The dictionary's §45 claim that phase transitions need infinite ramification is a literature theorem (Gefen–Mandelbrot–Aharony 1980–84), retrodicted, not found. The part that is not settled is §43's reading of it — that on a fractal the universality class is set by the room (lacunarity, how the holes are arranged) and not by the Hausdorff dimension. Two Sierpinski carpets of identical dimension log 8/log 3 but opposite lacunarity: do they percolate in the same class?

Sealed 2026-09-28 before any percolation on carpet U was computed (codifier). Known at sealing: the dictionary's §45 numbers (carpet H generations 3–4, crossing near 0.68; gasket crossing drifting toward 1); the Gefen–Mandelbrot–Aharony theorems on finitely vs infinitely ramified fractals (titles and abstracts only); that percolation thresholds are non-universal. Not read: any study of percolation or Ising exponents on carpets of equal dimension and different lacunarity — deliberately, so that the prediction is the lens's and not the literature's. Machine: the M5, after run 048.

Why these two carpets. Equal Hausdorff dimension is exact, not approximate: 64 of 81 at scale 9 is 8 of 9 at scale 3, squared. Generation 2 of carpet H is itself a 9×9 generator keeping 64 cells — with one 3×3 hole and eight singletons. Carpet U keeps the same count with seventeen singletons. Everything but the arrangement of the holes is held fixed; the matched sizes 81 and 729 compare the two at identical L.

What the outcomes mean. Confirmed: on fractals the room labels the class at fixed dimension, as §43 says, and "universality = symmetry + room" survives its first test where the room varies alone. Falsified: dimension labels the class; ramification decides only whether a class exists (the literature's statement) and §43's room reading is a restatement of dimension on regular lattices, where the two cannot be separated. Unanticipated in (0.03, 0.10): the classes differ weakly — crossover or slow convergence at these sizes — and the run repeats at carpet U generation 5 on BernoullisBook or a cluster.

Note under the seal, 2026-09-28 (codifier; before any carpet-U result was reported). The seal's description of carpet U is wrong in one respect and its operational definition stands. The explicit list of 17 removed cells is the definition; the gloss "no hole larger than one cell" is false: (2,4)–(1,4), (4,2)–(4,1), (6,4)–(7,4) and (4,6)–(4,7) are edge-adjacent, so U has four domino holes and nine singletons (found by Lab_Tree_Model's generation-2 checks). Dimension is unaffected (64 of 81 kept) and the contrast the run is built on survives — U's largest hole is 2 cells against H's 9-cell block — but U is less uniform than the seal says. The run proceeds on the sealed list. A D₄-symmetric 17-singleton set exists (Lab_Tree_Model found 12); if the chat runs one, it is an unsealed observation reported beside the result, not a substitute for it.

Scored 2026-09-28 (codifier; R27_tree_groups/carpet_perc/RESULTS.md, Lab_Tree_Model). Withdrawn as written: the sealed instrument cannot resolve the square lattice at its largest size, and the control was built to catch exactly that. The grid-free numbers are an unsealed observation.

What they show, held with conviction and marked unsealed: H and U — one 9-cell hole against four dominoes and nine singletons — percolate in the same class (1/ν 0.546, 0.526; D_f 1.841, 1.831). On this contrast the dimension, or something both carpets share, labels the class, and §43's "room" reading gets no support. But U′ does not fit. Lab_Tree_Model's all-singleton set, also exactly 64 of 81 per block and connected, gives p* = 0.866, 0.877, 0.881 — rising with generation like the gasket's — and 1/ν = 0.28. At equal dimension, rearranging seventeen single-cell holes so that they nearly wall off blocks along diagonals moves the threshold from 0.75 to near 0.88 and still drifting, and the fitted exponent to half the others'. That is the room acting — but on whether a transition exists at these sizes, the literature's statement (Gefen–Mandelbrot–Aharony), not on which class it belongs to. The honest one-line reading: ramification decides existence; at equal dimension, among carpets that do percolate, no class difference was seen.

What would settle it. A reseal with a p-grid of spacing 0.0005 in the crossing window (or the grid-free estimator named as the statistic), H and U as sealed, and U′ promoted from observation to arm with its own prediction: p*(L) → a limit below 1 or → 1. U′ is the arm that can still move the dictionary's §43.