Tyranids against the Auspex list (2026-10-02)

T-524 rounds 4 and 5 (Jordan, 2 Oct, via the supervisor). Made by node tools/tiertest.js --round4 … --round5 … --tyr4-out; round 4's section is also in reports/tiers/2026-10-02-space-marines.md, where V4 and step g are explained. A report: nothing here changes a published score or page.

Round 5 (2 Oct): V5, eight variables one at a time, Tyranids first

T-524 round 5 (Jordan, 2 Oct, via the supervisor). Jordan's rule: iterate on Tyranids against the Auspex list, Tyranid list share and his reading, one variable at a time, until the Tyranid table is explained. V5 is V4 plus the eight variables he named, each a switch measured on its own and stacked in his order, plus a ninth found on the way (Banner's smooth start, below). Space Marines and Orks run once at the end as a check.

In plain words.

• V5 agrees with Auspex more than V4 does, and with what Tyranid players take:
  • order (ρ with Auspex): 0.19 → 0.38;
  • the same letter as Auspex: 14 → 16 of 44, and within one letter: 26 → 31;
  • ρ with Tyranid list share: 0.12 → 0.24.
• Today's live letter still agrees most (ρ 0.65, 21 the same, 37 within one), because it borrows popularity: both it and Auspex follow what wins.
• What moved the table most:
  • The small fixes (8). Biovores' Spore Mines make them a top Runner (D → C), the Hive Tyrant's and Swarmlord's CP count, and one-use units stop being Runners.
  • Detachment context (7). The Norns become S, through Singular Purpose: re-rolls into a chosen target, or OC 15 and Feel No Pain 5+ on an objective.
  • Survival absolute (1). On its own it makes the order worse (0.19 → 0.15) but the letters better (16 and 29 of 44). The big monsters live longer (the Norn Assimilator C → A; the Hive Tyrant's and Swarmlord's Hammer rise 15 points), and cheap elites fall (Hive Guard S → A, Zoanthropes B → C, Carnifexes A → B).
• Arrival rules (2) do less than expected, for a reason in the rules:
  • The Red Terror leaves after its own fight, so it's off the table when the enemy shoots. Its Hammer rises 5 points.
  • Raveners leave at the end of the enemy's turn, after being shot. Re-arriving 9" away gives a 28% charge, against a near-certain one from where they stand. So Death From Below buys position (Runner) in our five-turn game, not life or damage.
• The meta's fire (3): small arms 8%, special 20%, plasma-class 37%, elite 7%, anti-tank 22%, mortal wounds 6%. Weighting by it, with the sixth attacker that ignores saves, helps a little on its own (0.19 → 0.20; list share 0.12 → 0.14).

Does V5 explain the Tyranid table? Partly.

• Explained now:
  • the Norns (S, as Auspex);
  • Hive Guard (A, as Auspex, so not a Sleeper);
  • Biovores, up a letter;
  • the Hive Tyrant and Neurotyrant, C → B;
  • Ripper Swarms, our aura error, A → C;
  • Carnifexes, closer (A → B; Auspex C).
• Still unexplained, and why:
  • Raveners, Hyperadapted Raveners and the Red Terror (S, S and A against D, C and D). They live about 2.5 turns, and their Deep Strike charge comes off 28% of the time. The detachment movement rules weren't in detfx.json when this ran: Subterranean Assault's tunnels and Vanguard Onslaught's charge after an advance. A what-if with tunnels lifts the Red Terror's Hammer from 54 to 71, still a D. What's left is likely the stratagems we don't model: re-deploying and Rapid Ingress, played every round, and choosing the target each time.
  • Zoanthropes, the Swarmlord, the Hive Tyrant, the Maleceptor and the Neurolictor (S against B, B, B, C and B). Their value is what they do for others: Synapse, CP, Shadow in the Warp, Battle-shock and debuffs. Our worths for those are stated guesses: a CP at 15 pts, a Battle-shocked enemy unit-round at 10% of a median unit, a debuff on half the attacks. Doubling them is the next test, and Jordan's reading should set them.
  • Biovores (S against C). Spawning makes them a top Runner (83rd percentile), but the Runner job's S and A cuts are the highest of the four.
  • The Tyrannocyte (C against S). Our error remains: V2c's transport-capacity credit. It values a one-use pod by what it adds to Warriors or Pyrovores per point, and 0 of 23 lists take it.

Each variable, alone and cumulative

Measured on the Tyranid units only, against a pool that doesn't move: each run's percentiles and letter cuts come from one all-units run at its scale (variables 1 and 3 rescale every unit, so the pool is rescored with them; the other variables are Tyranid rules, ranked against V4's pool). With every switch off this reproduces V4 exactly: ρ 0.19, 14 the same letter, 26 within one. Columns: Spearman's ρ with Auspex's order (44 graded units), the same letter and within one letter of Auspex's (global letters, of 44), and ρ with Tyranid list share (the 52 Tyranid units, 23 lists). Higher is better on every column.

The numbers behind them. The meta's incoming fire (variable 3), as the share of the expected damage winning lists' weapons deal into the five stand-in targets: Small arms 8%, Special 20%, Plasma 37%, Elite 7%, Anti-tank 22%, Mortal wounds 6%. The sixth attacker is 2 mortal wounds a hit, no save, Feel No Pain counts (Feel No Pain against mortal wounds too: the Norn Emissary's 4+, from docs/data/tierfx.json). An enemy unit fails a Battle-shock test 36% of the time at no modifier, 51% at −1 (the Leadership of winning lists' units). A Synapse unit shares Synapse with the 5.7 Synapse units an average Tyranid list carries. Constants (V5C): mortal Damage 2; an enemy-side aura touches 50% of the attacks on the unit and its two nearest friends; a Battle-shocked enemy unit-round is worth 10% of the median unit (120 pts), ×2 while the Neurolictor is near; 10 enemy units take an army-wide test; a CP is worth 15 pts; a spawned unit's Runner counts at 50%; Synapse range 75% of the game; tunnels 7"; a marked target takes 50% of two friends' attacks.

The named units, before and after

V4 and V5 are global letters; V5 is the ceiling under the Context rule (its best detachment), with the floor (no detachment, the Even lens: no meta weights) as a range where it differs. The four jobs are V5's percentiles. "Moved by" is the variable that changes its composite most on its own (composite points; a letter change where there is one).

What each rule adds to Banner's raw number (points lent per point of the unit, V2c's scale: an aura stand-in was 2 × the median Leader lift × the median unit's points; where a rule now models an aura, it replaces that stand-in):

Leader pairs (variable 6: shown, never in the letter)

Each Tyranid squad's best leader by what its effects (data/leaderfx.json; the Hyperadapted Raveners' from unitfx.json, since its rule covers its unit) add to the squad's own damage, the five targets' mean on the default wargear; an effect that needs a Battle-shocked target counts a fifth of the time. Leaders come from BSData's links and from the joins written in winning lists. Jordan's rule: a leader doesn't lift the squad's letter (Banner credits the leader); this is the line the squad's page would show.

Detachment context (variable 7): what detfx.json held at run time

Each Tyranid unit is scored under every Tyranid detachment and keeps the best (the Context rule's ceiling); its floor is no detachment under the Even lens. Dice effects with a condition the engine can't check count for a share of the game: within Synapse range 75%, below half strength 35%, below Starting Strength the same for units of more than one model (never for one), a Battle-shocked target 20%; a pick made each battle round counts each option for one round in five, and a pick made once takes the option with the most effects. Singular Purpose is the Norns' own rule but runs here, as the Talons of the Norn Queen's swap needs it: re-roll hits and wounds into the best chosen target, or OC 15 and Feel No Pain 5+ on an objective; with the swap, half the game each. The detachment data is another builder's (the Tyranid coverage card); this run used what was there, and the movement rules weren't yet:

What-ifs for the movement rules not yet in the data (not V5; each with V5's other switches, no other detachment effect): Subterranean Assault's tunnels (reserves arrive 7" away once a burrower is down, a 58% charge where Deep Strike's 9" is 28%) and Vanguard Onslaught's charge after an advance for VANGUARD INVADER units.

Hive Guard and Zoanthropes, left alone. Hive Guard: Auspex A, V4 S, V5 A (1 of 23 lists). V5 no longer rates it above the list, so it isn't a Sleeper candidate: absolute survival takes it from S to A (${hg.m} T${(recOf(hg).stats || {}).T} bodies on ${hg.p} pts live fewer turns than their price suggests). Zoanthropes: Auspex S, V4 B, V5 B.

The end-of-round check: Space Marines, Orks, list share

• Space Marines (Tista Minis, 30 units): ρ 0.04 → −0.01; mean distance within the army 1.33 → 1.27. Absolute survival on the big infantry: Centurion Assault Squad C → C, Centurion Devastator Squad B → A, Terminator Squad B → A.
• Orks (Sprues & Brews, 45 units): ρ −0.19 → −0.12; mean distance within the army 1.38 → 1.31.
• List share (every unit 5 or more winning lists could field): ρ 0.104 → 0.129.

Engine kinds (T-540, T-544, 2 Oct): heal, mortal and aura from the coverage data, Damaged X from the datasheet

Made with --kinds on top of round 5's command. V5 with every round-5 switch on, then each kind the dice engine now plays switched in, alone and cumulative, on the Tyranid units against round 5's pool. The kinds come from the coverage files (docs/data/unitfx.json, detfx.json, armyfx.json) with their numbers from docs/data/kindfx.json until the coverage files carry them; the calculator plays the same rules (To the death, the volley, a joined character's aura). A report: nothing here changes a published score or page.

Mortal wounds, unified: the Hammer's mortal wounds outside attacks come from the coverage data's numbered mortal rules for 5 Tyranid units in the pool (Harpy: Spore Mine Cysts, 4 a use, as mortalfx.json had it; Mawloc: Terror from the Deep, 2 a use, as mortalfx.json had it; Mucolid Spores: Floating Death, 1.9 a use, as mortalfx.json had it; Spore Mines: Floating Death, 1 a use, as mortalfx.json had it; Toxicrene: Hypertoxic Miasma (Aura), 1.6 a use, as mortalfx.json had it); mortalfx.json stays for the 101 other units of the pool it covers, whose armies' coverage has no numbers yet. Rare triggers (a failed Battle-shock, an enemy falling back) are left out, as mortalfx.json left them. The same numbers within rounding, so nothing moves: the change is where they're kept.

Buff auras, from the coverage data: 4 Tyranid bearers in the pool read their aura from unitfx.json (range, to, not) with its effects from the rule or kindfx.json, run through the engine on the two squads most often listed with them that they reach, ×2 squads as round 5: Hive Tyrant (Onslaught (Aura, Psychic) on Biovores +7% (with it in 5 lists); Onslaught (Aura, Psychic) on Zoanthropes +5% (with it in 4 lists)); Tervigon (Brood Progenitor (Aura, Psychic) on Termagants +49%); Venomthropes (Foul Spores (Aura) on Biovores +27% (with it in 1 lists); Foul Spores (Aura) on Neurolictor +27% (with it in 1 lists)); Zoanthropes (Warp Field (Aura, Psychic) on Raveners +4% (with it in 10 lists); Warp Field (Aura, Psychic) on Hormagaunts +12% (with it in 9 lists)). tierfx.json's aura stays for the marks (the Exocrine, the Psychophage). The same squads and lifts as tierfx.json's auras gave in round 5, so nothing moves: the change is that they're read from the coverage rule (and its not: the Venomthropes' spores skip a MONSTER, which none of their two squads is).

Damaged X, from the datasheet (T-544): 14 single-model Tyranid units in the pool have it (bsdata.js reads it from BSData: the Orks' core rule's number, or an older datasheet's "1-X wounds remaining" bracket with its -1 to hit). Its attacks take -1 to hit once typical fire has taken W − X of its W wounds, at turn-time T × (W − X) ÷ W of its T turns alive (ev.js damagedAt), each later turn's ranged and melee damage scaled by what -1 to hit leaves against each target. The calculator plays the same -1 in Fight back and To the death by the wounds the model has left. The Hammer of a unit that lives long enough to be damaged falls; a fragile one, gone before, is unchanged. The pool's other armies are scored without it for now, so the Tyranid monsters' fall against them is an upper bound until the pool is re-scored with it.

The new vocabulary (T-545): -1 Toughness (tMinus, no longer written as +1 Strength), +1 Toughness, a Save improved to no better than 3+, one hit and one wound re-rolled a volley, a mixed unit's majority Toughness, and the conditions "the attack's Strength above (or at most) the target's Toughness" and "Rapid Fire weapons only". Their numbers for the rules the coverage reports name are in kindfx.json (Death Guard's contagion, the Great Unclean Ones' Nurgle's Rot, Take Cover! and First Rank, Fire!, the Fire Prism's and the gunships' re-rolls, the Kill Teams, and 24 S-against-T conditions the player used to vouch for). Tyranid units in the pool they reach: Tyranid Warriors with Melee Bio-Weapons (an option switched off, the Warriors' +1 Toughness, which the tier test leaves out as it leaves out every effect marked off). Re-run with them, no Tyranid letter or composite moves: nothing here touches a Tyranid unit's own dice in the tier test, which reads the units' own abilities from unitfx.json as written.

What moves: 4 letters; the table also lists composites that move by 2 or more.

Engine kinds, later cards (re-run with --kinds after each)

• T-548, marks on an enemy unit (chosen, codex): no letter moves, no composite moves. The Norns' Singular Purpose prey is a chosen effect now (kindfx.json), but the tier test plays Singular Purpose from tierfx.json (every target in turn as the prey, or the objective), so the table is the one above.
• T-549, resource pools (swapSix, pool): no letter or composite moves: no Tyranid rule spends a pool, and the tier test reads no army's Miracle or Fate dice.
• T-550, tests a unit takes on itself (selfTest): no letter or composite moves: no Tyranid rule has one, and Hazardous keeps its numbers.
• T-551, a friendly unit nearby (lend, Bondsman and War Dog auras): no letter or composite moves: the tier test reads the Tyranid auras and marks from tierfx.json and the coverage aura objects as before; the lent marks (the Exocrine's, the Psychophage's, the Pyrovores') are the calculator's only for now.
• T-552, every army scored: the Tyranids' own abilities and every kind in one run is the table's last row (--own-kinds); every army carrying its rules is the next section.

Every army scored (T-552, 2 Oct): own abilities, kinds, army rules and the best detachment for every unit

Made with --every on top of round 5's command with --kinds. Until now only the Tyranid units carried their own abilities, kinds and detachments, and the other 1109 units of the pool were scored bare, so every percentile and quota was set against bare units. Here every unit carries its own abilities (unitfx.json; effects marked off left out, the same shares for the conditions the engine can't check), the engine kinds (heals, mortal wounds outside attacks, its auras on the squads its army takes most, Damaged X, T-545's vocabulary through kindfx.json), its army rules (armyfx.json through DetFx, Synapse's pilot shape left to variable 5; a pick's best option per target) and its army's best detachment (the Context rule), each a switch, cumulative. Base: round 5's V5 with every kind for the Tyranids and the other units bare (the "Engine kinds" table's last row). No constant changes. A report: nothing here changes a published score or page.

Letters moved by army (base → every switch on): 423 of 1161 units move, 197 up and 226 down.

The Tyranid table, every army scored: 28 Tyranid letters move against the base (the pool they're ranked in now carries its rules too); composites that move by 2 or more are listed.

Round 11 (T-565, 2 Oct): the workbench's fixes made permanent, every army

Round 11 (Jordan, 2 Oct: best legal size, turns capped at 5, a CP 30 pts). The Tyranid workbench's fixes made permanent in tools/tiertest.js as o.r11 switches, on every army: Banner counts only what a leader lends (its attacks go to Hammer inside the squad), a character at its real model count, Anvil's turns as the harmonic mean, the re-arrivers' two rules, every lend on the two squads it lifts most, Stealth in survival, the best legal size, the best leader, and effects on the loadout the damage uses. Against round 10 on today's data: Tyranids ρ 0.52 → 0.66, Space Marines −0.07 → 0.14, Orks −0.19 → −0.15; the per-army median 0.139 → 0.173. Within one letter of Auspex: 31 of 44 on the within-army ruler, 36 on Auspex's own shape; the 85% line (38) is not reached. Full table and sections: reports/tiers/2026-10-02-tyranids-round11.md (/lab/reports/tiers/2026-10-02-tyranids-round11/).

Round 10 (T-563, 2 Oct): Anvil as OC-turns per point, the final rank, the list-share headline

Round 10 (Jordan, 2 Oct, from the method appraisal). Anvil = OC × min(turns alive, 5) ÷ points, no gate; units over 2,000 pts out of the pool; the published score is the pool percentile of the composite, lettered on frozen cuts (S ≥ 92.02, A ≥ 77.02, B ≥ 49.96, C ≥ 20.03: 8/15/27/30/20%). Tyranids ρ 0.49 → 0.51, Space Marines −0.10 → −0.07, Orks −0.12 → −0.18; every unit's list share 0.077 → 0.157, the new headline (the per-army median ρ) 0.113 → 0.165, AUC 0.536 → 0.582; no unit over 1,000 pts is S. Also --explain, one unit against its peers, run on the eight Tyranids two or more letters off. Full table and sections: reports/tiers/2026-10-02-tyranids-round10.md (/lab/reports/tiers/2026-10-02-tyranids-round10/).

Round 9 (T-562, 2 Oct): Anvil divides by points once, three panels

Round 9 (Jordan, 2 Oct, from the independent critique). Round 8's run with Anvil = hits to destroy × OC ÷ points (one division, not two) × a survival gate (×0.5 removed in one typical enemy turn, ×0.8 in two): Tyranids ρ 0.47 → 0.49 (SE 0.13), Space Marines −0.18 → −0.10, Orks −0.12 → −0.12, every unit's list share 0.162 → 0.077; Neurogaunts' Anvil 99 → 89, the Norns' 96 → 100. Also the one-at-a-time sweep of every hand-set constant (--sweep; Runner's points gate, the Advance and the objectives' distances move most), every version since V2b against all three panels with a chart, and proposed frozen cuts (S ≥ 98.79, A ≥ 94.49, B ≥ 86.13, C ≥ 71.15; not adopted). Full table and sections: reports/tiers/2026-10-02-tyranids-round9.md (/lab/reports/tiers/2026-10-02-tyranids-round9/).

Round 8 (T-560, 2 Oct): step grading on fixed cuts

Round 8 (Jordan, 2 Oct). Round 7's run graded by steps (the best job's percentile, +5 for each other job at 90 or above, +2 for each at 80 to 89, never capped; ties by the mean of the others), lettered on fixed cuts over every army (S ≥ 95, A ≥ 88, B ≥ 80, C ≥ 65) and, as a second view, by position within the army (10% S, 15% A, 25% B, 30% C, 20% D): ρ with Auspex 0.50 → 0.47 (SE 0.13), strict on our ruler (within-army view) 20 → 20 of 44 (chance 10.6), raw 12 → 19 (all armies), within one 29 → 31, Tyranid list share 0.39 → 0.41, every unit's list share 0.164 → 0.162; 246 units are S on the fixed cuts. The full table, both views and the S count per army: reports/tiers/2026-10-02-tyranids-round8.md (/lab/reports/tiers/2026-10-02-tyranids-round8/); the bars on sliders: /lab/reports/tiers/sliders/.

Round 7 (T-559, 2 Oct): best-only grading

Round 7 (Jordan, 2 Oct). Round 6b's run graded best-only (the best job's percentile plus 0.25 a point for each other job above 60, at most 100), with Feed the Swarm's lend capped and the Tyrannocyte delivering once: ρ with Auspex 0.39 → 0.50 (SE 0.12), strict on our ruler 19 → 20 of 44 (chance 10.3), raw 12 → 12, within one 27 → 29 (chance 20.4), Tyranid list share 0.32 → 0.39, every unit's list share 0.171 → 0.164. The full table and every yardstick: reports/tiers/2026-10-02-tyranids-round7.md (/lab/reports/tiers/2026-10-02-tyranids-round7/).

Round 6 (T-557, 2 Oct): Jordan's settings and the movement rules

Round 6b (Jordan, 2 Oct): a CP worth 30 points. Round 6 run again with V6C.CP_PTS = 30: ρ with Auspex 0.34 → 0.39, same letter 12 → 12, within one 26 → 27, Tyranid list share 0.27 → 0.32, every unit's list share 0.169 → 0.171; the Hive Tyrant C → B, the Swarmlord B → A, the Lictor B → A. Side by side, with the full table: reports/tiers/2026-10-02-tyranids-round6b.md (/lab/reports/tiers/2026-10-02-tyranids-round6b/).

The full report is reports/tiers/2026-10-02-tyranids-round6.md (/lab/reports/tiers/2026-10-02-tyranids-round6/). Made with --round6 on top of the "Every army scored" command.

What changed:

• A CP is worth 0 (T-558 will price a CP from stratagems).
• Battle-shock stays at 10% and enemy debuffs at half face value, both marked tunable.
• The highest Toughness in a mixed unit.
• The detachments' movement rules are read in full (move). Their reach feeds three jobs through a new position track built on the delivery modes:
  • Hammer's turns in reach;
  • Anvil's position (the share of the game it holds the midfield objective and the one in the enemy's half);
  • Runner's reach.
• Oath at full share, army rules counted as on.

What moved the numbers:

• The position track alone lifts ρ to 0.41 and Tyranid list share to 0.36.
• Tunnels, surge, advance-and-charge and the Deep Strike grant hold or improve that.
• Two rules cost agreement:
  • Crusher Stampede's +2 OC (counted at 0.9 of the game) lifts every monster's Anvil.
  • Feed the Swarm's regeneration, lent to a 30- or 40-point Harvester, makes Pyrovores and Ripper Swarms B Banners (Auspex: A and B).
• Read in Hammer only, with no position, the same rules give 0.27, the old what-if's 0.28.
• Biggest movers: Pyrovores D → B and Ripper Swarms C → B (Feed the Swarm); the Psychophage, Mawloc, Maleceptor and Trygon D → up a letter (Crusher Stampede's OC); Biovores C → B (tunnels). The Hive Tyrant drops B → C (its CP Banner) and the Parasite of Mortrex B → C (Runner's reach from the track).

Round 4: V4 and every domain

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, V5 16 / 31.

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.

Coverage re-run (T-524, 2 Oct): V5 with every kind and every effect switched in

The tier test re-run once T-542 had written every army's unit abilities, detachments and army rules, and T-540 and T-543 to T-547 had made the engine read them. Made with round 5's full command plus --kinds and --own-out (the commands in reports/handoff.md), once on today's data files and once with docs/data/unitfx.json, detfx.json, armyfx.json and leaderfx.json as they were before the sweep (commit 993903b, T-541's report), each with and without --armyfx. The tool has no flag that stacks --own-out on --kinds or writes V5's letters for every unit, so those two were read from a scratch copy of tools/tiertest.js that only adds a dump (no formula or constant touched; the committed tool is unchanged). A report: nothing here changes a published score or page.

In plain words.

• The sweep moves nothing outside the Tyranids, because the tier test doesn't read it yet. Today's data files against the pre-sweep ones, everything else equal: 0 of the pool's 1,161 units change letter, and no composite moves. The tier test reads the units' own abilities and the engine kinds (heal, mortal, aura, damaged) for the Tyranids only, --armyfx only in the Tyranid pilot's shape (an aura with from, and ld), and the other armies' detachments only through V2b's detachment credit, which the sweep didn't change.
• On the Tyranids, full coverage moves 5 letters against V5 and 1 against the last V5 + kinds run. Venomthropes C → B: their own Foul Spores (cover for themselves). Hyperadapted Raveners gain 3.6 composite points (Alpha Invader's Sustained Hits) but stay C. The other four are the kinds already reported: Damaged X takes the Trygon A → C and the Haruspex and Toxicrene A → B; Old One Eye's heal takes him A → S.
• The yardsticks barely move: ρ with Auspex 0.39 → 0.39, the same letter 17 → 18 of 44, within one 30 → 30, ρ with Tyranid list share 0.24 → 0.23.
• --armyfx should stay off for now. With it, 100 letters move: 26 Tyranid letters go up (Pyrovores D → B, Raveners and the Red Terror D → B, 12 units to S), and 74 other units move only because the cuts move. Three reasons it isn't a coverage effect:
  • only the Tyranid pilot (Synapse) is in the shape it reads, so the Tyranids gain an army rule against a pool that has none;
  • Synapse's +1 Strength is already counted in V5's variable 5, so it counts twice;
  • its Battle-shock on three dice lifts every Tyranid's Runner and Anvil. With it, ρ with Auspex falls to 0.33 and ρ with list share rises to 0.30.

The yardsticks

On the Tyranid units against round 5's pool. Columns as in round 5: ρ with Auspex's order (44 graded units), the same letter and within one letter of Auspex's, of 44; ρ with Tyranid list share (52 units, 23 lists). "Full coverage" is V5 plus heal, mortal, aura and damaged plus each unit's own abilities from unitfx.json. The "less" rows drop one kind from full coverage.

Each kind's share:

• Heal is the only kind that lowers agreement (dropping it gives 0.40 / 18 / 31 / 0.24). It lifts Old One Eye and the Red Terror, and Auspex rates both lower than we do.
• Damaged X is the kind that helps most.
• Mortal and aura add nothing on top: the same numbers as the files they replaced.

Letters that move against the last V5 + kinds run

No other Tyranid composite moves by a point or more between the last run and this one.

The Tyranid table with full coverage

Every Tyranid unit in the pool, in Auspex's order and then by lists. The named units are in bold. Before is the last V5 + kinds run; full coverage is the letter under the Context rule (its best detachment), with the floor (no detachment, the Even lens) where it differs. The four jobs are full coverage's percentiles. Offense and Defense are the published rankings' raw numbers (per 100 pts, percentile in brackets), which the coverage doesn't touch. Delivery is the way in the Hammer chose and the detachment it's scored under. The last column is the same run with --armyfx, for the record.

The end-of-round check

• Space Marines (Tista Minis, 30 units): ρ −0.01, the same as round 5's.
• Orks (Sprues & Brews, 45 units): ρ −0.12, the same as round 5's.
• List share over every unit (5 or more winning lists could field it): ρ 0.129, the same as round 5's.
• With --armyfx these read −0.06, −0.10 and 0.150. That is the cuts moving under the Tyranids' Synapse, not any Space Marine or Ork rule.

The frozen-pool caveat from T-544 still applies, in full. The 1,109 non-Tyranid units in the pool are scored exactly as before the sweep. For them the tier test reads none of these:

• their units' own abilities;
• their heals, mortal wounds, auras or Damaged X;
• their army rules.

So the Tyranids' Damaged X fall (Trygon, Haruspex, Toxicrene) is still an upper bound. So are the heals' rise (Old One Eye) and the Venomthropes' own-ability rise: the units they're ranked against don't carry the same kinds.

Coverage effect sitewide

• Coverage data alone, V5 with today's data files against V5 with the pre-sweep ones:
  • 0 of 1,161 units change letter, and 0 composites move, in every army.
  • The same holds with --armyfx on both sides: the 27 other armies' army rules that T-542 wrote are calculator effects, not the pilot's aura/ld shape the tier test reads.
• Coverage switched in (own abilities and kinds, the tier test's Tyranid-only switches):
  • Tyranids: 5 letters change. Every other army: 0.
  • The biggest moves are the table above plus the kinds already reported. Nothing else moves 2 composite points.

What the tool needs before the sweep can count (not built here: tools/tiertest.js belongs to another builder, so these are named, not added):

1. Every army's own abilities and engine kinds in the pool's scoring (poolAt), not the Tyranids' alone.
2. --armyfx reading every army's rules through detfx.js armyEffects, in place of the pilot's shape, and the Synapse Strength counted once.
3. A combined own + kinds run and a dump of V5's letters for every unit as flags.

Verdict for Jordan

Does the Tyranid table look right to a Tyranid player now? Better at the edges, no better at the core. With every Tyranid ability the engine can play switched in, 18 of 44 letters match Auspex (17 before) and 30 are within one. The order agrees about as well as before (ρ 0.39), and agreement with what Tyranid players take dips slightly (0.24 → 0.23). The monsters read sensibly: the Damaged brackets pull the Trygon, Haruspex and Toxicrene down, the Venomthropes rise for their spores, and the Norns and Hive Guard stay where Auspex has them.

What's still unexplained is the same list as round 5:

• Biovores, Raveners, Hyperadapted Raveners and the Red Terror (S, S, S and A against C, D, C and D);
• the Neurolictor, Zoanthropes, the Swarmlord, the Hive Tyrant and the Maleceptor (S against B or C);
• the Tyrannocyte (C against S);
• Old One Eye (B against S, now higher through his heal).

None of these gaps is a dice effect we left out. They are of two kinds:

• Engine kinds not built yet, from the T-542 handoff:
  • fighting after being slain;
  • Fights First from an ability or enhancement;
  • Precision and Anti as mods;
  • damage re-rolls;
  • marks one unit puts on an enemy for others;
  • auras on other units;
  • resource pools;
  • keyword exclusions in when;
  • detachment rules' single re-rolls and ignore-hit-modifiers;
  • plus, for the Tyranids, the stratagems: Rapid Ingress and re-deploying.
• Constants Jordan must set:
  • what a Command point is worth (15 pts now);
  • a Battle-shocked enemy unit-round (10% of a median unit);
  • the share of attacks an enemy-side debuff touches (50%);
  • the worth of the detachment movement rules (tunnels, advance-and-charge, written under move).

These four are what separate the support monsters and the Ravener family from Auspex.

And one tool gap before any of it is fair: the other 27 armies are still scored without their coverage.