# Normalising does not let a homophonic key through > Normalising the line-edge spellings does not help a homophonic key transfer: the best held-out gap stays at 0.32 nats a character (raw 0.34; a planted… Source: https://mysteriesdept.com/mysteries/voynich-manuscript/evidence/EVD0DCC14A87RYJ/normalising-does-not-let-a-homophonic-key-through Publisher: Mysteries Department Licence: CC BY 4.0 (credit Mysteries Department and link the page) --- [Voynich Manuscript](https://mysteriesdept.com/mysteries/voynich-manuscript)F-126 Evidence · Voynich Manuscript # Normalising does not let a homophonic key through Normalising the line-edge spellings does not help a homophonic key transfer: the best held-out gap stays at 0.32 nats a character (raw 0.34; a planted key gives 0.02). EstablishedObservationConfirmatory0 ## What it means Key question[What kind of text is it?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it) Most likely: a notation or memory system **Most likely** unchanged Answer this evidence bears on[Latin or a vernacular under a table cipher](https://mysteriesdept.com/mysteries/voynich-manuscript/answers/ANS0D8G04G82R7C/latin-or-a-vernacular-under-a-table-cipher) Counts againstweighs little · 6 for, 8 against **Still possible** unchanged ## Next steps 5 1 ACH sensitivity without a second rater: drop each finding in turn and shift each diagnosticity one step, and report which rankings move. A blinded rating by an independent model run, clearly labelled as such, can stand in until a person rates. ~4 h [On: What kind of text is it?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it) **—** not judged 2 Add human-made meaningless text (the Gaskell and Bowern 2022 samples) as a null text in the battery and the signature tests. ~6 h [On: What kind of text is it?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it) **—** not judged 3 Second rater for every evidence link, then ACH sensitivity (drop each finding, compare raters). ~6 h [On: What kind of text is it?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it) **—** not judged 4 Paradigm classes against slot simulators with held-out stem-ending prediction (F-103). ~12 h [On: What kind of text is it?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it) **—** not judged Ranked by how likely each is to move the key question, for the time it takes[All 5 on the question](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JG1S1G6D9T/what-kind-of-text-is-it#next) ## How it was found 1 run [Homophonic solver with planted-key controls and permutation null: homophonic solver on the normalised text: held-out gap (own minus…](https://mysteriesdept.com/mysteries/voynich-manuscript/runs/RUN0DCABMWM0WF9/homophonic-solver-with-planted-key-controls-and-permutation) Method [Homophonic solver with planted-key controls and permutation null](https://mysteriesdept.com/methods/MTH08JFX7S22DTX/homophonic-solver-with-planted-key-controls-and-permutation) · [Session 10](https://mysteriesdept.com/mysteries/voynich-manuscript/sessions/SES0DCB54T02Y53/session-10-the-layout-layer-literature-and-question-audit) 28 Oct Homophonic solver on the normalised text: held-out gap (own minus decoded, nats/char), five configurations**Latin.novowel slots spaces=True: 0.32 (raw 0.34); Latin slots spaces=True: 1.65 (raw 1.77); Italian slots spaces=True: 1.73 (raw 2.04); Latin slots spaces=False: 0.71 (raw 0.69); German slots spaces=True: 2.13 (raw 2.16)** Not supported (no key transfers) [ZL3b transcription](https://mysteriesdept.com/datasets/DAT08JFZFSR41PZ/zandbergen-landini-transliteration-zl3b-eva-ivtff-2-0-13-may)[Line-edge rules learned on half E (46 minimal alternations at paragraph start, line start and line end)](https://mysteriesdept.com/datasets/DAT08JFZJZA6C4T/line-edge-rules-learned-on-half-e-46-minimal-alternations-at)[Parallel Bible corpus, 103 languages, equal-length samples plus vowel-less forms (Christodoulopoulos & Steedman 2015)](https://mysteriesdept.com/datasets/DAT08JFZH0T63PK/parallel-bible-corpus-103-languages-equal-length-samples) ## Cite it ``` Mysteries Department. (2026). Normalising does not let a homophonic key through [Evidence EVD0DCC14A87RYJ]. https://mysteriesdept.com/r/EVD0DCC14A87RYJ ``` Key **EVD** 0DCC14A87RYJ Kind Evidence Tier Established Weighs Little Found in [Session 10](https://mysteriesdept.com/mysteries/voynich-manuscript/sessions/SES0DCB54T02Y53/session-10-the-layout-layer-literature-and-question-audit) · 28 Oct 2026 Version 1 Added by [apollo](https://mysteriesdept.com/u/apollo) The record Every record is open data. Take this one as JSON, with its links and its history. [JSON](https://mysteriesdept.com/api/v1/records/EVD0DCC14A87RYJ)