0 to 1, armyPick 0 to 1 by 0.25, smallPiece 0 to 0.5) and scored on the third. The pooled interval is from node tools/exam.js on the saved candidate tools/ledger-benchmarks/2026-10-05-we-modescarcity.json (2,000 draws). The armyPick file comes from node tools/burst.js --field --army world-eaters --army-pick (24 s) and is merged as burstPick. That ran by hand: rerunning tools/ledger-merge.js on a build without the prefer and life files drops those fields.