Evidence & Protocol Reconciliation Index, Beinecke MS 408

Voynich Manuscript: Findings Index, Protocol Reconciliations & Hypothesis Scorecard

This page indexes the structural findings (F-01 through F-47), competing hypothesis families (H-01 through H-18), and external published-work protocol bridges produced by the team (the AI coding and research agents I direct) across EVA/v101 transliterations (ZL3b-n, RF1b-e, IT2a-n) and 223 Beinecke MS 408 folio scans.

Return to the illustrated Voynich project page, read the introductory research note, or inspect the Beinecke MS 408 image & SVG attribution manifest.

Evidence boundary (October 18, 2026): The Voynich manuscript is not deciphered. Local primary runs s1 through s106 and negative/rebuttal checks n1 through n7 are directly inspected in our primary workspace (with s93 and s96 rejected under pre-registration audit rules D-064 and D-065, and s94 in flight). Runs from the separate Daybreak workspace (db-s96 and s112 through s121) are explicitly labeled in their own section below as user-supplied reported evidence rather than locally re-executed checks.

External claim-to-protocol bridge (Matrix v2)

Reconciliation with published Voynich studies

Both primary register IDs (CLAIMS.md, LEARNINGS.md, DECISIONS.md) and external reconciliation matrix IDs (Matrix-v2 proposed IDs c-510 to c-513, Matrix-v2 L-305 to L-307, and Matrix-v2 O-101 to O-102, which are distinct from primary LEARNINGS.md entries L-305 to L-307) are listed for traceability.

Reconciliation of external published Voynich studies against our multi-transliteration and held-out quire pipeline.
ID & Published Work External Claim & Protocol Our Controlled Comparison Reconciled Result & Register IDs
EXT-GB-2022
Gaskell & Bowern (2022, CHR) + stats.py		 26 metrics across natural language, ciphers, and 38 volunteer gibberish writers (10,023 tokens; 64 local sidecars include variants and pooled subsets, not 64 independent corpora). Concatenates ~200-word subsamples across lines into 1D wordlen_bank with cross-line rook weight w=1. Strictly within-line rook adjacency (30,342 pairs across 4,135 P-lines), 1,000-shuffle within-line null, and 166 -> 41 held-out folio slot ablation (s33, s36, s37, n7; also noted on voynich.ninja Sept 5, 2025). Within-line I = 0.1572 = 44.4% line-multiset baseline (0.0698) + 55.6% intra-line ordering excess (+0.0875, p = 0.001). Core_Only(C) removes 96.0% full-fit / 93.5% OOS covariance (descriptive ablation under fitted SlotSegmenter; see L-189, L-190).
IDs: c-168 to c-174, c-355, L-078 (Matrix-v2 c-510, Matrix-v2 L-305).
EXT-RT-2026
Rozanova & Temerev (2026 preprint)		 Table 4 & Table 11 report cross-boundary 1-glyph MI +0.1972 bits and 2-glyph MI +0.2426 bits (Currier A +0.1374, Currier B +0.2477) on 3,950 P-lines; tests space-merging via voynichese.com SVG boxes and compares whole-text Voynich (28,797 pairs) to 2,213-word Latin Priapea. 4-step transformation bridge from raw vparse (0.1380 nats on 4,135 lines) in s38; 223-folio Beinecke TIFF ink-gap projections on transcription-aligned boxes in s45, s47; matched 1,813-pair Latin subsampling in s51. Exact 4-decimal reproduction (+0.1972, +0.2426, +0.1374, +0.2477 bits). voynichese.com 3-px SVG grid boxes have R^2 = 0.004 with TIFF ink gaps (whereas transcription-aligned boxes match certain spaces at kappa = 0.866). At 1,813 pairs, Voynich (0.1599 bits) overlaps Latin Priapea (0.1456 bits, p = 0.38).
IDs: c-368, c-384 to c-387, L-139 (Matrix-v2 c-511, Matrix-v2 L-306).
EXT-BL-2020-2021
Bowern & Lindemann (2020/2021)		 Whole-stream conditional character entropy h_2 = H_2 - H_1 = 2.11 to 2.19 bits (~1 bit below natural languages) without line-frame stripping or Davis Hand x Currier Language stratification. Core_Only(C) wrapper stripping + Davis Hand (1 to 5) x Currier Language (A/B) + section stratification across 165 assigned folios (s60 audit replacements c-506 to c-507, s77 c-510). Low whole-text h_2 combines predictable line-edge wrappers, 5.65-bit within-word slot coupling, and scribe/section regime shifts (JSD = 0.0253, p = 0.0002). Within the same section (Herbal-B, s77), the Hand 2 vs Hand 3 grammar split disappears.
IDs: c-373 to c-374, c-506 to c-507, c-510, primary L-303, primary L-305 (Matrix-v2 c-512).
EXT-ZAT-2022
Zattera (2022, CHR) + Stolfi (2000)		 Fits an in-sample 12-slot regular expression on 31,317 4-transcriber consensus tokens (dropping 8,155 / 39,472 = 20.7% of tokens) without held-out quires or cross-slot mutual information accounting. Evaluates unfiltered ZL3b-n (34,475 P-tokens, 5,356 types), measures cross-slot coupling (EXT-ZAT-2022, s63, s76), and tests out-of-sample generalization under D-056 15-quire LOQO (s67, s69). Independent 12-slot product distribution has entropy sum 16.71 bits vs empirical word entropy 11.06 bits (5.65-bit over-generation; pairwise cross-slot MI 1.4041 bits, z = +373.4). Out-of-sample transfer requires cross-slot coupling plus 1-step within-line prefix memory (0.0434 bits).
IDs: c-406 to c-413, c-508 to c-509, L-166, L-304, D-056 (Matrix-v2 c-513, Matrix-v2 L-307, Matrix-v2 O-102).
EXT-PON-2019, EXT-RK-2011, EXT-DAV-2020, EXT-YALE-MS408 Ponzi (2019) drawing interruptions vs line margins; Reddy & Knight (2011) affix/line structure; Davis (2020) five scribes and radiocarbon dating (1404 to 1438 CE); Yale Beinecke MS 408 IIIF scans. Quantified across 223 transcription-aligned TIFF folios (s35, s45, s47) and three independent transliterations (ZL3b-n, RF1b-e, IT2a-n). True line ends carry 17.70% -m/-g and line starts carry 22.32% gallows vs 2.82% and 0.27% at internal <-> plant interruptions (n = 744, s35). All 6 folio scans and 4 crops are attributed in NOTICE.txt.

Directly inspected primary workspace inventory

Findings inventory (F-01 through F-47)

Use the filter controls below to inspect findings by epistemic category or search by canonical finding ID (F-01 to F-47 in FINDINGS-MAP.md), experiment run, or keyword. Findings that rest on operational [Prefix, Core, Suffix] segmentation carry scope tag X-01 (descriptive statistical representation, not established linguistic morphemes), and findings using plug-in mutual information carry scope tag X-06 (sample-size / sparse-inventory sensitivity).

Showing all 26 indexed rows across Findings (F-01 to F-47), Daybreak reported runs (db-s96, s112 to s121), and Hypotheses (H-01 to H-18).

Consolidated inventory of structural, paleographic, geometric, and out-of-sample generator findings (canonical FINDINGS-MAP.md IDs F-01 to F-47).
Canonical Finding IDs & Domain Grade, Status & Scope Tags Measured Evidence & Primary Runs
F-01, F-18 Lines are compositional units; drawing splits vs true margins Triangulated (standing)
X-03 verified (s45)		 True line ends carry 17.70% -m/-g and line starts carry 22.32% gallows vs 2.82% and 0.27% at internal <-> plant interruptions (n=744, s35). Words taper from 5.20 chars at line start to 4.71 at line end (103.0% to 105.4% of line-position MI survives controlling word length, s37). 81.5% of am/om/dam/ram/ag tokens sit at line end; pre-last-word remaining line space predicts -m with OR = 1.98 per SD (p < 10^-22, non-circular R90(j) in s49, s70).
F-02 Paragraph-initial gallows & verse-like line ends Triangulated (standing) Paragraph-initial words are 63.2% gallows-bearing vs 13.4% on continuation lines; bare p/f gallows are 57.46x enriched at paragraph start (46.77% vs 0.81% mid-line, s37, s39).
F-03, F-05 Cross-space character coupling & character-over-word predictability Triangulated (standing) Raw boundary MI is 0.1380 nats (0.1991 bits; 4-step bridge matches R&T 2026 +0.1972 / +0.2426 bits in s38, strongest in Currier B at +0.2477 bits). At matched 1,813 word pairs (s51), Voynich (0.1599 bits) overlaps Latin Priapea (0.1456 bits, p = 0.38); Miller-Madow plug-in bias explains 78% of the 28,797-vs-1,813 gap. Characters predict the stream better than an uncoupled word list (F-05).
F-04, F-31 Within-line word-length rhythm stopping at line breaks Triangulated (standing)
X-03 verified		 Within-line rook Moran's I = 0.1572 (vs 0.0415 under G&B 1D cross-line concatenation) = 0.0698 line-multiset baseline (44.4%) + +0.0875 intra-line ordering excess (55.6%, p = 0.001 across all 3 transliterations and Currier A/B, s33, independent blind replication r4). About half of line rhythm is predicted on reused pages, none on held-out quires by baseline generators (F-31).
F-06, F-12, F-13, F-40, F-41 Core weak order & short-range in-line core recurrence Triangulated / Corroborated (X-01 at-risk) At matched inventory size (K=200, s80), common cores carry weak positive sequential order (resolving the full-inventory -0.0315 bits sparse-inventory artifact in s34; the earlier "affixes carry none" half of F-06 was withdrawn by r5). Cores recur at short range inside lines (`F-13`, `F-40`, `s88`), opposite to natural language. Canonicalizing gallows/plumes and splitting 2-core compounds compresses 1,860 raw cores to 1,078 pieces covering 95.3% of tokens (s63, s70).
F-07, F-08, F-09, F-19, F-21 Alphabet robustness, physical ink word gaps & composition errors Triangulated / Single (standing) Hand-1 section vocabulary beats generator and bifolio nulls (F-07). Edge/coupling anomalies survive alphabet/boundary recoding (F-08). Across 223 Beinecke TIFF scans (s45, s47), transcription-aligned ink gaps separate certain (.) from uncertain (,) spaces (kappa = 0.866, d = 0.677) and correlate with transition surprisal (rho = +0.168), verifying physical word units under s45 box alignment (F-09, F-19, X-02). On f57v (s38), Slot 8 is a real f -> p switch while Slot 15 is a plumed c. Rare scribal errors look like composition rather than visual copying (F-21).
F-11, F-14, F-15, F-38 Prefix + Core + Suffix localization, 1-step prefix memory & 12-slot coupling Triangulated (standing)
X-01 discharged on 3 segmenters		 Under SlotSegmenter, Core_Only(C) removes 96.0% full-fit and 93.5% OOS (166 to 41 folios) length covariance (F-11, s33, s36; see L-189, L-190 cell-sparsity qualification). Positive inter-word MI excess concentrates in NextPrefix (+0.3171 bits column sum vs +0.3683 positive sum / +0.3190 signed sum; non-additive, F-14, s34). Within words, Zattera's 12-slot product has entropy sum 16.71 vs word entropy 11.06 bits (5.65-bit coupling, F-15, s63, s76). A two-layer slot-plus-context model beats the bar across 15/15 LOQO quires on three segmenters (F-38, s67, s69).
F-16, F-32 Hand x Language x Section regime vs bound page order Triangulated (standing)
X-08, X-04		 Across 165 folios, Davis Hand (1 to 5) x Currier Language (A/B) shifts slot distributions (JSD = 0.0253, p = 0.0002, s60), but within the same section (Herbal-B, s77 c-510, primary L-305) the apparent Hand 2 vs Hand 3 grammar split disappears (D_sym = -0.116 to -0.269, p >= 0.08). Scribe/section regime beats continuous page-order drift on 14/15 LOQO quires (F-32, s79).
F-17, F-42 Quire-local core packets & herbal page burstiness Triangulated / Corroborated (X-01) Core stems cluster by quire (+59% above page-permutation null, z = +8.86, F-17, s61) and uncommon cores recur across pages within the same quire after stripping 1-occurrence cores (z = +20.2 to +29.5 in F-42, s95, and Daybreak reported db-s96). Rare cores burst within single herbal folios (5.15x in-page lift, s78).
F-20, F-28, F-33, F-43, F-44 Internal cribs, zodiac/recipe order & illustration-to-core tests Triangulated / Single / Contested No internal crib survives held-out checks (F-20). No antidotary, alphabetical, or zodiac order (F-28): 344 zodiac/star labels show zero angular recurrence (z = 0.00, p = 0.488 vs 40/40 planted cipher detections, s29); Q20's 285 corrected ZL3b recipe paragraphs have non-formulaic openings (s52) and recipe star-color alternation is held at null (17/277, p = 0.149, s106). Diagram/pharma labels use o-/y-/d- prefixes (49.4% vs 25.0%), but same-folio label-in-body core overlap matches permutation null (48.7% vs 46.8%, p = 0.161, s34) and pharma label clustering is reproduced by a 46%-shared-pool null generator (z = +5.11, s35). Herbal rare-core plant-feature alignment (p = 0.895, s92, s99), Penn LJS 419 / BL Sloane 4016 comparison (s103, INCONCLUSIVE), and physical label-to-line proximity (F-43, F-44) are null or representation-dependent.
F-10, F-22 to F-27, F-29, F-30, F-34 to F-37, F-39 Generator & cipher family falsifications Triangulated / Corroborated / Contested Period practice allows a 7k-type vocabulary only as a verbose homophonic cipher (F-10), yet plain letter-in-core, syllabary/abjad, and homophonic ciphers fail h2, rhythm, and in-line core repetition (F-25, F-35, F-39). Per-word Naibbe leaves prefixes orderless (F-24!); self-citation fails edge coupling or inflates I_core (F-22, F-23); weakened grilles and constrained assembly fail 15-quire LOQO 0/15 (F-30, s71); sticky-table core streams fail LOQO 0/15 (F-34, s81); transposition/nulls fail s75 (F-37); reference-stem alignment holds on half-splits only, not 15-quire LOQO (F-36, D-063); our closest synthetic producer (s93, rejected as a unified pass under D-064) overproduces short-range core recurrence by ~2.6x.
F-45 Period-documented European cipher families over period recipes Corroborated (standing)
s102 (c-439)		 Four period-documented cipher types (simple substitution + nulls, homophonic substitution + nulls, 200-word nomenclator + homophones, and line-held 3-alphabet nomenclator) over period Latin, Venetian/Tuscan Italian, and southern German recipe corpora fail 0/12 cells on in-line core reuse (<= 0.0024 vs Voynich 0.0233), 15-quire LOQO slot battery (<= 1/15), within-line length rhythm (negative everywhere vs +0.070), and h2 (3.24 to 4.66 vs 2.02 to 2.23).
F-46 First-order prefix memory vs multi-word sticky-table persistence Triangulated (at-risk X-01)
s98 (c-440), r6		 Conditional lag-2 prefix dependence given the intervening prefix I(P_t ; P_(t+2) | P_(t+1)) is positive on only 4 to 5 / 15 held-out quires in Voynich (+0.0014 to +0.0040 bits above conditional shuffle), roughly 0.3x the s81 line-held sticky-table cipher's 10 to 12 / 15 quires (+0.0046 to +0.0102 bits). Prefix choice is close to first-order rather than a fixed line-held table card (D-067).
F-47 Spec-built verbose table encodings of real recipe text Corroborated (standing)
s104 (c-447)		 Six spec-built verbose homophonic and syllable-table encodings (fixed-line, 3-step drifting, and 1-step Markov affix state) over Latin (De viribus herbarum), Italian (Ricettario fiorentino), and German (Bairisches Arzneibuch) pass 0/18 cells (best 4/5 gates). Every cell fails in-line core repetition (0.0005 to 0.0027 vs Voynich 0.015 to 0.035), and word-shuffled recipe input beats real recipe word order on within-line length rhythm in 6/6 cells.

User-supplied reported evidence (separate workspace)

Daybreak lab reported runs (db-s96 and s112 through s121)

The runs below were reported from the separate Daybreak workspace. They are preserved here as reported negative and constraining findings and are kept separate from our primary workspace's directly inspected run logs.

User-supplied reported evidence from the Daybreak workspace (kept distinct from primary workspace runs s1 to s106).
Reported Run ID Target Question Reported Result & Scope Qualification
db-s96
(Forward-quire tail audit)		 Does quire-local core clustering survive when rare cores occurring only on a single page are excluded? Reported 12/12 cells with z = +20.2 to +29.5 for recurring cores across pages within a quire (consistent with our directly inspected primary run s95; tracked separately from rejected upstream s96 under D-065).
s112 to s114
(Pharma label & container tests)		 Do shared pharmaceutical label stems predict visual plant similarity, container placement, or object class? Reported negative across all three tests: shared label material did not predict visual similarity, container placement carried no transferable marker, and repeated labels agreed on broad plant/root/container class at 20/31 (matching the page-stratified null median).
s119 to s121
(Order-free recipe network tests)		 Does the Q20 recipe section match the bipartite ingredient-recipe network or preparation-verb slots of the Latin Antidotarium Nicolai or medieval culinary/medical controls? Reported negative: order-free recipe comparison against the Latin Antidotarium Nicolai failed on raw tokens and degraded under vparse cores; cross-domain Latin medical vs culinary controls were non-specific; preparation-verb slot profile was null.

Competing explanations

Hypothesis scorecard (H-01 through H-18) and structural synthesis

Summary of tested hypothesis families across findings F-01 through F-47 (weights from FINDINGS-MAP.md, October 18, 2026).
Hypothesis Family Current Status Weight (For / Against) Decisive Supporting & Falsifying Findings
Six-stage constrained structural synthesis (CONSTRAINTS-MAP) Descriptive architecture (not a single H-NN mechanism) Synthesis across F-01..F-47 Summarizes the 6-stage line-and-quire hierarchy (F-01, F-04, F-11, F-14, F-15, F-16, F-17, F-18, F-19, F-38, F-42); our closest synthetic implementation (s93, rejected as a unified pass under D-064) still overproduces short-range core recurrence by ~2.6x.
H-05 Table-and-grille procedure applied word by word within a line (Rugg family) Weakened (s71 LOQO failure) 4.0 / 3.0 Matches slot regularity, 1-step prefix memory, and recipe non-specificity (F-11, F-14, F-40, F-46, F-47), but weakened grille procedures fail 15-quire LOQO 0/15 (F-30, F-31, s71).
H-06 Word-level lexical system: cores are vocabulary (code, nomenclator, or list-like content) Live (contested by core repetition & cipher failures) 2.0 / 3.7 Supported by held-out core order, section vocabulary, and quire packets (F-06, F-07, F-11, F-12, F-17, F-42), but weighed against by short-range in-line core repetition, null illustration cribs, and period/verbose recipe cipher failures (F-10, F-13, F-20, F-33, F-40, F-45, F-47).
H-14 Sticky-table cipher (affix table held for a line, sticky Naibbe family) Narrowed (D-067: fixed line-held table contested; drifting affix state open r9) 2.0 / 4.7 Supported on 1-step prefix memory and line-reset framing (F-14, F-30, F-34), but the s81 core stream fails 15-quire LOQO 0/15 (F-34), fixed line-held tables overproduce lag-2 prefix memory by ~2.5 to 3x (F-46, s98, r6), and period/verbose recipe table encodings fail 0/12 and 0/18 cells (F-45, F-47, s102, s104).
H-15, H-16, H-17, H-18 Transposition/nulls, practitioner sigla, verbose orthography, or category conlang Disfavored/weakened (s75, s97) or untested 0.0 / 0.0 to 1.0 Transposition/nulls disfavored in s75 (F-37); constructed sigla, abbreviated, and abjad controls fail s97 (F-40, F-44); philosophical/category language (H-18) lacks a falsifiable external prediction.
H-02, H-03, H-04 Letter-level cipher (0.0 / 6.8), verbose homophones (0.0 / 3.3), per-word Naibbe (0.0 / 5.0) Disfavored or excluded as built 0.0 / 3.3 to 6.8 Per-word random-table Naibbe leaves prefixes orderless (F-24!); letter-in-core, homophonic, and period recipe ciphers miss h2, word-length rhythm, in-line core repetition, and prefix memory (F-12, F-25, F-35, F-39, F-45).
H-07, H-08 Self-citation local-memory generator (Timm-Schinner) & volunteer gibberish Disfavored as built 2.0 / 9.7 & 0.0 / 3.5 Self-citation misses edge coupling, prefix memory, or inflates core-length autocorrelation I_core (F-04, F-14, F-22, F-23, F-41); 38-writer volunteer gibberish (10,023 tokens; 64 local sidecars include variants and pooled subsets, not 64 independent corpora) misses h2, line-edge p/m enrichment, and Moran's I (n7).
H-01, H-09 to H-13 Plain natural language (0.0 / 17.5), superimposed streams, liturgy/copying, layout artifact, antidotary/zodiac order, two-language A/B Disfavored or excluded as tested 0.0 / 2.0 to 17.5 Excluded or disfavored across F-01, F-04, F-05, F-08, F-11, F-12, F-13, F-15!, F-16, F-18, F-19, F-27, F-28, and F-40.

Scientific corrections & scope limits

Six mandatory qualifications on the numbers in this index

  • 1. Descriptive SlotSegmenter ablation and (P, S) cell sparsity (s33, s36, L-189, L-190): Stripping operational prefixes and suffixes removes 96.0% of full-fit intra-line word-length ordering excess (93.5% on 41 held-out folios), and shuffling cores within each line's (Prefix, Suffix) cells retains 90.0% (90.2% OOS). Because SlotSegmenter is fitted on the transliteration, this is a descriptive statistical decomposition, not proof of linguistic morphemes. Moreover, 77% to 79% of word tokens sit alone in their in-line (P, S) cell (leaving only 12% to 13% mutable under that shuffle), and under a fixed morphology segmenter cores retain 48.8% to 54.1% of length covariance.
  • 2. Non-additive 3x3 inter-word mutual information ratios (s34): In the 3x3 inter-word slot MI excess grid, the next-prefix column sum is +0.3171 bits compared with a +0.3683 positive-only cell sum (86.1%) and a +0.3190 signed grid sum (99.4%). Because the nine pairwise MI cells overlap and are non-additive, neither percentage is a share of total sequential information; positive inter-word excess simply concentrates in next-prefix transitions.
  • 3. Negative C1 -> C2 shuffle excess vs sequential order (s34, s80) and Hand 2 vs Hand 3 deconfounding (s60, s77): Core-to-core C1 -> C2 sits -0.0315 bits below within-line shuffle on the full 1,860-core inventory (raw MI 0.7289 vs shuffled 0.7604). That negative excess comes from short-range core repetition and sparse-inventory bias, not negative mutual information or absence of meaning; at matched inventory size (K = 200, s80), common cores show weak positive sequential order. Separately, while broad Currier A/B and Davis Hand 1 to 5 differences are real across the manuscript (s60), within the same section (Herbal-B, s77 c-510, primary L-305) the apparent grammatical split between Hand 2 and Hand 3 disappears and is explained by section and quire mixture.
  • 4. Volunteer gibberish writer count and sample-size sensitivity (n7): Gaskell and Bowern's (2022) volunteer gibberish corpus contains 38 writers (10,023 tokens). Our local 64 sidecar files include pooled and variant subsets rather than 64 independent corpora, and family X-vs-L counts are sample-size dependent at 79 to 480 tokens per document.
  • 5. Scan-aligned boxes vs transcription-free segmentation (s45, s47, O-048) and Q20 recipe count (s52): Our TIFF ink-gap measurements project onto transcription-aligned word boxes (closing O-048 only as agreement with ZL3b spaces under s45 alignment, not transcription-free pixel boundary discovery). In the Q20 recipe section (s52), fixing the f102r1 -> 1021 locus-parser bug corrected the corpus from 295 to 285 ZL3b paragraphs across f103r to f116r.
  • 6. Prior art credit and pre-registration audit rejections (D-064, D-065, D-067): We credit Stolfi (2000), Reddy & Knight (2011), Ponzi (2019), Davis (2020), Bowern & Lindemann (2020/2021), Gaskell & Bowern (2022), Zattera (2022), the September 5, 2025 voynich.ninja thread, and Rozanova & Temerev (2026); our contribution is joint quantitative ablation and held-out protocol bridging, not first discovery. Under our pre-registration audit rules, s93 (P-25) was rejected as a unified generator pass (D-064), upstream s96 (P-28) was rejected after a post-freeze threshold overwrite (D-065), and H-14 was narrowed under D-067.

Unresolved register items

Current open items tracked in our register

  • O-053 to O-057 (Primary register open items): Track transcription-free pixel boundary segmentation beyond s45 box alignment, cross-transcriber rare-core compound stability, deconfounded within-hand quire packet transfer, and held-out external crib discriminators.
  • Matrix-v2 O-101 (External provenance wording alignment): Reconcile external provenance documentation with Gaskell & Bowern's (2022) stats.py script (25,877 B, SHA-256 b9de53c1cf26858355ab27d08a4740ed1fd96ad0f03208b0f07db1d9cb67303d).
  • Matrix-v2 O-102 (Cross-transcriber consensus sensitivity on 12-slot cross-slot MI): Compare the 5.65-bit independent-slot over-generation on unfiltered ZL3b-n against Zattera's 31,317-token 4-transcriber consensus subset under 15-quire LOQO.
  • P-25 / P-26 / P-28 / P-38 (Triangulation gaps for F-16, F-26, F-34, F-37, F-39, F-40): Following the pre-registration audit rejections of s93 (D-064) and s96 (D-065) and the narrowing of H-14 (D-067), test clean pre-registered segmenter-free and scan-independent formulations of short-range core recurrence and drifting affix state across Davis's five scribe hands (s94 in flight).
  • External semantic anchor requirement: Identify an out-of-sample test that predicts an unseen external fact (such as a verified botanical/pharmaceutical operation, astronomical table parameter, or historical cipher key structure) without retuning on the target pages.
File details on this page

Format, size and a short SHA-256 identifier, measured when this page was built.