S1 s1_h3_tree_metrics: distances, order reversals, MAP with virtual letters, float inversions and exact cost in the octahedral H^3 tree (TreeEmbedding3D, t = 3/4, tau = ln 7) Settles: why ranking must be exact; the bit budget; L4's float-inversion controls; the virtual-letter tie effect behind L6(c)'s doubt. Decides no claim. Synthetic words only. ~ node v22.23.4, HPCore 2.0.0, HPDecode3D 2.0-a, linux-x64 1. Distances in the octahedral tree (exact cosh d = -_L; d = acosh, 4 decimals) t = 3/4, cosh tau = 25/7, sinh tau = 24/7, tau = 2 atanh(3/4) = ln 7 = 1.9459 (|tau - ln 7| < 1e-15: true) P.children([0]) = [0,2,3,4,5]: child 0 is straight (local +x), children 1-4 are local +y, -y, +z, -z parent-child [0] [0,0]: cosh d = 25/7, d = 1.9459 siblings at 90 deg (straight vs bent) [0,0] [0,2]: cosh d = 625/49, d = 3.2375 (law of cosines with cos theta = 0: equal) siblings at 90 deg (bent +y vs bent +z) [0,2] [0,4]: cosh d = 625/49, d = 3.2375 (law of cosines with cos theta = 0: equal) siblings at 180 deg (bent +y vs bent -y) [0,2] [0,3]: cosh d = 1201/49, d = 3.8918 (law of cosines with cos theta = -1: equal) root children at 90 deg [0] [2]: cosh d = 625/49, d = 3.2375 (law of cosines with cos theta = 0: equal) root children at 180 deg [0] [1]: cosh d = 1201/49, d = 3.8918 (law of cosines with cos theta = -1: equal) straight grandchild (no bend) [0] [0,0,0]: cosh d = 1201/49, d = 3.8918 (law of cosines with cos theta = -1: equal) bent grandchild (one 90 deg bend) [0] [0,0,2]: cosh d = 625/49, d = 3.2375 (law of cosines with cos theta = 0: equal) bend then straight [0] [0,2,0]: cosh d = 1201/49, d = 3.8918 (law of cosines with cos theta = -1: equal) all 936 edges of the radius-4 ball: cosh d = 25/7 exactly on 936/936 bend loss of one 90 deg bend between two edges: 2 tau - acosh(625/49) = 3.8918 - 3.2375 = 0.6543 the loss grows to ln 2 = 0.6931 for long edges (asymptotically 2 tau - d -> -2 ln sin(90 deg / 2) = ln 2): in this tree, one bend between straight runs of k edges, loss 2k tau - d: k = 1: 0.6543, k = 2: 0.6923, k = 3: 0.6931, k = 4: 0.6931, k = 8: 0.6931, k = 16: 0.6931 one bend between single edges of length tau, t = (m-1)/(m+1) with m = 7^e: tau = 1 ln 7: 0.6543, tau = 2 ln 7: 0.6923, tau = 4 ln 7: 0.6931, tau = 8 ln 7: 0.6931 a word that bends at every step (always P.children(w)[1]), length 40: n tau - d = 23.9150, 0.6132 per bend the 20 random words of length 40 (Rng(90040)): 641 bends, total loss 446.0305, 0.6958 per bend ~ item 1: 154 ms 2. Order reversals: hyperbolic against tree distance, 937-word ball of radius 4 around the root sources (Rng(90000).sample(937, 10)): [0,3,3,0] [2,0,5,1] [5,5,0,2] [3,4,0,3] [1,5,0,0] [3,5,2,0] [0,2,4,0] [5,3,0,4] [0,0,2,4] [2,4,1,1] comparisons (b, c) with d_T(a,b) < d_T(a,c), binned by hops (d_T(a,b), d_T(a,c)); reversed: d_H(a,b) > d_H(a,c); tied: d_H equal hops 1 < 2: 50 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 3: 50 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 4: 250 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 5: 250 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 6: 1250 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 7: 1250 comparisons, 0 reversed (0.00%), 0 tied hops 1 < 8: 6250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 3: 250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 4: 1250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 5: 1250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 6: 6250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 7: 6250 comparisons, 0 reversed (0.00%), 0 tied hops 2 < 8: 31250 comparisons, 0 reversed (0.00%), 0 tied hops 3 < 4: 1250 comparisons, 5 reversed (0.40%), 0 tied hops 3 < 5: 1250 comparisons, 0 reversed (0.00%), 0 tied hops 3 < 6: 6250 comparisons, 0 reversed (0.00%), 0 tied hops 3 < 7: 6250 comparisons, 0 reversed (0.00%), 0 tied hops 3 < 8: 31250 comparisons, 0 reversed (0.00%), 0 tied hops 4 < 5: 6250 comparisons, 126 reversed (2.02%), 0 tied hops 4 < 6: 31250 comparisons, 6 reversed (0.02%), 0 tied hops 4 < 7: 31250 comparisons, 0 reversed (0.00%), 0 tied hops 4 < 8: 156250 comparisons, 0 reversed (0.00%), 0 tied hops 5 < 6: 31250 comparisons, 862 reversed (2.76%), 0 tied hops 5 < 7: 31250 comparisons, 4 reversed (0.01%), 0 tied hops 5 < 8: 156250 comparisons, 0 reversed (0.00%), 0 tied hops 6 < 7: 156250 comparisons, 5986 reversed (3.83%), 0 tied hops 6 < 8: 781250 comparisons, 561 reversed (0.07%), 0 tied hops 7 < 8: 781250 comparisons, 39802 reversed (5.09%), 0 tied by hop difference d_T(a,c) - d_T(a,b): +1: 976550 comparisons, 46781 reversed (4.79%), 0 tied +2: 846300 comparisons, 571 reversed (0.07%), 0 tied +3: 195250 comparisons, 0 reversed (0.00%), 0 tied +4: 169000 comparisons, 0 reversed (0.00%), 0 tied +5: 38750 comparisons, 0 reversed (0.00%), 0 tied +6: 32500 comparisons, 0 reversed (0.00%), 0 tied +7: 6250 comparisons, 0 reversed (0.00%), 0 tied total: 2264600 comparisons, 47352 reversed (2.09%), 0 tied (0.00%); first source recounted with direct Q comparisons: equal tau * d_T - d_H over the 10 x 936 (source, word) pairs, by hops (min .. max): 1 hops: 0.0000 .. 0.0000 2 hops: 0.0000 .. 0.6543 3 hops: 0.0000 .. 1.6244 4 hops: 0.0000 .. 2.7245 5 hops: 0.0000 .. 3.8819 6 hops: 0.0000 .. 5.0654 7 hops: 0.6931 .. 4.6620 8 hops: 0.6931 .. 5.8486 all: 0.0000 .. 5.8486; exactly 0 (cosh d_H = cosh(d_T tau), a geodesic tree path) on 32 of 9360 pairs ~ item 2: 330 ms 3. Graph-edge MAP, exact Lorentz ranking, ties pessimistic (subtree root [0]; neighbours = parent and children; candidates = all other nodes) (a) perfect 5-ary, depth 3: 156 nodes, words of up to 4 letters, letters per level [1,1,1] MAP (pessimistic ties) = 1.000000 = 1/1; queries with AP < 1: 0/156 MAP (optimistic ties, contrast only) = 1.000000 farthest neighbour of any query: d = 1.9459 (cosh 25/7); nearest non-neighbour of any query: d = 3.2375 (cosh 625/49); queries whose nearest non-neighbour is as near as or nearer than their farthest neighbour: 0/156 exact ties between a neighbour and a non-neighbour: 0/156 queries, 0 (neighbour, non-neighbour) pairs; queries with any exact tie among candidates: 156/156 AP recomputed with Q Lorentz products and an explicit sort on 156 queries: equal ~ 97 ms (b) root with 25 children, m = 2, then 5, 5: 776 nodes, words of up to 5 letters, letters per level [2,1,1] MAP (pessimistic ties) = 0.995703 = 941107/945168; queries with AP < 1: 25/776 MAP (optimistic ties, contrast only) = 0.999141 farthest neighbour of any query: d = 3.8918 (cosh 1201/49); nearest non-neighbour of any query: d = 3.2375 (cosh 625/49); queries whose nearest non-neighbour is as near as or nearer than their farthest neighbour: 25/776 exact ties between a neighbour and a non-neighbour: 25/776 queries, 485 (neighbour, non-neighbour) pairs; queries with any exact tie among candidates: 776/776 AP recomputed with Q Lorentz products and an explicit sort on 111 queries: equal ~ 522 ms (b) root with 125 children, m = 3, then 5, 5: 3876 nodes, words of up to 6 letters, letters per level [3,1,1] MAP (pessimistic ties) = 0.995276 = (54-digit / 54-digit rational); queries with AP < 1: 126/3876 MAP (optimistic ties, contrast only) = 0.995351 farthest neighbour of any query: d = 5.8377 (cosh 58825/343); nearest non-neighbour of any query: d = 3.2375 (cosh 625/49); queries whose nearest non-neighbour is as near as or nearer than their farthest neighbour: 126/3876 exact ties between a neighbour and a non-neighbour: 120/3876 queries, 580 (neighbour, non-neighbour) pairs; queries with any exact tie among candidates: 3876/3876 AP recomputed with Q Lorentz products and an explicit sort on 126 queries: equal ~ 5305 ms (b') 25-ary at every level, m = 2, depth 2: 651 nodes, words of up to 5 letters, letters per level [2,2] MAP (pessimistic ties) = 0.260362 = 9869264220139/37905932634570; queries with AP < 1: 650/651 MAP (optimistic ties, contrast only) = 0.845473 farthest neighbour of any query: d = 3.8918 (cosh 1201/49); nearest non-neighbour of any query: d = 3.2375 (cosh 625/49); queries whose nearest non-neighbour is as near as or nearer than their farthest neighbour: 650/651 exact ties between a neighbour and a non-neighbour: 525/651 queries, 3260 (neighbour, non-neighbour) pairs; queries with any exact tie among candidates: 651/651 AP recomputed with Q Lorentz products and an explicit sort on 131 queries: equal ~ 457 ms why (item 1 distances): with one letter per child every neighbour is 1.9459 away and the nearest non-neighbour 3.2375. With m = 2, a child is 3.2375 from its parent if its last letter bends and 3.8918 if it is straight: the same distances as its own grandchildren (bent or straight last step) and its siblings under the same virtual prefix (90 or 180 deg), which therefore tie with the parent or come before it. With m = 3 the parent can be 3 tau = 5.8377 away. ~ item 3: 6381 ms 4. Float against exact, by depth n = 1..40: 20 random words from Rng(90000+n) and the straight word [0,...,0] ranking: 21-26 candidates per word (parent, grandparent, siblings, uncles, children, children of the straight child, children of the first sibling), ordered by the Lorentz product with the word: exact (Q), f64 point() (float placement), f64 exact (Q.toNumber), f32 (Math.fround throughout) a float order: inv = some pair of consecutive exact levels inverts; tie = no inversion, but two exact levels meet in a float tie; nf = a non-finite float32 product (and no inversion); fail = any of the three decoding: decode(point(w), n) and decode of the rounded exact point, on all 50 words of Rng(90000+n); decodeExact(pointExact(w), n) on every ranked word per depth: random = inv/fail counts of the 20 words, f64 point() | f64 exact | f32; straight = status, same order; decode = point() | rounded exact n= 1 random 0/ 0 | 0/ 0 | 0/ 0 straight ok | ok | ok decode random 50/50 | 50/50, straight ok | ok n= 2 random 0/ 0 | 0/ 0 | 0/ 0 straight ok | ok | ok decode random 50/50 | 50/50, straight ok | ok n= 3 random 0/ 0 | 0/ 0 | 0/ 0 straight ok | ok | ok decode random 50/50 | 50/50, straight ok | ok n= 4 random 0/ 0 | 0/ 0 | 0/ 0 straight ok | ok | ok decode random 50/50 | 50/50, straight ok | ok n= 5 random 0/ 0 | 0/ 0 | 6/ 6 straight ok | ok | inv decode random 50/50 | 50/50, straight ok | ok n= 6 random 0/ 0 | 0/ 0 | 18/18 straight ok | ok | inv decode random 50/50 | 50/50, straight ok | ok n= 7 random 0/ 0 | 0/ 0 | 20/20 straight ok | ok | inv decode random 50/50 | 50/50, straight ok | ok n= 8 random 0/ 0 | 0/ 0 | 20/20 straight ok | ok | inv decode random 50/50 | 50/50, straight ok | ok n= 9 random 0/ 0 | 0/ 0 | 20/20 straight ok | ok | inv decode random 50/50 | 50/50, straight ok | ok n=10 random 0/ 0 | 0/ 0 | 20/20 straight inv | inv | tie decode random 50/50 | 50/50, straight ok | ok n=11 random 0/ 0 | 0/ 0 | 20/20 straight inv | inv | tie decode random 50/50 | 50/50, straight ok | ok n=12 random 3/ 4 | 1/ 2 | 20/20 straight inv | tie | tie decode random 50/50 | 50/50, straight ok | ok n=13 random 14/15 | 13/14 | 20/20 straight tie | tie | tie decode random 49/50 | 49/50, straight ok | ok n=14 random 19/19 | 19/19 | 20/20 straight tie | tie | tie decode random 32/50 | 34/50, straight ok | ok n=15 random 20/20 | 20/20 | 20/20 straight tie | tie | tie decode random 24/50 | 29/50, straight ok | ok n=16 random 20/20 | 20/20 | 20/20 straight tie | tie | tie decode random 8/50 | 6/50, straight ok | ok n=17 random 20/20 | 20/20 | 20/20 straight tie | inv | tie decode random 2/50 | 3/50, straight ok | ok n=18 random 20/20 | 20/20 | 20/20 straight tie | inv | tie decode random 0/50 | 1/50, straight ok | ok n=19 random 20/20 | 20/20 | 20/20 straight tie | inv | tie decode random 0/50 | 0/50, straight ok | FAIL n=20 random 20/20 | 20/20 | 20/20 straight tie | inv | tie decode random 0/50 | 0/50, straight ok | ok n=21 random 20/20 | 20/20 | 20/20 straight tie | inv | tie decode random 0/50 | 0/50, straight ok | ok n=22 random 20/20 | 20/20 | 20/20 straight tie | tie | tie decode random 0/50 | 0/50, straight ok | ok n=23 random 20/20 | 20/20 | 20/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=24 random 20/20 | 20/20 | 20/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=25 random 20/20 | 20/20 | 20/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=26 random 20/20 | 20/20 | 20/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=27 random 20/20 | 20/20 | 19/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=28 random 20/20 | 20/20 | 20/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=29 random 20/20 | 20/20 | 18/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=30 random 20/20 | 20/20 | 18/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=31 random 20/20 | 20/20 | 17/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=32 random 20/20 | 20/20 | 3/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=33 random 20/20 | 20/20 | 4/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=34 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=35 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=36 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=37 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=38 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=39 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok n=40 random 20/20 | 20/20 | 0/20 straight tie | tie | nf decode random 0/50 | 0/50, straight ok | ok first depth with an inversion, random words: f64 point() 12, f64 exact 12, f32 5 first depth with an inversion, straight word: f64 point() 10, f64 exact 10, f32 5 first depth at which the float order fails (inv, tie or nf), random words: f64 point() 12, f64 exact 12, f32 5 first depth at which the float order fails (inv, tie or nf), straight word: f64 point() 10, f64 exact 10, f32 5 all 20 random words fail at every depth from: f64 point() 15, f64 exact 15, f32 7 first depth with a non-finite float32 product: 23 why the straight word mostly ties instead of inverting: [0]^n sits at (cosh n tau, sinh n tau, 0, 0), and the two coordinates are different doubles only at depths 1-9, 19 (rounded exact) and 1-9 (point()), for n = 1..42; at depth 19, 7^19 / 2 lies exactly halfway between two doubles (checked), so cosh rounds up and sinh down. Where both points have equal coordinates the float product is exactly 0: the products with all 4 candidates on the word's own line (exactly -25/7 or -1201/49) are 0 at every depth from 12 with point() and from 22 with the rounded exact points, so those exact levels meet in float ties. first depth at which float decode fails, random words: point() 13, rounded exact 13 first depth at which float decode fails, straight word: point() none to 40, rounded exact 19 decodeExact(pointExact(w), n) returned the word on 840/840 ranked words; the adjacent-level and all-pairs inversion tests agreed on every ranking ~ item 4: 30902 ms 5. Exact point cost by depth n = 1..60 (pointExact; max over the 4 coordinates and the 20 random words, or the straight word) every coordinate is an integer over 7^k, k <= n; columns: max numerator bits / max denominator bits n= 1 random 5/ 3 straight 5/ 3 log2 7^n = 2.8 n= 2 random 11/ 6 straight 11/ 6 log2 7^n = 5.6 n= 3 random 15/ 9 straight 16/ 9 log2 7^n = 8.4 n= 4 random 21/ 12 straight 22/ 12 log2 7^n = 11.2 n= 5 random 28/ 15 straight 28/ 15 log2 7^n = 14.0 n= 6 random 31/ 17 straight 33/ 17 log2 7^n = 16.8 n= 8 random 41/ 23 straight 44/ 23 log2 7^n = 22.5 n=10 random 52/ 29 straight 56/ 29 log2 7^n = 28.1 n=12 random 60/ 34 straight 67/ 34 log2 7^n = 33.7 n=14 random 71/ 40 straight 78/ 40 log2 7^n = 39.3 n=16 random 81/ 45 straight 89/ 45 log2 7^n = 44.9 n=18 random 91/ 51 straight 101/ 51 log2 7^n = 50.5 n=20 random 101/ 57 straight 112/ 57 log2 7^n = 56.1 n=22 random 111/ 62 straight 123/ 62 log2 7^n = 61.8 n=24 random 119/ 68 straight 134/ 68 log2 7^n = 67.4 n=26 random 131/ 73 straight 145/ 73 log2 7^n = 73.0 n=27 random 134/ 76 straight 151/ 76 log2 7^n = 75.8 n=28 random 139/ 79 straight 157/ 79 log2 7^n = 78.6 n=30 random 149/ 85 straight 168/ 85 log2 7^n = 84.2 n=32 random 159/ 90 straight 179/ 90 log2 7^n = 89.8 n=36 random 182/102 straight 202/102 log2 7^n = 101.1 n=40 random 197/113 straight 224/113 log2 7^n = 112.3 n=44 random 220/124 straight 247/124 log2 7^n = 123.5 n=48 random 240/135 straight 269/135 log2 7^n = 134.8 n=52 random 259/146 straight 291/146 log2 7^n = 146.0 n=56 random 275/158 straight 314/158 log2 7^n = 157.2 n=60 random 297/169 straight 336/169 log2 7^n = 168.4 numerator growth, depth 20 -> 60: random 4.90 bits per letter, straight 5.60 (2 log2 7 = 5.61); denominators 2.80 (log2 7 = 2.81) nonzero coordinates whose reduced denominator is exactly 7^n: 2520/4821; every denominator divides 7^n: true every Lorentz product checked in items 1-4 is an integer over 7^(n_a + n_b): 31269 products ~ n=10: 0.25 ms per exact point (random), 0.23 ms (straight); float point(): 1.4 us ~ n=20: 0.77 ms per exact point (random), 0.26 ms (straight); float point(): 2.5 us ~ n=28: 1.63 ms per exact point (random), 0.37 ms (straight); float point(): 3.0 us ~ n=40: 4.30 ms per exact point (random), 0.61 ms (straight); float point(): 4.8 us ~ n=60: 11.19 ms per exact point (random), 1.00 ms (straight); float point(): 6.8 us ~ item 5: 4266 ms 6. Exact ties item 2: comparisons with d_T(a,b) < d_T(a,c) and d_H(a,b) = d_H(a,c) exactly: 0 of 2264600 item 4: rankings with an exact tie among distinct candidates: 840/840 (5052 tie groups; 20937 candidates in 5993 exact levels); rankings whose nearest exact level is exactly the parent and the 5 children, all at cosh d = 25/7: 840/840 item 3 [5,5,5]: queries with an exact tie among candidates 156/156; with a neighbour/non-neighbour tie 0/156; a float evaluator (acosh of the float64 product of the rounded exact points, ties broken by node index) gives a different AP from the pessimistic exact evaluator on 0/156 queries (0 of them have a neighbour/non-neighbour tie) item 3 [25,5,5]: queries with an exact tie among candidates 776/776; with a neighbour/non-neighbour tie 25/776; a float evaluator (acosh of the float64 product of the rounded exact points, ties broken by node index) gives a different AP from the pessimistic exact evaluator on 25/776 queries (25 of them have a neighbour/non-neighbour tie) item 3 [125,5,5]: queries with an exact tie among candidates 3876/3876; with a neighbour/non-neighbour tie 120/3876; a float evaluator (acosh of the float64 product of the rounded exact points, ties broken by node index) gives a different AP from the pessimistic exact evaluator on 104/3876 queries (104 of them have a neighbour/non-neighbour tie) item 3 [25,25]: queries with an exact tie among candidates 651/651; with a neighbour/non-neighbour tie 525/651; a float evaluator (acosh of the float64 product of the rounded exact points, ties broken by node index) gives a different AP from the pessimistic exact evaluator on 433/651 queries (433 of them have a neighbour/non-neighbour tie) SETTLES: - Distances at t = 3/4: parent-child tau = ln 7 = 1.9459 (cosh 25/7); siblings at 90 deg 3.2375 (cosh 625/49); siblings at 180 deg and the straight grandchild 2 tau = 3.8918 (cosh 1201/49). One 90 deg bend between two edges loses 0.6543; the loss tends to ln 2 = 0.6931 for long edges; on long paths it is 0.6132 per bend when every step bends and 0.6958 on random words. - Hyperbolic distance reverses tree-distance order in 47352 of 2264600 comparisons (2.09%) on the radius-4 ball, only at hop differences +1 and +2; tau * d_T - d_H ranges from 0 to 5.8486. - Ranking must be exact. Float64 Lorentz ranking of the rounded exact points (L4(b)) first inverts an exactly strict order at depth 10 on the straight word and 12 on random words, and fails (inversion or float tie) on all 20 random words at every depth from 15; with the float placement point() the depths are 10, 12 and 15. Float32 first inverts at depth 5 (straight) and 5 (random); its products overflow from depth 23. Float decode(point(w)) first fails at depth 13 on random words (none to 40 on the straight word); decodeExact never failed (840/840). - Bit budget: at depth 28, <= 139-bit numerators (random), 157 (straight) and 79-bit denominators (7^28); at depth 60, 297 / 336 / 169. Every coordinate at depth n is an integer over 7^n and every Lorentz product an integer over 7^(n_a + n_b), so ranking by an integer key is exact. - Graph-edge MAP with pessimistic ties: perfect 5-ary subtree of depth 3 = 1/1 (0 neighbour/non-neighbour ties); root with 25 children (m = 2) = 0.995703; root with 125 children (m = 3) = 0.995276; 25-ary at every level (m = 2) = 0.260362 (optimistic 0.845473). Neighbour/non-neighbour exact ties appear only with virtual letters (25, 120 and 525 queries): the virtual-letter tie effect behind the doubt in L6(c). - L4 controls: float32 already inverts at depth 5; exact neighbour/non-neighbour ties exist in virtual-letter trees, and there a float evaluator with index tie-breaking differs from the pessimistic exact AP on 562 queries (0 in the perfect 5-ary tree). ~ total runtime 42.0 s