S3 s3_wordnet_shape: the WordNet 3.0 verb forest and Owl Academy's canon tree under the L3 address rule: shape, exact H^3 points (TreeEmbedding3D, t = 3/4), exact-tree graph-edge and closure MAP with pessimistic ties Settles: the value behind L6(c)'s prediction rule; the bit and time budgets; query sampling; the kernel's 25-letter tile limit. Decides no claim. Verbs and the canon tree only; nouns are counted, never addressed or ranked. ~ node v22.23.4, HPCore 2.0.0, HPDecode3D 2.0-a, linux-x64 0. Data and metadata (counts only for nouns) wordnet.zip (nltk_data, WordNet 3.0): 10775600 B, sha256 cbda5ea6eef7f36a97a43d4a75f85e07fccbb4f23657d27b4ccbc93e2646ab59 (pinned yes); data.verb sha256 29cc96ed80c9f47d94fe75e332a9df80f4b1c737205f92d2f433d63c6da2ab51 canon/canon.manifest.json (Owl-Academy, read only): 61955 B, sha256 2ddc35b1f1c1ae8bb839d1fda176d32f1d5b705110408f1cb44143d7a749db1f; format owl-canon/1, 145 tome records synsets (recounted as map sizes, equal to wordnet_v1.meta.json): noun 82115, verb 13767, adj 18156 (10693 satellites), adv 3621 echoed from wordnet_v1.meta.json, not recomputed: noun closure edges (hypernym + instance hypernym) 743241; mammal.n.01 (01861778) closure 1182 nodes, 6542 edges, spanning subtree 1176 nodes ~ load 737 ms 1. Shape under the L3 rule (a) VERB forest: 13767 synsets, 559 roots (no '@' pointer; 225 of them isolated: no children), 13239 '@' pointers, 0 '@i'; 31 synsets have more than one '@' (the spanning parent is the first; the closure keeps all) virtual root [1] with 559 children, m = 4 letters each, so every verb root has a 5-letter word; P.children([1]) = [1,2,3,4,5] depth (top-level nodes = 0) histogram, with word length min..max (mean) at each depth: depth 0: 559 nodes, word length 5..5 (mean 5.00) depth 1: 3280 nodes, word length 6..9 (mean 7.58) depth 2: 3902 nodes, word length 7..12 (mean 9.53) depth 3: 3013 nodes, word length 8..14 (mean 11.40) depth 4: 1595 nodes, word length 9..16 (mean 13.05) depth 5: 735 nodes, word length 10..18 (mean 14.75) depth 6: 366 nodes, word length 12..20 (mean 16.36) depth 7: 150 nodes, word length 13..21 (mean 18.29) depth 8: 60 nodes, word length 14..23 (mean 20.35) depth 9: 63 nodes, word length 19..25 (mean 21.65) depth 10: 32 nodes, word length 20..26 (mean 21.94) depth 11: 11 nodes, word length 21..23 (mean 22.64) depth 12: 1 nodes, word length 22..22 (mean 22.00) internal nodes (>= 1 child): 3309; children per internal node, c:count: 1:1462 2:599 3:338 4:237 5:147 6:93 7:57 8:62 9:38 10:41 11:28 12:29 13:26 14:18 15:16 16:8 17:9 18:16 19:8 20:12 21:7 22:4 23:6 24:5 25:2 26:2 27:1 28:3 29:1 30:1 32:1 33:3 34:1 35:1 36:2 37:1 38:2 39:1 41:2 44:1 45:1 53:1 56:2 59:1 62:1 72:1 73:1 77:1 83:1 92:1 96:1 98:1 112:1 122:1 137:1 192:1 399:1 internal nodes by letters per child m = max(1, ceil(log5 c)): m=1 (c <= 5): 2783, m=2 (6..25): 485, m=3 (26..125): 38, m=4 (126..625): 3 nodes with more than 5 children: 526; more than 25: 41 (plus the virtual root, 559) single-child share: 1462/3309 internal nodes have exactly one child (44.18%); 1462/13767 nodes are only children (10.62%) letters per node (on the edge from its parent, the virtual root included): m=1: 5357, m=2: 5195, m=3: 1928, m=4: 1287; mean 1.9379 letters per edge; 12912 virtual letters (letters beyond the first on an edge) word length histogram, length:count: 5:559 6:445 7:1447 8:1577 9:1985 10:1650 11:1690 12:1376 13:1041 14:729 15:388 16:300 17:148 18:132 19:82 20:90 21:51 22:23 23:27 24:13 25:9 26:5 maximum word length 26; words with <= 25 letters: 13762/13767 (99.96%); <= 27 letters: 13767/13767 (100.00%) the 5 longest words (26 letters), offset lemma depth word: 00513757 pick_up depth 10 14000211120312002402125020 00513936 wave_off depth 10 14000211120312002402125000 00514069 foreground depth 10 14000211120312002402125021 00557537 raise depth 10 14000211120312002402125022 00840264 soft-pedal depth 10 14000211120312002402125002 largest families: 00126264 change (c = 399, m = 4); 00109660 change (c = 192, m = 4); 02604760 be (c = 137, m = 4); 01835496 travel (c = 122, m = 3); 00173338 remove (c = 112, m = 3); 02327200 supply (c = 98, m = 3) (b) CANON tree (canon root > dept > artifact > tome; root at [0]; children by globalIndex, checked equal to I..X, artifact number and clusterIndex order): 169 nodes = 1 root + 10 depts + 13 artifacts + 145 tomes; P.children([0]) = [0,2,3,4,5] depth (top-level nodes = 0) histogram, with word length min..max (mean) at each depth: depth 0: 1 nodes, word length 1..1 (mean 1.00) depth 1: 10 nodes, word length 3..3 (mean 3.00) depth 2: 13 nodes, word length 4..4 (mean 4.00) depth 3: 145 nodes, word length 6..6 (mean 6.00) internal nodes (>= 1 child): 24; children per internal node, c:count: 1:9 4:1 9:9 10:1 16:4 internal nodes by letters per child m = max(1, ceil(log5 c)): m=1 (c <= 5): 10, m=2 (6..25): 14, m=3 (26..125): 0, m=4 (126..625): 0 nodes with more than 5 children: 14; more than 25: 0 single-child share: 9/24 internal nodes have exactly one child (37.50%); 9/169 nodes are only children (5.33%) letters per node (on the edge from its parent): m=1: 13, m=2: 155; mean 1.9226 letters per edge; 155 virtual letters (letters beyond the first on an edge) word length histogram, length:count: 1:1 3:10 4:13 6:145 maximum word length 6; words with <= 25 letters: 169/169 (100.00%); <= 27 letters: 169/169 (100.00%) example words: canon 0; dept-I 000; artifact-I-1 0000; dept-I/clifford-toroid-proof 000000; dept-X 025; artifact-X-4 0253; dept-X/expression-16 025340 ~ section 1: 63 ms 2. Exact H^3 points (HPDecode3D.TreeEmbedding3D.pointExact, t = 3/4, cosh tau = 25/7, sinh tau = 24/7) verbs: pointExact on all 13767 nodes; sheet identity exact on 13767/13767; every coordinate is an integer over 7^n (n = word length); equal to the BigInt frame product (step matrices x 7, second way) on 13767/13767 max bits per coordinate (lowest terms): numerator 133, denominator 73; by word length n, max numerator/denominator bits: 5:28/15 6:32/17 7:39/20 8:43/23 9:49/26 10:54/29 11:59/31 12:64/34 13:70/37 14:74/40 15:80/43 16:85/45 17:90/48 18:96/51 19:101/54 20:105/57 21:110/59 22:115/62 23:120/65 24:124/68 25:127/71 26:133/73 the virtual root [1] (not a node): (25/7, -24/7, 0/1, 0/1) canon: pointExact on all 169 nodes; sheet identity exact on 169/169; every coordinate is an integer over 7^n (n = word length); equal to the BigInt frame product (step matrices x 7, second way) on 169/169 max bits per coordinate (lowest terms): numerator 33, denominator 17; by word length n, max numerator/denominator bits: 1:5/3 3:16/9 4:22/12 6:33/17 ~ pointExact: verbs 3527 ms for 13767 points (0.256 ms per point), 1.406 ms per point at n = 26; canon 17 ms for 169 ~ section 2: 3791 ms 3. Exact-tree MAP on the VERB forest (H^3, L3 rule, exact integer keys, ties pessimistic) queries: 13542 (every real verb with a graph-edge neighbour; the 225 isolated roots are candidates only); 13766 candidates per query; the virtual root is neither graph-edge neighbours: 26416 in all (mean 1.9507), max 399 (00126264 change) closure neighbours (ancestors and descendants in the '@' closure): 35079 ancestor-descendant pairs; per query 70158 in all (mean 5.1808), max 1703 (00126264 change); 0 non-queries have closure neighbours (an extra '@' parent) integer keys: points scaled to 7^26 (<= 133-bit coordinates); key = cosh d * 7^52; largest key of any (query, candidate) pair 257 bits, of any neighbour 200 bits VERB GRAPH-EDGE MAP (exact H^3, L3 rule) = 0.6309 over 13542 queries graph-edge MAP exactly: 205-digit / 205-digit rational, sha256("n/d") 2e428dd7f198b81b0ee0e824404aefaf40ed3542a292c761eb7664968ad0efb2; truncated 0.6308581090844327 queries with AP < 1: 7893/13542; with a neighbour tied exactly with a non-neighbour: 5396/13542; optimistic-tie MAP (contrast only) 0.8149 closure MAP = 0.4206 (2052-digit / 2053-digit rational, sha256 f274ad0b43f3ea8fb489efda272cf8799ec68ccf1a856e2a264b3b50afd49e85; truncated 0.4206252510130392) queries with AP < 1: 12919/13542; with a neighbour/non-neighbour exact tie: 9716/13542; optimistic-tie MAP (contrast only) 0.5390 graph-edge MAP by the letters of the query's own children (leaf = no children): leaf 10233 queries, MAP 0.5683 m_out=1 2783 queries, MAP 0.8607 m_out=2 485 queries, MAP 0.6238 m_out=3 38 queries, MAP 0.7368 m_out=4 3 queries, MAP 0.7265 graph-edge MAP by the letters on the query's edge from its parent (m_in = 4: the 334 verb roots with children, and the 728 children of the 3 families with more than 125 children): m_in=1 5357 queries, MAP 0.9974 m_in=2 5195 queries, MAP 0.4087 m_in=3 1928 queries, MAP 0.3217 m_in=4 1062 queries, MAP 0.4301 query sampling (all 13542 queries were run, so nothing above is sampled; for comparison): Rng(90000).sample(13542, 2000): graph-edge 0.6247, closure 0.4200; Rng(90001).sample(13542, 1000): graph-edge 0.6262, closure 0.4215 20 replicate 1,000-query samples Rng(90010..90029): graph-edge 0.6114..0.6427 (sd 0.0085), closure 0.4028..0.4347 (sd 0.0084) second way: AP recomputed with Q Lorentz products and an explicit sort on 11 queries (Rng(90002).sample(13542, 10), the most graph-edge and the most closure neighbours), graph-edge and closure: equal exact MAP against the float64 mean of the APs: |difference| < 1e-12 (graph-edge and closure) ~ exact ranking: 35.5 s for 13542 queries x 13766 candidates (2.62 ms per query, 191 ns per pair, both MAPs in one pass); exact MAP sums 147 ms; Q-path verification 3.6 s (163 ms per AP) ~ extrapolation at this cost per pair (noun words are longer, so a lower bound): 1,000 queries x 82,115 noun candidates ~ 16 s; 200 queries ~ 3 s ~ section 3: 39524 ms 4. Exact-tree MAP on the CANON tree (all nodes are queries and candidates; ties pessimistic) canon graph-edge MAP = 0.3544 over 169 queries (exactly 14815261627/41800814055; truncated 0.3544251939090615); AP < 1 on 155; neighbour/non-neighbour ties on 123; optimistic 0.8559 canon closure MAP = 0.3435 (exactly a 33-digit / 34-digit rational; truncated 0.3435433135098868); AP < 1 on 168 graph-edge AP by level: root 1.0000 (1), dept 0.7158 (10), artifact 0.9261 (13), tome 0.2738 (145); closure AP by level: root 1.0000 (1), dept 0.5465 (10), artifact 0.9122 (13), tome 0.2740 (145) contrast only, the canon root treated as virtual (as the verb root is): 168 queries, graph-edge MAP 0.3675, closure MAP 0.3774 second way: every canon AP (both variants, graph-edge and closure) recomputed with Q Lorentz products and an explicit sort: equal ~ section 4: 508 ms 5. Python recomputation (s3_wordnet_shape.py, python3 -B -I, hpstep imported in place; exact ints and Fractions; independent of this file) py| S3 python: independent recomputation of the verb forest and the canon tree (hpstep exact frames, int keys, Fractions) ~ py| python 3.13.18, numpy 2.5.3 py| wordnet.zip sha256 cbda5ea6eef7f36a97a43d4a75f85e07fccbb4f23657d27b4ccbc93e2646ab59 (pinned: yes); data.verb sha256 29cc96ed80c9f47d94fe75e332a9df80f4b1c737205f92d2f433d63c6da2ab51 py| verb forest: 13767 synsets, 559 roots, 3309 internal; virtual root [1], 4 letters per root; max word length 26 py| verb.synsets = 13767 py| verb.roots = 559 py| verb.maxlen = 26 py| verb.words.sha256 = ec39f99762d2814876ee1339664903308a7c36a0c47ba535e7279cec85e9c3d7 py| points: integer frame products on all 13767 nodes; equal to hpstep point_exact on 69 nodes (Rng(90003).sample(13767, 64) and the 5 deepest) py| verb.points.sha256 = c230e715293fa61a4f96b8ea214f0b288508a82e7f86415d38a05881096b53b9 ~ py| points: 113 ms for 13767 integer frames; point_exact check 44 ms for 69 nodes py| queries: 13542 (every verb with a graph-edge neighbour), 13766 candidates each; closure pairs 35079 py| verb.queries = 13542 py| verb graph-edge MAP = 0.6308581090844327 (truncated; exactly a 205-digit / 205-digit rational) py| verb.graph.ap.sha256 = 65b4d27e237d90c65fe5bbbec41cff9ae2124bc91c96e2c733a3f741bd629b49 py| verb.graph.map.sha256 = 2e428dd7f198b81b0ee0e824404aefaf40ed3542a292c761eb7664968ad0efb2 py| verb.graph.map.trunc16 = 0.6308581090844327 py| verb closure MAP = 0.4206252510130392 (truncated; exactly a 2052-digit / 2053-digit rational) py| verb.closure.ap.sha256 = 6f9f6dbbe6d647257b2c24945914df8b6c4392d171ed9faa284249ed037b20fc py| verb.closure.map.sha256 = f274ad0b43f3ea8fb489efda272cf8799ec68ccf1a856e2a264b3b50afd49e85 py| verb.closure.map.trunc16 = 0.4206252510130392 py| query samples, graph-edge MAP: Rng(90000) 2000 queries 0.6247051130304867, Rng(90001) 1000 queries 0.6261891982235723 py| verb.graph.sample2000.sha256 = e23654ac205da3bce4f22d74d1af8fa900dee66e7b464692fb1eaa070eb36297 py| verb.graph.sample1000.sha256 = 3b565d87fd8778b33f9d0329a2d92d7db19a78337a17685534cf1500a1441b05 ~ py| ranking: 68.3 s for 13542 queries x 13766 candidates (5.04 ms per query, both MAPs) py| canon.manifest.json sha256 2ddc35b1f1c1ae8bb839d1fda176d32f1d5b705110408f1cb44143d7a749db1f; 145 tomes py| canon tree: 169 nodes (root [0] > 10 depts > artifacts > tomes), max word length 6 py| canon.nodes = 169 py| canon.words.sha256 = e9e5f44ba5f2bfc0009e20c0e9b8793c82d727efbac298cc1d989261d5b97e85 py| canon.points.sha256 = b9e27813d64511edd3b1530d1e50c0146aa20ce9eb95f673ecd6a8c209b89d4a py| canon graph-edge MAP = 0.3544251939090615, closure MAP = 0.3435433135098868 (truncated), over 169 queries py| canon.graph.ap.sha256 = 50e298eb5c70701dceb8c179fd01c542c07dda2d1bea6e9829742e432d94700d py| canon.graph.map.sha256 = 3e6f8a7fd01d02a39120f4a7b70c77625b1576c379679be7b36a5419497f6ad0 py| canon.closure.ap.sha256 = 92d29d888f048234e1e7299b5f86942459eb1a152070c31754b70fec52334bf4 py| canon.closure.map.sha256 = a3bd34f1f5b4c1362fabc78fea3da1287ecf662168cf2b61ce31cd8ca702c15b ~ py| python total 68.8 s comparison with this script: 21/21 values identical (verb.synsets, verb.roots, verb.maxlen, verb.words.sha256, verb.points.sha256, verb.queries, verb.graph.ap.sha256, verb.graph.map.sha256, verb.graph.map.trunc16, verb.closure.ap.sha256, verb.closure.map.sha256, verb.closure.map.trunc16, verb.graph.sample2000.sha256, verb.graph.sample1000.sha256, canon.nodes, canon.words.sha256, canon.points.sha256, canon.graph.ap.sha256, canon.graph.map.sha256, canon.closure.ap.sha256, canon.closure.map.sha256) verb graph-edge MAP: node 0.6308581090844327, python 0.6308581090844327 (truncated to 16 decimals); the exact rationals are equal (same reduced n/d, by SHA-256), so |difference| = 0 ~ python wall time 68.9 s (concurrent with sections 1-4); section 5 waited 24282 ms SETTLES: - L6(c)'s rule: VERB GRAPH-EDGE MAP (exact H^3, L3 rule) = 0.6309 over all 13542 verb queries (exact rational, sha256 2e428dd7f198b81b..., 0.6308581090844327 truncated), reproduced exactly by the Python reference. Against the rule's threshold 0.90, L6(c)'s registered prediction would read "refuted" (a prediction fixed by the rule, not a decision of L6(c)). Verb closure MAP on the same queries 0.4206. Canon tree (all 169 nodes): graph-edge 0.3544, closure 0.3435. Ties pessimistic: 5396 of 13542 verb queries have a neighbour tied exactly with a non-neighbour; the optimistic MAP would be 0.8149. - Shape: verbs 13767 synsets under a virtual root [1] with 559 children (4 letters each); depth <= 12 below the verb roots; 1462/3309 internal nodes (44.18%) have one child; 526 have more than 5 children and 41 more than 25; 12912 virtual letters. Canon: 169 nodes, words of 1, 3, 4 and 6 letters (depts 2 letters, artifacts 1, tomes 2). - Tile limit: the longest verb word has 26 letters; 13762/13767 verb words (99.96%) fit the kernel's 25-letter id59 tile and 13767/13767 (100.00%) fit id64 (<= 27); 5 verb words need a continuation tile. Every canon word fits (max 6). - Bit budget: verb points have <= 133-bit numerators and <= 73-bit denominators per coordinate (lowest terms); over the common 7^26 they are <= 133-bit integers; the ranking keys (cosh d * 7^52) have <= 257 bits. - Query sampling: not needed for verbs; all 13542 queries were ranked against all 13766 candidates. A 1,000-query sample would have moved the verb graph-edge MAP by up to 0.0195 (20 replicates, sd 0.0085); the Rng(90000) 2,000-query sample gives 0.6247. ~ - Time budget (this machine): pointExact 0.256 ms per verb point on average, 1.406 ms at n = 26; exact ranking ~ 2.62 ms per verb query (191 ns per pair) in Node, 35.5 s for all; Python 68.3 s for all; 1,000 noun queries x 82,115 candidates ~ 16 s or more in Node. ~ total runtime 68.9 s