Tactical Reroll
⋯

Bench 4: five cheap and mid-cost Tyranids (2026-10-02)

From reports/tiers/2026-10-02-bench-4-cheap.md , rendered when the site is built.

T-564's workbench on Zoanthropes, Von Ryan's Leapers, Neurogaunts, Pyrovores and Tyranid Warriors with Ranged Bio-Weapons. Baseline: all ten switches on except bestSize. What-ifs ran in a scratch copy of tools/tierbench.js over the tracked raw table (reports/tiers/2026-10-02-tyranids-raw.json, which reproduces round 10). Another process rewrote build/tier-raw-tyranids.json at 19:16 UTC during this run (baseline ρ 0.771 there, counts unchanged), so the tracked copy was frozen. Rules come from Wahapedia's 11th-edition pages, in our own words. Lists: grimstat-corpus (CC BY 4.0), 23 Tyranid lists. A report: nothing here changes a published score.

In short

1. Yardsticks

Shape: our 44 graded units ranked by the tie key and lettered with Auspex's own counts. This is a yardstick, not a proposed ruler (decisions.md: nothing is fitted on Auspex).

Run (on the baseline)Within oneSame, our rulerRaw sameShape: within oneShape: sameρTyr list share ρ
Baseline34221338260.760.61
+bestSize (all ten)30211339240.710.57
+listedSize35241441250.770.65
+minOrListed35241442240.780.66
+bestSize, fire scaled to points34241138260.760.65
Stealth in survival34231239250.780.61
OC falls as models die32181439230.750.56
Feed the Swarm, median of 5 co-listed35241239290.780.63
Package: minOrListed + Stealth + median-535261241250.810.69

With the package, nine units are still two letters off on our ruler: Exocrine, Hive Tyrant, Maleceptor, both Norns, Pyrovores, Tyrannofex, Winged Hive Tyrant and Zoanthropes. On Auspex's shape only three are: Neurogaunts, Carnifexes and Hive Tyrant.

2. bestSize: who it helps and hurts

Gap = Auspex minus ours in letters.

UnitAuspexSizes listedBest / listedBaselinebestSizeminOrListed
Hive GuardA6:16 / 6D (−3) 72.8B (−1) 93.3B (−1) 93.3
Warriors, MeleeB3:2, 6:46 / 6C (−1) 78.4B (0) 93.2B (0) 93.2
ZoanthropesS3:7, 6:166 / 6B (−2) 85.6B (−2) 90.5B (−2) 90.5
NeurogauntsB11:221 / 11A (+1) 96.5S (+2) 99.5A (+1) 96.5
Warriors, RangedC3:26 + Prime / 3B (+1) 86.6A (+2) 94.9C (0) 86.6
PyrovoresA2:4, 3:31 / 2C (−2) 85.2C (−2)C (−2) 85.2 (listed alone: D 73.1)
Hive Tyrant, Tyrannofex, Norn Emissary, Old One Eye1 / 1each −1 letter, passed by climbersHive Tyrant −1, rest unchanged

Helps: the three graded units the lists field large: Hive Guard, the Melee Warriors and Zoanthropes. Hurts: the units fielded small but scored large (Neurogaunts, the Ranged Warriors, and Gargoyles, Termagants and Barbgaunts, which are ungraded), and every single-model unit they pass. Why: per point, a bigger squad deals and soaks the same. Only turns alive grow, because the typical fire doesn't grow with the target. With fire scaled to points, best size falls back to the smallest (34; Hive Guard picks 3 again). So "best size" by our measure always means the biggest. Use the most-listed size, per unit, never below the minimum (minOrListed). Most-listed size alone sends Pyrovores to 2 models (−12, C → D), because their Banner per point favours one model. Caveat: Hive Guard's 6 rests on one list.

3. The five units

Zoanthropes (90 pts / 3, 190 / 6; Auspex S, 16 lists)

Check. T5, 5+, 4+ invulnerable, W3, Fly, Synapse; two Psychic Warp Blast modes, scored as alternative profiles. Warp Field (a 6+ invulnerable aura) is in the baseline through auraToNeighbours. Spirit Leech is noted: it needs the enemy's Battle-shock as a dice result and is worth a few wounds a game. Psychic attacks ignore hit modifiers and cover: worth nothing on our no-cover table (+6.5 in residuals-1's terrain run).

UnitPts (models)Auspex · listsOurs (shape)H/A/R/BDmg /100 pts G/M/T/LV/HV
Zoanthropes + Neurotyrant (baseline)220 (4)S · 16B (B) 85.684/83/12/22232 238 221 344 253
Zoanthropes at 6190 (6)S · 16B (A) 90.588/86/15/11265 188 219 413 296
Hive Guard at 6160 (6)A · 1B (S) 93.391/83/27/0224 153 204 475 329
Tyrannofex170 (1)S · 7B (A) 90.288/87/15/0237 217 197 403 403
HypothesisKindResultWithin one / shape / ρ
Z1 scored at the size most lists take (6), minOrListedconsensus right, unmeasured (size)90.5; B, shape A35 / 42 / 0.78
Z2 reward generalists: Hammer peak 0.25 (+minOrListed)weight91.2; B35 / 41 / 0.78
Z3 the S prices the Neurotyrant brick (15 of 16 lists)popularitythe pair at 6 + 1 scores 87.9, below 6 alone–

Verdict: size, then the ruler. At 6 models they are 23rd of 52 (Hammer 88). The quota's 5 S places hold them at B; on Auspex's shape they are A. The rest of the S is the Synapse brick, whose Battle-shock help we don't credit. Change: minOrListed: it also moves Hive Guard +20.6 (D → B), Broodlord and Genestealers +15.2, Melee Warriors +14.7 (C → B) and the Neurotyrant +2.0.

Von Ryan's Leapers (55 pts / 3; Auspex S, 12 lists)

Check. M10", T5, 4+, 6+ invulnerable, W3; 6 attacks each at WS 3+, S5, AP −1. Infiltrators is modelled (reach 95; the walk way starts 12" away). Stealth and Fights First are not modelled: tiertest.js reads neither. Pouncing Leap (a cheaper Heroic Intervention outside the once-a-phase limit) is noted, CP only. Synapse: the Leadership part is 0 for every Tyranid.

UnitPtsAuspex · listsReachSurvival turns (pct)LdUnits /100 ptsFootprintRunner (pct)
Von Ryan's Leapers55S · 12951.61 (32)0925750.4 (94)
Lictor60– · 111002.16 (42)6877156.9 (97)
Neurolictor80S · 17953.26 (60)6637155.4 (97)
Deathleaper80A · 7953.69 (67)6637157.1 (98)

Survival is the whole gap; the Lictors' Lone Operative doubles theirs. The Leapers' damage, the best of the five per point (141 per 100 pts into GEQ), isn't in Runner.

HypothesisKindResultWithin one / shape / ρ
L1 Stealth: BS one worse for ranged attackers (×4/3 hits needed on the four ranged classes), 98 pool unitsconsensus right, unmeasured95.0; B → A34 / 39 / 0.78
L2 the ruler: 15th of 52order agreesS on Auspex's shape already–
L3 Fights Firstconsensus right, unmeasuredneeds the melee exchange in the engine–

Verdict: a missing quantity. Stealth is on the datasheet and gives the benefit of cover, a rule we already apply. Change: Stealth in survival (Anvil and Runner). It moves Gargoyles +3.4 (A → S), both Raveners +2.3, the Neurolictor +1.6 (A → S) and the Deathleaper +1.4. Biovores, Hormagaunts and both Norns drop across the S and A lines, so within one stays at 34.

Neurogaunts (45 pts / 11; Auspex B, 2 lists)

Check. T3, 6+, W1, OC 1. Neurocytes (Synapse near a Synapse unit) is noted; residuals-3 measured the relay at almost nothing. Data: build/units.json has the larger size as 21 models; the datasheet has 22.

UnitPtsAuspex · listsOurs (shape)H/A/RAnvil rawOC / pt
Neurogaunts45B · 2A (S) 96.524/97/4949.70.24
Termagants60– · 4S 98.748/99/4463.20.33
Hormagaunts70S · 13S (S) 97.725/98/7157.20.29

Anvil raw 20% below Hormagaunts' is 1 percentile point lower: every horde sits in the pool's top 4%.

HypothesisKindResultWithin one / shape / ρ
N1 OC falls as models die, Anvil × (m+1)/2m, every unitconsensus right, measured96.3; A (B with minOrListed)32 / 39 / 0.75
N2 the two-letter gap with all ten on is bestSize's 21 modelswrong sizeA (+1) under minOrListed35 / 42 / 0.78
N3 Termagants (a gun) and Hormagaunts (M10") fill the slot: 4 and 13 lists against 2popularitynothing per point separates them–

Verdict: no change of its own. minOrListed removes the two letters. N1 lowers every horde together and costs 2 within one. The +1 that remains is popularity between near-equal hordes, and the lists agree with Auspex.

Pyrovores (40 / 70 / 100 pts for 1 / 2 / 3; Auspex A, 4 lists)

Check. T6, 3+, W5; a 12" Torrent flamer with Ignores Cover and Twin-linked, modelled. Burning Spray (strips cover) is noted and worth nothing on our table; Deadly Demise is not counted. Harvester: in Assimilation Swarm, each Harvester unit regenerates one unit within 6" in each Command phase (wounds back, a model back, or 3 for an Endless Multitude). tiertest.js credits this as a lend capped at 38.4 pts for every Harvester at every size. The lists' use: 2 or 3 models (sizes 2:4, 3:3) beside the Norn Assimilator, Hive Tyrant and Biovores, in Assimilation Swarm with Talons of the Norn Queen. They field durable Harvesters with an Overwatch flamer, not cheap action bodies.

UnitPts (models)Auspex · listsOurs (shape)H/A/R/BDmg /100 pts GEQTurnsBan raw
Pyrovores40 (1)A · 4C (B) 85.213/33/41/85121.140.96
Pyrovores70 (2)A · 4C 73.149/40/73/02042.170.55
Ripper Swarms30 (1)B · 1B (S) 92.58/0/55/9341.001.40

Regeneration's worth (EV.healLift) on the squads the lists put beside them: Neurotyrant 102 pts, Neurolictor 71, Hormagaunts 61, Hive Tyrant 42, Norn Assimilator 29, Biovores 16.

HypothesisKindResultWithin one / shape / ρ
P1 Feed the Swarm on the Harvester's co-listed squads: median of the five most co-listed (42 pts)consensus right, measured badly88.0; C → B, shape A35 / 39 / 0.78
P1b best of the two most co-listed (71 pts)variant95.7; A34 / 39 / 0.78
P2 a Harvester lends only while it lives (× turns ÷ 5)consensus right58.3; D33 / 39 / 0.76
P3 at the most-listed size (2)size73.1; D35 / 41 / 0.77

Verdict: a quantity measured badly. The cap gives a 30-pt Ripper Swarm and a 40-pt Pyrovore the same lend, whatever they heal. Change: P1. It holds under any tie order; P1b's partner flips between three squads with 3 lists each. It moves Ripper Swarms −19.1 (B → C; their co-listed squads, from 1 list, gain little); no other composite moves. Two engine notes. A lone Pyrovore dies before its 12" gun reaches range (its damage per point is a seventeenth of two models'). And P2 shows why the lists field 2 or 3: a lone Harvester's lend should end when it dies. Next test: P1 with P2 at the lists' size.

Tyranid Warriors with Ranged Bio-Weapons (60 pts / 3; Auspex C, 1 list)

Check. T5, 4+, W3, OC 2 a model, Synapse; devourers by default (venom cannon or strangler, 1 per 3, isn't in the table). Adaptable Predators (shoot and charge after Falling Back) is noted. Data bug: build/units.json gives them Stealth, as it does the Melee Warriors, Venomthropes and Genestealers. The 11th-edition datasheet has none: BSData hides the link unless the attached leader takes the Chameleonic enhancement. docs/bsdata.js:986 collects rule links without reading hidden modifiers (its Fights First check, line 1065, does). The live rankings show a "stealth" trait; the tier test doesn't read it.

UnitPtsAuspex · listsOurs (shape)H/A/R/BDmg /100 pts GEQOC / ptTurns
Warriors, Ranged60C · 1B (A) 86.635/87/48/101200.101.83
Warriors, Melee75B · 4C (B) 78.419/78/42/10210.081.83
Venomthropes55B · 1C (C) 76.033/71/59/761260.051.95

Why B with 1 list: Anvil (OC 6 for 60 pts, the 87th percentile), and nothing else.

HypothesisKindResultWithin one / shape / ρ
W1 scored at the size the lists take (3), minOrListed: other units climb pastsizeB → C (exact); shape B35 / 42 / 0.78
W2 OC falls as models dieconsensus right, measured79.6; C32 / 39 / 0.75
W3 consensus wrong: per point they out-shoot the Melee Warriors (120 against 21) for less, yet Auspex and the lists (4 against 1) prefer the melee unitpopularity, melee-centred playHammer's melee delivery is beyond this bench–

Verdict: size, no change of its own. With all ten on they were scored as 6 with a Prime (A); at the listed 3 they are C. W2 also reaches C, but costs 2 within one.

4. Bugs and gaps

  1. Data, live: docs/bsdata.js:986 ignores hidden modifiers on rule links. A leader's enhancement (Chameleonic) becomes unconditional Stealth on four Tyranid units in the rankings' Utility; other armies' conditional links are likely affected.
  2. Data: Neurogaunts' larger size is 21 models in build/units.json; the datasheet has 22.
  3. Engine: Feed the Swarm is a flat 38.4 pts per Harvester, blind to the squad healed and to the Harvester's survival.
  4. Engine: absolute survival drives size: a lone short-ranged walker delivers almost nothing (Pyrovores ×17 from 1 to 2 models), and the same effect makes bestSize pick the largest size.
  5. Not modelled: Stealth, Fights First, Spirit Leech, Burning Spray, Deadly Demise.

5. Method

The scratch bench is tools/tierbench.js plus the size switches, a per-row hook and per-row multipliers (hits to destroy, Hammer, Anvil), applied to pool and Tyranid rows alike. Stealth comes from build/units.json without the four false Tyranid entries. Co-listing and sizes come from reports/tiers/2026-10-02-tyranids-colist.json and -sizes.json.