Tyranids, round 7: best-only grading (2026-10-02)

T-559 (Jordan, 2 Oct, via the design session). Round 6b's run (CP worth 30, every other setting unchanged) with three changes behind one switch, r7 (tools/tiertest.js, the round-6b command plus --r7json): best-only grading, Feed the Swarm's lend capped, the Tyrannocyte's one drop. Both columns come from the same run; round 6b's numbers reproduce exactly. A report: nothing here changes a published score or page.

What changed

1. Best-only grading (Jordan's rule). Score = the best allowed job's percentile + 0.25 × (p − 60) for each other allowed job above 60, at most 100 (V7C.BAR, V7C.BONUS). 92/10/10/10 scores 92; 80/80/80/80 scores 95; the Tyrannofex's 93/89/16/0 scores 93 + 7.25, capped at 100. A weak job never lowers a unit. The bonus counts only the jobs a unit may hold (no Anvil for OC 0, no Anvil or Runner for a joining Support character or a one-use unit), as the best job already did. Letters stay by rank quota split by job demand; since the scores bunch near the top, the cuts sit higher than round 6b's (B starts near 88).
2. Feed the Swarm (Assimilation Swarm). The healed squad keeps the regeneration's worth; the Harvester is lent at most what V2c credits an aura for one squad (V7C.REGEN_UNITS = 1: the median Leader's lift, 32%, on the median unit's 120 pts, 38 pts; one squad a phase). Pyrovores: lent 71 pts → 38, Banner 89 → 62, letter B → D. Ripper Swarms: 61 pts → 38, Banner 94 → 83, letter B → C. The healed squads' own Anvil doesn't count the heal yet.
3. The Tyrannocyte delivers once. Its transport credit in Banner was V2c's ride (12" on turns 1 and 2, then on foot), capped at 2; now its best cargo is set down once on turn 2, as Deep Strike sets a unit down (deliver's drop), so the credit is 0.32 (Banner 91 → 21). As a one-use unit (tierfx.json oneUse) it may hold neither Anvil nor Runner, and under best-only those no longer reach its score through the mean: score 86.1 → 42.1, letter A → D.

The yardsticks

Against Auspex's 44 graded Tyranids (ungraded units out of all three). Strict on our ruler: Auspex's letters re-cut at our quota (S 7%, A 15%, B 28%, C 30%, D the rest over its 44 units, each of its letters one tied block, a block that straddles a cut taking the letter most of it falls in), which makes its 13 S units A, its A units B, its B and C units C and its D units D; then the same letter as ours. Raw: the same letter on Auspex's own letters. Lenient: within one letter on raw letters. Shuffle baselines: the mean over 2,000 random permutations of our letters against theirs, what chance alone gets. ρ's SE is Bonett and Wright's.

The other two panels on the same three yardsticks (6b → 7):

The Tyranid table, all 44 units Auspex grades

Sorted by Auspex's tier, then name. Letters are global (the quota over every unit of every army). Score is round 7's; the four job percentiles are round 7's (each unit's best detachment). Lists: the Tyranid winning lists (of 23) that took the unit.

The 8 Tyranids Auspex doesn't grade

Two letters or more from Auspex on our ruler: 9 units

Gap: our round-7 letter against Auspex's re-cut at our quota (+ we rate it lower, − higher). Reasons from the round-4 diagnostics where they hold, our numbers otherwise.

Every army, round 6b → round 7

291 of 1161 letters move, 150 up and 141 down (armies with no letter moving left out). Outside the Tyranids only the grading changed: units strong in one job and weak in the rest rise, all-rounders whose mean carried them fall, and the quotas shift under both.