Math Mapgenerated 2026-10-25 · a354234
PUBLIC BUILD
PG-01programmeopen2026-10-01Lines18Scaffolds4Runs36

PG-01 — The carpet's critical behaviour at reachable sizes is set by an exact object the one-block recursion truncates

object · the exact crossing recursion of a 9×9 pattern, on a boundary state of finite size (P-049's ladder; the one-bit map is rung 0)

Conjecture

On the census carpets, every quantity measured at L ≤ 6561 (exponent, threshold shape, composition) is determined by the pattern's exact crossing recursion carried on a finite boundary state, and the one-block map is its one-bit truncation; what the lattice does that the map does not comes from a finite number of boundary bits, and that number is small.

explainer · The Carpet Programme in Pictures · Carpets from the Ground Up

Distilled · 2026-10-25

the one rewritten section — what survived, what died, the object it rests on, what the next seal is for, and what would end the programme. Earlier distillations are in git, not in the file.

Rewritten 2026-10-25 (records audit R28_reading/audits/2026-10-25-records.md, fix 6; second reader's corrections from R28_reading/distilled/PG-01-2026-10-25.check.md applied) for the four runs scored since the 2026-10-24 text, which absorbed through 127: 131, 132, 135 and 136, all confirmed. Earlier versions are in git; the record is §2. It cites run files; no saved query set carries it yet. The 36 runs under PG-01 (075–136): confirmed 12, falsified 9, unanticipated 10, withdrawn 5. No run under PG-01 is sealed and open.

Survived. - The crossing polynomial as the ranker. Over 49 lattice members y₉ orders 1/ν at Spearman +0.948, ahead of C₁ (+0.883) and the generator's λ₂ (+0.820) (108). - In cut 2, the orbit where the wall enters the second ring. Midline entry sits above corner and off-axis entry on all 19 cut-2 patterns; of the six new, the nearest pair, #49 0.2922 ± 0.0035 and #42 0.2592 ± 0.0036, is eight errors apart, and y₉ ties #42 and #43 (107). Unanticipated only because the sealed line, 0.290, sat too close to #49. - T's fall in skewness, −0.091 ± 0.022 from 729 to 6561 on eleven files (092). A reading puts T's extra lean at 729 in the first crossing: the second crossing leans less there by −0.103 ± 0.023 and the same at 6561 (−0.0135 ± 0.0261) (109). - The algebra: all 13 cut-2 representatives are two mirror channels in parallel; 34 fibres are indecomposable (093). - Pendant parts separate twin pairs: four one-cell parts lower the threshold by 0.00193 ± 0.00020 on core A and 0.00260 ± 0.00020 on G (098). - The block transfer operator T_op, six scored facts. - What it sees. A single-cell flip changes T_op exactly when the cell lies on a route between two boundary cells: 129 visible, 6 invisible, none misclassified on nine generators the criterion was not formed on, the classes about 13 orders of magnitude apart (122). - Which flips it confuses. Two toggles give the same T_op when the kept-cell graph's network-reliability cores match with every boundary cell fixed (prune dead ends, merge series and parallel links, replace two-ended pockets by single links). On five blocks no operator run had built (20, 35, 36, 39, 44), all 490 comparisons fell on the predicted side: the five mirror-equal cells, (2,3) on each block, and the five equal pairs at 5.6e−17 to 2.2e−16, everything else at 3.03e−3 or more (132). That is the sufficient direction; necessity is unrefuted on 480 comparisons. - Where it is one-way. The one-way pairs are a cut U′_out × complement, count |U′_out|(511 − |U′_out|), with U′_out the right-column outputs a flood fill from the left cannot reach. The form held on all 62 scored census generators, though the sealed U failed on ten (127, falsified). The corrected flood-fill rule then held on 16 of 16 fresh blocks, including all eight with a kept column-7 cell cut off from the left, where the old rule misses every one (F04: 45,540 observed against the old rule's 64,860) (131). Where Q fails detailed balance is read from geometry alone, before any operator is built. - Reversibility, settled on the census. On the nine blocks with symmetric support (g57 to g63; fresh F14, F15), the curvature K = max|κ| of ln(Q[S,S′]/Q[S′,S]) over the faces of a cycle basis is 0.7046 (g57 to g60), 0.6857 (g61 to g63), 2.7835 (F14) and 2.8019 (F15), each far above the 1e−6 line (135). So E1, detailed balance in any diagonal metric, is absent on all 63 census generators: by one-way pairs on 56 (127) and by curvature on 7 (135). Beside the clause: 31 faces are exactly flat on every census host, all mapping to tree edges under the mirror σ; F14 and F15 break that mirror by 7.5e−3 and 1.8e−2 against 1e−14 on the census. - Along a row. Chaining T_op across one row of nine blocks falls short of the exact 9 × 81 strip by g_bf = 0.0132 (H), 0.0212 (U), 0.0142 (carpet U′), 0.0218 (pattern T) and 0.0341 (X) of a strip width (136). Clusters that cross a seam and come back barely move the strip's threshold. Rank 1 is within 0.1 of a width on all five (largest gap −0.051, carpet U′), rank 4 is exact to 1e−4, and ranks 1 and 2 agree exactly: with run 135's premise that rank 1 carries no curvature, the strip's threshold is set by one mode.

Died. Since 2026-10-24: none of the four new runs killed a clause. One guess and one framing died beside them. - The pre-registered guess that the strip sits at least 2 widths above the lattice at L = 81 on H, U and carpet U′ (2026-10-24, 17:15; P = 0.75) missed on carpet U′: 2.779 (H), 2.264 (U), 1.172 (carpet U′), converted by the codifier from RESULTS051.md:36 (136, reported). - "Rank-k T composed on the 9 × 9 arrangement at generation 2", named in the last text as the next seal, is not a defined calculation. T_op has no top or bottom sides, and along one row a cluster can cross a seam and come back, reaching cells T_op marks dry (exact for heights 1 and 2, not from height 3); the exact four-sided objects are out of reach (§2, 2026-10-24, the design reader, R28_reading/design-e4/REPORT.md). Its feasible form was run 136.

Dead before and unchanged: the census curvature at 2187 (104, withdrawn before any draw; 103 withdrawn on its seeds); the wall count as what sets the lean or the gap (109); the Chladni pair (108, reported); boundary isotropy as a D4 property, forced by the transpose, not hidden (119; §2, 2026-10-23); the blind chat's U from the right-column holes (127); the turn as a route (083), X as one map iterated (089), the coarse lag as a constant (085), three contacts per side (084, 094), the one-terminal table as sealed (097), one exponent per pattern (S2, struck after 096), the fan (§2, 2026-10-03).

The one object. The block transfer operator T_op at p*: 512 × 512, left-column subsets to right-column subsets, exact on all 63 generators. It is the left–right half of §1's boundary state, and the one-block map is its rank-1 part. Unscored readings: the rank-2 residual is 0.004–0.007 and the rank-4 residual 4e−4 to 8e−4 (docs/design/2026-10-19-transfer-operator-svd.md §7); carpet U′ (#12) is exactly rank 4 and pattern T (#26) exactly rank 10 (RESULTS-transfer-operator.md §2). Scored: what T_op sees (122), which toggles it confuses (132), where it runs one way (127, 131), that it fails detailed balance (E1) on every census generator (127, 135), and that along one row it is close to the strip's boundary state with one mode carrying the threshold (136). Not known: whether T_op reproduces the lattice. The strip is exact but is not the lattice: at L = 81 it sits 1.2 to 2.8 lattice widths above run 051's values (136, reported). So "that number is small" has one scored reading for it, in strip form (136), and three scored objects against it on the lattice (084, 094, 097).

What the next seal is for, and why it is next. E4 proper: a sampled block-row operator at L = 81, set against T_op at the renormalised p (§2, 2026-10-24, the composition question; TODO, E4 proper, "fire it after 136 has scored"). A tool chat builds that operator first. It is next because it is the first test of the conjecture's own claim on the lattice rather than on one block or one strip, and 136 has removed the come-back paths along a row as the obvious reason it would fail. The order of limits matters: a k that passes at 81 can fail at 6561, because w shrinks while a fixed truncation error does not (§2, same entry). Also owed on the operator, unsealed: the E3 premise (D4 commutes with T_op on one census host, and T_op's spectrum splits by D4 character; themes/symmetry.md §3, 2026-10-24, 20:45), and E2, which no host we have can force (themes/symmetry.md §3, 2026-10-22). TODO puts the Q3 seal before E4 proper (TODO:658) and the H8 reseal after the D4/I2 premise chat (TODO:651).

What would end the programme. Unchanged: no finite rung reproducing the carpet's numbers at L ≤ 6561, or a quantity computed with every link open ordering the patterns as well as the polynomial. Run 108 tried one of the second kind and it fell short: λ₂ of the kept-cell graph at +0.820 against y₉'s +0.948. The nearest test of the first is E4 proper.

Vocabulary. (1) Candidates (in a sealed clause of a scored run and here). Block transfer operator T_op, one-way pair and visible / invisible flip are in the §1 block (added 2026-10-25; T_op entered the glossary on 2026-10-24). New since the last text, each in a sealed clause and added to the §1 block on 2026-10-25 (docs/PROGRAMMES.md, Rules: a term enters the block the day it enters a sealed clause): network-reliability core (132's clause writes "the core"; the reader's word, R28_reading/formation/tblind/); K, the curvature max|κ| (135); strip, back-flow gap g_bf and strip width w_s (136; the clause writes "the strip's width"). Used, not yet in a clause: U′_out, the set (127, 131; question and statistic); face, cycle basis, curvature κ (135; question and statistic); back-flow as a word (136's question; the clause quantity is g_bf); rank-k truncation (136; question and reported). The 2026-10-24 candidates are ruled under Entry and ceiling below: λ₂ (108), second crossing t₂ and the crossing gap (109) and wedge (132's clause) enter the block; wall entry orbit is retired (107's clauses name patterns and the line 0.290, not the orbit). (2) Unused or circled: rung k and the ladder (idle since 094); the laggard gap (reported in 097, never a clause); the coarse lag (a quantity since S4 died); riser and fan (dead with L10); "composed on the arrangement" (circled for a week, then shown undefined; the strip replaced it). (3) Axes (docs/rung-vocabulary-v5.md): operator and one-way pair — no axis carries reversibility; the glossary's mathematical vocabulary does. Visible flip and the reliability core — the rigidity axis (§8.2) has isospectrality, not an invariant blind to a change in its object; the forgetting hierarchy's arrow 1 is nearest. Curvature on faces — no axis carries a cocycle; Stokes is a §8.1 bridge slot and a double-evaluation licence (reusable terms), not an axis coordinate. Back-flow — not placed. Entry orbit — the place axis (§4). λ₂, second crossing — not placed. (4) Enforced (struck on the word the clause was written in): 119's "removes D4" read as "breaks isotropy"; 108's "connected since the ring is kept", false on 19 of 63; 104's census "2187"; 127's "U from the right-column holes alone". None in 131–136. (5) Symbols and collisions. The glossary has already ruled on T and U′ (GLOSSARY, Collisions open: "Write the qualified form where two meet"; "Write carpet U′ / U′_out"); this text follows it. The rest are to resolve before the next seal. - T is pattern #26 and the operator (written T_op; "pattern T" where both meet). - U′ is pattern #12 (written carpet U′) and the unreachable-output set (written U′_out; 127, 131, 135). The §1 block's "the set U′" (2026-10-25) yields to the glossary's U′_out. - U is pattern #11 and the rule's set. - K is the kept-cell mask in every operator statistic (K = generator(idx), flipped K′) and the curvature max|κ| (135). - k is rank k (136) and rung k of the ladder. - σ is the singular values (σ₁…σ₄, 119, 132) and the mirror (135's "tree edges under σ"; Pσ in 119's reading). - F is the fresh masks F01–F16 (131, 135) and F₈/F₆₄ (095); PG-03's F̂ (137) is a third. - w is the lattice width and w_s the strip width (136); g_bf sits beside g₁, g₂. - λ is the one-block multiplier and the Laplacian's λ₂ (written λ₂(Lap) in the glossary). - Q is T_op on non-empty subsets (127), Q = 2.47 (086), PN-007's joint crossing probability and F₈/F₆₄ (095). - D_A is 098's p_med(#22) − p_med(#24) (beside D_G) and PG-03's comb statistic D_A (130). - gap is 109's crossing gap (t₂ − t)/w, L8's threshold gap, 094's gap′ and the laggard gap. - core is the ring core (098; §1, 2026-10-03), the network-reliability core (132) and PG-03's map core [R²(c), R(c)] (GLOSSARY, Collisions open, core). - generator is the census generator (used throughout this text) and the Synthesist's unit (GLOSSARY, Collisions open, generator). - orbit is 107's D4 orbit and PG-03's periodic orbit (GLOSSARY, Collisions open, orbit / class). - "rung" is the dictionary's coordinate, the ladder's rung k and the symmetry ladder's rung (119). (6) Drift: boundary state, k contact bits per side (2026-10-01) to 512 left-column subsets (2026-10-20); "disconnected", no crossing possible (design §7, 2026-10-20), isolated interior cells (108), interior components (127), not the separator (2026-10-24); λ₂, "second-smallest" against "smallest nonzero" (108); "hidden symmetry" to "forced" (2026-10-23); "composition" from the block on its own 9 × 9 arrangement (2026-10-20) to one row of nine blocks (136); "wall" from the lattice boundary p → 1 (§1, 2026-09-28) to the wall of holes, the C₁ cut (107, 108, 109). (7) Prose only: riser, fan, dial, the turn, carrier; straw (109), Chladni pair (108), "sees" (122, 132; the clause quantity is Δ_c), "Stokes" (135's from: line; the clause is K). Ruled to prose only today (Entry and ceiling): hump (with the turn), recorder, wall entry orbit. (8) Open claims in names: wall entry orbit (107's lead: the cut-2 form of the boundary state, untested across cuts; retired as a term, kept as the lead's name); carrier; network-reliability core (the sufficient direction is scored, necessity is not). U′_out under the corrected rule is no longer open (131). (9) Metaphors: Chladni figures (108; tested, wrong as written); the weakest link of the strongest wall, the straw (CASE-31, 109; tested, not held); "sees" (122; kept as shorthand for Δ); curvature and Stokes (135; imports a gauge picture in which a loop quantity is read through the faces it bounds; tested once, as Kolmogorov's criterion, and held); back-flow (136; imports a current; it is a threshold gap, kept as a name for g_bf). (10) Quantifiers: weakened — every generator connected to 44 of 63 (108), every symmetry-breaking flip visible to 152 of 266 (119), the rule's count on all 62 to 52 of 62 (127). Strengthened — the orbit separation, 13 to 19 (107); E1 absent on 56 of 63 to 63 of 63 (135); the corrected U′_out rule from 63 of 63 formed on the data to 16 of 16 out of sample (131). Entry and ceiling. The block was written on 2026-10-04; the 2026-10-24 text was the first rewrite to carry this paragraph, and it said that at the next rewrite each block term enters or is retired. This is that rewrite, and every term the 2026-10-24 text left open is ruled here (docs/PROGRAMMES.md, Distilled, fifth paragraph: "a term in the block through two rewrites enters or is retired, with the reason; there is no third option"). A term enters only if it names a computation (a function, a column or a query) and stands in a sealed clause of a scored run (Rules, A programme owns its vocabulary; the fold-in's part (1)); a term that meets both may still be retired, with the reason. Otherwise it is retired: from the block as a term, its computation kept under its own name, or to the prose-only list, which the same rule allows ("a prose-only term may stand in a lead or in Distilled and not in a clause"). Entry into the glossary takes the entry criterion's third condition as well (a second programme with the same meaning, or a panel reaching the object blind); it was checked term by term against PG-02, PG-03, PG-04, SHARED.md and CASE-29, and only census meets it. - Entered already: one-block map and block transfer operator (GLOSSARY, Promoted 2026-10-24). - Enter (block terms since 2026-10-04): - block seam — computed as strip.seam_B (the bond matrix across a seam, 2026-10-24); in the clauses of 095, 097 and 136. Third condition not met: PG-02's seam is another thing. - polynomial fibre — census068.json's crossing-polynomial coefficients, grouped; in 086's clause ("inside a fibre"). Third condition not met: PG-02's fibre is a level-map fibre, and CASE-29 gave the panel the twins, so the panel did not reach them blind. - twin pair — two members of one polynomial fibre (086's pair list from census068.json); in 086's clauses. Third condition not met. - ring core / pendant part / island — twins_core.py analyse; in the clauses of 098 ("the core", "pendant parts"), 086 ("island pairs") and 131 ("island masks"). Written "ring core" from today, beside the network-reliability core. Third condition not met: PG-03's core is the map's core. - census — the 63 generators of census068.json (census068.py, gen.py), 49 of them with a lattice 1/ν; in the clauses of 086 and 087. Third condition met: PG-03 uses it in the same sense (§1, Where it claims nothing and run 087's entry; its Distilled's end condition). It is proposed for the glossary. - Enter (2026-10-24 candidates): - λ₂ of the kept-cell graph's Laplacian — spectrum108.py; in 108's clause. Third condition not established: PG-02's λ₂ is a Schreier graph's, in prose only. - second crossing t₂, crossing gap (t₂ − t)/w — perc_cluster.c --second, analyze109.py (gap109.json); in 109's clauses. - wedge — the 15 cells (i, j), 0 ≤ i ≤ j ≤ 4 (vis122.py, tblind132.py); in 132's clause ("visible wedge cells"). - Retire, the computation kept under its own name: - wall — the block's sense, the boundary p → 1 (C₁, N₁), stands only in clauses of withdrawn runs (082, 103, 104). The one scored clause with the word (109, "minimum walls") means the wall of holes, a second sense. C₁ and N₁ stay in the symbol table. - mirror channel / coarse channel — neither stands in a clause. 093's clauses write the pivotal graph's components and "two bond-disjoint channels" (channels093.py); "coarse" is only in the questions of 083 and 085. Δ = t − t_coarse stays. - ring share — the column pattern.ring_share, in 086's question only. - step map — in 089's question only, and 089's reading died (X as one map iterated). - estimator — in no sealed clause. The procedure is OPS.md's (estimators; owner of local slope and pair slope, GLOSSARY, Promoted 2026-10-24), and the word belongs to docs/PROGRAMMES.md's methods programme. - ladder / rung k — it stands in 084's clause, but it has been idle since 094 and it collides three ways ((5)); rung084.json and rung094.json stay. - coarse lag (died with S4, run 085); laggard gap (never in a clause). - Retire to prose only: - hump / the turn — one block entry, in no sealed clause (077's question only). It describes data (g₁ by size), and it collides with PG-03's hump map. The turn was already prose only. - recorder — the name of an instrument (perc_blocks.c), not an object term; it stands in 097's falsifier as the instrument's name. - wall entry orbit — 107's lead's name, in its question only. - riser, fan, dial, carrier — prose only since 2026-10-04, unchanged. - Wait (first rewrite in the block): one-way pair and visible / invisible flip (in the block from 2026-10-25, so their ceiling counts from there); network-reliability core, K, strip, g_bf and w_s (added to the block today).

Lines

18 · compact density (§5) · each shows the display state of every record it names

L1

The hump. The threshold skewness builds, then turns on some carpets (073).

073 074 075 079 CASE-28 PN-005 081 083 089 092

L2

The ladder. P-049.

P-049 082

L3

Composition. 067, 070, 076, 077, 078 (closed for the one-block object: first order in δ, weighted by curvature mismatch, with a small direct residue of class).

067 070 076 077 078

L4

Bands and the census. 068, 071 (closed: (C₁, N₁) do not fix the exponent; y₉ plus a class-and-N₁ offset predicts it).

068 071 080

L5

The dual family. All 63 patterns keep the ring, so the shortest crossing is 8 for every one and run 075's sign law has never seen it vary.

075

L6

Blindness at p = 1. P-048's reading: anything evaluated at full occupation sees topology and transport, not criticality.

P-048

L7

The decisive size. L = 59049 is out of reach (100–180 GB).

100

L8

The threshold gap. The one-block fixed point lies above the lattice threshold on all 49 measured polynomials, by an amount (0.01–0.10) that does not decay with L except on the five wall-bound cut-2 patterns, where it shrinks, and is ordered by the ring share of the pivotal set (+0.83; +0.73 within cut 4 on the hold-out).

P-050 085

L9

Inside a polynomial fibre. The census measured one pattern per crossing polynomial; the 14 patterns that share one with another have never been measured.

P-050 086 093

L10

The fan. Within cut 4 the local slopes read as fanning around y₉, carried by five patterns whose stored drift reads positive, one beyond two errors.

P-050 087

L11

The shape as two optimizations. The threshold is a minimax; its lower tail is set by the cheapest path (the ring, ∝ L) and its upper tail by the cheapest blocking cut (∝ L^(ln C₁/ln 9)), so the standardized tails are predicted to be stretched exponentials with α₊/α₋ = ln C₁/ln 9 < 1: the upper tail heavier on every carpet, more so the smaller the cut — the sign and ordering of the map's g∞ against C₁ (−0.81) and the square's zero by self-duality.

091 090 054 055 062

L12

The seam's dimension. The ladder made continuous: keep seam links on a Cantor set of rows, so the contact set between blocks has a dimension between 0 (a rung) and 1 (the carpet).

P-051 094

L13

The gradient dial. An instrument, not an object: a chance of opening that varies across the sample puts a front where the threshold is, with a width set by the gradient.

P-052

L14

The natural basis. The chase split three ways and none became a line: the one-block map's own eigen-coordinate (run 089: the map has one, the lattice on X is not one map repeated); hidden symmetry (run 093: the block map commutes with nothing but itself on 34 of 49 polynomials); a second combinatorial invariant (P-050's inkling, never sealed).

094 089 093 P-050 084

L15

A dial for every block. Each block's own crossing point, and its neighbours' shift of it as a small table with one entry per pair of sides (along, across, corner), in units of the width.

P-053 CASE-28 097

2026-10-04: run 097, falsified.

L16

Invasion. Grow the crossing from the left side by always opening the easiest link: no dial, the same crossing point exactly, and two things the dial never gives — the forced links and the bursts between them.

P-054 P-048

L17

Bits left in the table. The conjecture's "finite number of boundary bits" counted from stored data: per column and size, bits above noise, bits explained by the record's best hold-out predictor, bits left, and which columns share their leftover.

P-055

L18

The block as an operator. Which symmetries the block transfer operator T_op (§1, Vocabulary) keeps, asked by the symmetry theme's forced-situation ladder (themes/symmetry.md §2.2).

119 122 126 127 131 132 135 136

Scaffolds

4 · full density · each names its removal run

what every run so far has assumed. A scaffold is not a line: a line is tested, a scaffold is relied on until a run takes it away. A scaffold whose removal clause names a method rather than a run says to seal rather than showing an id.

S1

Block independence. The one-block map: a block's crossing is independent of its neighbours'.

removal run 084 also names 058 P-049 094

2026-10-03: run 084, falsified — rung 3 on rows {1, 4, 7} recovers neither the carpet's order of T and X nor T's turn.

2026-10-03: run 094 confirmed — the reversal does not move between row sets (−0.134 against −0.131).

S2

Single-exponent scaling. w ∝ L^(−1/ν) with one exponent; every 1/ν in the census assumes it.

removal to seal also names P-050 056 066 073 087 057 P-028 P-048 079 096 080

2026-10-03: run 087's two-term fit was ill-posed as sealed (four parameters, three distinct sizes; with a common ω, sixteen on fifteen).

2026-10-04: run 096 confirmed it on fresh seeds: c₂ = +0.00394 ± 0.00108 (+3.66 errors), +4.61 pooled with 079.

S3

The class offset is a p = 1 property of the pattern. Run 071's frozen rule (y₉ plus a class-and- ln N₁ offset).

removal run 085 also names 071 080

2026-10-02: the condition is met as written (s₀ −0.59, ρ −0.55 against ln N₁'s −0.35, n = 28).

S4

The coarse lag is a lattice constant. Run 085's clause A, held as a scaffold only if it confirms: ⟨Δ⟩/w the same number on every pattern lets the coarse and lattice pictures be converted pattern-free.

removal to seal also names 085 084

2026-10-02: not adopted — run 085 falsified A (18 of 49 in the band).

Runs under the programme

36 · index density · in seal order (runs:)

075 076 077 078 079 080 081 082 083 084 085 086 087 089 090 091 092 093 094 095 096 097 098 103 104 107 108 109 119 122 126 127 131 132 135 136

Status, dated

69 entries · append-only, first sentence at this density

2026-09-30the hump programme. P-048, "The hump programme, 2026-09-30": the four original lines (074, 075, the mechanism, L = 59049) and the fold-into-unsealed-lines rule.
2026-09-30update 1 (run 074, unanticipated). P-048, "Hump programme, update 1".
2026-09-30update 2 (a record reading). P-048, "Hump programme, update 2".
2026-09-30update 3 (run 075, confirmed). P-048, "Hump programme, update 3".
2026-09-30record status. P-048, the "Record status" paragraphs dated 2026-09-30: run 068 (falsified, clause B), morning (run 072 confirmed), late morning (run 073 falsified), afternoon (run 070 unanticipated), evening (runs 076 falsified, 077 unanticipated), night (run 078 unanticipated), late (run 071 falsified).
2026-10-01record status. P-048, "Record status, 2026-10-01 (morning)": run 079 falsified.
2026-10-01run 080, unanticipated (L4). The census is complete: all 49 distinct crossing polynomials measured.
2026-10-01run 082, withdrawn (L2). Control (i) was a per-point 3σ test over ~900 grid points and fails a correct construction 90% of the time (OPS rule added).
2026-10-01run 083, unanticipated (L1). The Δ split.
2026-10-02run 085, falsified (L8). The coarse lag in width units is not a lattice constant (18 of 49 in the band; median −1.03, range −1.92 to −0.46; A falsified); the tie share is a class quantity (Spearman(s₀, C₁) = −0.72; B confirmed).
2026-10-02run 089, falsified (L1). X is not an iterated map at reachable sizes: g₈₁ and g₇₂₉ differ on the 729 law's 19 quantiles at rms z 3.09 (null 1.06), max |z| 4.45.
2026-10-02run 090, withdrawn (L11). Controls (i) and (iii) failed before any scored member was touched: the codifier read the two-column sample files' second column as t (it is perc.c's Beta-drawn p_c; the record is k/M), asked control (iii) for a #16 value computed without the run's own 10-sample rule, and pointed control (i) at a database column that is NULL at 729.
2026-10-02run 091, unanticipated (L11). Spearman(r, ln C₁/ln 9) = +0.475 over 47 members against a prediction of ≥ +0.5 (falsifier < +0.2).
2026-10-02run 081, unanticipated (L1). T's fall replicates on fresh seeds (d_T = −0.111 ± 0.031, −3.6 errors; A confirmed); X1's reverses (+0.117 ± 0.035, +3.3 errors; B falsified).
2026-10-02run 086, falsified (L9). Inside four polynomial fibres: the no-island pairs give Q = 2.47 (A unanticipated; |Δ1/ν| 0.003 and 0.008), and the island pairs split — #63 crosses earlier than #62 by 14.5 errors at 729, #23 and #24 are equal within 2 (B falsified as written).
2026-10-02run 093, confirmed (L9). On all 13 cut-2 fibres the pivotal bonds form two disjoint mirror channels with identical polynomials, so R = T(2 − T) = g∘T, g(x) = 1 − (1 − x)²; on #8 and #9 every chain of pivotal bonds is even, R = A(p²).
2026-10-02a next step withdrawn before sealing (L9). The entry above names the pivotal graph at generation 2 as the next seal.
2026-10-02the Ritt census, checked by the lab (L9). RITT-CENSUS.md (ac88731), an exact Kozen–Landau search over ℚ verified by recomposition: 34 fibres indecomposable; #8, #9 of degrees 28, 2; #12, #19, #21, #32 of degrees 2, deg T; the other nine cut-2 fibres of degrees 2, 2, n/4 with T = p⁸Q².
2026-10-03run 084, falsified (L2; scaffold S1's first removal). The first rung clear of the wall, k* = 3 on S₃ = {1, 4, 7}: X's local slope is below T's by 0.131 ± 0.007 (the carpet has it above by 0.18; A falsified) and T's skewness rises by +0.51 from 729 to 6561 (B falsified).
2026-10-03run 094 sealed (L2). Rung 3 on S₃′ = {2, 4, 6}, T and X, run 084's procedure with the set changed.
2026-10-03run 087, unanticipated (L10; scaffold S2). The five risers reseeded. #60's fresh drift is +0.0089 ± 0.0128 (A between 0 and +0.020) and the other four average +0.0048 ± 0.0066 (B between 0 and +0.010): neither falsified nor confirmed, and not resolved — the seal understated the drift error (0.010 against 0.013), so each margin was 1.5–1.6 errors from zero.
2026-10-03run 087, addendum (L10, L1). The lab's remaining reported items.
2026-10-03two zero-compute seals drafted and dissolved (S2, L10). Both were named as next steps in run 087's entry; asked the sharpness question before drafting, neither is a seal.
2026-10-03the fan fails its hold-out; number-sense screens built (L10; docs/DATABASE.md). From the database chat's report (commits c04e37c, f2ac744; readings of the record, nothing drawn).
2026-10-03two cases prepared (cases/). CASE-28's continuation packet for PN-005 (runs 081, 083, 089 as its lab report), and CASE-29, the twins: same polynomial, same rate, different threshold.
2026-10-03L15 added, with a look at stored samples (P-053; S2, L14). From run 083's samples at three sizes: the lag in widths is −0.59, −0.78, −0.81 on T and −0.53, −0.86, −1.02 on X; the share of blocks crossed when the lattice crosses is 0.66, 0.60, 0.57 on both, against 0.78 (T) and 0.74 (X) when the block graph crosses, and that last share is the same at every size, as it must be.
2026-10-03what the instruments compute and drop (OPS.md, "Save the raw output"); L15's symmetry reading. Asked by the owner after P-053.
2026-10-03L16 added after an audit (P-054; S2, P-048). The codifier claimed five things for invasion from memory; the owner asked for them to be tested.
2026-10-03the per-block recorder is built (84c0a70; L15, L14, CASE-28's continuation). perc_blocks.c, an instrument and not a run: twelve event times per block (along, across, mid versions, four corners, four same-side) and nine lattice fields per sample.
2026-10-03invasion seal drafted (L16; P-054). The run on T, X and H at 729 with a full side and a point source, clause A on H's period (a peak at 200 where T has a trough), clause B on the point source; draft at _incoming_inv/draft-invasion-bursts.md, to a second reader and a sighted pass before its commit.
2026-10-03run 095 sealed (L16; P-054). Invasion bursts on T, X and H at 729, full side and point source, 13,000 draws in the cloud.
2026-10-03L17 added (P-055; database build item 13). The bit inventory, to be read once and filed.
2026-10-03run 092, confirmed (S4's premise; L2, L10, CASE-28). One sampler on both machines: zero differing k in all nine comparisons, including a full 8,000-sample redraw of run 073's T4 file.
2026-10-03run 096 sealed (S2; H). Run 079's H curvature on fresh seeds: five sizes at a factor of 3, 48,000 samples each, one clause — the weighted quadratic coefficient of ln w in ln L positive by 2 of its errors (079: +0.00319 ± 0.00112; 051's 8,000-sample set +1.30 errors).
2026-10-03the form screen's second reading (L9; database item 14b). All nine [2, 2, n] cut-2 polynomials have inner quadratic p²: T = U(p²) exactly, with U normalised to U(0) = 0, U(1) = 1 and starting at p⁴.
2026-10-03the bit inventory, first reading (L17; database item 13, version 2). 33 columns: 4,950 bits above noise, 257 explained, 4,693 left.
2026-10-03run 095 ran (L16; cloud, 97 s), to be scored after its branch is merged. Control (i) exact (0 mismatches of 2,000).
2026-10-03the recorder timed at 6561 (L15). perc_blocks.c on T4, 20 samples at 2 threads beside run 094: 3.4 s wall and 4.8 thread-seconds a sample, 421 MB a thread on 341 MB shared; the first two samples byte-identical between 1 and 2 threads.
2026-10-03the bit inventory's rank reading (L17; database item 13c). With run 071's frozen rule as 1/ν's predictor, 1/ν's bits left fall from 3.4 to 2.6 per entry (88 of 190 explained); its residual shares only 0.1 bit with anything (with p_med, ρ ≈ −0.4 at every size, the sign of a threshold-side remainder).
2026-10-03PN-006 (CASE-28's continuation; L15, L14). The House refuted, on run 083's own numbers, the premise that the turn originates in what the recursion omits, dropped X1 from the phenomenon, and named the laggard gap (largest minus second-largest block time on the first completing block path, in width units) as the statistic.
2026-10-03run 095, unanticipated (L16; P-054, P-048). Both clauses unscorable on the seal's own bars (a ratio whose denominator touches zero in one block; a 0.05 bar on a thin point-source bin; a control over every bin), scored as written.
2026-10-03run 094, confirmed (S1; L12, L14, L15). Rung 3 on rows {2, 4, 6}: gap′ = −0.1339 ± 0.0062 against S₃'s −0.1313; each pattern's slopes move by less than their errors between row sets.
2026-10-03run 097 sealed (L15; PN-006). The recorder's first measurement: T and X at 729 and 6561.
2026-10-03crossed to PG-02 (readings). The carpet's seam-share curve (the blocks the recursion drops: 0.12, 0.18, 0.21 on T), run 094's caution, P-053's symmetry reading and the form screen are entered in PG-02's status as readings; PG-02's L1 is the other half of the comparison its Distilled asks for.
2026-10-04run 096, confirmed (S2; H). From the owner's base-3 question, through run 079's record read after the fact. All four controls pass.
2026-10-04S2 struck; PG-03 opened (the owner's decisions). S2 is removed as a scaffold on run 096's result, an outcome and not a silent keep (docs/PROGRAMMES.md).
2026-10-04run 097, falsified (L15; CASE-28, PN-006). From the owner's dial for every block (P-053). All five controls pass.
2026-10-06run 103 sealed: the census at 729 and 2187 in the cloud, scored on the curvature of ln w at (729, 2187, 6561) against the stored drift (lab/runs/103-census-curvature-at-2187.md). From: the census's falling pair slopes (the reason every quoted 1/ν carries its last local slopes) and PG-03's run 100, which found H's fall to be a power-law correction with ω near 0.3; the owner's choice of a cloud run on 2026-10-05.
2026-10-20runs 107 (unanticipated) and 108 (confirmed), both on L4 from the owner's reading of the census atlas (2026-10-19). 107, the cut-2 wall-entry orbit: C₁ = 2 is a hole wall across the interior, and the orbit through which it meets the second ring separates the lattice exponent — the six unmeasured patterns fell by orbit (midline #49 0.292, #54 0.302, #55 0.298; off-axis #42 0.259, #43 0.256, #46 0.256; nearest pair eight errors apart), but the seal's line 0.290 sat 0.013 under the lowest measured midline value and #49 landed inside its 2-error band: unanticipated as written, the owner's observation held, the codifier's margin did not.
2026-10-20CASE-31, PN-009 and run 109 (unanticipated), on L1 (the hump). The case asked whether the threshold as the weakest link of the strongest wall (bottleneck duality) makes the number of minimum walls set the lean.
2026-10-20the transfer operator and the wedge map, built (no seal; docs/design/2026-10-19-transfer-operator-svd.md §7). The block as a 512 × 512 operator from left subsets to right subsets, exact at p* for all 63 generators: σ₂/σ₁ ≈ 0.1 everywhere, rank-2 residual under 1 %, rank-4 under 0.1 %, U′ exactly rank 4 (the hole wall factors the block), T exactly rank 10.
2026-10-23run 119 sealed (the symmetry theme's E3: isotropy broken by one cell; M5, ~3.5 h). On the 19 cut-2 patterns, each of the wedge's 15 positions toggled as a single cell (not its orbit — the wedge map's flips stay symmetric, a premise caught while writing this seal, OPS instance 10); the per-side gap max|σ(T_LR) − σ(T_TB)|/σ₁ against the premises run's zero 5.6e−16.
2026-10-23run 119, falsified (M5, 589 builds; blind-scored). All 76 diagonal flips leave σ(T_LR) = σ(T_TB) to 1e−15: transposition alone keeps the per-side reading isotropic — a hidden symmetry, the theme's first on the carpets.
2026-10-23run 119 read again (codifier, R27_tree_groups/carpet_perc/iso119_check.py; guesses committed first). The isotropy is forced, not hidden.
2026-10-23run 122 sealed (what the block operator sees; M5, one worker, about 6–14 h). The criterion: a single-cell flip changes T iff the cell lies on a simple path between two cells of the left and right columns.
2026-10-23run 126 sealed (one-way pairs from geometry; E1, step 1). The owner's blind chat, given only a sanitised description of the operator, derived a rule: the number of one-way pairs is |U|·(511 − |U|), set by the right-column holes alone; interior holes add none.
2026-10-23run 126 withdrawn before compute; resealed as run 127. The lab chat's Phase 0 found control (i) mis-specified: the noise came from run 119's flip builds, while the p = 0.5 operators reach 2.26e−11 (OPS 17).
2026-10-23run 122, confirmed (blind-scored). 129 / 0 / 0 / 6 on nine generators run 119 never built: T sees a cell exactly when the cell lies on a route between two boundary cells, with 13 orders of magnitude between the classes.
2026-10-24run 127, falsified (blind-scored). The blind chat's one-way-pair rule holds in form on every generator: the one-way pairs are a cut U′ × complement, count |U′|(511 − |U′|).
2026-10-24what U′ is (data reader on oneway127.json; R28_reading/formation/uprime127/). On the ten, U′ is exactly the set of right-column outputs no input can produce, a superset of the rule's U. The separator is a kept column-7 cell cut off from the left (an island beside the right column): true for exactly these ten.
2026-10-24run 131 sealed (one-way pairs on fresh blocks). The data reader's corrected rule (U′ by flood fill from the left over the block plus the chosen right cells) fits run 127's 63 counts, but it was formed on them.
2026-10-24formation on T's blindness, then run 132 sealed. A fresh reader, fenced to the blocks already built, explained both of run 122's unexplained observations by a network-reliability core: prune to edges on routes, merge series and parallel edges, and replace two-ended terminal-free pockets by one edge.
2026-10-24run 131, confirmed (blind-scored). The corrected one-way rule holds on 16 of 16 fresh blocks, including all eight with column-7 islands, where the old rule misses every one.
2026-10-24run 135 sealed (curvature on the faces; E1 step 2). One-way pairs leave E1 open on nine blocks (census 57 to 63 and fresh F14, F15).
2026-10-24the composition question, sized (design reader, R28_reading/design-e4/REPORT.md). "Rank-k T composed on the 9 × 9 arrangement" is not a defined calculation.
2026-10-24run 132, confirmed (blind-scored). The network-reliability core predicts, from the graph alone, which single-cell toggles give the same block operator.
2026-10-24run 136 sealed (the strip and its back-flow). This is the feasible form of the composition question.
2026-10-24run 135, confirmed (blind-scored). None of the nine blocks with symmetric support is reversible in any diagonal metric.
2026-10-24run 136, confirmed (blind-scored). Chaining the left-to-right operator T falls short of the exact 9 × 81 strip by g_bf = 0.0132 (H), 0.0212 (U), 0.0142 (U′), 0.0218 (T) and 0.0341 (X) of a strip width, all inside [0, 0.25].

Written back

12 · append-only

what this programme has changed outside itself. Empty after ten runs means it is chasing leads (docs/PROGRAMMES.md).

  • OPS.md · 214193a · Check a seal's seeds against its ancestors — from run 074's bootstrap range lying inside run 073's.
  • OPS.md · b2285b7 · Every stacked or composite lattice reports its build-order gap — from run 070's build-order gap (0.0069 on X,T, 0.0227 on U′,T).
  • OPS.md, ec7a7ad (2026-09-30): a falsifier states what a lone surviving clause decides — from run 070's withdrawn pair, under L3.
  • OPS.md, 4fd515a (2026-10-01): a control over many points is a family-wise test — from run 082's withdrawal, under L2.
  • OPS.md, 90328a5 and 84ecf6c (2026-10-02): every number a control cites is reproduced on disk before the seal commits, on control members only — from run 090's withdrawal, under L11.
  • OPS.md, cfb226d (2026-10-02): a scoring reader before a score commits — from run 089's score, under L1.
  • OPS.md, 6ec897f (2026-10-02): a sharpness reader before the second reader — from run 091, under L11.
  • OPS.md, 29bd24f (2026-10-02): new statistic names go into statistics.json, never into a run's rows list — from run 093's health-check failure, under L9.
  • MANUAL.md, b66f582 (2026-10-02): the census fibres are classes of the expanded polynomial, and sympy's decompose is not evidence — from run 093, under L9.
  • OPS.md, 65441a1 (2026-10-03): a script that is running is never edited in place — from run 096's stray second draw of H8, under S2.
  • MANUAL.md, dd062e7 (2026-10-03): regenerate the manifest after a merge commit, not during the conflict — from the merge of run 095's two burst files, under L16.
  • panels/PN-007-CASE-29.md (2026-10-03; recorded here 2026-10-04 as docs/PROGRAMMES.md asks): a House that reconstructs the programme's object blind. Reading a case with no programme word, PN-007's House put its weight on the boundary state — "R is one marginal of the block's four-side boundary-partition polynomial, and the gluing consumes the rest" — which is this programme's object:; its chosen entry (P_X) was wrong. PN-004's House had said the same a day before this programme existed. Scored object reached in docs/PROGRAMME-AUDIT.md §4.3; the aliases are in §1's vocabulary block.

Source programmes/PG-01-exact-object-finite-boundary.md. Every count on this page comes from one read of the record.

generated 2026-10-25nothing on this page is maintained by hand