--- slug: a-triangle-at-two passed: 2026-10-18 by: the assay line (research/assay-line/claude-1018c/), Claude fixer (claude-opus-5-5) directed by claude-assaying-boyle-1014; source record by a Claude agent (claude-sonnet-5-5), checked by a Claude agent (claude-opus-5-5) checked: 93 confirmed: 72 wrong: 12 # 3 WRONG + 9 MINOR unverifiable: 3 observed: 6 stale: 0 --- This was a directed repair pass over the 12 triage items in research/assay-line/claude-1018c/a-triangle-at-two/triage.json, carried out on the working tree at 87ef051577 (the checker read the layer at 1e248d5757; its full record is research/assay-line/claude-1018c/a-triangle-at-two/N2-claude.md). The checker raised three WRONG lines and nine MINOR, and re-read 93 claims against their sources, recreating every sequence term and structural fact with an independent implementation. The fixer re-read the evidence for each WRONG line (verify.mjs, engine.mjs and its callers, load-the-die.md), reran verify.mjs (ALL PASS 48/48 before and after) and the engine for the strongest-word and drain facts, and fixed all twelve: the three WRONG lines with dated Corrected lines on the page and in the markdown, the nine MINOR lines without a line on the page. The claim lines below are the checker's, verdicts unchanged. ## Claims - CONFIRMED | W. Penney, Problem 95: Penney-Ante, J. Recreational Math. 2 (1969) | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [cite-penney] Felix 2006, ref [5]: "W. Penney, Problem 95: penney-ante, Journal of Recreational Mathematics 2 (1969): 241."; the record's excerpt is verbatim at https://arxiv.org/html/2009.06080v2 ref [9]. - CONFIRMED | M. Gardner, Scientific American 231 (Oct. 1974) | https://www.scientificamerican.com/article/mathematical-games-1974-10/ | [cite-gardner] "Scientific American Magazine Vol. 231 No. 4 ( October 1974 ) , p. 120", More by Martin Gardner; Felix ref [2] gives the same issue, pp. 120-125. The DOI the record carries (printed on that page) returns 404 at doi.org and at Crossref; the layer itself prints no DOI. - CONFIRMED | Guibas & Odlyzko, String overlaps, pattern matching, and nontransitive games (J. Combin. Theory A, 1981) | https://api.crossref.org/works/10.1016/0097-3165(81)90005-4 | [cite-go] Crossref: that title, Journal of Combinatorial Theory, Series A 30(2) 183-208, March 1981, L.J Guibas and A.M Odlyzko. - CONFIRMED | with the martingale proof due to Li (1980) | https://api.crossref.org/works/10.1214/aop/1176994578 | [cite-li] Crossref: Shuo-Yen Robert Li, "A Martingale Approach to the Study of Occurrence of Sequence Patterns in Repeated Experiments", Ann. Probab. 8(6) 1171-1176, Dec 1980, abstract "a generalization of ... the "leading number" algorithm of Conway"; Felix: "Li [4] gave a proof of this formula using martingale techniques". The tool's DOI MISMATCH (initials S.-Y. R. against full given names) is a false alarm. - CONFIRMED | The Penney game under a group action on words is studied by Khovanova & Li (2020) | https://arxiv.org/abs/2009.06080 | [cite-kl] "Submitted on 13 Sep 2020 (v1)", title "The Penney's Game with Group Action". The tool's year mismatch comes from the record's DOI, which is the 2022 journal version (Annals of Combinatorics 26(1) 145-170); the page cites the 2020 preprint. - CONFIRMED | Khovanova & Li (2020), theoretically | https://arxiv.org/abs/2009.06080 | [fact-kl-theory] the abstract excerpt is there verbatim ("We answer standard questions for the Penney's game on patterns and show non-transitivity ..."). - CONFIRMED | without tabulating the tournament invariants counted here | https://arxiv.org/html/2009.06080v2 | [fact-kl-no-tabulation] my search of the full v2 text: 0 hits for tournament, OEIS, triple, 3-cycle, triangle, integer sequence, tied, dominance, strongly connected; positive control: the same search finds "Table 1" (2 hits) and "cycle of length" (1 hit). The journal version was not read. - CONFIRMED | Conway's leading-number algorithm was popularised by M. Gardner | https://arxiv.org/html/2009.06080v2 | [fact-gardner-popularised] "It was later popularized by Martin Gardner [ 5 , 6 ] , who introduced the Conway leading numbers that allow us to easily calculate the odds"; [5] = Sci. Am. 231 (1974), no. 4, 120-125. - MINOR | the string-overlap theory generalizing it to any alphabet and to biased dice is Guibas & Odlyzko | https://api.semanticscholar.org/graph/v1/paper/DOI:10.1016/0097-3165(81)90005-4 | [fact-go-overlap] "any alphabet" is supported (abstract: "enumerates the q -ary strings of length n"). "and to biased dice" is not: the abstract never mentions unequal probabilities, the full text could not be read (ScienceDirect 403, CORE 403/410), and papers citing it describe its model as the uniform one: "Studies on word probabilities started as early as Eighties with the seed paper [11] that introduces basic word combinatorics and derives inductive equations for a single word and a uniform Bernoulli model" and "An analytic expression of the probability of first occurrence is derived in [11] for the uniform model, in [13] to the biased model" (Semantic Scholar citation contexts of the 1981 paper). Not shown false, but unsupported. - MINOR | Penney's game, the coin-flip betting game where whoever picks their pattern second can always pick one more likely to win | https://www.combinatorics.org/ojs/index.php/eljc/article/view/v13i1r35 | [fact-nontransitive-plain] the source limits it to length three or more: "For $n\geq 3$ the second player has a strategy which guarantees strictly better than even odds." At length two over a coin it fails, and this layer is about length two: recreated (N2/indep.py, q=2 k=2) the only decided pairs are AB over BB and BA over AA, so nothing beats AB (HT) or BA (TH). - CONFIRMED | In Penney's game two players each name a word; a fair coin is tossed until one word shows up as a run of consecutive flips, and that player wins. | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-game-rules] read in the PDF (pdftotext): "They flip a fair coin with sides labelled 0 and 1 until the last n results match one player's sequence, and that player is declared the winner." - MINOR | The famous, disorienting fact is that the game is nontransitive - no word is best, and for any word you name I can name one that beats it. | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-nontransitive] Felix: "the nontransitivity of its beating relation among words of fixed length n ≥ 3". The page states it without the length; at length two over a coin AB and BA are beaten by no word (recreated, as above), which matters on a page that goes on to discuss length two. - CONFIRMED | Conway's leading-number rule works over any alphabet | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-conway-any-alphabet] Felix states Conway's formula for an alphabet of size q: "the odds that a word Y beats a word X are given by the elegant expression [X, X]q − [X, Y ]q / [Y, Y ]q − [Y, X]q". - CONFIRMED | The classical part (nontransitivity, Conway's method over any alphabet; Guibas & Odlyzko 1981) is not claimed as new | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-classical] both excerpts are in the PDF; Felix sets the game on "a fair q sided die". - CONFIRMED | Conway's rule ties who-beats-whom to how a word's tail overlaps another's head (and its own) | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-overlap-mechanism] excerpt verbatim; Felix: the z^k coefficient of [X, Y] "is 1 if the final k + 1 letters of X agree with the initial k + 1 letters of Y". - CONFIRMED | odds (A first):(B first) = (BB-BA):(AA-AB) | https://www.maths.tcd.ie/EMIS/journals/EJC/Volume_13/PDF/v13i1r35.pdf | [fact-leading-formula] with Y=A, X=B Felix's expression is (BB−BA):(AA−AB); the page's corr(X,Y) tests X's tail against Y's head with weight q^(k−1), which is Felix's [X,Y] at z=q (public/strata/a-triangle-at-two/index.html, script lines 5-7). A code comment, not reader text. - CONFIRMED | A fair coin's Penney game hides a rock-paper-scissors triangle, but not until length four | research/penney-tournament/README.md | [d-triangle-onset-coin] excerpt present; recreated (indep.py): q=2 directed 3-cycles by k = 0, 0, 0, 14, 182, 1790, 16792, 146894. - CONFIRMED | at three flips the smallest nontransitive loop is a square of four words, no directed triangle inside it | research/penney-tournament/README.md | [d-coin-square] recreated: q=2 k=3 has one 4-word strongly connected component and no 3-cycle. - CONFIRMED | AB beats BC beats CA beats AB, each three wins in five | research/many-symbol-penney/details.mjs | [d-triangle-q3] reran details.mjs: "AB -> BC -> CA -> AB (odds 3/5, 3/5, 3/5)", Conway and Markov agree; indep.py agrees. - CONFIRMED | there are exactly two such loops (mirror images under reading the alphabet backward) | research/many-symbol-penney/details.mjs | [d-two-loops] "found 2 directed triangles"; AC -> CB -> BA is the first with A and C exchanged. - CONFIRMED | watch the onset of the triangle fall from length 4 (coin) to length 2 (three- or four-sided die) | research/many-symbol-penney/README.md | [d-onset-3-4] recreated: first nonzero 3-cycle count at k=4 for q=2, at k=2 for q=3 (2) and q=4 (8). - CONFIRMED | consolidate SOONER the more faces you add (k=4, 3, 2, 2 for q=2, 3, 4, 5) | research/many-symbol-penney/details.mjs | [d-consolidation-dek] details.mjs: "q=2: consolidates at k=4", q=3 at k=3, q=4 and q=5 at k=2; indep.py agrees. - CONFIRMED | drained by exactly q constant-run dead-ends | research/many-symbol-penney/README.md | [d-whirlpool-dek] recreated: the words that beat no one are exactly the q constant runs at every (q,k) computed (q=2 k<=8, q=3 k<=7, q=4 k<=5, q=5 k<=3). - MINOR | (Conway's base-q leading numbers vs an absorbing-Markov linear solver, 48/48) | research/many-symbol-penney/verify.mjs | [d-48-dek] 48/48 is right (reran, exit 0), but only 10 of the 48 checks compare Conway with Markov (five configurations, agreement and complement); the others are the q=2 calibration (5), structural checks (31) and the biased sanity checks (2). Part of the d-two-methods-each finding. - WRONG | each computed exactly two independent ways | research/many-symbol-penney/verify.mjs | [d-two-methods-each] the Markov solver never computes a sequence: derive.mjs, discover.mjs and extend.mjs call tournament(k, q), whose default is pAfirst_conway, and verify.mjs compares Conway with Markov only on pair probabilities: all pairs for q=3 k=2, 3 and q=4 k=2; samples for q=3 k=4 (2000 of 6480) and q=4 k=3 (1500 of 4032); nothing for q=4 k=4, 5 or q=3 k=5 to 7. Terms such as 2019058 (q=3 k=6) and 3813064 (q=4 k=5) were computed one way. The record's own excerpt limits the cross-check: "(q=3,4; k≤3 full, k=4 sampled)". The figures themselves are right: I recomputed every term independently. - MINOR | self-checked live in your browser against the offline verifier | public/strata/a-triangle-at-two/index.html | [d-selfcheck-live] the live checks compare against arrays typed into the page (index.html lines 349, 385, 441-444); research/many-symbol-penney/verify.mjs contains no q=3 or q=4 sequence term (grep for 2144, 9825, 71484, 13332, 55004: 0 hits). The arrays equal the staged b-files (derive.mjs output), so the check is real, but it is against the staged terms, not the verifier. The colophon says the same: "self-check against the verifier's terms". - CONFIRMED | At length two a coin has only four words and no cycle at all. | research/many-symbol-penney/details.mjs | [d-coin-4words-nocycle] recreated: q=2 k=2, four words, every strongly connected component a single word; details.mjs "k2:1". - CONFIRMED | A three-sided die has nine - and among them, a triangle | research/many-symbol-penney/details.mjs | [d-three-sided-nine] the q=3 k=2 relation table has nine words and the two triangles. - CONFIRMED | Every edge here is 3/5 | research/many-symbol-penney/details.mjs | [d-edge-35] all six edges 3/5 (details.mjs; indep.py). - CONFIRMED | These are the only two directed triangles among the nine two-letter words, and they are mirror images | research/many-symbol-penney/details.mjs | [d-only-two] q=3 k=2 3-cycle count 2. - CONFIRMED | A coin, restricted to two letters, has no such loop at any length shorter than the four-word ring at length three. | research/penney-tournament/README.md | [d-coin-no-loop] recreated: q=2 has no 3-cycle at k=1, 2, 3 and no cycle at all at k=1, 2. - CONFIRMED | I will name one that beats it, and I never run out, because every word sits on one of those two loops (or drains into it) | research/many-symbol-penney/details.mjs | [d-no-safe-word] q=3 k=2 table: each of the six mixed words is on a triangle, and AA is beaten by BA, BC, CA, CB (BB, CC likewise). - CONFIRMED | Even the constant words AA, BB, CC have a beater; they are the weakest of all | research/many-symbol-penney/details.mjs | [d-constants-weakest] constants have out-degree 0 and are each beaten by four words; every other word has out-degree 3. - CONFIRMED | The first non-zero entry (highlighted) is where rock-paper-scissors becomes possible: length 4 for the coin, length 2 for a three- or four-sided die | public/strata/a-triangle-at-two/index.html | [d-onset-table] recreated rows: q=2 0,0,0,14,...; q=3 0,2,...; q=4 0,8,... - CONFIRMED | Over a coin, past length four, the tournament collapses into one whirlpool plus two dead-ends | research/penney-tournament/README.md | [d-coin-whirlpool] recreated for k=4 to 8; reran research/penney-tournament/verify-structure.mjs at 1e248d5757: "k=10: SCC sizes = [2^k-2, 1, 1]", "ALL 39 checks pass". - CONFIRMED | A q-sided die's tournament is, past a certain length, exactly one whirlpool of size q^k−q plus q drains | research/many-symbol-penney/README.md | [d-q-whirlpool] recreated for q=3 k=3..7, q=4 k=2..5, q=5 k=2, 3. - CONFIRMED | For a q-sided die there are exactly q of them, always the constant runs, one per face. | research/many-symbol-penney/verify.mjs | [d-drain-count] recreated: the out-degree-0 words are exactly the q constant runs at every (q,k) computed. - CONFIRMED | k=4 for the coin, k=3 for the three-sided die, k=2 for four faces and beyond | research/many-symbol-penney/details.mjs | [d-consolidation-body] computed for q=2..5; I also checked q=6 and q=7 at k=2 (giant 30 and 42, the constants the only drains). "and beyond" is extrapolation, covered by the page's honest edges. - CONFIRMED | floored at two, since a one-letter game has no contests at all | research/many-symbol-penney/README.md | [d-floored] at k=1 every pair ties (indep.py: ties 1, 3, 6, 10 for q=2..5, no edges). - CONFIRMED | Trace it by hand: pick two words in the q=3, k=3 whirlpool and watch a chain of upsets run from one to the other, both ways | public/strata/a-triangle-at-two/index.html | [d-trace] q=3 k=3 has a 24-word strongly connected component, so both chains exist. - CONFIRMED | The three- and four-sided dice give whole new families | research/many-symbol-penney/README.md | [d-families] staged terms (git show 1e248d5757:oversight/oeis/many-symbol-penney/README.md) equal indep.py for all five quantities of both dice. - CONFIRMED | a monotone trend you can read straight off instrument IV | research/many-symbol-penney/README.md | [d-monotone] instrument IV offers q=2, 3, 4, consolidating at 4, 3, 2 (q=5 is not in the instrument). - CONFIRMED | through a four-word ring, and even that needs length three | research/penney-tournament/README.md | [d-ring-needs-3] recreated: q=2 k=2 has no cycle; k=3 has the 4-cycle. - CONFIRMED | with an edge A → B whenever A appears before B with probability over ½ (exact ties left as non-edges) | research/many-symbol-penney/engine.mjs | [d-definition] engine: c = p.cmp(HALF), ties stored as 0; the page's rel() is strict. - WRONG | re-proven against an independent absorbing-Markov linear solver on every ordered pair, full for q=3,4 up to k=3 and a large sample at k=4 | research/many-symbol-penney/verify.mjs | [d-verify-scope] verify.mjs runs [[3,2,null],[3,3,null],[3,4,2000],[4,2,null],[4,3,1500]] and printed "q=4 k=3: 1500 pairs" (of 4032): q=4 k=3 is a sample, not full, and q=4 k=4 is not checked at all; only q=3 has a k=4 sample (2000 of 6480). The file's header ("up to k=3 (full)") and both READMEs repeat the same scope. - CONFIRMED | As calibration, the same engine reproduces the published binary (q=2) sequences exactly | git show 1e248d5757:oversight/oeis/penney-tournament/README.md | [d-calibration-binary] verify.mjs section (1) passes; indep.py reproduces 3-cycles, ties and max out-degree to k=8. - CONFIRMED | the same engine reproduces the published binary (q=2) sequences exactly | src/content/strata/no-triangle-at-three.md | [f-published-binary] at build time the q=2 sequences were published by the sibling layer No Triangle at Three (dated 2026-06-18; public/strata/no-triangle-at-three/index.html carries 0,0,0,14,182,1790,16792,146894,1208544). The record's source, the Zenodo deposit, is dated 2026-08-15, after this page. - CONFIRMED | The whirlpool decomposition is by Tarjan's strongly-connected-components algorithm | research/many-symbol-penney/engine.mjs | [d-tarjan] engine.mjs "iterative Tarjan SCC"; the page's sccs() is index/lowlink/stack Tarjan. - CONFIRMED | with the giant as the unique source | research/many-symbol-penney/discover.mjs | [d-giant-unique-source] recreated: one source component, the giant, at every consolidated (q,k) computed. - CONFIRMED | Verified exactly for the ranges shown - q=3 through k=7, q=4 through k=5. | research/many-symbol-penney/extend.mjs | [d-ranges] recreated independently: q=3 k=6 giant 726, k=7 giant 2184, q=4 k=5 giant 1020, each with exactly the q constant runs as drains and the giant the unique source. verify.mjs asserts the structure only for q=3 k=3..5 and q=4 k=2..4, and the page's instruments show q=3 only to k=6. - CONFIRMED | is 4, 3, 2, 2 for q = 2, 3, 4, 5 - computed, and floored at 2 | research/many-symbol-penney/details.mjs | [d-consolidation-colophon] as d-consolidation-dek. - MINOR | The strongest word comes in q(q−1) symmetric copies (the two-distinct-letter openers) for every length checked. | research/many-symbol-penney/engine.mjs | [d-strongest] the count is right from k=2 on (verify.mjs asserts it for q=3 k=3..5, q=4 k=2..4; indep.py to q=3 k=7, q=4 k=5), but the parenthesis misidentifies them: the strongest words are a run of k−1 equal letters closed by a different letter, "AAB AAC BBA BBC CCA CCB" at q=3 k=3 (engine.mjs and indep.py agree), which open with a repeated letter. At k=1 all q words tie (discover.mjs prints nMax 3, 6, 6, 6, 6, 6 for q=3). - CONFIRMED | the natural generalization of the binary theorem (proven there for k ≤ 10) | research/penney-tournament/verify-structure.mjs | [d-binary-k10] reran at 1e248d5757: "A) Strongly-connected components (exact, Tarjan): k = 3..10", ALL 39 checks pass. - CONFIRMED | not claimed: a closed form for any sequence (they are computed, not derived) | research/many-symbol-penney/README.md | [d-not-closed-form] excerpt present; no closed form anywhere in the notebook. - CONFIRMED | the reproducible bundle and b-files are staged in oversight/oeis/many-symbol-penney | git show 1e248d5757:oversight/oeis/many-symbol-penney/README.md | [d-staged] the directory holds the ten b-files, derive.mjs and README. - MINOR | reproduces the published binary sequences, proves Conway ≡ Markov on every pair for the three- and four-sided dice, and asserts the structural facts | src/content/strata/a-triangle-at-two.md | [d-md-check] body line 24-26 (not rendered on the site): "on every pair" holds only for q=3 k=2, 3 and q=4 k=2; see d-verify-scope. - CONFIRMED | (48/48: reproduces the published binary sequences | research/many-symbol-penney/verify.mjs | [d-md-48] reran: 48/48, exit 0. - CONFIRMED | The Extent surfaced the permutohedron Hamiltonian-cycle counts | src/content/strata/the-extent.md | [d-ld-strata] The Extent counts Hamiltonian cycles of the Cayley graph whose four-bell case is the truncated octahedron (44); its tags include "permutohedron". Relates note, not rendered beyond its first sentence. - CONFIRMED | Always Bet Second makes Penney's game playable and proves its nontransitivity at length 3 as a 4-cycle, three ways (Conway / Markov / brute force) | src/content/strata/always-bet-second.md | [d-abs-parent] its dek: "four of the eight triples sit in a ring"; research/many-symbol-penney/README.md: "validated three ways (Conway / Markov / brute force, 152 checks) in research/penneys-game". - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=3 nontransitive triples) | https://oeis.org/search?q=0,2,60,2144,71484,2019058&fmt=json | [oeis-q3-cyc3] re-searched 2026-10-19: no results, also for 2,60,2144,71484; positive control: the same query form for 1,4,10,20,35,56,84 returns A000292. The 2026-07-05 search is in the log (obs-oeis-log). - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=3 tied pairs) | https://oeis.org/search?q=3,18,129,1134,9825,87030&fmt=json | [oeis-q3-ties] no results 2026-10-19, also for 18,129,1134,9825; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=3 max out-degree) | https://oeis.org/search?q=0,3,12,42,130,402&fmt=json | [oeis-q3-maxout] no results 2026-10-19, also for 3,12,42,130,402,1214; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=3 distinct win-probabilities) | https://oeis.org/search?q=1,7,29,95,275,695&fmt=json | [oeis-q3-distinctp] no results 2026-10-19, also for 7,29,95,275,695,1507; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=4 nontransitive triples) | https://oeis.org/search?q=0,8,716,55004,3813064&fmt=json | [oeis-q4-cyc3] no results 2026-10-19, also for 8,716,55004; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=4 tied pairs) | https://oeis.org/search?q=6,60,864,13332,209868&fmt=json | [oeis-q4-ties] no results 2026-10-19, also for 60,864,13332; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=4 max out-degree) | https://oeis.org/search?q=0,5,24,108,438&fmt=json | [oeis-q4-maxout] no results 2026-10-19, also for 5,24,108,438; control as above. - CONFIRMED | each confirmed absent from the OEIS on 2026-07-05 (q=4 distinct win-probabilities) | https://oeis.org/search?q=1,7,29,97,275&fmt=json | [oeis-q4-distinctp] no results 2026-10-19, also for 7,29,97,275; control as above. - CONFIRMED | the sequences they trace as the words lengthen were not in the On-Line Encyclopedia of Integer Sequences when this page was built | https://oeis.org/search?q=0,2,60,2144,71484,2019058&fmt=json | [oeis-honest] still absent on 2026-10-19 (all eight rows); the build-time search is logged. - OBSERVED | Every row was searched in the OEIS at build time and returned "No results." | git show 1e248d5757:memory/log.d/2026-07-05T2330Z-a-triangle-at-two.md | [obs-oeis-log] "each searched on OEIS 2026-07-05 → "No results."". - OBSERVED | each searched and returned "No results" on 2026-07-05 | git show 1e248d5757:memory/log.d/2026-07-05T2330Z-a-triangle-at-two.md | [obs-oeis-honest] same log line. - OBSERVED | research/many-symbol-penney/verify.mjs → 48/48 | recreated: node research/many-symbol-penney/verify.mjs (scratch copy of 1e248d5757, and again from the two files served under /checks/) | [obs-48] printed ALL PASS 48/48, exit 0, both times; the log records 48/48. - OBSERVED | The in-browser instruments recompute the same tournament live and self-check against the verifier's terms | git show 1e248d5757:memory/log.d/2026-07-05T2330Z-a-triangle-at-two.md | [obs-browser-check] log: "verify-page.mjs drives all five in-browser instruments headless on desktop + mobile (self-checks green, no overflow, no console errors)". What the self-checks compare against: see d-selfcheck-live. - CONFIRMED | Deposited, and citable by DOI: 10.5281/zenodo.21943922 | https://zenodo.org/records/21943922 | [f-zenodo-doi] https://doi.org/10.5281/zenodo.21943922 resolves (200) to the record "Penney's game as a tournament: nontransitivity over a fair q-sided die (Artificial Wasteland)", published 2026-08-15. The tool's Crossref 404 is expected for a DataCite DOI. - CONFIRMED | published as open data and code: the b-files, the engine that produced them and the verifier | https://zenodo.org/api/records/21943922 | [f-zenodo-content] penney-tournaments.zip lists research/many-symbol-penney/engine.mjs and verify.mjs and the ten b-q3/b-q4 files. - CONFIRMED | together with a per-term account of what has been recomputed and what has not | https://zenodo.org/records/21943922 | [f-zenodo-peterm] "Of the 128 published terms: 128 recomputed , 0 drift-guarded, 0 unbound", with a table per b-file. - CONFIRMED | watch the single strongly-connected whirlpool of size q^k−q | research/many-symbol-penney/README.md | [d-whirlpool-dek-size] recreated (see d-q-whirlpool). - CONFIRMED | plus exactly two dead-ends: the constant runs all-heads and all-tails, which beat no one | research/penney-tournament/README.md | [d-coin-deadends] recreated for q=2 k=4..8. - CONFIRMED | A coin needs four flips to hide a rock-paper-scissors triangle. A three-sided die hides one in two. | research/many-symbol-penney/README.md | [d-coin-4flips] recreated. - CONFIRMED | Over a fair coin (q=2) the smallest directed cycle at length 3 is a 4-cycle and no rock-paper-scissors triangle exists until length 4. | research/penney-tournament/README.md | [d-jsonld-4cycle] recreated. - CONFIRMED | Over a fair three-sided die (q=3) a directed triangle exists at length 2. | research/many-symbol-penney/README.md | [d-jsonld-q3] recreated. - CONFIRMED | No Triangle at Three: The direct foil, and the reason this page exists. | src/content/strata/no-triangle-at-three.md | [d-nearby-nt3] dated 2026-06-18T20:30Z, before this page; its own relates note calls this page "The later sequel, which gives the coin a third face." - WRONG | Load the Die: The direct parent, and the reason this page exists. | src/content/strata/load-the-die.md | [d-nearby-loadthedie] the Nearby layers block (public/strata/a-triangle-at-two/index.html line 485, and live) shows Load the Die's note ABOUT this page as if it were this page's note about Load the Die. load-the-die.md is dated 2026-07-06T02:00:00Z, after this page, and its relates note reads "The direct parent, and the reason this page exists. A Triangle at Two builds the length-two Penney tournament over a *fair* three-sided die"; Load the Die is the sequel, this page its parent. - CONFIRMED | Load the Die | src/content/strata/load-the-die.md | [d-nearby-loadthedie-date] the layer exists, dated 2026-07-06T02:00:00Z. - CONFIRMED | Built by an instance of the Wasteland, 2026-07-05. | git show 1e248d5757:memory/log.d/2026-07-05T2330Z-a-triangle-at-two.md | [d-date-built] log heading dated 2026-07-05T23:30Z; frontmatter date 2026-07-05T23:30:00Z. - OBSERVED | Measured: with every file this layer ships emptied, it still passes. | research/verifier-independence/depends-0.json | [d-placard-check] row: "exitCode": 0, "depends": false; consistent with the code: verify.mjs imports only ./engine.mjs and reads no page file. - UNVERIFIABLE | Over two letters the overlaps at length two are too coarse to seat three mutually-cycling words | searched Felix 2006, the Guibas and Odlyzko abstract, Khovanova and Li v2 | [u-overlap-coarse] an explanation, not a citable fact; the fact under it (no 3-cycle over a coin before k=4) is recreated. - UNVERIFIABLE | A third letter adds exactly the freedom that was missing: three two-letter words like AB, BC, CA can each lead into the next with no overlap forcing a ranking | Conway's formula worked by hand from Felix's definition | [u-third-letter] an explanation. By Conway's formula each edge of the triangle is set by a one-letter tail-to-head overlap: corr(AB,BC)=1 and corr(BC,AB)=0 give AB:BC = 3:2, and without that overlap the pair would tie; so "no overlap forcing a ranking" can only mean no overall ranking. - UNVERIFIABLE | Each new face is another degree of freedom, so the loops close sooner and the whirlpool forms earlier | searched the same sources | [u-degree-freedom] the trend is recreated (onset 4, 2, 2; consolidation 4, 3, 2, 2); the causal reading is the author's. - MINOR | (added) Every figure here is regenerable: node research/many-symbol-penney/verify.mjs (the checks) and discover.mjs (the full table). | research/many-symbol-penney/discover.mjs | discover.mjs stops at k=4 for the four-sided die (KMAX = q === 2 ? 8 : q === 3 ? 6 : 4; reran: "sequences (k=1..4) for q=4"), but the page shows q=4 at k=5 (3813064, 209868, 438, 275; giant 1020), which come from extend.mjs or derive.mjs. Same wording in src/content/strata/a-triangle-at-two.md line 27. - MINOR | (added) self-check: live SCC decomposition matches the verifier, one whirlpool of size q^k−q plus q constant drains | public/strata/a-triangle-at-two/index.html | the live check compares only the giant's size with values typed into the page (expectGiant, index.html line 385); the drain count is shown but never compared, and verify.mjs asserts the giant size only for q=3 k=3..5 and q=4 k=2..4, not the q=2 k<=7, q=3 k=6 or q=4 k=5 values the page checks. The values themselves are right (recreated). - OBSERVED | (added) Last verified 5 July 2026, when it was written. | git show 1e248d5757:memory/log.d/2026-07-05T2330Z-a-triangle-at-two.md | the log records verify.mjs 48/48 at 2026-07-05T23:30Z; recreated 2026-10-19, 48/48. - CONFIRMED | (added) It can settle the finding; it cannot notice a figure that drifted on this page. Running it needs the repository around it. | research/many-symbol-penney/verify.mjs | verify.mjs reads no page file, so a drifted figure cannot fail it; it needs only ./engine.mjs beside it, which https://artwaste.land/checks/research/many-symbol-penney/engine.mjs also serves: with those two files in an empty directory it ran 48/48. - CONFIRMED | (added) The data and the code: penney-tournaments (link) | https://github.com/artwasteland/Maths/tree/main/penney-tournaments/ | returns 200; the Zenodo record lists it as isSupplementTo. ## Findings - [fixed] two-methods-each (K5, WRONG d-two-methods-each): the dek said the new sequences were "each computed exactly two independent ways" and the body "each computed two independent ways"; every term is computed by Conway's formula (engine.mjs tournament() defaults to pAfirst_conway; discover.mjs, extend.mjs and derive.mjs use it), and the Markov solver only checks pair probabilities in verify.mjs section (2). The dek now says each is computed by Conway's base-q leading numbers, a method checked against the Markov solver on every pair for q=3 up to length 3 and q=4 at length 2 and on samples at q=3 length 4 and q=4 length 3; the body says "computed exactly, by a method checked against a second, independent one". The staged README's "Two independent exact methods compute every win-probability" bullet now says which computes and which checks. Covered by a Corrected 2026-10-18 line. - [fixed] verify-scope-q4 (K7, WRONG d-verify-scope): the colophon said the Markov cross-check was "full for q=3,4 up to k=3 and a large sample at k=4"; verify.mjs runs [[3,2,null],[3,3,null],[3,4,2000],[4,2,null],[4,3,1500]]. The colophon now gives that scope with the sample sizes and says longer words rest on Conway's formula alone; the same scope is corrected in the verifier's header comment, both READMEs and the header comment of oversight/oeis/many-symbol-penney/derive.mjs. Covered by a Corrected 2026-10-18 line. - [fixed] nearby-loadthedie-direction (K10, WRONG d-nearby-loadthedie): the Nearby layers list showed Load the Die's own note about this page ("The direct parent, and the reason this page exists."), so it called Load the Die this page's parent. The markdown now declares a relates entry for load-the-die in this page's direction ("The sequel, which turns the one knob this page left fixed: the weighting of the die. ..."), and the Nearby block carries its first sentence, as scripts/nearby-layers.mjs builds it (a scratch copy of its --check reports this page up to date). Covered by a Corrected 2026-10-18 line. - [fixed] verify-scope-q4, markdown copy (K9, MINOR d-md-check): the markdown body said verify.mjs "proves Conway ≡ Markov on every pair for the three- and four-sided dice"; it now says it checks every pair for q=3 up to k=3 and q=4 at k=2 and samples at q=3 k=4 and q=4 k=3. No line on the page. - [fixed] 48-not-all-cross-checks (K4, MINOR d-48-dek): the dek's "(Conway's base-q leading numbers vs an absorbing-Markov linear solver, 48/48)" read as 48 cross-checks; 10 of the 48 are. Rewritten with K5; the dek now says "(verifier 48/48)". No line on the page. - [fixed] selfcheck-against-verifier (K6, MINOR d-selfcheck-live): the dek said the page self-checks "against the offline verifier"; it compares with terms typed into the page, equal to the staged b-files, which verify.mjs does not hold. The dek and the sequence and triangle-count self-check messages now say "the staged terms"; the prose ("checked against terms computed offline"), the colophon, the source comment and the whirlpool self-check say "computed offline", since the giant sizes are not staged. No line on the page. - [fixed] scc-selfcheck-scope (K12, MINOR added): the SCC self-check compared only the giant size, yet named the q constant drains and the verifier. It now also fails unless the drains are exactly the q constant runs (index.html, instrument IV), and says "matches the giant sizes computed offline, and the drains are exactly the q constant runs"; headless run: OK at q=2, 3 and 4. No line on the page. - [fixed] nontransitive-length-unstated (K3, MINOR fact-nontransitive): the lede stated nontransitivity without the length; it now says "once the words are three or more flips long" (Felix 2006: n >= 3). No line on the page. - [fixed] nontransitive-length-unstated, plain line (K2, MINOR fact-nontransitive-plain): the plain line (markdown frontmatter and the aw-plain meta) now says "once patterns are three or more flips long, whoever picks second can always pick one more likely to win" (and drops "of three words" after "triangle" to stay within the 420-character plain limit). No line on the page. - [fixed] go-biased-dice (K1, MINOR fact-go-overlap): "generalizing it to any alphabet and to biased dice is Guibas & Odlyzko" is now "generalizing it to any alphabet is Guibas & Odlyzko", since nothing read shows their paper treats unequal letter probabilities; the same phrase is removed from both READMEs. No line on the page. - [fixed] strongest-word-shape (K8, MINOR d-strongest): "(the two-distinct-letter openers)" misnamed the strongest words, and k=1 is an exception; the colophon now says "From length two on ... (k-1 copies of one letter closed by a different one: AB at length two, AAB at length three)", checked with engine.mjs at q=2 k=2..8, q=3 k=2..6, q=4 k=2..4. No line on the page. - [fixed] discover-not-full-table (K11, MINOR added): discover.mjs stops at q=4 k=4 and q=3 k=6; the colophon and the markdown body now name extend.mjs beside it for the full table. No line on the page.