Mysteries Department
Sign inRequest access
Run · Voynich Manuscript

Naibbe key fitted on the training quires, read on the held-out quires (Latin)

Held-out decode (seed 1): aarifiminretiuceitauttreaccreqtumencmfrissireaoimoutaesesiiigtaedrirbereldlsueemiaisiiuritsritoiisiuireciiithaturilluima... Letter-like strings, no Latin words. The fitted key's training gap (1.04) is also far from the control's (-0.05), so the problem is not transfer: no Naibbe-structured key reads even the pages it was fitted on.

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 keymanuscript: 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, 0.360

Result

RunMeasuredResultAgainstDecisionDate
RunHeld-out gap (nats/letter) / dict5 / restart agreement, manuscript vs its words shuffled across the bookmanuscript: 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, 0.36Planted-key control: gap -0.018, dict5 0.543, agreement 0.997 · permuted fitted key: gap 2.79-3.06Not 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

What it used and found

Parameters

text = paragraph text, parsed with Greshko's grammar, swaps = 3000000, restarts = 4, solver_seeds = [1, 2, 3], test_symbols = 10962, train_symbols = 28328, null_shuffle_seeds = [3, 4, 5]

Cite it

Mysteries Department. (2026). Naibbe key fitted on the training quires, read on the held-out quires (Latin) [Run RUN0DC7QYS5KSXZ]. https://mysteriesdept.com/r/RUN0DC7QYS5KSXZ
Private beta Downloads are for beta members

We're in a private beta while we get things right. Everything here is free to read; downloading datasets, sources and code is open to members for now.

It takes a couple of minutes. We'll reply by email.