| Naibbe as published, laid into the manuscript's pages: statistics outside the cipher's spread | Statistics (of 21: T15 FIT and HELD, layout MI first/last, glyph H1, word-length sd, top-10 share) with |z| > 2 | whole book: 20/20/20 of 21 (genre Latin / Pliny 16 / Dante); Currier B: 20/20/19; Currier A: 19/19/19 | Plain Latin of the genres in the same lines: 20 (book), 17 (B), 21 (A) | Not supported: the published cipher reproduces the manuscript on no statistic except repeated 4-word phrases (rare in both) and, for A and for Dante on B, the top-10 word share |
| Naibbe as published against Currier B: glyph entropy, word length and top-word share | H2 glyph units / H1 glyph units / mean word length / share of 10 commonest words: manuscript vs Naibbe mean | h2 2.105 vs 2.167; H1 3.845 vs 3.851; mean length 4.66 vs 4.80; top-10 0.147 vs 0.152 | Plain Latin in the same lines: h2 3.349 · H1 3.973 · mean length 5.61 · top-10 0.128 | Close but outside the spread (z -12, -3, -13, -3): the cipher moves Latin most of the way to the manuscript's glyph statistics, as Greshko reports, without matching them |
| Naibbe as published: vocabulary richness and word-order information | Share of word types seen once (hapax) / bits per word saved by word order / repeated 3-word phrases / word-length sd | Currier B: hapax 0.693 vs 0.452; order gain 0.314 vs 0.125; repeated 3-grams 103 vs 27; length sd 1.69 vs 1.51. Whole book: hapax 0.705 vs 0.393; order gain 0.404 vs 0.143; 3-grams 145 vs 56; sd 1.77 vs 1.51 | Plain Latin in the same lines (Currier B): hapax 0.633, order gain 1.73, 3-grams 1,097, sd 2.57 | Fails (z 33, 8, 16, 22 on B): the cipher's vocabulary is too closed (too few one-off words) and its word sequence carries even less information than the manuscript's, whose weak word order is itself far below Latin |
| Naibbe as published: page locality, form-function coupling and edge-glyph dependency | Adjacent-page vocabulary z / Spearman(form, context) / near-twin rate within 10 words / MI last glyph to next first glyph (bits) | Currier B: locality 8.93 vs 1.11; coupling 0.144 vs 0.014; near-twins 0.263 vs 0.142; edge MI 0.265 vs 0.010. Whole book: 16.2 vs 1.7; 0.149 vs 0.019; 0.262 vs 0.140; 0.212 vs 0.008 | Plain Latin of the genres in the same lines (Currier B): locality 7.89, coupling 0.036, near-twins 0.056, edge MI 0.049 | Fails (z 27, 12, 52, 297 on B). The plaintext has topical page structure (plain Latin locality 7.9, the manuscript 8.9); after Naibbe encipherment almost none survives (1.1), because card-drawn homophones and random respacing break every recurrence of a plaintext word |
| Naibbe as published: information of line position about the first and last glyph | MI(line position; first glyph) / MI(line position; last glyph) / JS divergence of line-initial first glyphs, bits | Currier B: 0.133 / 0.062 / 0.264 vs Naibbe 0.001 / 0.001 / 0.001. Whole book: 0.112 / 0.056 / 0.198 vs 0.001 / 0.001 / 0.001 | Plain Latin in the same lines: 0.002 / 0.001 / 0.002 | Fails, as it must: no step of the cipher depends on the line, so it has none of the manuscript's line-edge layer (S-007, 46 rules) |
| Share of the manuscript's words that Greshko's grammar can read | Tokens that are a unigram type or split into prefix + suffix / of which unigram types / word types that parse | whole book 78.2 % (41.7 % unigram), types 36.5 %; Currier B 81.3 % (47.0 %); Currier A 72.4 % (30.8 %) | Naibbe ciphertext 100 % (about 47 % unigram) · plain Latin 2 % | Descriptive: a fifth of the running text is outside the cipher's word grammar |
| Naibbe key search, planted-key controls: Pliny NH 16 and Paradiso under an unknown Naibbe key | Held-out letters decoded correctly / held-out gap (nats/letter) / dict5 / restart agreement | Latin: 98.0 % / -0.018 / 0.543 / 0.997. Italian: 99.1 % / 0.069 / 0.443 / 1.000 | Same key with letters permuted within each class: gap 2.93 (Latin), 3.31 (Italian) · dict5 0.003, 0.001 | Control passes: the solver recovers a Naibbe key from ciphertext alone and the key read on the held-out quires gives the plaintext |
| Naibbe key fitted on the training quires, read on the held-out quires (Latin) | Held-out gap (nats/letter) / dict5 / restart agreement, manuscript vs its words shuffled across the book | manuscript: gap 1.231, 1.222, 1.259; dict5 0.040, 0.038, 0.042; agreement 0.40, 0.38, 0.45. Shuffled: gap 1.299, 1.279, 1.242; dict5 0.033, 0.030, 0.040; agreement 0.44, 0.36, 0.36 | Planted-key control: gap -0.018, dict5 0.543, agreement 0.997 · permuted fitted key: gap 2.79-3.06 | Not supported: the held-out gap is 1.2 nats/letter worse than the control and no better than on the order-free shuffle (mean difference 0.04, ranges overlap); dictionary coverage equals the shuffle's; restarts do not agree on a key |
| Naibbe key fitted on the training quires, read on the held-out quires (Italian; Currier B alone) | Held-out gap (nats/letter) / dict5 | whole book, Italian: 1.622 / 0.009 (shuffled 1.618 / 0.013). Currier B only: Latin 1.389 / 0.034; Italian 1.709 / 0.012 | Italian control: 0.069 / 0.443 | Not supported: Italian does no better on the real word order than on the shuffle, and restricting to Currier B, the pages the tables were built for, does not help |
| T38 grid: held-out statistics missed by the best Naibbe-family variant in each section (fresh seeds) | Held-out statistics (of 12) with |z| > 2 for each section's best variant, re-scored on fresh seeds | text B 5 (stay95, codebook, shuffle, layer); herbal B 6 (stay90, codebook, layer); pharmaceutical A 7 (per-unit, codebook, layer); stars B 8 (per-unit, codebook, layer); herbal A 9 (stay50, codebook, layer); biological B 10 (stay95, shuffle, layer) | Picked from the grid on the selection seeds: 4, 5, 7, 8, 7, 9 · plain Latin in the same pages 9-12 · the section's words shuffled across its pages 10-12 | Not supported: no variant reproduces any section; the best settings differ by section and lose one to two statistics on fresh seeds |
| T38 grid: page clustering against how long the scribe stays on one table | Burst z (topical clustering of the 200 commonest words over pages, no page order used) | manuscript: stars B 38.7, biological B 17.2, herbal A 15.2, text B 14.1, herbal B 6.3, pharmaceutical A 6.0. Naibbe per unit (published) -0.7 to 2.2; keep table p = 0.90: 3.4 to 13.5; p = 0.95: 6.8 to 28.3; p = 0.99: 39 to 94; one table per page: 87 to 265 | Plain Latin and published Naibbe · the section's own words shuffled across pages give 0 by construction | Partly supported: keeping a table for about 10 to 20 draws (p 0.90 to 0.95) gives every section its page clustering, which the published per-unit draw cannot |
| T38 grid: word-to-word glyph dependency and near-twins, every variant, as written and without the first glyph | MI between a word's last glyph and the next word's first (edge MI, bits); share of words with a one-glyph variant within the next 10 words (twin10); edge MI on word bodies | edge MI: manuscript 0.19-0.32 by section, the 72 variants 0.01-0.10; on bodies 0.11-0.27 against 0.01-0.12. twin10: manuscript 0.24-0.33, variants 0.13-0.19 | The section's own words shuffled across its pages: edge MI 0.02-0.08, twin10 0.18-0.27 | Not supported: no variant in any section comes within the spread on either statistic |
| Word-initial glyph layer: published Naibbe against each section | Share of words beginning with the four commonest glyphs / entropy of word-initial glyphs (bits) | top-4 share manuscript vs Naibbe: stars B 0.644 vs 0.638, pharmaceutical A 0.661 vs 0.650, text B 0.616 vs 0.632, biological B 0.688 vs 0.637, herbal B 0.603 vs 0.644, herbal A 0.596 vs 0.642. Whole book: o 20 %, ch 16 %, q 16 %, d 9 % begin words (61 %); qo- 15 % | Plain Latin: top-4 share about 0.37, entropy 4.05 bits | Descriptive: the cipher reproduces the concentrated word openings within |z| 2 in three of six sections; it is not where Naibbe fails, and removing first glyphs does not rescue the statistics it misses |
| T39 round 1: which period-cipher settings fit the herbal training pages (2,000 prior draws) | Posterior share among the best 5 % (prior share in brackets), Herbal A / Herbal B | spacing stream 0.66 (0.35) / 0.48 (0.32), word 0.34 / 0.52, sign-as-word 0.00 / 0.00; nomenclator 400 words 0.44 (0.26) / 0.40 (0.26); no syllable signs 0.58 (0.34) / 0.48 (0.34); homophone column kept 98 % 0.32 (0.18) / 0.24 (0.18); copying nulls 30 % 0.56 (0.26) / 0.28 (0.24); vowels dropped 0.34 (0.24) / 0.54 (0.25) | Prior shares · best distance 1.45 (Herbal A), 1.36 (Herbal B), median 2.2 | Exploratory: signs written inside words (stream or word spacing), sticky homophones, a large nomenclator and copied nulls are favoured; one sign per written word is excluded. All draws still write words too long and too varied (Herbal A best 5 %: mean length 5.8 against 4.3, TTR 0.64 against 0.40, glyph h2 3.1 against 2.37) and too few common word openings (top-4 share 0.48 against 0.60) |
| T40 rung A, planted control: key recovery with the true sign boundaries | Held-out key accuracy by use / character similarity to the true plaintext / dictionary coverage (true plaintext's) | 300k iterations: 0.627 / 0.634 / 0.294 (0.697); 1.5M iterations: 0.706 / 0.728 / 0.336 (0.697) | Protocol bar: 70 % of the plaintext's dictionary coverage (0.49) or 50 % word accuracy | Partly passes: the key is largely recoverable when the boundaries are known (held-out decode '...sce mpesr ibusme a ... risior e u a smon t iusex e...' for '...in campestribus melior autem ... crispiora quae in montibus excepti...'), but coverage stays below the bar |
| T40 rung B, planted control: key recovery with the sign boundaries learned | Held-out key accuracy by use / character similarity / dictionary coverage (true plaintext's) | V 450, signs up to 6 glyphs: 0.114 / 0.062 / 0.232 (0.697); V 200, signs up to 3 glyphs: 0.228 / 0.082 / 0.271 (0.697) | Rung A (true boundaries): 0.706 / 0.728 / 0.336 | Fails: with learned boundaries the key is not recovered, so by the protocol the manuscript is untested beyond rung A for this family |
| T39 confirmation: best period-cipher settings re-scored on held-out herbal pages (fresh seeds) | Held-out statistics missed (|z| > 2, of 12), and of all 30 including fit and word-body statistics, on held-out pages | Herbal B best: 3 / 12 held-out, 9 / 30 all (stream spacing, no nomenclator, abbreviation, vowels dropped, simplified spelling, 3 homophones per vowel and 2 per consonant kept 95 % of the time, 15 % copied nulls, line layer 0.5). Herbal A best: 5 / 12, 15 / 30 (stream spacing, 50-word nomenclator, 1 homophone per vowel and 4 per consonant kept 90 %, simplified spelling, 30 % copied nulls, line layer 0.5) | Naibbe family in the same sections (T38, fresh seeds): 5-10 / 12 · published Naibbe 9-12 / 12 · plain Latin 9-12 / 12 | Closest fit in the programme, still not a fit: the remaining misses in Herbal B are glyph predictability (h2, z -5.7), one-off words (hapax, -5.0), the edge dependency (+3.8), held-out bigram gain (-3.0) and glyph context gain (+2.2); in Herbal A also word length, TTR, form-function coupling and the line-initial measures |
| T40 exploratory: Herbal A decoded with learned sign boundaries, against its own words shuffled | Held-out element-trigram score (nats per element; Latin's own -2.418) / dictionary coverage of the decoded held-out text | Herbal A -2.563 / 0.296; shuffled -2.824 / 0.216 | Planted control under the same rung-B pipeline: -2.767 / 0.271 (true plaintext 0.697) | Exploratory, not a reading: the real word order beats its shuffle by 0.26 nats per element, but the decoded text is not Latin ('iumu ae s io s e i nea n un a li d uee u uei i u n tu e n ti oru u') and the pipeline fails its control at this rung |