Panels from node tools/scratch/tyr-full-panels.js (the source table reports/tiers/sources/2026-10-03-tier3-other-factions.csv); reviewer agreement from node tools/scratch/tyr-full-reviewers.js; the ledger from node tools/ledgerstep.js --base tools/ledger-benchmarks/2026-10-03-step6-baseline.json --lists …; the consensus check from node tools/scratch/diagnose-full.js --quick beside --summaries --quick (tools/diagnose.js unchanged). Step 24 is proposed: the yardstick (ledgerstep.js ARMIES, diagnose.js's five lists) is unchanged until the supervisor adopts it.
Adopted (4 Oct, kept as a yardstick step): the seven lists here are the headline everywhere (ledgerstep.js ARMIES, diagnose.js LISTS, ledgerfit.js PANELS, progress.js SEVEN, roles.js, the ledger page's first columns). The wrapper tools/scratch/diagnose-full.js is gone: node tools/diagnose.js --quick is its run, --summaries --quick the summaries'. The new reference numbers are in step 24's row of reports/tiers/ledger-steps.md.
The panels
| List | Summary (superseded, kept for the frozen inputs) | Transcription | Key | Matched | Letters scored by the ledger |
|---|---|---|---|---|---|
| Auspex Tactics (12 Sep) | tyranids-auspex.json, 44 units | 2026-09-12-tyranids-auspex-full.json | auspexFull | 52 of 52 | 52 |
| HivemindHobbies (21 Jun) | 2026-10-03-tyranids-hivemind.json, 49 | 2026-06-21-tyranids-hivemind-full.json | hivemindFull | 52 of 52 (Sky-slasher Swarms, Legends, unmatched) | 48 (4 dropped for points moves) |
| Into the Hive Mind (27 Jul) | 2026-10-03-tyranids-second.json, 44 | 2026-07-27-tyranids-second-full.json | secondFull | 52 of 52 | 48 (4 dropped for points moves) |
Letters as the summaries had them: Auspex's six bands fold to five (both halves of Tier 3 are B; o keeps the six), HivemindHobbies' and Into the Hive Mind's five rows are S to D. Every panel keeps the grade as heard (heard), the name as heard where it differs (match) and our own paraphrase of the note in why.
Patch alignment: Auspex's video follows the 5 Sep update, the last Tyranid points move, so nothing is dropped. The two summer lists keep the HivemindHobbies file's stand-in commit (80568cc, 2 Jul) and its four drops (Pyrovores, Neurotyrant, Tyrannocyte, Toxicrene); Into the Hive Mind's summary dropped three, and Pyrovores, now graded, takes the fourth. Its video is 27 Jul, after the July dataslate; our patch history starts at 5 Sep, so some of those moves may predate it (the reviewer mentions the Neurotyrant's rise): a drop then only costs a unit. The ledger's comparators now respect every panel's patchDrop (tools/ledger-merge.js; before, only the other armies' did, and no Tyranid or Marine comparator had one).
What changed, list by list
Auspex Tactics. One shared letter differs: the Tyranid Prime with Lash Whip, C → S (graded leading Hormagaunts, as a pair). Eight units the summary left out: the Lictor S, the Broodlord and Genestealers A, Gargoyles and Termagants B, Barbgaunts, the Toxicrene and the Winged Tyranid Prime C. The summary's Barbed Hierodule (unmatched) isn't in the transcription.
HivemindHobbies. All 49 shared letters the same. Three added, all D: the Harpy, the Hive Crone, Mucolid Spores.
Into the Hive Mind. Eight of 44 shared letters differ (summary → transcription): Deathleaper B → A, Neurolictor B → S, Screamer-killer C → B, Hive Guard B → A, Hyperadapted Raveners B → A, Tyranid Prime with Lash Whip B → S, melee Warriors A → B, ranged Warriors B → C. Four of them (Neurolictor, Lash Whip Prime, Screamer-killer, Hyperadapted Raveners) were silent placements read from a low-resolution thumbnail of the final board (their why says so). Eight added: Old One Eye, the Maleceptor and Pyrovores A (Pyrovores dropped for points), the Trygon B, the Hierophant and Termagants C, the Harpy and the Hive Crone D.
The ledger against each list (step-6 baseline)
| List | n | Pairwise (pairs) | 2+ tiers apart (pairs) | ρ |
|---|---|---|---|---|
| Auspex summary | 44 | 80% (742) | 87% (480) | 0.662 |
| Auspex transcription | 52 | 81% (1,044) | 88% (642) | 0.688 |
| HivemindHobbies summary | 45 | 79% (743) | 91% (311) | 0.622 |
| HivemindHobbies transcription | 48 | 80% (869) | 91% (410) | 0.664 |
| Into the Hive Mind summary | 41 | 77% (630) | 88% (282) | 0.587 |
| Into the Hive Mind transcription | 48 | 79% (889) | 88% (454) | 0.644 |
| Maelstrom (unchanged) | 44 | 71% (744) | 81% (368) | 0.571 |
| Astrategas (unchanged) | 48 | 66% (911) | 70% (555) | 0.381 |
| Space Marines Auspex full (unchanged) | 84 | 63% (2,440) | 74% (859) | 0.284 |
| Space Marines Tobias (unchanged) | 79 | 55% (2,375) | 57% (1,180) | 0.197 |
Mean over the seven lists: old (summaries) ρ 0.472, pairwise 70.1%; new (transcriptions) ρ 0.490, pairwise 70.8%.
On the summary's own units the transcriptions give ρ 0.697 (Auspex, 44), 0.622 (HivemindHobbies, 45: the letters are the same) and 0.607 (Into the Hive Mind, 41). So, unlike the Marines' step 11, the summaries didn't flatter the ledger: their letter changes help a little, and the units they never named help more.
The reviewers against each other (the letters the ledger scores)
ρ / pairwise (pairs) / n shared / same letter.
| Auspex | HivemindHobbies | Into the Hive Mind | Maelstrom | Astrategas | |
|---|---|---|---|---|---|
| Auspex | – | 0.73 → 0.83 · 90% → 95% · n 38 → 48 | 0.67 → 0.81 · 86% → 94% · n 35 → 48 | 0.84 → 0.82 | 0.75 → 0.64 |
| HivemindHobbies | – | 0.54 → 0.76 · 80% → 92% · n 40 → 48 | 0.82 → 0.85 | 0.58 → 0.64 | |
| Into the Hive Mind | – | 0.73 → 0.77 | 0.52 → 0.49 |
Leave-one-out ceilings (each list against the mean letter of the other four):
| List | ρ, summaries | ρ, transcriptions | Pairwise, summaries → transcriptions |
|---|---|---|---|
| Auspex | 0.86 (n 44) | 0.86 (n 52) | 92% → 92% |
| HivemindHobbies | 0.77 (45) | 0.86 (48) | 88% → 93% |
| Into the Hive Mind | 0.68 (41) | 0.79 (48) | 82% → 88% |
| Maelstrom | 0.92 (48) | 0.93 (48) | 96% → 97% |
| Astrategas | 0.79 (48) | 0.71 (48) | 87% → 84% |
| Mean | 0.80 | 0.83 |
The three 11th-edition lists now agree with each other as well as any pair does; Astrategas, the 10th-edition list, is the one left apart (0.49 to 0.64 against the three).
The consensus (tools/diagnose.js, --quick: 2 repeats, 1 scramble)
The consensus is each unit's mean letter over the five headline lists. With the transcriptions in place of the summaries:
| Summaries | Transcriptions | |
|---|---|---|
| Ceiling (each list against the mean of the other four, pairwise) | 88.5% | 90.4% |
| (a) the ledger as it is, in sample, vs the consensus | 73.3% | 75.3% |
| (a) held out, vs the consensus (± SE) | 73.3% ± 0.2 | 74.6% ± 2.2 |
| (a) held out, vs the lists | 74.4% ± 1.6 | 75.0% ± 2.3 |
| (c) the best flexible model, held out vs the consensus | 74.4% ± 0.4 | 72.5% ± 0.8 |
Per list, held out, (a): Auspex 81.0% → 80.2%, HivemindHobbies 78.4% → 79.5%, Into the Hive Mind 76.6% → 79.1%, Maelstrom and Astrategas unchanged (71.8%, 64.5%). A quick run, so a point or two either way is noise; the ceiling moves 1.9 points and the ledger about as much, so the gap to the ceiling stays where it was (about 15 points).