5 Oct 2026. A review: no model, number, setting, benchmark, data file or frozen input changed. Base: the confirmed step-33 baseline, tools/ledger-benchmarks/2026-10-05-step33-baseline.json (seven headline lists ρ 0.550, pairwise 73.7%; Tyranids 77.4%, Space Marines 64.5%; the 14 never-fitted armies 59.7%; 16 armies 61.1%). Every number below was run read-only through docs/lab/ledger/ledger.js's compute(). The previews in section 4 re-value each unit's forms after compute() and pick its best form again; they are previews, not steps: no switch exists for them, no re-fit and no held-out run was done, and two of them carry a constant chosen after looking (said where it applies). Our own words and numbers; unit names only; no rules text. Tournament shares are from grimstat-corpus (CC BY 4.0), July to September, via docs/data/inclusion.json.
The short answer
- The ledger ranks units by how much they kill per point in one go, times a few multipliers for their rules. The multipliers are where most of the misses come from. Support rules (Synapse, command points, healing, auras) are applied as a multiplier on the unit's whole worth, which is mostly its own damage. So a hard-hitting Synapse creature has its damage doubled (Winged Hive Tyrant ×4.1, Zoanthropes ×2.4, Tervigon ×2.4, both Warriors ×2.2, Old One Eye ×1.8 for healing), while a support piece that barely fights gets almost nothing from the same rule (Neurolictor, Neurotyrant), and a unit with no such rule looks worse by comparison (Tyrannofex, Exocrine, Genestealers, Raveners, Gargoyles, Tyrant Guard, Termagants: multiplier 1.0 to 1.12).
- The pattern is almost clean. Every one of the 8 "too high" units carries a multiplier of ×1.8 or more. Of the 9 "too low", 7 carry none (or ×1.12), and the other 2 are support pieces whose own damage is too small for a multiplier to matter. Take the support multipliers away and 12 of the 17 named units move closer to the reviewers, 7 of them to within 3 places (Winged Hive Tyrant 4 → 22, Tervigon 21 → 38, Tyrannofex 26 → 14, Exocrine 19 → 8, Genestealers 29 → 16, Raveners 27 → 15). But the Hive Tyrant, the Swarmlord, the Neurolictor and the Neurotyrant then collapse. The multipliers are holding those four up, and they lift the wrong units as a side effect.
- Two of the fitted numbers are labels, not game effects. The Leapers' "command points" rule is really a cheaper counter-charge; it still multiplies their damage by 1.88. Fights First is ×2.64 on melee damage, fitted on three units (Lictor, Leapers, Deathleaper), and the reviewers love one of them (the Lictor) for other reasons. Our own rules notes say every charger strikes first in 11th edition, so Fights First barely matters on the charge.
- Some misses aren't ours to fix. The reviewers grade the Leapers anywhere from S to C and the Deathleaper from S to D, and 52% and 30% of tournament lists take them. Zoanthropes are in 70% of lists. A "consensus rank" for these units is an average of disagreement, and in the crowded middle of the table a 20-place gap is about half a letter.
- What was wasted. About 15 model steps added lines on top of the multiplier structure, so none could move the misses it caused. Steps 13 and 14 (sharing a joined unit's worth between its members) were spoiled because a leader's multipliers also multiply its squad's damage. Since step 25 there is a second trap for any re-run: joined units still use the old damage measure while units alone use Burst. The mission models (27, 28) priced scoring well enough but couldn't separate cheap characters, because nothing in the mission deck separates them.
- What to build, in order (section 3, with previews):
- Battle-shock pressure priced in points, only for rules that force enemy tests without needing to do damage first (the Neurolictor's, the Neurotyrant's). Preview: +1.1 points on the headline (±0.9), Neurolictor 28 → 12, Neurotyrant 37 → 30, nothing else moves more than a place, Space Marines and the 14 armies unchanged.
- Support as a credit, not a multiplier: each support rule adds a fixed amount per point instead of multiplying the unit's own damage. Preview at half the current strength: +1.1 (±0.9); together with (1), +2.3 (±1.4, positive in 98% of resamples), Tyranids 77.4% → 80.7%, the 14 armies −0.2. The strength was picked from a grid after looking, so it has to be re-fitted and judged on the 14 armies before anything is kept.
- Joined units measured on the same damage as single units, then squads take their fair share. It lifts exactly the right squads (Genestealers 29 → 15, Raveners 27 → 17, Tyrant Guard 45 → 30), but Carnifexes led by Old One Eye jump to 5th. It's blocked until we know why that pair is worth so much.
- When to stop: after these three. If the 14 never-fitted armies are still near 60%, stop building lines. A ledger built only from the rules seems to top out near 80% on Tyranids, against 86% for simple list share. From there it is better used to explain a tier list than to set it.
1. A blunt appraisal
What is sound
- The frame and the discipline. Worth in points, line by line, per point, against a stated meta; one change per step, written predictions, the movement view, the 14 never-fitted armies as the honest gauge. Steps 29, 30 and 33 (pulling thinly fitted numbers back toward neutral, scoping the Biovore dial) were the right kind of step and they transferred.
- Burst Kill (step 25) is the one model change that improved Space Marines by a lot and carried to the 14 armies. Only finished targets count; that matches how players think.
- The diagnoses were right. The independent review (T-648) and the hedonic fit (T-651) both say the misses are the units whose worth isn't in their own dice. That still holds. This critique adds why the ledger gets them wrong in a particular direction: the multipliers.
What has been wasted
- Lines stacked on a broken base. A rule's multiplier on the whole worth acts mostly on Kill (91% of Value). Every new line since step 6 was judged with that in place: synapse at the margin (6), leaders leading (10), joined shares (13, 14), arrival (18), scarce answers (19), the mission lines (27–28b). Each one moved its target units and broke others, because the "others" were held up by multipliers on their damage.
- Joined-unit steps judged on a mismatched measure. Since step 25, units alone read Kill as Burst while joined units keep the old mean Kill. So any "what the pair adds beyond its members" partly measures the gap between the two damage measures. Even before that (steps 13 and 14 ran on the step-6 base, before Burst), it partly measured the leader's multiplier applied to the squad: Old One Eye's heal ×1.8 on the Carnifexes' damage, or the Winged Prime's Synapse ×2.18 on the Gargoyles'. Steps 13 and 14 were never a fair test of the idea.
- Fitting multipliers by tag. "Command points" covers four very different rules: a cheaper stratagem every round, a point for each character killed, a cheaper counter-charge, and the Swarmlord's several. "Battle-shock pressure" covers a test forced every round with no damage needed, and tests that only follow damage the unit already deals. One number per tag, fitted on the units the reviewers like, leaks to every other carrier of the tag.
Assumptions that may be wrong, checked where we could
| Assumption | What we found today | Verdict |
|---|---|---|
| Units ranked at their best solo form, joined forms excluded | Tyrant Guard (joined 3,041 vs solo 431 a hundred points), Gargoyles (1,273 vs 351), Genestealers (2,165 vs 948), Raveners (2,507 vs 965), Neurotyrant with Zoanthropes (1,737 vs 599) can't reach their grades alone. A squads-only share (leaders keep their solo row) moves four of them to within a few places of the reviewers, but only once joined units use Burst too, and Carnifexes break it (section 4) | Right as a default; wrong for squads that are taken with a leader. Fix the measure first (proposal 3) |
| Per point (Value per 100 points) | Grading by total worth instead (Value × points ÷ 100) loses 2.9 points on Tyranids but gains 1.5 on the 14 never-fitted armies at half strength (γ 0.5; 1.1 at full). Tyranid reviewers grade closer to per point than the others do | Keep per point; ask Jordan whether a tier means "per point" or "worth taking in a list" (section 5) |
| The typical enemy army | Not the source of these misses: no named unit's gap is explained by its target classes | Fine for now |
| One-turn Kill | Favours alpha strikers (Lictor, Leapers, Deathleaper: Infiltrators plus Fights First) over units that kill every turn. Tested as Kill × a per-point life share: within ±0.6 points, Tyrannofex 26 → 25. With support off, the Tyrannofex is already 14th and the Exocrine 8th, so their gap is mostly the multiplied units above them | Not the main cause |
| Reviewers grade per unit in a list | They grade a unit in its usual role (Gargoyles with a Winged Prime, Tyrant Guard with a Tyrant). Mean letters of 5 correlated reviewers make a crowded middle: Leapers (S B A B C) and Deathleaper (A S A B D) average out to "22nd" | Report the gap in letters as well as places; treat high-disagreement units as contested |
| Detachment choice (best of any) | The over-high melee infiltrators (Lictor, Leapers, Deathleaper), the Winged Hive Tyrant and Zoanthropes all take their best form in the arrival detachment, whose extra arrival rule is credited through the charge model | Generous but not tested here; a card (best form in the two most-played detachments) |
| 10th-edition lists in the yardstick (Maelstrom, Astrategas) | Astrategas is the weakest list (ρ 0.45). In steps 13–14 the two 10th-edition lists rose while the three 11th-edition lists fell | Report the 11th-edition-only mean beside the headline; keep them for now |
2. The misses, unit by unit, grouped by cause
How to read: "ours → reviewers" is the rank among the 52 graded Tyranids → the consensus rank (mean letter over the five lists). "×" is the form's multiplier on its whole worth. Value is per 100 points. "Support off" is the rank with the army-wide support multipliers set aside (a diagnostic, not a proposal). Share is the share of the 23 Tyranid tournament lists that take it.
Group A: support multipliers on the unit's own damage (too high)
| Unit | Ours → reviewers | Letters, share | What it does in a game (our notes) | The line that over-credits it | Support off | The mechanic that fixes it |
|---|---|---|---|---|---|---|
| Winged Hive Tyrant | 4 → 25 | A B B B B, 4% | A flying Synapse monster with a psychic debuff and a cheaper-stratagem aura; can't take a bodyguard, so it dies fast for its cost | Synapse ×2.18 × command points ×1.88 = ×4.10 on a middling melee Kill (Value 3,273; without them about 800) | 22 | Support as a credit (proposal 2) |
| Zoanthropes | 9 → 19 | S A B B B, 70% | Psychic shooting with a strong save, a 6+ save aura for nearby friends, Synapse; often led by a Neurotyrant | Synapse × aura ×2.44 on its ranged Kill (and role scarcity ×1.76) | 17 | Proposal 2; the aura as a Soak credit for friends would be the honest form (later) |
| Tyranid Warriors (melee) | 13 → 28.5 | B B B B B, 17% | Synapse bodies, decent melee, cheap OC | Synapse ×2.18 on Kill | 21 | Proposal 2 |
| Tyranid Warriors (ranged) | 16 → 43 | C C C D C, 4% | Synapse bodies that shoot; six of them hold an objective | Synapse ×2.18 on Kill, Score 86 and Hold 134 | 29 | Proposal 2 |
| Tervigon | 21 → 40 | B C C C D, 0% | Spawns Termagants, Synapse, a buff aura; slow and big | Synapse × aura ×2.44 on its small Kill, plus Spawn | 38 | Proposal 2 |
| Old One Eye | 12 → 28.5 | B B A B C, 17% | A melee monster that heals; leads Carnifexes; Jordan: gets melted by anti-tank | Heal ×1.805 on its whole worth (healing should add to its Soak, not to its damage) | 12 (role scarcity ×2.33 holds it up) | Proposal 2 (heal as a Soak credit); role scarcity's monster tilt is the rest |
Group B: labels and a fit that leak (too high, and contested)
| Unit | Ours → reviewers | Letters, share | What it does | The line | The fix |
|---|---|---|---|---|---|
| Von Ryan's Leapers | 2 → 22 | S B A B C, 52% | Cheap infiltrating melee pack, fights first, counter-charges cheaply | Fights First ×2.64 on melee Kill and command points ×1.88 from a rule that is really a cheaper counter-charge (filed as "cp" in the effects file) | Re-file the counter-charge rule (a data correction: Value ÷ 1.88, about 11th by our Value table); Fights First at its game meaning (below) |
| Deathleaper | 6 → 22 | A S A B D, 30% | Lone, hidden character killer with a Leadership aura; the reviewers split | Fights First ×2.64 and command points ×1.88 from "a point per character killed" | Fights First as a defensive effect; that command point counted as the few it yields, not ×1.88 |
| (Lictor, for contrast) | 1 → 3 | S A S S S, 48% | Cheap lone scorer, and it lets friends arrive from reserves cheaply (our data counts that as the Lictor's own arrival) | The same two multipliers: the fit was pulled up by this unit | The friends' arrival rule as a support credit (it's measured on the wrong unit now) |
Fights First: design/rules-notes.md says every charger gets it in 11th edition, so the ability matters mainly when the unit is charged. That's a defensive effect, which belongs in Soak's melee side, not ×2.64 on the unit's own damage. At ×1 the three fall about 12 places each (Leapers 2 → 14, Deathleaper 6 → 18, Lictor 1 → 12). The headline is unchanged (−0.1), and the review found the 14 armies lose 0.9, because the multiplier stands in for "a good melee unit" there too. So it shouldn't be changed alone: it needs the Lictor's real support credited first (above).
Group C: support with little damage of its own (too low)
| Unit | Ours → reviewers | Letters, share | What it does | Why the ledger misses it | Fix |
|---|---|---|---|---|---|
| Neurolictor | 28 → 1.5 | S S S – S, 74% | Infiltrates, can't be shot from afar, Synapse; forces an enemy unit near it to test Battle-shock every round and makes shocked enemies easier to wound and worse at hitting | Its Kill line is about 55 a hundred points (the army's median about 160), so ×4.10 on it buys little; the forced tests (OC denied on objectives) are on no line (debuff kind at ×1) | Battle-shock pressure (proposal 1) |
| Neurotyrant | 37 → 17 | A – – A B, 74% | Makes the army's Battle-shock tests harsher, extends Synapse; leads Zoanthropes with +1 to hit and wound | Same: small Kill × 2.18; its leading worth sits in a joined form we don't rank | Proposal 1; its joined form (proposal 3) |
Group D: no multiplier, so passed by the multiplied (too low)
| Unit | Ours → reviewers | Letters, share | What it does | The line | Support off |
|---|---|---|---|---|---|
| Tyrannofex | 26 → 6.5 | S A A A S, 30% | The army's specialist anti-tank gun on a very tough body (first on Soak by far: 1,272 a hundred) | Kill unmultiplied (×1); Soak is only 2% of Value at these weights | 14 |
| Exocrine | 19 → 6.5 | A A A S S, 17% | Elite-infantry killer; marks a target for friends | ×1.12 only | 8 |
These two need nothing new. Once support stops multiplying the units above them, they rise (proposal 2: 26 → 23 and 19 → 17 at half strength; 14 and 8 with support off). What's left is that the ledger's single turn of Kill undersells a gun that fires every turn from a body that lives the whole game. Our per-point life preview couldn't fix that without lifting the Hierophant and Harridan too.
Group E: worth in a joined unit (too low)
| Unit | Ours → reviewers | Letters, share | Usual partner (best joined form, Value a hundred) | Squad-share preview (proposal 3) |
|---|---|---|---|---|
| Tyrant Guard | 45 → 32.5 | B B A C C, 13% | The Swarmlord (3,041) or a Hive Tyrant: bodyguards | → 30 |
| Gargoyles | 46 → 19 | B B B A S, 13% | Winged Prime (1,273); also shoot-then-move, which the ledger files as a melee-landing multiplier on a unit with no melee, so it does nothing | → 43 (38 with today's multipliers) |
| Genestealers | 29 → 14 | A A A A B, 48% | Broodlord (2,165) | → 15 |
| Raveners | 27 → 14 | S B A B A, 52% | Hyperadapted Raveners (2,507); return to reserves and re-arrive | → 17 |
| Termagants | 38 → 28.5 | B B C B A, 17% | A Prime (1,754); cheap OC and screens | → 37 (34 at base) |
Two more data corrections sit in this group: shoot-then-move rules on ranged units (Gargoyles; Termagants carry a movement rule there too) act on melee landing, which is zero for them. The rule should act on reach for scoring, which the mission model has.
3. The three most promising changes
Proposal 1: Battle-shock pressure, priced in points (switch shockPressure)
- In game terms. A Battle-shocked enemy unit has OC 0 for a round, so the objective it stands on can flip or be denied. Some rules force those tests with no damage needed: one forces a test every round on an enemy unit near it, and another makes the whole army's tests harsher. Others only test a unit the carrier has just hit, killed or engaged. Their worth is mostly the damage, which Kill already counts.
- Formula. Worth W points a game for each unconditional rule, from the VP bridge (2,000 ÷ 65 ≈ 31 points a VP): W = (enemy units in reach) × (chance a test fails) × (rounds) × (VP lost per shocked holder) × 31. Value per 100 = W ÷ the form's points × the army's median Value a hundred (the median unit is worth its points; step 27's bridge). It is added to Value, outside the weights and multipliers. Conditional triggers count 0.
- The constants, by reasoning (set before the preview). A test forced every round from a forward deployer: about 2 enemy units in 12", a failure chance about 0.35, 4 rounds, about 1 VP lost per shocked holder: 2.8 VP, W ≈ 86. A rule that makes tests harsher: the once-a-game army test plus others, about 10 units × about 0.14 more failures each, so 1.4 VP, W ≈ 43. No fitted number.
- Predicted (and previewed) moves. Neurolictor 28 → 12, Neurotyrant 37 → 30; everything else a place at most. No Space Marine or other-army form carries an unconditional rule in today's data, so they don't move.
- Risk. As built it is a two-unit line, close to a labelled hand dial. Before it's adopted, every army's battle-shock rules in docs/data/unitfx.json need classifying as unconditional or after-damage (a data card, in our words), so it is a rule of the game and not a fix for two units. The tag-wide version shows the danger: crediting every "Battle-shock pressure" rule lifts the Deathleaper to 1st, the Parasite 36 → 14, the Chaplain 64 → 3 and the Reivers 17 → 2 (preview V, section 4).
- How it differs from what failed. Step 6 priced our own Synapse coverage at the margin (small, since the list already has plenty). This prices the enemy's lost objective control. Steps 27 and 28 priced what a unit scores, not what it takes away from the enemy.
Proposal 2: support as a credit, not a multiplier (switch supportCredit)
- In game terms. An aura, a Synapse bubble, a cheaper stratagem or a heal is worth about the same to the army whether the carrier hits hard or not. Today it multiplies the carrier's own damage, and in a joined unit its squad's too.
- Formula. For the army-wide kinds (Synapse, command points, buffs, debuffs, heals), the form's Value = (its Value without those multipliers) + κ × Σ(m − 1) × the army's median support-free Value a hundred, m being each kind's multiplier today. The detachment kind stays a multiplier, since a detachment does change the unit's own output. In a joined form the credit attaches to the form, not to the squad's damage.
- Constants. One new number, κ (the credit's strength, keeping the five multipliers' ratios). Reasoned candidates: κ = 1 (the multiplier read at the median unit) and a "same total support in the army" rule (κ 1.24 on Tyranids) both preview at −0.9. The best of the grid {0, 0.1, 0.25, 0.5, 1} is κ 0.5, chosen after looking, so it is fitted. The step should fix the shape, re-fit κ with the five multipliers through tools/ledgerfit.js (ridge toward 1, as the review advised), and judge on the 14 armies and held-out Tyranid units.
- Predicted moves (κ 0.5 preview). Winged Hive Tyrant 4 → 6, Zoanthropes 9 → 11, Tervigon 21 → 27, ranged Warriors 16 → 20, Neurolictor 28 → 24, Exocrine 19 → 17, Tyrannofex 26 → 23, Hormagaunts 17 → 13, Biovores 7 → 1. Space Marines: Captains down (Captain 7 → 20, Captain in Terminator Armour 14 → 23), Infiltrators 23 → 36 (away from the reviewers).
- Risk. Cheap Space Marine characters. A support credit that doesn't shrink with the carrier's damage favours cheap carriers: at κ 1 the Captain in Phobos Armour goes 74 → 35, the Apothecary 75 → 41 and the Rhino 79 → 49, all three around 70th of 84 by the reviewers. The per-unit form (a fixed amount for each carrier, whatever its cost) is far worse: Apothecary 1st, Captain in Phobos Armour 11th, headline −5.5. Keep the credit per point and κ modest. A separate Space Marine fix isn't needed at κ 0.5 (those three move 5 to 7 places).
- How it differs. Steps 29–31 shrank the multipliers' size and kept their shape. Step 6 replaced Synapse with a marginal measurement. This changes the shape: the credit no longer scales with the carrier's own damage. That's the cause in groups A and D.
Proposal 3: joined units on the same Kill as single units, then squads take their share (switches joinedBurst, then squads-only joinedShapley)
- In game terms. Gargoyles, Tyrant Guard, Genestealers and Raveners are graded as players field them, with a leader. The squad's fair share of the pair is what a reviewer is grading.
- Formula. First, Burst for joined forms (tools/burst.js run on them, so a pair's Kill is measured like its members'), with proposal 2 in place so a leader's multipliers no longer multiply its squad. Then a squad (never the leader) ranks at the better of alone or its even share: its own worth plus half of what the pair adds beyond its two members. Leaders keep their solo row, which is how step 10 found Marine reviewers grade characters.
- Constants. None (the even split is Shapley's).
- Preview (with a stand-in for joined Burst: each pair's Kill scaled by its members' Burst ÷ old-Kill ratio; κ 0.5). Genestealers 29 → 15 (14), Raveners 27 → 17 (14), Tyrant Guard 45 → 30 (32.5), Hormagaunts 17 → 6 (11), Gargoyles 46 → 43 (19); Carnifexes 31 → 5 (40). Headline −0.5, the 14 armies −0.8.
- Risk and blocker. The Old One Eye and Carnifex pair is worth about twice its members alone by our engine, with or without the heal multiplier and with the Burst stand-in. Until a diagnosis card says which part of that is real (the leader's dice buffs, the pair's role-scarcity factor, the stand-in), this step will fail as 13 and 14 did. Space Marine squads mostly move toward their grades (with the stand-in at today's multipliers: Bladeguard 16 → 3, Intercessors 22 → 8, Sternguard 36 → 17). Captains fall 15 to 20 places, because squads now pass them.
- How it differs. Steps 13 and 14 gave both members a share, measured with the leader's multiplier on the squad. This is squads only, on matched Kill, after proposal 2.
4. Previews (read-only; labelled previews, not steps)
How they were run: each army's step-33 compute() output, then every form re-valued by the change and each unit's best solo form picked again (the base reproduces 73.7% / ρ 0.550 / the 14 at 59.7% exactly). Pairwise and ρ are as tools/ledgerstep.js scores them (the shape on S to D, ρ with F). The 14 and 16 armies are scored as tools/ledgercarry.js does, on build/ledger. Uncertainty: a paired bootstrap over units within each headline army (300 resamples). No re-fit and no held-out units were run. The scripts were scratch files; the formulas are in section 3, enough to repeat them.
| Preview | Headline pairwise (ρ) | Tyranids | Space Marines | the 14 | Notes |
|---|---|---|---|---|---|
| Base, step 33 | 73.7% (0.550) | 77.4% | 64.5% | 59.7% | |
| 1. Battle-shock pressure, unconditional, W as reasoned | 74.8% (0.571) | 78.9% | 64.5% | 59.7% | +1.09 ± 0.92, positive in 87%; Neurolictor 28 → 12, Neurotyrant 37 → 30; 13 closer, 11 farther, Tyranid gap 8.9 → 8.5 |
| 1 at half / double W | 74.4% / 75.0% | 78.4% / 79.2% | 64.5% | 59.7% | at double: Neurolictor → 7, Neurotyrant → 20 |
| Battle-shock credit for every tagged rule (what not to do) | 73.8% (W 40) to 73.1% (W 86) | 77.6% / 77.0% | 64.3% / 63.3% | 58.8% / 57.4% | Deathleaper → 1st, Parasite 36 → 14, Chaplain 64 → 3, Reivers 17 → 2 |
| 2. Support as a credit, κ 0.5 (fitted by looking) | 74.7% (0.570) | 78.9% | 64.3% | 59.5% | +1.07 ± 0.93, positive in 94%; 18 closer, 12 farther |
| 2 at κ 0 / 0.1 / 0.25 / 1 / by budget | 71.3% / 73.0% / 74.1% / 72.8% / 72.8% | 74.1 / 76.5 / 78.0 / 77.2 / 76.4% | 64.1 / 64.1 / 64.4 / 62.0 / 63.7% | 58.0 / 58.2 / 58.9 / 59.3 / 59.7% | κ 0 = support off |
| 2 for Synapse only, κ 0.5; command points only, κ 1 | 74.2%; 73.8% | 78.1%; 78.3% | 64.5%; 62.7% | 59.7%; 59.5% | |
| 2 as a fixed amount per unit (not per point), κ 0.05–0.25 | 72.0–72.3% | 75.1–75.5% | 64.1–64.2% | 58.0–58.3% | Neurogaunts 32 → 4, Apothecary 75 → 1 (at κ 1) |
| 1 + 2 together | 76.0% (0.592) | 80.7% | 64.3% | 59.5% | +2.32 ± 1.35, positive in 98%; Neurolictor → 4, Neurotyrant → 22, WHT → 7, Tervigon → 28; 20 closer, 12 farther, gap 8.9 → 7.8 |
| 3. Squads take a Shapley share (joined Burst stand-in, κ 0.5) | 73.2% (0.540) | 76.8% | 64.2% | 58.9% | Genestealers → 15, Raveners → 17, Tyrant Guard → 30; Carnifexes 31 → 5 |
| 3 at base, no Burst stand-in | 72.4% | 75.5% | 64.6% | 58.8% | |
| Both members share (step 14's shape), at base | 70.7% | 75.2% | 59.6% | 58.0% | |
| Threat draws fire: Kill × life, life ∝ Soak ÷ Kill (median 2 / 3 / 4 rounds) | 69.5 / 70.6 / 72.0% | 73.3 / 74.0 / 74.8% | 60.0 / 62.4 / 64.9% | 57.6 / 57.3 / 57.4% | punishes every cheap melee unit (Lictor 1 → 14, Broodlord 11 → 25) |
| Durable per point: Kill × life, enemy fire ∝ points (φ 0.5 / 1 / 2) | 74.1 / 73.4 / 73.0% | 77.7 / 76.5 / 76.9% | 64.9 / 65.8 / 63.3% | 59.3 / 58.0 / 58.6% | inside noise; Tyrannofex barely moves |
| Graded by total worth: Value × (points ÷ 100)^γ, γ 0.25 / 0.5 / 1 | 73.3 / 71.5 / 67.6% | 76.6 / 74.5 / 70.6% | 65.0 / 63.9 / 60.3% | 60.5 / 61.2 / 60.8% | the one change that lifts the 14 armies; see section 5 |
| (ledgerstep) command points at 1 | 73.0% | (review: −2.3) | Neurolictor 28 → 42; Captains −30 to −48 | ||
| (ledgerstep) Fights First at 1 | 73.6% | (review: −0.9) | Leapers 2 → 14, Deathleaper 6 → 18, Lictor 1 → 12 |
Reading:
- The best change rules can make so far is 1 + 2: +2.3 points on the headline and Tyranids at 80.7%. But proposal 2's κ was chosen by looking, and neither change moves the 14 armies (proposal 1 has no carriers there; proposal 2 gives −0.2). By the review's own test that is not yet evidence that they transfer.
- Nothing rules-based in this round lifts the 14 armies except grading by total worth, and that drops the Tyranid lists. The review had found the same: what the 14 armies' reviewers see is mostly not in our lines.
5. The order to build them, and when to stop
0. Hygiene first. No headline number should move; each is its own small step with a written prediction.
- Re-file the Leapers' counter-charge rule. It's a cheaper stratagem, not command points. Also log which rules in each "command points" and "Battle-shock pressure" tag are unconditional and which only follow damage (data, in our words).
- Credit the Lictor's "friends arrive cheaply" rule to the army, not to its own arrival. And stop shoot-then-move rules on units with no melee from acting on melee landing.
- Run tools/burst.js on joined forms, so joined and single units read Kill the same way (no switch on it yet).
- Report the 11th-edition-only Tyranid mean and the gap in letters beside the headline. Mark contested units (reviewers' letters three or more apart: Leapers, Deathleaper) so no step is aimed at them.
1. Proposal 1, Battle-shock pressure. W fixed by the reasoning above, written in the prediction. Judged under the provisional keep rule, and on the 14 armies once the classification gives them carriers. Expected from the preview: kept.
2. Proposal 2, support as a credit. The shape is fixed. κ and the five support multipliers are re-fitted by tools/ledgerfit.js (prior: ones) inside the step. It is judged on the 14 armies and held-out Tyranid units, with the headline only as a guard. Expected: the headline holds or rises. The 14 armies are a coin flip, which is the real test.
3. Diagnose the Old One Eye and Carnifex pair (where a pair's worth beyond its members comes from, form by form). Then proposal 3, squads only.
Ask Jordan, once (in his question format): "Is a tier about a unit's worth per point, or about how good it is to take in a list?" Grading by total worth helps every army but his own (the 14 armies +1.5 at γ 0.5). If his answer is "worth taking", the yardstick, not the ledger, should change.
When to stop. Stop building ledger lines after step 3 if any of these holds: the 14 never-fitted armies stay at or below about 60.5%; held-out Tyranid units haven't risen past noise (about ±0.6); or three more steps go by without one being confirmed. Beyond that, list share already orders the Tyranid consensus 86% right. Our best rules-only preview reaches about 81% on the five Tyranid lists (not quite the same measure), near the high-70s ceiling the review guessed. Then the ledger's job is to explain a tier list ("why the Tyrannofex is worth it, in points"), with tournament inclusion as the order's prior and the reviewers as the check, rather than to set the order alone.