Tier system v2, tested: Space Marines (2026-10-02)

T-524, for T-523 (Jordan, 1 Oct (Pacific), via the design session). Made by node tools/tiertest.js. A report, not the site: nothing here changes a published score or page.

In short

Against a public list (added 2 Oct, once the network allowed it)

The list: Tista Minis' "Best Space Marine Units Right Now" (22 Jul 2026), a ranked list of 11th-edition Space Marine units, each with a reason.

• Mapping to letters: ranks 1 to 4 are S; ranks 5 to 12 and its best characters are A; its "other strong" units are B; its "good but not automatic" are C.
• Matched: 30 of our Space Marine units.
• What we keep: only names, our letter and our own paraphrase of each reason (reports/tiers/panel/2026-10-02-space-marines-tistaminis.json).
• Not used: Goonhammer and Tabletop Battles answer with a bot check, and Reddit refuses.

None of our systems agrees well with this list, and V2 agrees least. V1's fitted weights, fitted on list share without ever seeing this list, do beat equal weights here (1.43 against 1.47), so the fit carries over a little.

Where they rate units higher than we do:

• Eradicators and Infernus (their S; our B to D). Both are close-range guns, melta and flamers. Our neutral table assumes targets beyond half range, so melta's bonus is off, and it doesn't count Overwatch. Their reasons are delivery by transport, the Biologis and the Forgefather's Seekers detachment, all synergy we don't score. V2's damage-before-death also punishes three-model squads that die quickly.
• Scouts, the Lieutenant with Combi-weapon, Jump Pack Intercessors (their S and A). Mission work. V2's Runner does see it (88 to 89), but the quotas and composite still leave them B or C.
• Land Raider Redeemer, Repulsor (S, A). Transports and Overwatch, neither modelled.
• The Apothecary Biologis, the Librarian, the Jump Pack Captain (A). They support a squad or unlock a detachment (the Librarian opens Librarius Conclave). This comparison exposed a bug: the Banner tool skipped all 48 Support characters, the Biologis among them. They're scored now, and the Biologis still rates low on our numbers.

Where we rate units higher than they do: Centurions, Terminators, Outriders, the Impulsor, Assault Intercessors, the Land Raider Crusader. Efficient on paper, but their reason is cost against speed and delivery: slow, hard to transport, or outdone by cheaper scorers. We don't price movement or delivery against damage.

Their reasons behind the units V2 under-rates, counted: durability 5, anti-tank 4, leader support 4, mission work 4, then transport, Overwatch, synergy and deployment (2 each).

What this adds to the shopping list for Utility v2:

1. Overwatch;
2. close-range rules at their range (melta's bonus within half range for units built to get there);
3. delivery, meaning transports and what they carry;
4. leader and support synergy, which T-527 has started;
5. a price on slowness for expensive units.

Caution: this is one author's list, and their letters are our mapping of its sections, not tiers they gave. Treat the agreement numbers as a direction, not a verdict.

Our own data

Does the intuition hold? Partly. Quotas and jobs fix what's plainly wrong today. They don't yet make the letters agree with what players take, and the comparison with a public tier list couldn't be run from here.

• Too many S today. 264 units (23% of the pool) wear S, mostly because the meta lifts them; Value alone gives 80. With quotas S holds at 81 (7%) by design. Of the 44 units that wear S only through the meta with a Value in the 60s, the proposal keeps 6 at S and 9 at A; 15 drop to B and 14 to C or D.
• The job split. Winning lists put about 37% of their points into units whose best job is Hammer, 40% into Anvil and 22% into Banner. That gives S quotas of 30 / 33 / 18 (A: 65 / 70 / 39). Anvil taking the biggest share fits the edition's tilt to objective control and durability. Space Marines lean to Hammer (46 / 30 / 24).
• The fitted Banner weights: board control 0.7, reach 0.2, support 0.1, action economy 0. Agreement with list share (Spearman's ρ) rises from 0.153 with today's equal weights to 0.174. It is 0.126 without Banner and 0.113 for today's Value. Every one of these is weak.

V1 against V2 (Jordan's four jobs). V2 is built from how a game is won: Hammer as damage before death, Anvil with OC per point and Battle-shock, Runner for secondaries, Banner for force multiplication. Offense and Defense stay the published raw pair under both.

• Job demand under V2: Hammer 46%, Anvil 18%, Runner 16%, Banner 20% of the winning lists' points. That gives S quotas of 37 / 15 / 13 / 16.
• Agreement with what players take is lower for V2: ρ 0.127 with T-527's Banner (0.113 with the first Banner), against 0.174 for V1 fitted (0.153 for V1 with equal weights) and 0.113 for today's Value.
• Where V2 is better: it finally rates characters (25 of its 82 S units, none in V1), and it puts 2 of the review's 11 best Space Marine picks at S or A where V1 puts none.
• Where V2 overshoots: cheap leaders get S as Runners even though they join a squad (Chaplains in 0 of 84 lists). Mortarion is B with T-527's Banner (C with the first one).
• V2's Banner is T-527's (what a character adds to its best squad per point, against more bodies; reports/tiers/2026-10-02-banner.md) for the characters it scores; the report's V2 numbers use it.
• Recommendation: V2's Hammer (damage before death) and Banner (lift per point), with V1's simpler Anvil, and Runner scored only for units that act alone. Test that blend against the community list when one can be read.

The three biggest lessons

1. Characters are judged alone, and that's our biggest blind spot. Among the 75 units players take in 40% or more of their army's lists that the proposal grades C or D, Leaders are twice as common as they are among all units, and Auras 2.2 times. Azrael is in 22 of 23 Dark Angels lists and scores D. A Leader's worth is what it adds to its squad: its damage and durability bonus, re-rolls, an invulnerable save. Measuring that is the first item for Utility v2.
2. Our maths explains little of what players take (ρ 0.17 at best). Hammer, Anvil and today's Utility parts are not what decides a list. From the units we under-rate, the shopping list for Utility v2: Leader and Aura value, indirect fire (Biovores, 22 of 23 Tyranid lists, D), transport capacity (Venom, 13 of 13 Drukhari lists, D), Fights First, Deep Strike and Fly. Infiltrators, Scouts and Stealth are not over-represented. This list comes from the units players take, not from a community list's reasons; those still need reading.
3. Global quotas are harsh on one army's page. S and A together are 22% of every unit, spread over 36 armies, so a faction few winning lists take ends up with almost none. Space Marines units are in 0 to 2 of the 84 lists that could field them. Of the 11 units one review names as the best Space Marine pick in a role (seen only through a search summary, see below), today's letters and Value each put 3 at S or A and the proposal puts none: Eradicators C, Intercessors B, Gladiator Lancer C. If the letters must also read well inside an army, quotas per army (or a faction-relative letter beside the global one) are worth testing.

First run (blocked; done since, above). In the first run the public tier lists were blocked by this environment's network policy, so these sections were left for the list above:

• the agreement tables against a community list;
• the 15 biggest disagreements each way with their stated reasons;
• the count of recurring reasons;
• the test of the fitted weights on that list.

Allow one of those hosts in the environment's network settings and the comparison runs: node tools/tiertest.js --community <file>, where the file holds a list's unit names, letters and our own one-phrase paraphrase of each reason. Jordan's example is Mortarion (named by the design session). He wears S today only because of the meta (10 of 16 Death Guard lists), is B on maths alone, B under V1 (A with equal weights) and B under V2 (C before T-527's Banner). His job is Anvil, where he ranks 110th of 341 against an S quota of 33. What players take him for (mortal wounds each turn from Host of Plagues, the once-a-turn buffs to nearby units, the debuff zone) is what our numbers can't see. See "Mortarion, the acid test" below for each gap and what kind it is.

Choices made:

• Anvil is Defense's hits-to-destroy times the unit's OC. A product is the simplest blend, and it needs no weight.
• Job shares are counted as the lists taking a unit × its points at the smallest size. Our lists record which units, not how many.
• The fit is a grid over tenths, scored on units that 5 or more lists could field.
• Tied jobs go to the first in Hammer, Anvil, Banner order.

Round 2 (2 Oct): a fairer yardstick, V2b, and Mortarion

V2b is V2 with Hammer counting damage delivered before death, plus a detachment credit in Banner. Each turn of a five-round game on a 44" × 60" table, the model asks whether the unit reaches the enemy with its guns or a charge:

• Start: 30" between the units. Infiltrators start 9" away, Scouts move 6" before the game, and Deep Strike arrives on turn 2 at 9.5".
• Movement: each turn the unit moves its Move (+2" with Fly; Infantry in an army with Transports rides 10" on turns 1 and 2), and the enemy closes 3". A unit still out of range advances 3.5". A gun unit stops at its longest range, or at half range when that's worth over 15% more (melta, Rapid Fire). A melee unit keeps coming.
• Damage: its ranged damage when in range (the half-range figure inside half range), its melee damage times the chance of making the charge (2D6 against the gap left), and Overwatch on 33% of enemy turns from turn 2, hitting on 6s (a quarter of a 3+ hitter's damage), or in full with Torrent.
• Survival: each turn is weighed by the chance the unit is still alive (V2's turns alive under typical fire).
• Detachment credit: a character carrying the keyword a detachment's effects need (Librarians for a PSYKER detachment) shares that detachment's best lift (rankings.json's detachment synergy) with the army's other characters carrying it, on Leader-lift's scale (up to 1.5 points). It goes to armies where 6 or fewer characters carry the keyword.

Over every unit, V2b's ρ with list share is 0.087, against V2's 0.127 and V1's 0.174.

The yardstick, three ways, on the 30 units of the public list. (a) Spearman's ρ between our score and their order within the army (higher is better). (b) Our letters re-cut by quota within Space Marines's 84 units, against our mapping of their sections (mean distance; lower is better). (c) Our global letters, as in round 1.

V2 to V2b on the list's units (their letter, then our numbers):

Mortarion under every system (Death Guard, 375 pts, in 10 of 16 winning lists):

V2b's Hammer for Mortarion: 78 → 75 (he moves 10" with Fly and Deep Strike, so he arrives; his damage per point stays thin).

Round 3 (2 Oct): delivery modes, the bundle, and five fixes

Jordan, 1 Oct (Pacific), via the design session: the bundle rule (05:28Z), delivery modes (05:38Z) and the round-3 fixes (05:44Z). V3 is V2's four jobs with Hammer as damage delivered before death, scored under every way the unit may legally reach the fight; it keeps the best and names it:

• Walk: on the table from turn 1, 30" from the enemy. Scouts move 6" first. Infiltrators start 12" away: more than 8" from the enemy's deployment zone, with the enemy a few inches inside it.
• Deep Strike: off the table on turn 1, arrives on turn 2 9" away (the 11th-edition rule is more than 8"). It shoots at once if its guns reach. A charge from there needs 9 on 2D6 (28%, no re-roll assumed).
• Strategic Reserves: off on turn 1, arrives on turn 2 within 6" of a table edge and more than 8" from the enemy (11th-edition core rules), taken as 18" from the enemy. The 50% points cap on reserves is a list-level limit; we note it but don't enforce it per unit.
• Transport (the bundle): the unit plus its best legal Dedicated Transport, priced at both together (characters aren't bundled: they ride with the squad they join). The pair moves at the transport's Move on turn 1 and the unit is out on turn 2. The unit can't be shot for 1 round inside. The transport's own guns add to Hammer, and its hits to destroy and OC to Anvil, all per point of the pair. If it doesn't beat the unit alone, the transport isn't part of the unit's story.

The same rules as V2b after that: the enemy closes 3" a turn; a unit out of range advances 3.5"; guns stop at their range, or at half range when that's worth over 15% more; melee charges on 2D6; Overwatch 33% of enemy turns. Only turns the unit can be shot use up its life: a unit off the table or inside a transport keeps it. No terrain yet: every shot is taken to have a line (the exposure factor is 1, to be calibrated against the panel).

Who may ride what comes from BSData's transport entries at run time (tools/transports.js): the capacity, the keywords a passenger needs, the models it can't carry, and the ones that take two or three spaces. Only Dedicated Transports form bundles. With every transport allowed, Scouts chose a Land Raider Crusader and Boyz a Gunwagon, inheriting the big tank's guns. Those are units in their own right, scored alone. Allies are skipped for now.

The fixes, each switchable: (a) mortal wounds a round from abilities, our estimates read on 11th-edition datasheets (34 units, reports/tiers/2026-10-02-mortals.json); (b) a character that joins a squad (Leader or Support) and a unit with no OC can't be a Runner; (c) Battle-shock bites only below half strength, so Anvil's Leadership factor is 1 − 0.35 × (1 − the pass chance); (d) a transport's Banner counts its capacity ÷ 6, at most 2; (e) Indirect Fire is in range from turn 1. (d) lowered every yardstick, so V3 leaves it out; its row shows what it costs.

Each term on its own, against V3 with every term off (V2b's single path: ρ 0.107):

For comparison: V1 fitted 0.174, V2 0.127, V2b 0.087.

How units arrive under V3: walk 579, Deep Strike 220, Strategic Reserves 233, in a transport 129 (of 1161).

The 20 biggest gains from a transport (the unit page's line would read "Best delivered by …"):

Across the 129 bundles, the transport's own guns are 57% of the pair's damage at the median. For a passenger that does little itself (Krieg Combat Engineers, Pyrovores), the bundle's gain is mostly the transport. That is the rule as set ("when you pay for the transport you also get its guns"), but it means a weak passenger's letter partly measures its ride.

Round 1's over- and under-rated groups, and the arrival units (the public list rated the first two high and the Centurions and Terminators lower than we did):

Turn-by-turn profiles under the mode V3 chose: the distance to the enemy after moving, whether its guns reach (R), the charge chance (c), and the chance it's still alive. A dash means off the table or inside a transport.

Do the arrival-mode units move toward the public list? 19 of its 30 units arrive other than on foot under V3; their mean distance from the list's letters is 1.84 tiers under V2b and 1.79 under V3 (lower is closer).

Round 3, V2c (2 Oct): what the under-rated units have that we couldn't see

T-524 round 3 (the supervisor's brief, 2 Oct, with Jordan's refinements of steps b and d). Round 2 found that V2b ranked Space Marines against the public list (ρ −0.24) and that none of our structural maths explained much of list share (ρ ≤ 0.17). The brief: stop tuning the composite; add the things the under-rated units have that we can't see, one at a time, each measured. V2c is V2b with all six steps. The first build session ran the same round its own way as V3 (delivery modes, the bundle priced as a pair, the section above); its numbers sit beside ours in the table of every system. A report: nothing here changes a published score or page.

Verdict: partly. Does adding what we couldn't see bring our letters to what players and reviewers rate?

• On the Space Marine list, yes, up to a point.
  • ρ with their order goes from −0.24 to −0.08. That's the best of any system so far, though still not positive.
  • Mean distance with army letters goes from 1.57 to 1.43, level with V1.
• On list share, yes. ρ goes from 0.087 to 0.137, past V2 (0.127) but short of V1 fitted (0.174).
• On the Orks review, no. ρ falls from 0.06 to 0.01 (V1 0.22).
• On Auspex Tactics' Tyranid list (the third panel list, added at Jordan's request), partly. ρ goes from 0.07 to 0.21. Plain Value (0.33) and V1 (0.29) still do better, and today's letter (0.65), which carries list share, does best. Biovores (their S, 22 of 23 lists) are D under every maths-only system.
• Against V3:
  • V2c is better on the Space Marine list: −0.08 against −0.18.
  • V3 is better on list share (0.164 against 0.137) and on Orks (0.18 against 0.01).
  • On Tyranids both score 0.21.

Neither version wins everywhere. V3 lets each unit choose its best way to arrive. V2c counts durability per point. Those two ideas haven't been tried together yet.

• Almost all the Space Marine gain comes from one fix, and it's a correction, not a new feature: durability counted per point (step f). The five features we added are small by comparison:
  • mortal wounds, the joining-character rule, damped Battle-shock, Banner per point and Indirect Fire together move ρ with list share by +0.03;
  • they move the Space Marine ρ the wrong way.
• Mortarion goes from C to S (2nd of 36 in Death Guard), as the lists have him. But it's mostly survival lifting his Runner, not his mortal wounds. Right letter, partly the wrong reason.

The three lessons.

1. V2b was backwards because it mixed per-point damage with absolute survival. Hammer was damage per 100 pts × turns alive, and turns alive came from a unit's whole durability. A 150-pt squad lived twice as long as an 80-pt one only because it was bigger. Counting durability per point fixes it: the enemy's fire spreads over an army roughly by points. That one change moves the Space Marine ρ from −0.33 to −0.08. It also takes big centrepieces down: the Defiler goes S to C, Knights S to C. That's why the Orks review, which likes big Orks, now agrees less. How to price size is the open question.
2. What we couldn't see mostly isn't what players pay for. The brief's test cases still miss:
   • Biovores (22 of 23 lists): D, because Indirect Fire changed nothing; their 48" guns were always in range. Players take them for the Spore Mines they make.
   • The Venom (13 of 13 lists): D. The bundle test finds no squad it can carry at its smallest size that beats the Venom alone. Players take it for its rule that splits a ten-model squad into two to ride.
   • Eradicators (their S): D. They're still diluted by the five-target average.

All three are rules or archetype weights, not dice we're missing.

1. Survival belongs in Runner, but not on its own. It stopped cheap 4-wound leaders maxing Runner: 37 character S/A Runners and Anvils down to 23, and the Lieutenant with Combi-weapon keeps his place as a Lone Operative. But it now lifts big, durable models into Runner: Mortarion, Land Raiders, winged Daemon Princes. A Runner needs survival and to be cheap enough to spend on a mission. The two together, not either alone, is the next refinement.

What we'd do next: keep step f (durability per point, exposure, the bundle test) and steps a and c. Rework Runner as above. Test the archetype lens (T-529, T-531) on Eradicators and Infernus. Settle how to price size against a third panel list before anything moves on the site.

The steps, one at a time (each on top of the last)

Yardsticks: (1) Spearman's ρ with the public list's order within the army; (2) mean distance in tiers, our letters re-cut by quota within the army; (3) ρ with list share over every unit some 5 or more winning lists could field. Space Marines is the brief's yardstick; Orks is the second panel list, worth about half a ranked list; Tyranids, added at Jordan's request, is the third (its own section below).

a. Mortal wounds are damage

• Devastating Wounds were already in. docs/ev.js turns a critical wound into mortal wounds that skip the save, with Feel No Pain (and Feel No Pain against mortal wounds) still applying, and Offense and Hammer both come from it.
• New: docs/data/mortalfx.json, 83 datasheets' mortal-wound abilities as expected mortal wounds per use against each target, when they fire and how far they reach, in our own words and numbers. The effects files had twenty-odd of these as notes with no numbers ("not in the maths"), Mortarion's not among them; we read every 11th-edition ability that deals mortal wounds and wrote the ones that hurt the enemy. 107 rows of the pool carry one.
• Mortarion, Host of Plagues: after he moves, each enemy unit within 6" takes D3 on a 3+ (a 2+ when already afflicted; half assumed afflicted); with 1.5 enemy units in reach, 2.25 mortal wounds a turn once he's close. Damage delivered before death, per 100 pts: 12.7 → 14.5 (14% more; 1.8 of it mortal wounds). Hammer percentile 75 → 80.
• The first session's estimates (reports/tiers/2026-10-02-mortals.json, 34 abilities, used by V3) are per round with their own assumptions: Mortarion 1.4 a round against our 2.25 once he's in reach. Both are kept; each version uses its own.

The biggest gains:

b. Characters who join a squad

• The rule (Jordan's refinement, 2 Oct): a Support character (it joins a squad only through Supporting links; 41 in the pool) can't take Runner or Anvil as its job: it's scored by Banner and by its own attacks inside the squad. A Leader (271) may still be scored alone, since a Captain or a Lieutenant can run missions on his own, but Runner now counts survival: Runner = (reach + turns alive + Leadership + 0.5 × (units per 100 pts + small footprint)) ÷ 4, each a percentile, turns alive being Hammer's own term (at most 5; a Lone Operative's counts 2 times, since it can only be shot from close by). Cheapness and footprint together can no longer carry Runner. Lone Operatives (48) keep every job. A unit with OC 0 can't be an Anvil (73 units; their Anvil was already 0, so this changes nothing yet).
• S and A as a Runner or Anvil, characters, before → after: 37 → 23. 3 Support characters and 22 Leaders lose a solo S or A as a Runner or Anvil. S or A of any job, characters: 81 → 64; every unit: 256 → 255 (the quotas hold the total, so others move up).

Space Marines characters with S or A before or after (Runner = R, Anvil = A):

The public list's characters: Lieutenant with Combi-weapon (Lone Operative) C · Runner → C · Runner, Runner 89 → 86; Lieutenant (Support) B · Runner + Banner → C · Banner, Runner 87 → 30; Captain (Leader) C · Runner → D · Banner, Runner 69 → 28; Captain with Jump Pack (Leader) A · Runner → B · Runner, Runner 99 → 89; Apothecary Biologis (Support) C · Runner → D · Banner, Runner 70 → 26; Librarian (Leader) C · Runner → C · Banner, Runner 76 → 20.

The Chaplains: Chaplain A · Runner + Banner → A · Banner (0 of 84 lists); Chaplain in Terminator Armour S · Runner + Banner → A · Runner + Banner (0 of 84 lists); Chaplain on Bike A · Runner → A · Runner (0 of 84 lists); Chaplain with Jump Pack A · Runner + Banner → A · Runner + Banner (0 of 84 lists).

c. Battle-shock bites only below half strength

• The choice: Anvil's Leadership term becomes 1 − b × (1 − the chance to pass), where b is the share of a five-turn game the unit spends below half strength. Under typical fire a unit that lives T turns loses about 1 ÷ T of itself a turn, so it's below half from turn T ÷ 2 until it dies: b = (min(5, T) − T ÷ 2) ÷ 5, never below 0. A unit that dies in a turn spends a tenth of the game below half; one that lives 5 turns, half; one that would last 10 or more, none. T is Hammer's turns alive, uncapped.
• Anvil's spread: the Leadership term ranged 42% to 83% of a unit's durability; damped it ranges 71% to 100%.

The Tyranid count (T-530's pilot, run with --armyfx docs/data/armyfx.json).

• V2: Tyranid S letters rose from 6 to 25 when Synapse went in. Damping Battle-shock alone takes that 25 to 22.
• V2c's chain: 20 S just before this step, 19 after it (6 without army rules).

So the damping is right, but Battle-shock in Anvil wasn't the main way the pilot inflated Tyranids. The rest comes through three other routes, not measured apart here:

• Runner's Leadership part;
• Banner's Synapse grant;
• the melee Strength bonus.

The pilot's lesson stands: no army's rules go into the letters until every army's are written.

d. Banner from what a unit adds to others, per point

• One scale for everyone: beats bodies. T-527's method (reports/tiers/2026-10-02-banner.md): what a unit adds to others, against spending its points on more of the squad it helps (1 = as good as bodies). Banner = a Leader's multiplier on its best squad (tools/banner.js; read from its JSON) + its auras + its transport capacity + the detachment credit, ranked over every unit.
• Auras: each aura (at most 2, a Synapse grant counting as one when army rules are on) lifts 2 units by the median Leader's lift on its squad (32%), units of the median unit's points (120), against the bearer's own points. A 50-pt banner bearer's aura is worth 1.54; Mortarion's (375 pts) 0.20.
• Transport capacity: for each squad of its army a transport can carry (tools/transports.js, the first session's reader, takes capacity, the keywords it takes, what it can't carry and what takes double room from BSData at run time), damage delivered before death riding against walking (the delivery model, with step f's exposure once that's on), against the same points spent on more of that squad. Its best squad's multiplier counts, at most 2. 92 transports in the pool; 23 carry a squad that gains.
• The Venom (13 of 13 Drukhari lists): best squad none, multiplier 0.00; Banner 83 → 0; C · Banner → D · Runner.

e. Indirect Fire doesn't need delivering

• 70 units carry an Indirect Fire weapon. In V2b they were already in range from turn 1 wherever its range is 30" or more (the gap at deployment); only 8 have a shorter one (Kairos Fateweaver, Kairos Fateweaver, D-Cannon Platform, Reductus Saboteur, Cthonian Beserks, The Silent King, …). 0 change letter.
• Biovores (22 of 23 Tyranid lists, 60 pts): D · Anvil → D · Anvil; Hammer 28 → 28. Their spore mine launcher reaches 48", so delivery never held them back. What players take them for is the Spore Mines they make and the actions a cheap unit can do, neither of which we score (Spore Mines' own bursts are now in mortalfx.json, but they belong to the Spore Mines unit, not the Biovores).

f. What was backwards in V2b

Five turns of each unit, before and after the fix: the gap to the enemy at the end of its move (– in reserve), the damage it delivers per 100 pts that turn, and the share of it still alive.

Who rides. V2b let 507 units ride: every Infantry unit in an army with any transport. Round 3 asks three things: may it ride (a Dedicated Transport of its army can carry it: capacity, the keywords it takes, what it can't carry and what takes double room; 312 units), does it ride, and what's it worth. Does it ride is Jordan's bundle test: the pair, passenger riding plus transport, must beat per point both the passenger walking and the transport alone, on Hammer's damage delivered before death per 100 pts (riding changes nothing else we score). 147 units ride under it (a first cut, a transport costing no more than its passenger, gave 202). Units whose transport alone is worth more than the pair: 0, by construction. Two rules, two jobs: the Dedicated Transport list says what may ride, and the bundle test stops a passenger taking a letter from a transport that's better on its own.

What was backwards, and the fix. Hammer is damage per 100 pts × turns alive. V2b's turns alive came from a unit's whole durability: its hits to destroy against half the median unit's. So a 150-pt Centurion squad lived 4.2 turns and an 80-pt Intercessor squad 1.7, and the per-point damage of every big unit was multiplied by its size. The first session's handoff names the same cause ("slow, tough units win on turns alive").

The fix (three parts, measured one by one in the step table):

• Durability per point. An army's fire spreads over the enemy roughly by points, so turns alive come from Defense's hits to destroy per 100 pts against half the median unit's. The median unit still lives 2 turns; at least 1, at most 5.
  • Centurion Assault Squad: 4.2 → 3.0 turns.
  • Intercessors: 1.7 → 2.2 turns.
• Exposure. A unit uses up its life only on turns it stands on the table outside a transport. A Deep Striker no longer loses a turn of life in reserve.
• Who rides: the bundle test above.

Two constants of V2b were wrong in the same direction:

• every Infantry unit rode;
• a reserve turn counted as a turn under fire.

Neither alone explains the −0.24; the per-point durability does (step table: −0.33 → −0.08).

Every system on the three yardsticks

Higher ρ is better; lower distance is better. "Army" letters are re-cut by quota within the army; "global" letters are our published scale. Today's letter carries the meta lift (it partly is list share), so its ρ with list share isn't a fair contest.

Tyranids: the third panel list (Auspex Tactics)

The list: Auspex Tactics, "Strongest Tyranids Units?", YouTube, 11th edition (after the latest balance update; Tyranids about a 51% win rate). Ranked by what is used to success now; the author stresses detachments. Read 2026-09/10 (video date not read; unit names via YouTube's summary, passed on by Jordan, 2 Oct). Tier 1 = S, Tier 2 = A, Tier 3 = B (usable or niche; its upper and lower halves not separated in the summary), Tier 4 = C, Tier 5 = D. Units the summary did not mention (Termagants, Genestealers, the Broodlord, the Lictor, Gargoyles and others) are not graded, not D. Reasons blank: the transcript was not read. Matched: 44 of our 52 Tyranid units; not placed: Barbed Hierodule. Name matches to check: Hyperadapted Raveners (named Hyper-adaptive Raveners); Hormagaunts (noted as led by a Tyranid Prime); Tyrannofex (either weapon); Tyranid Prime with Lash Whip (named as the Tyranid Prime, noted with melee Warriors; matched to the ground Prime); Hierophant (named Hierophant Bio-Titan).

This list ranks what wins now, so today's letter, lifted by list share, should agree best with it, and does. Among the maths-only systems V2c is the closest of the four-job versions, but plain Value (0.33) and V1 still rank Tyranids more like the list.

The named units: global letter / letter within Tyranids, under each system (their tier first).

Every graded unit under V2c:

The 20 biggest moves, V2b → V2c (556 units change letter; 275 up)

Ordered by letters moved, then by places in the overall order. "Mostly" is the step that, applied alone to V2b, moves the unit furthest in the same direction.

Space Marines' public list under V2c

Round 4 (2 Oct): V4, the two strands merged, Runner's points gate and solo characters

T-524 round 4 (the supervisor's brief, 2 Oct, with Jordan's step g). Round 3 left two strands, V2c and V3, each strong on different panels, and two known faults: Runner rewarded big, durable models, and Chaplains in no winning list kept A or S. V4 merges the strands into one system with every part a switch, adds Runner's points gate and the solo-character rule, and, at Jordan's request, step g: Hammer as enemy points destroyed, weighted by the meta. A report: nothing here changes a published score or page.

Verdict: partly. Does the intuition hold now?

• On the Space Marine list, yes, for the first time. ρ with Tista Minis' order is positive (0.04, from V2c's −0.08 and V3's −0.18), and both letter distances are the best of any system (1.33 within the army and on the global scale).
• On Tyranids, about level. 0.19 against 0.21 for both V2c and V3; plain Value (0.33) still ranks Tyranids more like Auspex Tactics.
• On the Orks review and on list share, no. Orks fall to −0.19 (V3 0.18), and ρ with list share to 0.104 (V2c 0.137, V3 0.164, V1 fitted 0.174).
• The named checks:
  • Mortarion is no longer a Runner: Hammer only, B (S under the merge alone).
  • The Lieutenant with Combi-weapon keeps his C.
  • The Chaplains don't fall: B, A, B and A, in 0 of 84 lists. Halving their turns alive moves them only within the bottom of survival. A quarter instead of a half changes nothing measurable. They are now carried by Banner, not Runner.
  • Step g lifts the specialists it was meant for: Eradicators D → C (anti-tank), Hive Guard A → S (anti-tank), Tyrannofex stays S (anti-tank).

How much of this is signal. A Spearman ρ over 30 units has a standard error of about 0.19, and over 44 or 45 about 0.15. Most differences between our systems on one panel list are smaller than that. Only ρ with list share, over 1,105 units (error about 0.03), separates them reliably, and on it V4 is worse than both strands. Read the panel columns as direction, not verdict.

The three lessons.

1. Merging didn't add the strengths up. The merge alone (V3's arrival modes plus V2c's parts) scores below V2c on Space Marines and below V3 on Orks and list share. The parts pull different panels different ways:
   • durability per point and Runner's gate help the Space Marine list;
   • the Orks review rates big warlords (Ghazghkull, Wazdakka) and loses them to the same two parts;
   • the Support/Leader rule is the one part that helps everywhere but Orks, which it leaves level: without it, list share falls from 0.104 to 0.070 and Tyranids from 0.19 to 0.10.
2. The gate does its job and exposes the size problem again. S or A Runners over 130 pts go from 27 to 0, and Mortarion, Magnus, Fulgrim and Angron stop being Runners. But a big model that was carried by Runner has nothing else to stand on: under durability per point its Hammer and Anvil per point are middling, so within Orks Ghazghkull ends at C and Wazdakka at D under V4, while the review has both at S. Pricing size is still open, and the gate makes it visible rather than hiding it.
3. The Chaplains are a Banner problem now, not a Runner problem. Beats-bodies credits a Chaplain with his best squad's lift: a Tactical Squad, which is in 0 of 84 lists, at ×2.0, against a Captain's ×0.57. Banner should count a lift only on squads worth fielding (weighted by the squad's own score) and against the other leaders that squad could take. Halving the Chaplains' whole Runner instead of their turns brings one to C and three to B (the table under "Solo characters"): still not all of them to C or lower.

On step g.

• The meta's points sit mostly in elite infantry and characters (TEQ 38%), then the two vehicle classes (22% and 23%), with chaff 10% and MEQ 6%. So anti-chaff specialists peak at half weight: Barbgaunts lose their Hammer tag and stay B as an Anvil.
• Points per wound run from 7 (chaff) to 13.75 (heavy vehicles), so most generalists also peak on vehicles: 348 of the 400 Hammer units are tagged anti-tank. The tag should name the target where the unit ranks highest among units, not where its points are largest.

What we'd change next:

• Banner on squads worth fielding, and leader competition.
• A size term for big centrepieces, tested on the warlords and the Knights.
• The anti-tank tag by within-target rank.
• Strategic Reserves at 18" from every target is generous: 290 units arrive that way.
• Above all, more panel lists, since one list can't separate systems that differ by less than 0.15.

Tyranids, Jordan's own army: every unit Auspex grades, and the 20 biggest disagreements, each with a diagnosis, are in "Tyranids against the Auspex list" below. The same section is on its own page, reports/tiers/2026-10-02-tyranids.md.

Every system on the yardsticks, grouped

Each faction cell reads: Spearman's ρ with the public list's order within the army (higher is better) · mean distance in tiers with our letters re-cut by quota within the army · mean distance with our global letters (lower is better). The last column is ρ with list share over every unit 5 or more winning lists could field. Panels: Tista Minis (Space Marines, 30 units), Auspex Tactics (Tyranids, 44), Sprues & Brews (Orks, 45, worth about half a ranked list).

Maths only (no list share inside the score; V1 fitted's Banner weights were fitted on list share, so it borrows a little):

With popularity:

The second group isn't comparable with the first: today's letter lifts a unit by its share of winning lists, and the community lists also rate what people play and win with, so the two share a source rather than one predicting the other.

V4's parts, measured

The merge first, then the two new terms on top, then V4 with each part switched off in turn (a part helps where switching it off makes the row worse).

How units arrive under V4: walk 580, Deep Strike 272, Strategic Reserves 290, in a Dedicated Transport 19 (of 1161; a bundle counts only where the pair beats both parts per point).

Runner's points gate

Runner × a factor of 1 up to 75 pts (the Light weight class's bound in tools/hill.js), falling in a straight line to 0 at 260 pts (twice the Middle bound, 130): 86% at 100 pts, 70% at 130, 32% at 200. Applied to Runner's raw mean before it's ranked, so cheap units take the places big ones leave. S or A Runners: 80 → 34; of them over 130 pts: 27 → 0.

The biggest units that lose an S or A as a Runner:

On the Orks review, the gate moves 5 of its 45 units' letters within Orks: Wazdakka Gutsmek (200 pts, theirs S: A → D); Bannernob (35 pts, theirs B: C → B); Stormboyz (70 pts, theirs B: D → C); Wartrakks (70 pts, theirs B: D → C); Ghazghkull Thraka (300 pts, theirs S: C → D).

Solo characters pay for being alone

A single-model character that joins squads (Leading or Supporting links) but is scored on its own, without Lone Operative and not a MONSTER or VEHICLE, is the enemy's first target: its turns alive × 0.5 (Hammer's delivery, Runner's survival and the Battle-shock share all use them). 272 units in the pool; S or A among them 53 → 52. Lone Operatives and big models are untouched.

Space Marines characters with S or A before or after:

Stronger versions, for comparison (not V4): the same yardsticks and the four Chaplains and the Lieutenant with Combi-weapon under each.

Step g: Hammer as enemy points destroyed, weighted by the meta

Jordan's addition (2 Oct, via the supervisor). Specialists are bought for one target, and the five-target average buried them; and a dead tank is worth more than dead chaff. So each target's damage delivered before death is counted as the enemy points it destroys (wounds × the stand-in's points per wound, tools/targets.js ppw: GEQ 7, MEQ 8, TEQ 10.67, Light vehicle 13.33, Heavy vehicle 13.75), per 100 pts of the attacker. The targets are weighted by the current meta: each winning-list unit's points (lists × points) fall in the stand-in class it resembles (vehicles, monsters, walkers and aircraft: heavy with 14 or more wounds, light below; everything else: chaff with 1 wound, TEQ with a 2+ save or 3 or more wounds, MEQ otherwise). Hammer = 0.75 × the peak target + 0.25 × the meta-weighted mean, the jobs' own peak-plus-breadth rule. The peak counts a target in full when its class holds at least an even share of the meta (20%) and in proportion below that, so a unit built for a class that's rare in the meta peaks lower. The Hammer tag names the peak's class: (anti-chaff), (anti-elite) for MEQ and TEQ, (anti-tank) for both vehicle classes.

Hammer units by what they kill under V4: anti-chaff 10, anti-elite 42, anti-tank 348.

The named units

Global letters. The community tier is the panel list's, where it grades the unit. The job is V4's.

Five-turn delivery profiles under V4

Each turn: the distance to the enemy after moving, whether its guns reach or the charge chance, whether it can be shot (exposed) or not (off the table or inside the transport), and the chance it's still alive at the start of the turn. Turns alive come from durability per point (the median unit lives 2 turns) and are used up only on exposed turns. The last column is Hammer's raw number under step g: enemy points destroyed before death per 100 pts, 0.75 × the peak target + 0.25 × the meta-weighted mean (a bundle's per 100 pts of the pair, its transport's guns included). A Rhino can't carry Intercessors under the BSData commit we read (it excludes TACTICUS models), so the Rhino row is a what-if; the Impulsor is the legal ride.

Tyranids against the Auspex list

In plain words. Our maths and Auspex agree on the big shooting monsters and on the bottom of the list, and disagree on the units Tyranid players take for their rules.

• Where we agree: the Tyrannofex is S on both, and the Harpy, Harridan, Hierophant, Spore Mines and Sporocyst are D on both.
• Where Auspex rates a unit higher, the reason is nearly always a rule we don't play out yet:
  • Biovores making Spore Mines;
  • Raveners and the Red Terror diving back into reserves to strike again;
  • the Neurolictor's Battle-shock games;
  • the Norns' chosen target;
  • the Hive Tyrant's CP discount and aura.

Those are rules, not dice, and they're the next things to model.

• Where we rate a unit higher, it's mostly one letter on tough monsters that are strong per point (Carnifexes, the Haruspex, Old One Eye). Two are our own errors:
  • the Tyrannocyte: we treat a one-use drop as a mission unit;
  • Ripper Swarms: an aura on a 30-pt unit looks huge per point.
• The Tyranid Prime is the one where we side with the players over the list: 11 of 23 winning lists take it.
• All four jobs are still scored for every unit; the tag names only the best. The table now shows all four, with the raw attack and defence per 100 pts behind them. Raveners and the Red Terror deal top-tier damage per point (88th and 89th percentile) but live about two turns, and they spend the first on a Deep Strike charge that comes off 28% of the time. Turns alive is what keeps their Hammer at D; the anti-tank weighting isn't. The section after the disagreement tables shows this unit by unit.
• Today's live letter still matches Auspex best (21 of 44 the same letter, 37 within one), because both follow what wins. V4 gets 14 the same and 26 within one.

The list: Auspex Tactics, "Strongest Tyranids Units?", YouTube, 11th edition (after the latest balance update; Tyranids about a 51% win rate). Ranked by what is used to success now; the author stresses detachments. Tier 1 = S, Tier 2 = A, Tier 3 = B (usable or niche; its upper and lower halves not separated in the summary), Tier 4 = C, Tier 5 = D. Units the summary did not mention (Termagants, Genestealers, the Broodlord, the Lictor, Gargoyles and others) are not graded, not D. Reasons blank: the transcript was not read. Global letters throughout; "Lists" is the Tyranid winning lists that took the unit. Value is today's maths-only number (its letter by the fixed bands). V4's tag carries what its Hammer kills (step g).

Legend. Hammer, Anvil, Runner and Banner are V4's four job percentiles over every non-Legends unit (0 to 100; the letter comes from 0.75 × the best allowed job + 0.25 × the mean of the four). Offense is wounds removed per 100 pts (the five targets' mean) and Defense hits needed to destroy the unit per 100 pts (the five attackers' mean), each with its published percentile in brackets. In reach: turns of the five its guns reach or it charges (a charge counts its chance). Alive: turns alive under typical fire from durability per point, used up only on exposed turns. Arrives: the arrival mode V4 chose.

Same letter / within one letter of Auspex's, of 44: today 21 / 37, Value 12 / 27, V3 19 / 27, V4 14 / 26.

Where Auspex rates a unit higher than V4 does (20 units; the 10 biggest gaps, then by lists):

Where V4 rates a unit higher than Auspex does (10 units):

Why six Tyranid units land where they do

Jordan, 2 Oct, via the supervisor: "Are we not scoring all domains anymore? For melee units with great movement, is mobility not reaching Hammer (the Red Terror, Raveners)?" We still score all four domains for every unit; the tag only names the winning one. The table above now shows all four, and the raw numbers under them. Below: each unit's five turns as V4 plays them, every arrival mode it could take, and what each Hammer term costs it.

The five-turn delivery profiles under V4. Each turn: the distance to the enemy after moving, whether its guns reach or the chance its charge comes off, whether it can be shot (exposed) or not (shielded: off the table or inside a transport), the chance it's still alive, and the wounds it removes that turn if alive (the five targets' mean; a charge counts by its chance, guns add a little Overwatch). Turns alive come from durability per point and are used up only on exposed turns: a unit that lives 1.8 turns is alive for all of its first exposed turn, 80% of its second and gone after. Delivered is the sum of each turn's wounds × alive. The last column is Hammer's raw number (enemy points destroyed per 100 pts under step g) and its percentile.

Every arrival mode each could take. V4 keeps the mode that gives the best composite. A Tyrannocyte bundle only counts when the pair beats both the passenger alone and the Tyrannocyte alone per point. For the five of the six that can ride one, the bundle fails, because the Tyrannocyte on its own scores more per point than the pair.

What each Hammer term costs. Each column changes one thing and keeps the rest of V4:

• Lives all five turns: durability per point removed.
• No exposure rule: life is used up every turn from turn 1, even off the table.
• Deep Strike arrives in combat: the turn-2 charge always comes off.
• Move 12": Jordan's figure. The 11th-edition datasheet and our BSData both give Raveners 10".
• The three step-g columns: step g off, step g counted in wounds rather than enemy points, and step g with the five targets weighted evenly.

Each cell: Hammer's raw number · its percentile · the letter it would give with the other three jobs held (S and A are cut within the job, B and C overall). The meta weights (the points share of winning lists by class): GEQ 10%, MEQ 6%, TEQ 38%, Light vehicle 22%, Heavy vehicle 23%; points per wound: GEQ 7, MEQ 8, TEQ 10.67, Light vehicle 13.33, Heavy vehicle 13.75. The composite's cuts in V4: B from 75.4, C from 60.3; in the Hammer job A from 80.1 and S from 86.7, in the Anvil job A from 84.9 and S from 87.6, in the Runner job A from 87.5 and S from 92.6 (S and A places go to each job by its points share of winning lists: Hammer 42%, Anvil 20%, Runner 15%, Banner 23%). The Hammer job's percentile for an S is 93 or more in V4 and the median unit's Hammer raw is 70.4.

Raveners: D · Anvil (Hammer 40, Anvil 52, Runner 32, Banner 0). Damage per point isn't the problem: at full reach they remove 4.9 wounds per 100 pts, the 88th percentile. Durability per point is. Their Defense is 27.2 hits per 100 pts, the 39th percentile, so they live 1.8 exposed turns, and the first of those is spent on the Deep Strike turn, where a 9" charge comes off 28% of the time. That leaves one full fight (6.2 wounds at 77% alive) and nothing after: 6.5 wounds delivered, against 26 if they lived all five turns.

• Lives all five turns: Hammer 93, the letter A. This is the term that sinks them.
• Arriving already in combat: Hammer 60, still D.
• Step g: worth about 2 points either way (39 in wounds, 38 with even weights, against 41), so the anti-tank weighting isn't it.
• Move 12": changes nothing. Deep Strike is the mode chosen, and arriving is the move.

With Hammer at 40, their best job is Anvil at 52, and the composite (46.8) sits below C's cut (60.3). What players buy them for is Death From Below: they leave at the end of the enemy's turn and come back where they're needed. We don't play that yet, and it would feed Runner (any objective, any round) and Hammer's choice of target more than it would keep them alive: they still face one enemy turn on the table per arrival.

The Red Terror: D · Hammer (anti-tank) (Hammer 46, Anvil 44, Runner 31, Banner 0). Its story is the same as the Raveners'.

• Damage per point: 4.8 wounds per 100 pts at full reach, the 89th percentile.
• Durability: 38.5 hits per 100 pts, the 38th percentile, so it lives 2 turns. The Deep Strike turn's 28% charge uses one, and one full fight (6.3 wounds) uses the other: 7.8 delivered.
• Lives all five turns: Hammer 93, A.
• Arriving in combat: 66, still D (with its other jobs held, a C needs a Hammer of about 69).
• Step g: helps it slightly (43 in wounds, 39 even). The "(anti-tank)" tag is a quirk: its points destroyed against a light vehicle (63.2) equal those against MEQ (63.7), and MEQ's class holds 6% of the meta, under an even share, so the peak goes to the vehicle.

Two of its rules would change this, and both are about life rather than damage:

• Subterranean Hunter: it leaves at the end of a Fight phase when nothing is left in Engagement Range. Unlike the Raveners' rule, this can take it off the table before the enemy shoots, so it would cut its exposed turns.
• Swallow Whole: it heals by eating infantry, which adds turns alive.

The Neurolictor: B · Runner (Hammer 34, Anvil 21, Runner 98, Banner 73). It is already near the top of Runner. Its composite, 87.4, misses the Runner job's A cut, 87.5, by 0.1. Runner gets 15% of the S and A places (its points share of winning lists), so its cut is the highest of the four jobs. Its claws remove 2.4 wounds per 100 pts (21st percentile). It infiltrates 12" away, moves to 4" and charges on turn 1 at 92%, so Hammer is low because it hits soft, not because it arrives late. The table below answers the brief's three questions.

Biovores: D · Runner (Hammer 33, Anvil 32, Runner 39, Banner 0).

• The gun: its own gun reaches every turn (In reach 5) but removes 2.2 wounds per 100 pts, the 14th percentile.
• The bodies: it is one model with OC 1 on 60 pts and Move 5", so Anvil and Runner are low too. Even living all five turns only lifts its Hammer to 70, a C.
• What we don't count: nothing in our dice is what the 22 of 23 winning lists take it for. Each turn it can spend its shooting on a new Spore Mines unit set up anywhere within 48", more than 8" from the enemy. That gives a fresh screen and action body every round. That's the first item on the list below.

Hormagaunts: A · Anvil (Hammer 36, Anvil 99, Runner 58, Banner 0). The nearest to Auspex's S of the six:

• Anvil: 99. Ten bodies with OC 2 on 70 pts and their revive make them nearly the best holders per point in the game.
• The S cut: their composite (86.3) is 1.3 under the Anvil job's S cut (87.6).
• The fight: they walk, advance to 16.5" on turn 1 and charge on turn 2 at 92%, but at T3 5+ they last 2.5 turns, so Hammer stays low.
• What would add the 1.3: a Banner for screening (bodies in the way) or a Synapse-range body count, which we don't score.

Von Ryan's Leapers: B · Runner (Hammer 68, Anvil 79, Runner 94, Banner 0). The best Hammer of the six per point. They infiltrate at 12", move 10" and fight from turn 1, delivering 7.2 wounds on 55 pts (Hammer raw 103.8 against a median of 70.4).

• Runner is still their best job (94).
• The cut: their composite (85.6) is under the Runner job's A cut (87.5).
• The mean of four: their Banner is 0. A Banner of 75 would add 4.7 points of composite, and they need 1.9. A cheap unit that's good at three jobs and has no support role pays the most for it.
• Living all five turns would make them S.

The Neurolictor's invulnerable save, Lone Operative and aura

• The 4+ invulnerable save counts in full in Defense, against every attacker but small arms. Without it, its hits to destroy fall from 24.3 to 19.7 per 100 pts and its turns alive from 1.9 to 1.2. In Anvil it's worth only 4 percentile points (21st against 17th). Anvil multiplies durability by OC per point, and one OC on 80 pts keeps it out of Anvil whatever its save: even doubled durability or OC 2 only reaches the 34th.
• Lone Operative counts in Runner, where it doubles the survival part: Runner 98 with it, 88 without. It isn't in Anvil, and doubling its durability there would still only reach the 34th percentile. So its Lone Operative is already its main job.
• Its aura reaches Banner, but only at a stand-in value. The Psychological Saboteur aura is credited at the median Leader's lift over two units (1.0), which puts its Banner at 73. A named second job needs 80, which is 1.3 raw, and an S-level Banner needs 2.0. Our unitfx entry gives Tyranids +1 to wound against a Battle-shocked target, but only as a manual chip. We count neither its forced test each Command phase nor the enemy's −1 to hit while shocked.
• What a little more would do: almost any credit for that pressure (0.1 of composite) turns its B into an A.

Tyranid rules to model (the shopping list)

From the units Auspex rates higher than V4. Each is a rule we don't play out, the way it would go into our effect vocabulary (our words, a proposed shape, nothing built), and the job it would feed. Carded as T-540.

The systems

• Current: today's letter. Value, lifted by the meta: a unit in a share s of its army's winning lists scores at least 40 + 100 × s (with 3 lists' doubt). Bands 80 / 65 / 50 / 35.
• Maths only: Value alone, the same bands.
• V1 (three domains), the first proposal: the three jobs below, composite = 0.75 × best + 0.25 × mean, letters by quota on the composite's rank (S 7%, A 15%, B 28%, C 30%, D 20%), S and A split across the jobs by their share of the winning lists' points. Banner's weights fitted (below).
• V2 (four jobs), from how a game is won (primary: OC on objectives when they score; secondary: actions and positions, which a Battle-shocked unit can't start or finish; denial: kills and screens). Hammer, Anvil, Runner and Banner, each a percentile over every non-Legends unit; the same composite (0.75 × best + 0.25 × mean of four) and quotas, S and A split by the four jobs' share of the winning lists' points.

The jobs' names are Jordan's (2 Oct) and our own words, not Games Workshop keywords: Hammer (hitting), Anvil (holding), Runner (running the secondaries, V2 only) and Banner (enabling others). A unit's tag reads like "S · Hammer" or "A · Banner".

Offense and Defense stay the published raw pair under either version (Jordan, 2 Oct: "quite solid math… more data is better"): every unit page and Rankings row keeps them with their percentile chips; the jobs are built from them and never replace them. V2 would add damage before death as a third shown number.

V2's jobs, as built here. Hammer: damage before death, the wounds it removes per turn per 100 pts (Offense's rule, the better of shooting and melee into each target, best loadout per target) × the turns it lives under typical fire (each attacker type's hits to destroy it over half the median unit's, so the median unit lives 2 turns; at least 1, at most 5). Durability counts here and in Anvil on purpose. Anvil: Defense per 100 pts × OC per 100 pts × the chance to pass a Battle-shock test (2D6 at its Leadership) × 1.15 where it heals or returns models (a returned model or a few wounds a turn restores roughly a tenth to a fifth of a unit over a game, so a modest lift). Runner: the mean of four percentiles: reach (Move, Fly, Deep Strike, Scouts, Infiltrators), units per 100 pts, a small footprint (models × a base size by type: infantry 1, mounted 2, walkers 3, light vehicles 4, monsters and battle tanks 5, aircraft 6, Titanic 12; swarms double) and Leadership. Strategic Reserves are open to nearly every unit, so they don't separate units and aren't counted. Banner: what it lends others per 100 pts: a Leader's lift on its best squad (T-062's measure), auras (up to 2), Transport, revives. Transport capacity and CP economy aren't in our data yet (BSData gives capacity as text), so a transport counts 1. V2's parts are equally weighted; nothing is fitted. Banner here uses T-527's scores for the 451 characters it scores (what each adds to its best squad per point against more bodies, tools/banner.js), mapped onto those characters' places so the order changes and their standing against other units doesn't.

V1's Anvil, as built here: the hits the unit takes to destroy per 100 pts (Defense's raw number, each of the five attacker types) × its Objective Control (OC × models), averaged over the attackers, then ranked. A product: a unit holds only if it is both hard to shift and counts on the objective (a fast tank with OC 0 or a fragile horde both score low). Ranked, a product orders units exactly as the geometric mean of the two would, so there's no weight to tune. Plain Defense stays a shown number. V1's Banner is today's four Utility parts; note board control (OC per 100 pts) appears in both Anvil and Banner, which the fit is free to weigh down.

Pool: 1161 rows (every non-Legends unit; a datasheet shared by several armies is a row in each, as on the tier list).

The job split (decides each job's S and A)

Each unit's job is its best domain; its share of the winning lists is the lists that take it (top-quarter lists, July to September, grimstat-corpus, CC BY 4.0) × its points at the smallest size. A unit taken twice in a list counts once (the lists we keep say which units, not how many).

With equal Banner weights the split is 34% / 42% / 23%.

V2, four jobs:

Banner's weights, fitted on list share

Every split of 1 into tenths over the four parts (286 in all), scored by how well the composite ranks 1105 units like their share of the winning lists that could field them (Spearman's ρ; units some 5 or more lists could field).

The fitted weights lift the composite's agreement with list share from 0.153 (equal weights) to 0.174; without Banner it is 0.126, and today's Value 0.113. Every one of these is weak: our numbers explain little of what players take, and the gain from fitting is small next to what's missing. The fit was made on list share only, so a community list can test it without circularity.

The meta-lifted S units with Value in the 60s

44 units wear S today only because the meta lifts them, with a Value in the 60s. Jordan's example, Mortarion, is among them (his own section follows).

Under V1: S 6, A 9, B 15, C 13, D 1. Under V2: S 3, A 8, B 10, C 17, D 6.

Mortarion, the acid test

What the design session expected and what happened. The design session expected Mortarion to land C or D as an Anvil. He lands B: his Anvil is 92, from Toughness 12, a 2+ save, a 4+ invulnerable save, Feel No Pain 5+ and OC 6. But he ranks 110th of 341 Anvils, and Anvil's S quota is 33, so he sits nowhere near S. Equal Utility weights give him an A, because his support part (65) counts for more. Today he wears S only through the meta (10 of 16 Death Guard lists); on maths alone he's a B. So the proposal takes away the S that players' choices gave him and gives back about what Value says.

Why the maths rates him lower than players do. The ability names below come from our data (BSData); the effects are read on Wahapedia and put in our own words.

Under V2 he's C with the first Banner, and B once T-527's Banner scores are used (his aura proxy lifts his Banner to 65, and Hammer's quota shifts). Taking the first Banner: Damage before death lifts his Hammer to 78: he lives all five turns under typical fire, so his thin damage per point adds up. But Hammer is V2's most crowded job (376 units, S quota 40), and he ranks 173rd. OC 6 at 375 pts makes his Anvil per point ordinary (74). Runner (77) and Banner (56) don't carry him either, since one aura over 375 pts is little per point. V2 still sees none of Host of Plagues or Lord of the Death Guard. So neither version passes the acid test; the gap is the same two unmodelled effects.

What the acid test says. The proposal does what it was meant to do: the meta no longer hands out S, and a unit's letter comes from what it does. Mortarion shows the cost. What players take him for (mortal wounds every turn without rolling to hit, a once-a-turn buff for the units near him, a big debuff zone) is exactly what Hammer, Anvil and today's Utility can't see. These two are the first items for Utility v2:

1. a hand-written "passive damage" effect for abilities like Host of Plagues, scored like a weapon;
2. a "lifts the units near it" part for lords and auras, measured the way T-062 already measures a Leader's lift on its squad.

Until those exist, a unit like Mortarion should show the crown (10 of 16 lists) next to his B, with the one-word read Workhorse: the lists take him more than our maths explains. Death Guard would be a good faction for the community comparison once a public list can be read here. None is open to this environment yet.

V1 to V2

834 of 1161 units change letter between V1 and V2 (450 up, 384 down). Domains are V1's Hammer / Anvil / Banner and V2's Hammer / Anvil / Runner / Banner percentiles.

The 15 biggest rises:

The 15 biggest falls:

Job flips: 669 units change job between the versions (V1's Banner counts as kept when V2 calls it Runner or Banner). From V1's job (rows) to V2's (columns):

What the moves say. V2 changes a lot: 834 units change letter and 669 change job. Two patterns account for most of it:

• Cheap characters climb. Chaplains, Icon Bearers, Tallymen, Psychomancers and Fireblades jump from D to S or A. Runner rewards a single cheap model with a small footprint and good Leadership, and Banner rewards a Leader's lift per point. Of V2's 82 S units, 25 are characters; V1 has none. This is the leader blind spot from V1 being fixed, though too generously. Most leaders join a bodyguard rather than run alone, so their Runner score should count only when they can act alone (Lone Operative, or no squad to lead). Several of these risers (the Chaplains, 0 of 84 lists) are never taken.
• Glass cannons and Titans fall. Damage before death halves the Hammer of units that die in a turn (Krieg Heavy Weapons, Retributors: A to D). OC per point drops Titans from A to D. Both are what V2 intends; whether players agree needs the community list.

What players take that our numbers miss (a proxy for the community's reasons)

75 units are in 40% or more of their army's winning lists but get C or D under the proposal. Which of our traits they carry, against every unit's rate (lift = how much more often):

Space Marines, every unit

The list: Tista Minis, 'Warhammer 40K 11th Edition: Best Space Marine Units Right Now' (David Batista, 22 Jul 2026), tistaminis.com. A ranked list with a reason for each unit, read on 2 Oct; only unit names, our letter for where it sits and our own one-phrase paraphrase of its reason are kept here. Ranks 1 to 4: S. Ranks 5 to 12 and the named best characters: A. 'Other strong units to watch' and the role tables' alternatives: B. 'Good but not automatic choices': C. Units it doesn't mention aren't scored. Chapter characters it names (Vulkan He'stan, Adrax Agatone, Marneus Calgar) are other armies' rows in our data and left out.

Current: mean distance 1.40 tiers, 6 of 30 the same letter.

Maths only: mean distance 1.43 tiers, 6 of 30 the same letter.

V1: mean distance 1.43 tiers, 4 of 30 the same letter.

V2: mean distance 1.67 tiers, 3 of 30 the same letter.

V1's fitted weights on this list: mean distance 1.43 tiers with the weights fitted on list share, 1.47 with today's equal weights (the fit never saw this list).

Where they rate a unit higher than V2 does (19 units, biggest gaps first):

Where V2 rates a unit higher than they do (8):

The reasons behind the units we under-rate (V2 lower than theirs), counted:

A stand-in check: one review's best picks by role

Units one public review names as the best Space Marine pick in a role (tistaminis.com, 'Best Space Marines in 11th Edition'), seen only through a web search's summary because the site is blocked from this environment: names and roles only, in our words. Not a tier list; a stand-in until a full list can be read.

S or A among these 11: current 3, maths only 3, V1 0, V2 2.