Full record

What each family has covered

Every bound the search has set, family by family: the family's status, then each finding with the experiments behind it, newest first. A finding says what was searched and what the scrambled copies of the carved text did under the same search.

198 experiments, 119 findings.

1 keyed periodic75 experiments

Keyed periodic systems at crib-surviving periods 27, 28, 29, 53+: Quagmire I-IV, autokey, progressive key

parked (five bounded sweeps since the pivot; re-enter with a new key source)

126 127 128Hand key schedules for an unknown key word (skip one letter every m, use each r times, forward then back, triangle or square steps, restart at each carved line) and the carved tableau rows as keys: no setting fits on standard or KRYPTOS tables; on free alphabets consistent at the coppers' rate, the skip rule's excess in a run cut short by its time budget (0126) gone with 30 fresh coppers (rank 6 of 31, 0127); tableau rows 6 of 24 known letters, coppers 5-8 (0128).
125The key forced at the known places read as running English under every dictionary keyword alphabet (plain, cipher or both sides, three rules, standard or keyword key alphabet; 3.7 million readings): K4 best -4.94, rank 18 of 51 among 50 coppers and below all 20 planted running keys from Carter (-3.82 to -4.78) (0125).
123 124Two unknown alphabets (Quagmire IV, any pair of orders), key length 1-26, exact: K4 fits only at 8, 13, 16, 19, 20, 23, 24 and 26, inside the 5-16 fitting lengths of 100 coppers; length 19 fits on only 3 of 100 coppers (0123), replicated as 26 of 1,000 with a look-elsewhere p of about 0.03, resting on the R under P at 28 and 66 (0124).
120At those two lengths the alphabet and the free key values climbed toward English with the known letters held (20 restarts x 200,000 moves): K4 -5.81 at 23 against 12 coppers -5.90 to -5.45, and -5.29 at 26 against -5.33 to -5.01; planted English beat every copper 3 of 6 times at 23 and 1 of 6 at 26 and was never recovered, so 23 is a weak test and 26 undecidable (0120).
119One unknown alphabet on both sides (any of 26! orders), periodic key 1-26, no move, decided exactly mod 13 and mod 2: in the added form no alphabet fits at any length, because S under S (33) and K under K (74) force key 0 in slots that another known letter shares at every length 1-26; in the Beaufort form only lengths 23 and 26 fit, against 0-15 and 1-11 fitting lengths on 100 coppers (0119).
118Ciphertext autokey continuing the carved stream from K3, lags 1-200, and the letter above on the plate: best 5 of 24, as on coppers (0118).
117The sculpture's texts as a running key plus any constant shift, with or without every type-1 move: no fit, deepest partial 8 as on coppers (0117).
109Keyword tables from any dictionary word (208,062 alphabets), plain, cipher or both sides, periodic key 1-26: nothing below 26 (0109).
103 104The copper or the message as its own key through K3's rule a + b i mod 97 (every a, b; mirror included): best 7 of 24 alone, as on coppers (0103); the message version dies on every setting (0104).
101An interrupted key, restarting at every word of the cribs (five splits), fits at no key length 1-26 (0101).
98The sculpture's running keys on an unknown mixed alphabet (Quagmire III, any of 26! orders), solved exactly: no fit on K4 at any of 80,676 settings, 0-57 on scrambled coppers (0098).
52 79 95Berlin-clock key schedules (0079): the Mengenlehreuhr's lit-lamp count, hour, minute, hour plus minute, five-minute row, or the HHMM digit stream, the clock advancing 1, 2, 3, 5, 10, 15 or 30 minutes per letter from every start minute of the day, added, subtracted or Beaufort with any constant, standard or KRYPTOS alphabets (725,760 trials): at most 8 of 24 residuals agree (16 mismatches), against 16-17 on ten shuffled coppers; no schedule fits. Weltzeituhr facet readings reduce to period 24 or a 24-letter ring, dead by the crib sweep and 0052. Riddle words (spelled numbers, units, compass and theme words, 93) dragged under all 376 crib-consistent periodic keys, clear of the cribs: none reaches 3 checks, random strings reach 4; low power (0095).
78Running keys from Sanborn's own writings (0078): his December 1989 letter to Agency employees (2,179 letters), the 1991 Cryptogram pages with Webster's dedication remarks and the Cryptolog story (4,887), and the August 2025 open letter (2,228), every offset, forwards and reversed, the 24 standard/KRYPTOS conventions and 61 vocabulary keyword alphabets on both sides (390 setups, 7.2 million trials): the fewest mismatches is 15 of 24, the same as on ten shuffled coppers (15-16); nothing within three.
77Both fingerprints in one construction (0077): every block of seven reshuffled by one of 5,040 orders, before or after ciphertext autokey with a constant shift, every lag, 32 classes, 30,965,760 exact cases: order A (autokey first) 5,760 / 168 / 6 at one / two / three checks against 5,640-6,792 / 168-282 / 0-24 on ten shuffled coppers; order B 8,640 / 426 / 12 against 7,872-11,400 / 198-438 / 0-36; nothing passes four checks; lag 21 never passes three.
73Plaintext autokey plus a periodic key, T[i] = s A(p[i-L]) + k[i mod P], every lag where two known letters pair (1-12, 30-52), periods 1-96, 32 standard/KRYPTOS classes plus 161 word alphabets (vocabulary, K1-K3 words, C13/C14 words) on both sides, 1,320 setups, 4,055,040 cases: survivors 3,171 / 135 / 12 / 0 at one / two / three / four checks against 3,069-3,784 / 76-152 / 0-6 / 0-1. The twelve three-check survivors are four coincidences counted at several periods: ROWS and SEVENTY alphabets at lag 8 each survive at periods 5, 8, 10, 20 and 40 (the same equations nest), plus DELIVERING at lag 47 and PRESENTLY at lag 33; lags 7 and 21 have no strong survivor (0073).
72A near-copy of the carved alphabet revives nothing. C15/C16 key source: alphabets keyword-mixed from the 32 compass points and the 2025 letter's words (POINT, POWER, RESIDES, WITHOUT, DIRECTIONS, BEARING and others), 40 new alphabets plus standard and KRYPTOS (1,764 pairs): periodic keys dead at periods up to 24; 25 none (controls 0-2), 26 in 87 cases (17-393); no word in 10,671 fragments at periods up to 29; ciphertext autokey with a constant 19 / 0 at one / two checks (7-19 / 0-1); autokey plus a periodic key up to 26: 998 / 27 / 0 at one / two / three checks (852-1,094 / 15-33 / 0-5) (0072).
71The substitution tally counts the fitted residuals as checks, so it overstates redundancy by one or two; the verdict does not change. The width-21 lead does not rescue ciphertext autokey through a single carving error. C27(c), a tableau slightly different from the carved one: the 925 alphabets one swap or one move from KRYPTOS, used for both sides, or against the standard alphabet on either side, Vigenere or Beaufort: periodic keys at periods up to 25 are dead under every variant; 26 survives in 21 cases, against 0-548 on ten shuffled coppers; no forced fragment among 16,671 at periods up to 29 holds a word or a K1-K3 window; ciphertext autokey with a constant shift, every lag, key letter on the variant or the standard alphabet: survivors 391 / 1 / 0 at one / two / three checks against 38-838 / 0-292 / 0-3 (0071).
68A letter-feedback stage tying the two fingerprints together is dead under the cribs. Ciphertext autokey with a constant shift, lags 1-96, 32 classes, the copper with one letter wrong (97 x 25, decided exactly: only the cribs at j and j + L can move), missing (98 insertions) or extra (97 deletions): no survivor with three or more checks; exact-copper survivors 3 at one check against 0-4 on ten shuffled coppers, shifted-copper survivors 225 at one check (16-480), substitution survivors 6 at two counted checks (2-14); lag 21 has none (0068).
67Two-lag ciphertext feedback, T[i] = s1 A(c[i-a]) + s2 A(c[i-b]) + k[i mod P], 1 <= a < b <= 74, four signs, P 1-30, 16 classes: 1,933,760 cases with at least one check (the record's 5.2 million counted cases with none); survivors by checks 8,203 / 254 / 15 / 1 at 1 / 2 / 3 / 4 checks against control means 8,209 / 231 / 7.7 / 0.5 (maxima 8,344 / 254 / 12 / 2); the 16 three-or-more-check survivors (listed after the run) are scattered over lags, periods and classes with no cluster; distances 1 and 21 survive only with one check at period 26 (0067).
62 65 66The width-21 lead from 0062 does not open ciphertext autokey on these tables. Ciphertext autokey at lag L plus a periodic key of period P (the two-component form 0065 points at), L and P 1-96, 32 classes (plain, cipher and key alphabets standard or KRYPTOS, Vigenere or Beaufort, either sign), solved exactly by subtracting the known carved letters: 221,184 cases with at least two equations; no case with three or more redundant checks survives, and the survivors with one or two checks match ten shuffled coppers (e.g. 5 at P up to 26 with two checks, against 0-6); lag 1 with period 7 and lag 21 at every period are dead; no fragment matches a word (0066).
64Family 1 stays closed with these alphabets too. Ciphertext autokey, lags 1-95 (at least six cribs past the primer), on the 102 alphabets from K1-K3 and C13/C14 words, key alphabet standard, KRYPTOS, plain or cipher, three modes: no setting satisfies every crib past the primer; the fewest mismatches is 3 (lag 68, six cribs checked), against 2-3 on ten shuffled coppers (0064).
53Periodic tableau, periods 1-52, three modes, plain and cipher alphabets keyword-mixed from the 100 new alphabets the K1-K3 plaintext words and the C13/C14 words (EGYPT, CAIRO, WALL, MESSAGE, DELIVER...) give, plus standard and KRYPTOS (10,404 pairs), against ten shuffled coppers: periods up to 25 and from 33 to 50 have no consistent case; 27-29 are consistent for every pair, as always; 26 and 52 run high (2,139 against a control mean of 1,166) because at those periods only places 22 and 74 share a slot and keyword alphabets tend to keep E,F and K,R equally spaced, so one equation is satisfied in bulk; no forced fragment at 26-29 matches a word or a K1-K3 window (0053).
49Periodic tableau keys at 27-29 are now dead under every alphabet pair the vocabulary can build. Shaped keys of length 27-29 built from the vocabulary (two words joined, joined with X or Q, a word and a mirror, a word doubled: 272 keys, far fewer than the record's estimate of 20,000), under 63 x 63 alphabet pairs, three modes and three key readings: 9.7 million trials, none within three mismatches (0049).
48Key completion at 27-29 with both alphabets keyword-mixed, all 3,721 pairs reached, 33,489 climbs: best -6.14 at 29, -6.36 at 27, -6.52 at 28, against -6.36 to -6.88 on shuffled K4; noise (0048).
47Carter's book and the sculpture's texts as running keys with one alphabet keyword-mixed and the other standard or KRYPTOS, key letters on either order, three modes, every offset both ways: 858 million trials, none within two mismatches (0047).
45Running key from the K1-K3 plaintexts and ciphertexts under keyword-mixed plain and cipher alphabets, key letters on the standard or KRYPTOS order, three modes, every offset both ways: 146.9 million trials, none within two mismatches (0045).
42The key alphabet drops out of decryption, so only plain-cipher pairs matter; both mixed at once is the one uncovered corner. Progressive key (per-repeat form), periods 1-30, steps 1-25, plain and cipher alphabets keyword-mixed from every vocabulary word, three modes, 8.9 million combinations: periods 1-24 die at every step under every pair; 25-30 survive at chance rates; no fragment matches (0042).
35These are the first candidates to pass the gates, by construction; the harness kept the period-95 text as champion with loss 4.4 because the key-cost term barely charges a 95-letter random key. Scorer case filed in #11. Same climb at 27-29 with one alphabet keyword-mixed from any vocabulary word and the other standard or KRYPTOS, 2,232 climbs: best -6.6 at every period, shuffled K4 -6.6 to -6.7 (0035).
34Key completed by English climb at the surviving periods under 24 conventions: at 27-29 the best text scores -7.3 to -6.7, the same as on shuffled K4 (-6.9 to -6.5) and far from English (-3.6); at 53-95 the climb reaches -3.3 on K4 and -3.2 on shuffled K4 by overfitting 29-71 free slots (0034).
31The copper off by one letter, 195 variants (a letter deleted at each position, an unknown inserted at each): the chained plaintext autokey has no consistent lag up to 25 under any variant and convention, and no two-word composition of total length up to 20 is consistent under any variant (0031).
30Self-referential keys: the ciphertext as its own key at every index rule a i + b (every lag and every mirror of the carved text), 223,488 rules x 24 conventions, none within five mismatches; the message keyed by itself backwards from any of 97 centres, chain broken under every convention (0030).
29Bean's closeness as a test over 184,512 alphabet pairs: the smallest largest circular shift among the 24 cribs is 8, the same as on ten shuffles of K4 (7-9); the closeness is a property of the ten KRYPTOS-letter cribs only and no alphabet pair extends it (0029).
25Both autokeys with keyword-mixed alphabets from the vocabulary (63 plain x 63 cipher x 3 key choices x 3 modes x 96 lags, 3.4 million cases): ciphertext autokey never within three mismatches; plaintext autokey chained closure dead at lags 1-24 and 33-49 under every triple, and the large closures at other lags read as noise against ten shuffled controls (0025; the run took 46 minutes, over budget, because the sweep did not check the clock).
22Porta, the reciprocal half-alphabet tableau, over 248 alphabet orders: no order puts all 24 crib pairs in opposite halves, the best leaves 7 in the same half (0022).
21The width-21 trace is not a period-21 or period-42 tableau key on a vocabulary alphabet. Running key from Carter's Tomb of Tut-ankh-Amen vol. 1 (284,914 letters as scanned), both directions, every offset, 24 conventions: 13.7 million trials, none within two mismatches (0021).
20Periods 1-52 re-swept with plain and cipher alphabets from the four Friedman mixings of every vocabulary word (248 alphabets, 184,512 pairs x modes): periods 21 and 42 have no survivor; 24 has one and 26 has 7,454 against 19 and 7,204 on shuffled K4, so chance; no forced fragment reads as English or matches a word (0020).
17Running key from the sculpture's own texts: the carved tableau by rows, by columns and with its index column, the Morse plaintexts in carved and reversed order, and the four ciphertexts as one 865-letter panel, each forward and backward, every cyclic offset, 24 conventions: 162,912 trials, none with fewer than four crib mismatches (0017; the record says 869 letters, the panel is 865).
12Period 13 is dead at every step. Open: alphabets from words outside the vocabulary; Plaintext autokey, chained to closure from the cribs: lags 1-25 and 30-52 dead under all 24 conventions; lags 26-29 force about 60 further letters that read as noise (checked after the run, not in the record); lags 53-96 force too little to decide (0012).
11The forced key is not a readable word under any alphabet built from the puzzle's vocabulary. Progressive key, per-repeat or per-letter, steps 0-25, periods 1-97, 24 conventions: periods 1-25 and 31-51 die at every step; 26 lives at 10 steps, 30 at 7, 27-29 at all; no fragment matches (0011).
10With plain and cipher alphabets mixed by keyword from any of 61 vocabulary words and key alphabets from all four Friedman mixings (248), 5.9 million stretches: the best, EFORMESTATU at -3.81, equals the best from shuffled K4 letters, -3.80 (0010).
9The 13-letter key stretch forced by EASTNORTHEAST and the 11-letter stretch forced by BERLINCLOCK, scored as English under all 24 conventions, with the copper exact or with one letter inserted or deleted anywhere: the best of 9,408 stretches is 2.3 standard deviations above noise and 3.0 below English, which is what the best of noise looks like (0009).
5Periodic tableau with separate alphabets (Quagmire I-IV): periods 1-26 and 30-52 excluded under all 24 conventions; 1,152 surviving (convention, period) pairs force key fragments that match no vocabulary word or K1-K3 window (0005).
4Running key from K1-K3 plaintexts and ciphertexts, both directions, all cyclic offsets, 24 conventions: 165,216 trials, zero survivors, best 8/24 (0004).
3Plaintext autokey excluded at lags 1-11 and 30-52; lag 12 consistent in one convention on one check; 13-29 and 53-96 undecided (0003).
2Ciphertext autokey excluded (24 conventions x 96 lags, 0002).

2 masked plaintext8 experiments

Masked plaintext, then a simple keyed cipher

active

92The message written into a grid along one of 30 hand paths (snake, spiral, diagonal, rows, columns, each reversed) and read along another, widths 2-96, 43,579 distinct routes, with a periodic key 1-26 before or after: nothing below period 21; distinct check sets at 21-23 inside the scrambled-copper range (0092).
75Doublets as nulls (0075): QQ (26-27), SS (33-34) and TT (68-69) carry the different known pairs NO, ST, IN, so no reading of the doublets as fillers or of 97 - 6 = 91 survives; one letter of BB (19-20), SS (43-44) and ZZ (47-48) struck out, in all 7 combinations, then a periodic key (periods 1-52), ciphertext autokey with a constant, or both, 32 classes: no combination passes more than chance; with a null between the cribs, period 26 is consistent in 8 cases only because the two cribs then share no slot (zero checks), exactly as on every shuffled copper.
74The record's second prediction, that nothing below 26 would appear, was wrong: striking letters before BERLINCLOCK brings it nearer and makes 24-25 consistent at the chance rate. Forced fragments at 24-25 were not matched against words, a gap left open. Masks that change letters, in their simplest form: every affine map of the standard or KRYPTOS alphabet (11 multipliers each, Atbash included; the additive part is absorbed by the key), as a decimated plain alphabet before the table or a decimated cipher alphabet after it, 110 pairings: periodic keys dead at periods up to 25, 26 in 8 cases (0-8 on ten shuffled coppers), no word in 668 fragments; ciphertext autokey with a constant 79 / 2 at one / two checks (33-56 / 0-5; the one-check excess comes from lags with only two equations, counted once per key alphabet); autokey plus a periodic key up to 26: 3,682 / 128 / 0 at one / two / three checks (3,723-4,152 / 41-151 / 2-16) (0074).
57Nulls chosen by letter: every copy of one letter (26) or of a pair (325) struck outside the crib ranges, cribs shifted, then a periodic tableau on standard/KRYPTOS alphabets, three modes, periods 1-52: smallest consistent period 24 for 18 masks, 25 for 117, 26 for 168, against 17.1, 104.6 and 174.3 per ten random strike-outs of the same size; the masks behave exactly like random deletions (0057).
23Masks that move the known letters without changing them, then a periodic tableau under 24 conventions: reversal, a null before every k-th letter for k = 2-24 at every phase, 1-24 nulls first, and reversal combined with each (648 masks): smallest live period 26, forced fragments at periods up to 30 score below ten shuffled controls, no vocabulary match (0023).

3 two component key18 experiments

Two-component key with a combiner (Scheidt's CKM habit)

active

115A puzzle word (171) plus any dictionary word of 3-24 letters, every start, 24 rules: removing the puzzle word never leaves a key repeating at 3-24, so no second word is forced (0115).
80 87 96A column word plus a row key on a grid of width 2-48 (key = a[col] + b[row]; 0087 called it the construction that lines doublets up by column, which 0096 showed false: the row letter cancels inside a row, so doublets fall as under a plain key of the column word's length), one alphabet, either sign, solved exactly: dead on the copper as carved at every width (width 7: 14 checks), at the scrambled-copper rate after or before 0080's moves (0087).
88The same with the second shift changing every m letters (m 2-50, any offset, word length 2-31), including one shift per 31-letter carved line: dead (0088).
61A word (vocabulary, K1-K3, C13/C14; 416 words) plus a short second key from the misspelt letters and the raised letters (every ordering of QUA, LOE, QUAE, LOEE, YAR, and DYAHR; 70 keys), either sign, letters on standard or KRYPTOS, plain and cipher standard or KRYPTOS, three modes: 451,440 fully determined settings, best 8 of 24 known letters, against 7-11 on ten shuffled coppers (0061).
59A one-letter carving slip does not rescue the two-word key on vocabulary tables. Two words in two passes on the 102 alphabets keyword-mixed from K1-K3 words and the C13/C14 words, total length up to 24: 1,747,872 cases, 2,042 consistent, the smallest at total length 21 (2), then 8 at 22, 306 at 23, 1,726 at 24, as chance predicts; no word match (0059).
50The copper off by one letter (each deletion, each inserted unknown: 48 distinct crib tuples) through the two-word test on 63 x 63 vocabulary alphabet pairs, total length up to 22: 22.1 million cases, survivors only at 21 and 22 (64 and 850) against 54 and 824 on a shuffled copper, which also reached 20; two word matches were 96-letter variants that fail the length gate (0050).
43Two words in two passes with keyword-mixed plain and cipher alphabets from the vocabulary, total length up to 24: 666,792 cases, survivors begin at total length 22 at chance rates, no word match (0043).
38Seventeen more digit readings from the story (dedication, Wall, Carter dates in two orders; 1990, 1989, 1986, 1922; the Morse E-counts; the four parts' letter counts; 97, 98): 46,400 cases, shortest consistent word length 26, forced words noise (0038).
24A word of length 2-30 plus the K2 coordinates as a repeating digit key, seven readings of the digits, every alignment, either sign, standard or KRYPTOS alphabets: shortest consistent word length 26, forced words below ten shuffled controls, no match (0024).
14Two key words in two tableau passes, lengths 2-30 with lcm outside the dead periods, plain and cipher alphabets standard or KRYPTOS, both plaintext signs, both combiner signs: 4,896 cases, 3,498 consistent, but none with total word length under 25, where the 24 cribs leave no redundant check; no survivor's words are in the vocabulary or K1-K3 (0014).

4 visual spatial8 experiments

Sanborn's "visual systems"

active

108K4 read from behind (each carved line right to left), with the line order reversed, and both, before or after the table: periodic keys dead below 26, no autokey lag with 3+ checks (0108).
39 40 89Free cubes searched by annealing reach 17 of 24, the same as on shuffled K4 and the same as the annealer manages on a planted cube, so free cubes stayed open until 0040 solved them exactly (0039 is the same run, whose summary line crashed on the empty result): the 24 known letters tie the cube's 81 coordinate variables into classes, and under the rule that all 27 symbols have different addresses no cube at all is consistent with K4 at any period 2-15 and any alignment; a planted trifid text is recovered exactly. Trifid is dead. The screen as a key by position: K4's carved lines laid over every window of four lines of the tableau plates (labels, extra L, header and footer) and the cipher plates, straight or folded at the centre, column shifts -4 to 4, 12 conventions, 10,800 overlays: best 5 of 24 known letters against 5-7 on ten scrambled coppers (0089).
33Playfair, bifid and the four-square family are dead because all 26 letters occur. Trifid, the 27-cell cube with '?' as the 27th symbol: cubes built from every vocabulary word by four mixings with '?' first or last, periods 2-20, every alignment, 104,082 decryptions, best 7 of 24 crib hits (0033).

5 gromark lfsr9 experiments

Gromark and LFSR variants beyond base 10

exhausted-bounded (two-term Gromark rules, any mixed alphabet: base 10 lags up to 6, bases 2-16 and 26 lags up to 5; three-term rules with a free alphabet closed by 0107)

107Three-term rules with any mixed alphabet, base 10, all 20 rules with lags up to 6, every primer: none consistent (0107).
69 76Free-alphabet Gromark in other bases (0076): digits mod B for B = 2-9 and 11-16 (two-term rules, lags up to 5) and B = 26 (lags up to 4), 146 base-rule settings, 15,135,782 primers, four forms: no consistent primer in any base, on K4 or on ten shuffled coppers; a planted base-7 text is recovered. With 0069, two-term Gromark is dead for every alphabet and every base tried.
69Gromark with a free mixed alphabet, base 10, every two-term rule with lags up to 6 (15 rules including Rogers' 4-5), every primer (5,432,100), four forms (M[P+d]=c, M[P-d]=c, M[C-d]=p, M[C+d]=p): no primer gives a one-to-one alphabet consistent with the 24 known letters, on K4 or on ten shuffled coppers; a planted Gromark text is recovered uniquely (0069).
54Gromark-type digit keys on the 102 alphabets keyword-mixed from K1-K3 words and the C13/C14 words: the smallest number base any of the 10,404 pairs admits under any mode is 17, against 17-19 on ten shuffled coppers, so no pair admits base 15 or less and no growth rule needs testing (0054).
46Two- and three-term rules with lags up to 13 (364 rules) in the bases 16-26 each triple still allows: no rule passes six checks under any triple (0046).
28Bases 2-9: under every one of the 184,512 alphabet triples the largest forced shift is at least 15, so every base up to 15 is dead before any recurrence check, including the base-2 and base-8 streams that reproduce the width-21 repeats (0028).
28Base-10 Gromark is dead on these tables as on the vocabulary ones (0028).
13Running numerical key k[i] = k[i-a] + k[i-b] mod base, every lag pair 1 <= b < a <= 12, bases 10 and 26, plain and cipher alphabets from standard, KRYPTOS and the four Friedman mixings of every vocabulary word (248 each), three modes: 184,512 alphabet triples x 66 rules; zero triples make all 24 shifts digits; zero rules fit the 13 consecutive shifts at 22-34 and the 11 at 64-74 with six or more checks (0013).
Same-letter cribs fix exact digit differences, so the test is decisive. Gromark in base 10 is dead whatever the alphabet.

6 transposition then tableau49 experiments

Matrix transposition stage(s), then the tableau last (Sanborn, CNN 2005: "double matrixes and triple matrixes... then went back to this table to do the ultimate")

active

113 114Rail fence and redefence (12,784 routes) and AMSCO under every dictionary word (891,360 routes), periodic key before or after: nothing below 21-22; AMSCO's period 23 above the coppers by distinct sets (15 vs 1-9) is within the expected count of chance cells (0113, 0114).
111 112Myszkowski keyed by every dictionary word with a repeated letter (171,995 keys), and double columnar with one dictionary word twice (229,634), periodic key before or after: nothing below 21 (0111, 0112).
110True keyword columnar from every dictionary word (229,634 orders), periodic key before or after: nothing below 21 (0110).
102The period-3 excess against 400 shuffles is p 0.043 (0.048 corrected), against 400 width-21 column permutations p 0.10: not a signal (0102).
100 102Replication of the marginal cells against 50 fresh coppers: all luck except Quagmire IV at period 3 after a move (12 distinct sets vs 0-11), near p 0.24 after selection; see 0102 (0100).
99Two quarter turns in mixed senses at every pair of widths (53,743 new routes), periodic key before or after: nothing below period 20, at the copper rate by settings and by distinct check sets (0099).
97Unknown mixed alphabets (Quagmire III, IV) with a periodic key 3, 7 or 21 after or before every type-1 move: at the copper rate except IV at period 3 after the move, 12 distinct equation sets against 1-11 (marginal); weak test at 7 and 21 (0097).
94One carved letter extra or missing at every place outside the cribs (96 or 98 letters), every single move on the true length, periodic key 1-26 before or after: nothing below period 20, all periods at the copper rate (0094).
93One carved letter wrong through the same library with a periodic key: nothing below period 14 with one miss allowed; smallest such period 20 after the move and 24 before, as on scrambled coppers (0093).
79 91The Berlin clock schedules of 0079 (60,480) joined to the same library, both orders: no full match, deepest partial 7 checks as on all ten scrambled coppers (0091).
90Running keys from the sculpture's texts, K1-K3, Sanborn's letters and Carter's book joined to every route of the type-1 library (27,852 distinct), before or after the table, 12 conventions: no full match, no whole crib stretch, deepest partial match 7 checks as on scrambled coppers (0090).
85 86Rerun with the true reading (0085): single and double keyword columnar from 201 word sequences, periodic key 1-26 either side and both autokeys at every lag, both orders: chance. The ? as a 98th symbol sent to the front by grids of 98 cells (any corner, one or two grids), K3's rule mod 98 or seven-column shuffles (300,651 routes), periodic key 1-26 or both autokeys: chance (0086).
84Type 2: double columnar with one stage of width 7 in all 5,040 orders and the other plain 2-30 or word-keyed (true keyword reading, repeats dropped or whole word), either first, 2,548,223 routes, periodic key 1-26 before or after: chance (0084).
83Three cw quarter turns at every width triple 2-96, forward and inverse (1,714,024 routes), periodic key 1-26 before or after: at the scrambled-copper rate; one coincidence at period 18 decrypts to noise with a non-word key (0083).
82Hand padding (blanks at the top of the grid, or the first grid's blanks carried as fillers through the second turn), one and two turns at every width and short-row-on-top columnar, 120,871 routes, with a periodic key 1-26 before or after: nothing live below period 22, as on scrambled coppers (0082).
81The same routes (27,752 distinct) joined to plaintext or ciphertext autokey, both orders, lags 1-96, one alphabet throughout: nothing at five checks, four at exactly four (chance), counts at the scrambled-copper rate (0081).
6 7 16 19 26 37 58 63 80 81Correction (issue #26): the keyed-columnar subsets of 0006, 0007, 0016, 0019, 0026, 0037, 0058, 0063, 0080 and 0081 read the columns in the inverse of the key order; their keyword bounds hold for that reading only.
80A repeating key on both sides of the move (substitute, move, substitute), same alphabet both stages (standard or KRYPTOS), either sign: K3's rule at all 9,312 settings, one quarter turn cw/ccw and two cw turns at widths 2-96, plain and vocabulary-keyed columnar, forward and inverse (26,158 distinct landing patterns), key lengths 2-14 each, solved exactly as a bipartite system: no case with five or more checks; survivors at every level at the scrambled-copper rate (0080).
70Family 6 re-entered once for the width-7 fingerprint: every block of seven reshuffled by one of the 5,040 orders, before or after a periodic tableau, periods 1-52, 8 classes (4,193,280 cases): periods up to 24 are dead under every order; 25 survives only twice (order B), against 0-3 on ten shuffled coppers; 26 survives at the control rate (A 1,800 against 1,200-2,160; B 212 against 128-276); 27-29 always; no period is singled out at multiples of 7; no forced fragment at 25-26 holds a word of five letters or more (checked after the run) (0070).
7 62 63The prediction that nothing below 26 would live was wrong by 16 pairs out of 355,008, with no control run to say whether that is chance; the fragments do not read as words. Transposition composed with ciphertext autokey, the one recipe that makes the width-21 repeats (0062), in both orders: 3,107 routes (0007's set plus keyed columnar from K1-K3 and C13/C14 words) x lags 1-95 x 24 conventions x 2 orders, 14.2 million exact checks: no full match; best 8 cribs past the primer with the transposition first, 9 of 24 with the autokey first (0063).
58Every transposition-only reading is excluded without enumerating keys. Periodic tableau then keyed columnar transposition, single or double, column orders from every K1-K3 word (3-20 letters) and the C13/C14 words, forward and inverse (14,792 routes x 24 conventions, 355,008 pairs): smallest live period 27 for 340,640 pairs, 26 for 14,352 and 25 for 16 (a few keyed doubles such as DESPARATLY then REMAINS, INFORMATION then ENCUMBERED); no forced fragment at 25-30 matches a vocabulary word or K1-K3 window (0058).
51Pure transposition, any permutation of the carved letters (C9a, KRYPTOS as a transposition key C8d): the 24 cribs need three Es and the copper has two, in the 97 and 98 readings and under every one-letter deletion; only a carved letter that should be an E rescues it (93 places) (0051).
41Route transpositions, spirals from every corner in either turn, snakes by rows and columns, diagonals, widths 2-48, forward and inverse (1,128 routes): before the tableau the smallest live period is 23 and the best forced key sits inside ten shuffled controls; after a substitution no route passes the one-to-one test, nor does any shuffle (0041).
37Three stages sampled: 200,000 random triples from 211 stage choices (plain columnar 2-30 either way, quarter turns, keyed columnar from vocabulary words), 8 conventions: smallest live period 20 (one triple; a shuffled control reached 20 too); best forced key INDERSH at -3.61 against thirteen full-size shuffled controls with maximum -3.63 and median -3.94, so chance by a whisker, one in fourteen (0037).
32A grid keyed by row, one key letter per row of width w for w = 2-48 and any key length, plain and cipher alphabets from the standard, KRYPTOS and 61 keyword-mixed alphabets, three modes: no width survives the within-row agreement test under any alphabet pair (0032).
26Two keyed columnar stages from every ordered pair of vocabulary words, each forward or inverse (15,376 routes), then the tableau: smallest live period 23; best forced key -3.80 against ten shuffled controls ranging -4.32 to -3.58 (0026).
16All 23,930 routes with the forced key scored as English at live periods up to 40: best UHAVETHE at -3.55 under a K3 rule at period 40; twelve shuffles of K4 give a maximum of -3.54 and a median of -3.81, so it is chance (0016, with ten extra shuffles run after the record).
8The period test has little power in this order; it needs English scoring or longer keys. K3's own rule with every shift (a + b i mod 97, 9,312 rules, both directions) and double rotations of a grid at widths 2-50, before the tableau: 504,600 pairs, smallest live period 22, zero matches (0008).
7Reverse order, tableau first then the same 2,905 routes: because the key slots follow plaintext positions, periods 1-25 die on structure alone, 26 survives in 4% of pairs by one coincidence, and no route gives a vocabulary or K1-K3 key at periods 26-30 (0007).
6Columnar routes (plain widths 2-96 in both column directions, forward and inverse; keyed columnar from every vocabulary word, both directions; every double of plain forward columnars at widths 2-50: 2,905 routes) then a periodic tableau with separate alphabets: over 69,720 route/convention pairs the smallest period the cribs allow is 23; no route allows a period under 23; forced fragments for live periods up to 30 match no vocabulary word or K1-K3 window (0006).
Family 6 rests again. Re-entered for type 1 of directions.md (2026-10-15 evening).

7 construction simulation6 experiments

Construction simulation: generate test ciphers by candidate systems and ask which reproduce K4's width-7 doublet column and width-21 repeat fingerprint (Oranchak's Zodiac 340 method)

active

65 96Recipes built to line doublets up in sevens, 4,000 samples each (0065): ciphertext autokey at lag 1 plus a period-7 key makes the width-7 doublet column in 1.5% of samples, against 0.1% for random letters and 0.3% for shuffled K4; with periods 14 and 28 it falls to 0.6-0.7%. A period-7 key with two equal neighbours, block-of-7 reshuffles in either order, a structured period-28 key and a width-14 read-out all stay at the calibration rate. The pre-registered 5% was not reached: ciphertext feedback at lag 1 with a 7 in the key is the only mechanism found that lines doublets up, and it does so weakly. Two-stage recipes from experiments 80-88 (31, 20,000 samples each): none makes both marks together more than 2 times in 20,000; column at most 1.4%, repeats at most 1.5% (0096).
18 27 62They still need a construction that reproduces them; 0018 and 0027 found none among the generators tried. Construction simulation aimed at the two confirmed fingerprints, 4,000 samples per generator from Carter's book and K1-K3: no generator left both marks in any sample. Only ciphertext autokey at lag 21 raises the width-21 repeats far above chance (5.4% against 0.2% for random letters), and Vigenere at period 7 raises them to 1.5%; nothing raises the width-7 doublet column above 0.3%, against 0.2% for random letters (0062).
4 5 56Look-elsewhere test of the two fingerprints against 10,000 letter-preserving shuffles, widths 2-48: the width-7 doublet column (5 of 6 doublets in one column) has nominal p 0.0005 and corrected p 0.005 over all widths (0.018 among the 729 shuffles that also have six doublets); the width-21 vertical repeats (11) have nominal p 0.0004 and corrected p 0.010. Both are K4's strongest widths. The pre-registered prediction (both unremarkable after correction) failed: the features are real properties of the carved text, not artefacts of scanning (0056).
27Round two, sixteen generators built to treat letters 7, 21 or 42 apart alike: none reproduces the width-7 doublet column (p at most 0.005); only Gromark base 2 reproduces the width-21 repeats (p 0.34), as Bean predicted for even bases with a five-digit primer; period-21 free alphabets reach only p 0.018 (0027).
18Thirteen generators, 5,000 samples each, ten statistics: no family is singled out; index of coincidence, doublets, top count and singletons fit every polyalphabetic family. The width-7 doublet column has p about 0.002 under every generator including random letters. The width-21 vertical repeats have p about 0.005 under every substitution-last generator and p near 1 under pure transposition and monoalphabetic, so they come from English neighbours kept 21 apart (0018).
2Ciphertext autokey is dead under the cribs at every lag (0002), so the repeats point at a ciphertext-feedback stage combined with something else, and the doublet column remains unexplained.

8 hill polygraphic10 experiments

Hill and other polygraphic ciphers (the HILL column, Bauer 2016-17; C21)

active

116Hill 2-4 with a periodic tableau key 1-26 in either order, solved exactly: nothing alive; fully decided for most key lengths of 2x2 and 3x3 and for 1, 2, 4, 8 of 4x4 (0116).
105Written as decryption, every known letter is one equation per row, partial blocks included: sizes 2-4 plain and affine, every alignment, four numberings, all dead; 5x5 affine underdetermined (0105).
60The K2 coordinates as a Hill matrix (C23b): 72 matrices, 2x2 and 3x3 from every window of the number and digit readings, forward and reversed, by rows and columns, four numberings, every alignment, both directions: at most 4 of 24 known letters, against a best of 7 (means 2.7 and 3.1) from 2,000 random invertible matrices per size (0060).
55Hill 2x2-4x4, plain and affine, both directions, every alignment, solved exactly from the whole crib blocks under 200 numberings (A=0 and A=1 on 100 alphabets keyword-mixed from K1-K3 words and the C13/C14 words): 7,200 settings; no 2x2 or 3x3 setting is consistent; 557 4x4 settings are consistent only because four or five blocks leave no redundancy, and every decryption fails the partial-block cribs (best 21/24) (0055).
52Structural check on the ciphertext: K4 uses all 26 letters (Y and M once each) and has 97 letters, so every system with at most 25 output symbols (Playfair, two-square, four-square, bifid and Polybius on 5x5, ADFGX, ADFGVX, a 24-letter Weltzeituhr ring, C12b) and every digraphic system without a null is excluded, whatever the key (0052; the record's walkthrough puts I at place 16, it is 17).
44With the letters numbered by the keyword mixing of every vocabulary word, 0- and 1-based (126 numberings): 4,536 settings, 369 consistent on their whole crib blocks, all 4x4 with no redundancy; the best decryption gets 21 of 24 known letters, so every one breaks the partial-block cribs (0044).
36Playfair is dead at both alignments: the ciphertext digraphs SS at 33-34, TT at 69-70 and QQ at 26-27 cannot occur in Playfair. Bauer's open variants, blocks chained to the previous carved block or message block, summed with the previous message block, or alternating two matrices, sizes 2-4, four numberings, every alignment: only two 4x4 ciphertext-feedback settings are consistent with their whole crib blocks, with no redundancy, and both break the eight partial-block known letters (0036).
15Hill 2x2, 3x3 and 4x4, plain and affine, four numberings (standard or KRYPTOS order, 0- or 1-based), every block alignment, matrix encrypting or decrypting, solved exactly from the whole crib blocks by elimination mod 13 and mod 2: all 2x2 and 3x3 settings inconsistent; the twelve consistent 4x4 settings give a singular matrix or break the eight crib letters in partial blocks (0015).

9 substitution then transposition4 experiments

Substitution first (monoalphabetic or homophonic over 26 symbols), then a transposition (the Zodiac 340 construction; suggested by K4's width-21 repeats, 0018)

exhausted-bounded for the route sets

106With 1.84 million further routes (type-1 library, grid paths, mixed double turns, all triple clockwise turns): no route fits (0106).
19Under every one of 23,931 routes (identity, columnar plain and keyed, double columnar, double rotation, K3 rule at every shift) the 24 crib images clash: some carved letter would have to stand for two different plaintext letters. Shuffled K4 gives the same, so the test is decisive, not lucky (0019).

Parameter applied to every family once it is otherwise exhausted: the copper may be wrong by one letter (K2 was). Re-run with one letter inserted or deleted at each position, and with the "?" counted (family G in knowledge/06-hypotheses.md). Applied so far: periodic key stretches (0009), chained autokey and two words (0031).

The same families as a tree of open and closed hypotheses are on the families page; the order they are worked in is on Priorities.

Rendered from the record on 2026-10-18 12:26 UTC; updated after every experiment.

view on GitHub