In plain English. Jordan's Votann notes say Thunderkyn and Hearthguard matter because nothing can hide from them: they come in by Deep Strike, Rapid Ingress, strategic reserves or a transport after the enemy has moved, and delete a key unit in one activation. Today's Burst assumes every gun reaches its best target, so it can't tell those units from ones that have to walk into range. We gave every unit a delivery score d (0 to 1): the share of the enemy's army it can put its burst on by our second turn, by its best way in. The new dial sureFire then adds a little Value to units whose delivery is surer than their army's middle unit, and takes a little from those whose delivery is worse.
- It helps on armies it was never tuned on: about +2.7 points held out (chosen on 10 armies, scored on the other 5, three splits, ± 0.6), on the exam and on consensus pairs alike; every split picks strength 0.5. Its placebo (the same credits handed to random units) is never picked and scores 0.
- It is not only "melee is overrated": a control that marks every melee unit down and every ranged one up loses (−0.5 held out); a control that compares each unit only with its army's units of the same kind (melee with melee, ranged with ranged) still gains +1.6.
- The prediction was wrong about who moves. Thunderkyn and Hearthguard rise only a place or two: under the rules every 24" gun can reach almost any target from strategic reserves, so most ranged units already score near 1. The gain comes from the other end: melee units that must walk (Aggressors, Genestealers, Eightbound, Votann's melee Steeljacks) fall, melee units that arrive where they choose (Exalted Eightbound, Terminators, the Deathleaper) rise, and so do long-ranged vehicles in armies whose middle unit walks (Votann's Hekaton and Sagitaur).
- Tyranids: +0.2 on the three 11th-edition lists at 0.5 (+0.7 on consensus pairs); −3.1 at strength 1.
- Without strategic reserves (a variant: every unit has them, so they might not separate units) the gain is larger, +3.5 held out, Tyranids +0.9.
Nothing is adopted; no ledger file, setting or value changed. For the next cascade: yes, at 0.25 and 0.5, with the no-reserves variant beside it.
Method
Track A (tools/ledger-benchmarks/2026-10-05-step34b-baseline.json, wCarry 0), snapshot d8b5a1bb14 (inputs hash c5e7d85b43f8). Each army is computed once and passed through tools/surefire.js's exam hook (T-685) at strengths 0.25, 0.5 and 1, re-sorted as the exam re-sorts, and scored on its 11th-edition lists: the exam's score (mean pairwise over the lists) and consensus pairs (two or more lists: the pairs every list grading both puts apart the same way; one list: its pairs). Pooled over the 15 armies; the Tyranids' three 11th-edition headline lists beside, never pooled or chosen on. Held out by army: popshare.js's three seeded splits of the 15 into 10 (choose the strength, 0 included) and 5 (score the gain over Track A); mean ± SE over the splits. Placebo: the same credits w × K × (d − d̃) dealt to the army's ranked units by a seeded permutation (seeds 1 to 3, the mean), so every move keeps its size and only which unit gets it is forgotten; a second placebo deals the offsets (d − d̃) instead, each unit keeping its own K. Intervals: tools/exam.js --post, 2,000 paired bootstrap draws.
Delivery d, by our turn 2, the best of a form's ways (tools/surefire.js's header has it in full). The enemy's units sit anywhere in its deployment zone and stay there (the charge model's assumption); our unit picks the one it wants, so d is the share of the field it can reach. Six maps by the field's event-layout shares (tools/deployments.json). No terrain or line of sight.
| Way | Ranged form (R: mean range of its guns, by how many it carries) | Melee form (a 9" charge in place of R) |
|---|---|---|
| Walk from the front of its zone | share within R + 2M | share within 2M + 9 |
| Scouts X" / Infiltrators | the same from X" further forward / from just over 8" off the enemy's zone | the same |
| Deep Strike, own Rapid Ingress, Aerial Seeding, turn-1 ingress | 1 (R > 8; Meteoric Descent R > 6) | 1 (Meteoric Descent 0: no charge that turn) |
| Strategic reserves (with or without the core Rapid Ingress) | share within R of the strip within 6" of an edge, outside the enemy's zone, more than 8" from the target | share within 9" of that strip |
| A transport in its army that fits it | share within 2M_t + 3 + R (Deep Strike or drop transports: 1) | share within M_t + 3 + M + 9 (out of a Deep Strike transport: 0) |
The charge's dice stay where they were, in the Kill line's lands: d is only how much of the field a unit can choose from. Teleport crest's Deep Strike (Hearthguard, Kâhl, Champion) is read here; the charge model doesn't read it for Votann. A tunnel isn't counted (as arrivalKill, design/ledger.md section 13).
The dial: Value′ = Value + w × K × (d − d̃), K the unit's Burst-line Value (its best form's Kill contribution to Value, which under Track A's burstKill is Burst), d̃ the median d of the army's ranked units. 0 is today to the bit (the run comes back untouched; tests/unit/surefire.test.js).
Results
Held out by army (mean ± SE over three splits; the strengths each split picks):
| Score | Real (picks) | ± SE | Placebo: credits at random (picks) | ± SE | Placebo: offsets at random | ± SE |
|---|---|---|---|---|---|---|
| Exam, as built | +2.72 (0.5, 0.5, 0.5) | 0.55 | +0.00 (0, 0, 0) | 0.00 | −0.24 (0, 0.5, 0) | 0.24 |
| Consensus pairs, as built | +2.74 (0.5, 0.5, 0.5) | 0.51 | +0.00 (0, 0, 0) | 0.00 | −0.23 (0, 0.5, 0) | 0.23 |
| Exam, without strategic reserves | +3.51 (0.5, 0.5, 0.5) | 0.47 | +0.00 (0, 0, 0) | 0.00 | −0.49 (0.25, 0.5, 0) | 0.42 |
| Consensus pairs, without strategic reserves | +3.42 (0.5, 0.5, 0.5) | 0.40 | +0.00 (0, 0, 0) | 0.00 | −0.42 (0, 0.5, 0) | 0.42 |
| Exam, control: melee against ranged only (d = 1 ranged, 0 melee) | −0.47 (0, 0.25, 0) | 0.47 | −0.28 | 0.18 | −0.18 | 0.18 |
| Exam, control: within mode (median of the same kind) | +1.64 (1, 0.5, 1) | 0.65 | +0.00 | 0.00 | −0.20 | 0.20 |
| Consensus pairs, control: within mode | +1.71 (1, 0.5, 1) | 0.68 | +0.00 | 0.00 | −0.19 | 0.19 |
In sample, each strength (pooled over the 15 armies; Tyranids beside):
| sureFire | Exam, 15 | Δ | Placebo (credits) Δ | Consensus pairs, 15 | Δ | Armies better | Tyranids, 3 lists | Δ | Tyranids, consensus | Δ |
|---|---|---|---|---|---|---|---|---|---|---|
| 0 (Track A) | 59.4 | 60.1 | 82.4 | 90.7 | ||||||
| 0.25 | 60.4 | +1.1 | −0.2 | 61.1 | +1.1 | 10/15 | 82.9 | +0.4 | 91.3 | +0.5 |
| 0.5 | 61.2 | +1.8 | −0.6 | 61.9 | +1.8 | 10/15 | 82.7 | +0.2 | 91.4 | +0.7 |
| 1 | 60.5 | +1.1 | −1.5 | 61.1 | +1.1 | 8/15 | 79.4 | −3.1 | 87.3 | −3.4 |
| no reserves, 0.25 | 61.1 | +1.7 | −0.2 | 61.7 | +1.6 | 12/15 | 83.7 | +1.3 | 92.1 | +1.4 |
| no reserves, 0.5 | 61.8 | +2.4 | −0.7 | 62.4 | +2.4 | 10/15 | 83.3 | +0.9 | 92.3 | +1.5 |
| no reserves, 1 | 61.5 | +2.1 | −1.7 | 62.0 | +1.9 | 9/15 | 77.9 | −4.5 | 85.9 | −4.8 |
| control mode, 0.5 | 58.8 | −0.6 | −0.4 | 59.5 | −0.5 | 6/15 | 82.2 | −0.3 | 90.9 | +0.2 |
| control within, 0.5 | 61.4 | +2.0 | −0.7 | 62.2 | +2.1 | 12/15 | 82.2 | −0.3 | 90.7 | +0.0 |
The exam's intervals at 0.5 (tools/exam.js, 2,000 paired draws): as built +1.8 [−0.0, +3.6] (armies resampled [+0.1, +3.5]), 10/15 armies better; without reserves +2.4 [+0.3, +4.6] ([+0.2, +4.6]); one placebo seed −0.1 [−2.2, +2.0], 5/15. Without Votann (14 armies) +1.3 [−0.4, +3.1]. Votann's 10th-edition lists beside: −0.7.
Per army at 0.5 (exam score; d's median and quartiles over the army's ranked units; how many are sure, d = 1):
| Army (lists; consensus pairs) | Track A | 0.5 | Δ | Placebo Δ | d median (quartiles); sure |
|---|---|---|---|---|---|
| Space Marines (2; 1515) | 60.1 | 59.8 | −0.3 | −0.4 | 1.00 (0.86–1.00); 55/84 |
| Adeptus Mechanicus (1; 322) | 71.4 | 70.8 | −0.6 | −0.9 | 1.00 (0.32–1.00); 19/34 |
| Agents of the Imperium (1; 223) | 56.5 | 52.5 | −4.0 | −1.0 | 0.82 (0.32–1.00); 8/29 |
| Astra Militarum (1; 1319) | 67.9 | 67.3 | −0.6 | −0.7 | 1.00 (0.36–1.00); 40/72 |
| Blood Angels (1; 124) | 61.3 | 67.7 | +6.5 | +1.6 | 0.32 (0.32–1.00); 9/20 |
| Dark Angels (1; 2325) | 53.6 | 56.4 | +2.8 | −0.4 | 1.00 (0.36–1.00); 56/93 |
| Death Guard (1; 476) | 55.0 | 56.9 | +1.9 | −1.9 | 1.00 (0.32–1.00); 21/36 |
| Drukhari (1; 207) | 58.0 | 54.6 | −3.4 | +2.6 | 0.81 (0.32–1.00); 14/37 |
| Emperor's Children (1; 191) | 69.6 | 70.7 | +1.0 | −3.5 | 0.91 (0.32–1.00); 10/23 |
| Genestealer Cults (2; 112) | 56.4 | 56.9 | +0.5 | +0.4 | 1.00 (1.00–1.00); 70/90 |
| Leagues of Votann (1; 105) | 61.0 | 69.5 | +8.6 | +2.9 | 0.96 (0.62–1.00); 6/22 |
| Orks (1; 606) | 63.5 | 65.0 | +1.5 | −3.2 | 0.36 (0.32–0.98); 11/55 |
| Space Wolves (1; 140) | 50.7 | 54.3 | +3.6 | −2.9 | 0.32 (0.32–0.36); 3/21 |
| Thousand Sons (1; 309) | 38.2 | 45.0 | +6.8 | +1.7 | 1.00 (0.40–1.00); 19/35 |
| World Eaters (3; 219) | 67.2 | 69.9 | +2.7 | −3.4 | 0.76 (0.32–1.00); 14/30 |
| Tyranids (3; 583; fitted on, not pooled) | 82.4 | 82.7 | +0.2 | −3.5 | 0.98 (0.32–1.00); 24/52 |
The named units, before and after at 0.5 (rank among the army's ranked units):
| Army | Unit | Reviewers (11e) | Mode, range | d (army median) | Best way | Rank Track A → 0.5 (no reserves) | Value Track A → 0.5 |
|---|---|---|---|---|---|---|---|
| Leagues of Votann | Brôkhyr Thunderkyn | A | ranged, 28" | 1.00 (0.96) | reserves 1.00 (Hekaton 0.93, walk 0.58) | 5 → 4 (3) of 22 | 1124 → 1145 |
| Leagues of Votann | Einhyr Hearthguard | A | ranged, 24" | 1.00 (0.96) | Deep Strike 1.00 (walk 0.40) | 15 → 13 (13) of 22 | 663 → 675 |
| Leagues of Votann | Cthonian Beserks | B | melee | 0.32 (0.96) | reserves 0.32 (walk 0.01) | 3 → 3 (4) of 22 | 1650 → 1154 |
| Space Marines | Eradicator Squad | A/B | ranged, 15" | 0.99 (1.00) | a transport 0.99 (walk 0.06) | 3 → 2 (2) of 84 | 1231 → 1223 |
| Space Marines | Aggressor Squad | S/C | melee | 0.54 (1.00) | a transport 0.54 | 7 → 22 (21) of 84 | 1072 → 844 |
| Space Marines | Inceptor Squad | A/B | ranged, 18" | 1.00 (1.00) | Deep Strike 1.00 | 5 → 4 (4) of 84 | 1125 → 1125 |
| Space Marines | Terminator Squad | A/D | melee | 1.00 (1.00) | Deep Strike 1.00 | 47 → 43 (38) of 84 | 695 → 695 |
| Tyranids | Lictor | S/A/S | melee | 0.95 (0.98) | Infiltrators 0.95 | 1 → 1 (1) of 52 | 4047 → 3996 |
| Tyranids | Deathleaper | A/S/A | melee | 0.95 (0.98) | Infiltrators 0.95 | 5 → 3 (3) of 52 | 3113 → 3075 |
| Tyranids | Zoanthropes | S/A/B | ranged, 24" | 1.00 (0.98) | Tyrannocyte 1.00 (walk 0.40) | 8 → 6 (5) of 52 | 2260 → 2285 |
| Tyranids | Genestealers | A/A/A | melee | 0.36 (0.98) | Scouts 0.36 | 30 → 34 (32) of 52 | 948 → 679 |
| World Eaters | Eightbound | B/A/S | melee | 0.45 (0.76) | Scouts 0.45 | 8 → 13 (13) of 30 | 988 → 843 |
Reading
- The dial works, but not where Jordan's notes point. The held-out gain beats both placebos and every split picks the same strength, so the ranking it adds carries something the reviewers share. Thunderkyn and Hearthguard barely move: d says almost every 24" gun can reach any target by turn 2 (strategic reserves arrive within reach of the whole enemy zone on every map), so being sure is the army's norm, not their edge. What they have that others lack is burst plus stratagems and detachment rules (T-713), and a protected ride (T-710).
- The movers are melee units. Walking melee is debited hard (Aggressors 7th → 22nd against S/C, Genestealers 30th → 34th against A/A/A, Eightbound 8th → 13th against B/A/S, all away from their reviewers; Votann's melee Steeljacks 9th → 15th toward their C). Melee that arrives where it likes is credited (Terminators 47th → 43rd toward A, Deathleaper 5th → 3rd). In Votann the gain is Hekaton 10th → 7th and Sagitaur 14th → 12th (both A) over the melee Steeljacks and the Iron-master (C) falling; Thunderkyn 5th → 4th and Hearthguard 15th → 13th are small parts of it. The mode control shows that melee against ranged alone loses; the within-mode control keeps +1.6, so the delivery part is real. Scouts melee (Genestealers, Eightbound) is the visible miss: d by turn 2 ignores that they reach everything by turn 3 and that the reviewers grade them as screens and scorers too.
- Strategic reserves blunt it. Without them d separates more (Votann's median 0.96 → 0.86, Thunderkyn 0.93 by Hekaton), and every score is higher: +3.5 held out, Tyranids +0.9. The reserves' cost (the 50% cap, a turn off the table, the CP of Rapid Ingress) isn't in d, which counts them as free; dropping them is the crude version of pricing that.
- Votann's +8.6 is noise-prone. One list of 105 pairs; one placebo seed gains +10.5 there too. The 14 armies without Votann still gain +1.3.
- Strength 1 overshoots (Tyranids −3.1, the placebo −1.5): a melee walker's whole Burst line shifted by up to 0.7 of itself is too much.
For the next cascade
Yes: sureFire at 0.25 and 0.5 (0 kept as the reference), and the no-reserves variant (reserves: 0) at the same two strengths beside it. Not 1. It passes the bar (held out above 0, beats its placebos) and stacks naturally with T-710 (transports by what only they carry), which credits the same delivery from the transport's side; measure them together before reading either alone.
Files
tools/surefire.js (the post module, --print for d per way, --measure for these tables; build/surefire.json), tests/unit/surefire.test.js. Reproduce: node --max-old-space-size=2500 tools/surefire.js --measure and node tools/exam.js tools/ledger-benchmarks/2026-10-05-step34b-baseline.json --post tools/surefire.js --post-args sureFire=0.5 --md.