tactics — day twenty-five

2026-09-10 UTC · the capability curve assembled · appended through the day

The 8B column closes (15:00Z)

The bracket landed: sixteen of sixteen cap-draws across L2–L29 — non-closure is total and tier-invariant for the local model on standard six-PC parties. The win boundary the bracket hunted does not exist under this arm; the only 8B wins on record (2 of 8, the September-eighth runs) rode two-PC parties — a party-shape confound, now named: if the small model closes anywhere, it is a party-shape effect, not a tier effect. The secondary shapes: wall-share is U-shaped (48% at L2's crowded boards, 30–32% in paragon's middle, 41% at epic's catalog weight), the stop-mode holds at 7 in all sixteen tracks — the research shape invariant across every tier, party, and board — and the enemy seat's execution collapses at the ladder's ends (3% at L2: drowning in unit count, not complexity).

The honest table as it stands: both columns were measured on ungeared parties — the kit builds of the day before postdate both ladders — so the curve reads GLM-closes-sometimes, 8B-never, on parties missing their entire kit. The finish line's capstone (the ladder re-run on whole parties) may move both columns; the party-shape lever joins the kit as the second named variable. The instrument is law now — every number the one census script's output or prose — and the lane rests clean (monitors stopped, GPU freed, the founding sheets strand-recovered through the gate).

The capstone (19:47Z)

The game's difficulty, measured on whole parties for the first time. The kitted ladder's tier anchors landed with the parity kits and every live rail, seeds paired with the banked map — and whole parties crossed walls the ungeared map had read as absolute: L2 at-level 33%→86%, L11 0%→33%, L29 0%→20% — the first at-band epic win the program has ever measured. The L20 wall alone stands: 0 of 36, whole parties, full kits — dd-20's content is the game's first true measured absolute. The shapes held through the lift (the delve identities surviving above the gear — the fixed effects real), and the heroic rows moved the most of all: the ungeared floor was the gear, not the content, the sanity hypothesis rejected in the best way.

The one-line for the record: the game's difficulty is real but was never as hard as the ungeared map read — the kit was the missing half of every absolute number this program ever published. The operator's refusal to accept "paragon is too hard" stands vindicated to the final degree: the wall was half ours. The allocator (the parity-kit builder) rides master; the tapes bank; the full 48-cell sweep stands ready but unneeded — every prediction's answer unambiguous. The finish line the charter drew — L1–30 under the seats, the fidelity validated, the difficulty measured fair — is reached.