16-05-2026, 06:07 AM
Hi Sean,
Thanks for the detailed write-up and for sharing the viewer. I spent some time with the zip file and tried to reproduce the central claim. A few notes, roughly in order of how central they seem to the framework.
On C10: "100% slot purity at lag=-1, 37670/37671 tokens"
I tested this against a ZLZI based transliteration with a four-slot prefix/gallows/core/suffix decomposition (37,465 tokens; within-line adjacency, n = 32,303 pairs). The natural readings I could derive from your description:
gallows[t] == gallows[t-1] (literal identity): 43.3%
gallows[t] given token[t-1] (best class given previous token): 66.4%
gallows[t-1] given (prefix, core, suffix) of token t: 64.9%
has_gallows[t] == has_gallows[t-1] (binary presence): 53.2%
gallows[t] == prefix[t+1] (literal display-register reading): 13.8%
shuffled-gallows null: 65.0% (SD 0.10 pp), so the best reading is 1.4 pp above null
naive first-character baseline given previous token: 33.4% entropy reduction, slightly stronger than the gallows version at 29.9%
The best reading I could construct is about 33 percentage points below 99.997%. The "1 exception out of 37,671" precision is also the part I'd most like to see derived: categorical data on a 9-value column doesn't naturally produce exactly one exception under any test I could construct from the text. If MACHINE_PROOF.md §8 is shareable, the rule and the result would both be straightforward to verify.
On manuscript_data.js
The file contains a type-consistent VMS to English substitution dictionary: 7,861 distinct word types, each mapped to one fixed English gloss, with zero word types showing more than one gloss and only 20 of 7,861 showing role variation. Multiple distinct VMS tokens collapse onto the same gloss: daiin = aiin = "reference"; chedy = shedy = "and"; ar = dar = "seek"; al = dal = "take"; qokeey = qokedy = "transform".
Applied to You are not allowed to view links. Register or Login to view. line 1 the result is "which seek type marker become reservoir formula reading ember become". Applied to You are not allowed to view links. Register or Login to view. line 1 it is "vessel purification ember the opening trace open vessel extract toward pathway vent outward quench fire vessel opening ember sustains". The output is not coherent English at the sentence level.
The forum post says "we cannot translate it"; the viewer file is, functionally, a complete word-by-word translation. Those two framings are contradictory, and it would help to know which one is current.
On the execution graphs
Role assignment in the data file is fixed per word type: every daiin is LBL, every qokeey is PRC, with only 20 of 7,861 types showing any variation. That makes the "FULL_EXECUTION" graphs a token sequence with one colour per word type rather than a per-position computation. If there is machinery that produces context-dependent role assignment, the token-to-role rule would be the thing to share, because the current visualisation cannot show it.
On the za'irajah comparison
Several of the 20 features in the Image 4 matrix ("Pharmaceutical use", "Humoral degree encoding", "Abjad numeral system", "Contemporary 1400–1440 CE") align with the proposed reading of the VMS by construction rather than independently. A version of the test restricted to device-independent geometric and procedural primitives, pre-registered before the VMS comparison, would be more informative.
Two further concerns with the historical comparator. David Link's reconstruction of the as-Sabti zairja (Scrambling T-R-U-T-H: Rotating Letters as a Material Form of Thought, in Variantology 4, eds. Zielinski and Fuerlus, Cologne: König, 2010) treats the procedure as deterministic and rule-governed but produces only a partial reconstruction; he writes in his Section 10 that he was unable to determine a coherent procedure for the reverse-table operation in the later cycles. The single example output in the Muqaddimah (the Idris / Holy Spirit answer) is one rhymed verse of around 47 characters produced from one question, not a 37,000-token continuous corpus, so the determinism claim and the corpus-generation claim are at different scales. Rosenthal's footnote (Muqaddimah trans. Rosenthal vol. 3, p. 197, fn. 880) describes the rationale of the letter-determination procedure and its relationship to the table as "by no means clear to me", and Link's reconstruction inherits that uncertainty for parts of the procedure.
I have also not found a primary-source attestation of the zairja as a pharmaceutical or medical device. Link's account, the standard scholarly reference, treats the device throughout as divinatory. If you are drawing on a medical-zairja tradition, a citation would help locate it.
Minor
The token count differs across your own materials. The forum post and the OpenSCAD comments cite 37,671; manuscript_data.js (generated 2026-04-07) gives 37,046. The C10 denominator does not match the data file. Worth reconciling.
None of this is a claim that the underlying structural intuitions are not worth pursuing. The four-slot decomposition in your wheel_spec (8 prefixes, 4 gallows, 9 named suffixes) tracks with what I've been working on independently - I'm curious how you landed on this configuration. The questions above are about the deterministic-generator claim and the za'irajah linkage specifically. If you can share the §8 rule and the source transliteration, I will run the same tests against your numbers and post the result.
Best,
Thanks for the detailed write-up and for sharing the viewer. I spent some time with the zip file and tried to reproduce the central claim. A few notes, roughly in order of how central they seem to the framework.
On C10: "100% slot purity at lag=-1, 37670/37671 tokens"
I tested this against a ZLZI based transliteration with a four-slot prefix/gallows/core/suffix decomposition (37,465 tokens; within-line adjacency, n = 32,303 pairs). The natural readings I could derive from your description:
gallows[t] == gallows[t-1] (literal identity): 43.3%
gallows[t] given token[t-1] (best class given previous token): 66.4%
gallows[t-1] given (prefix, core, suffix) of token t: 64.9%
has_gallows[t] == has_gallows[t-1] (binary presence): 53.2%
gallows[t] == prefix[t+1] (literal display-register reading): 13.8%
shuffled-gallows null: 65.0% (SD 0.10 pp), so the best reading is 1.4 pp above null
naive first-character baseline given previous token: 33.4% entropy reduction, slightly stronger than the gallows version at 29.9%
The best reading I could construct is about 33 percentage points below 99.997%. The "1 exception out of 37,671" precision is also the part I'd most like to see derived: categorical data on a 9-value column doesn't naturally produce exactly one exception under any test I could construct from the text. If MACHINE_PROOF.md §8 is shareable, the rule and the result would both be straightforward to verify.
On manuscript_data.js
The file contains a type-consistent VMS to English substitution dictionary: 7,861 distinct word types, each mapped to one fixed English gloss, with zero word types showing more than one gloss and only 20 of 7,861 showing role variation. Multiple distinct VMS tokens collapse onto the same gloss: daiin = aiin = "reference"; chedy = shedy = "and"; ar = dar = "seek"; al = dal = "take"; qokeey = qokedy = "transform".
Applied to You are not allowed to view links. Register or Login to view. line 1 the result is "which seek type marker become reservoir formula reading ember become". Applied to You are not allowed to view links. Register or Login to view. line 1 it is "vessel purification ember the opening trace open vessel extract toward pathway vent outward quench fire vessel opening ember sustains". The output is not coherent English at the sentence level.
The forum post says "we cannot translate it"; the viewer file is, functionally, a complete word-by-word translation. Those two framings are contradictory, and it would help to know which one is current.
On the execution graphs
Role assignment in the data file is fixed per word type: every daiin is LBL, every qokeey is PRC, with only 20 of 7,861 types showing any variation. That makes the "FULL_EXECUTION" graphs a token sequence with one colour per word type rather than a per-position computation. If there is machinery that produces context-dependent role assignment, the token-to-role rule would be the thing to share, because the current visualisation cannot show it.
On the za'irajah comparison
Several of the 20 features in the Image 4 matrix ("Pharmaceutical use", "Humoral degree encoding", "Abjad numeral system", "Contemporary 1400–1440 CE") align with the proposed reading of the VMS by construction rather than independently. A version of the test restricted to device-independent geometric and procedural primitives, pre-registered before the VMS comparison, would be more informative.
Two further concerns with the historical comparator. David Link's reconstruction of the as-Sabti zairja (Scrambling T-R-U-T-H: Rotating Letters as a Material Form of Thought, in Variantology 4, eds. Zielinski and Fuerlus, Cologne: König, 2010) treats the procedure as deterministic and rule-governed but produces only a partial reconstruction; he writes in his Section 10 that he was unable to determine a coherent procedure for the reverse-table operation in the later cycles. The single example output in the Muqaddimah (the Idris / Holy Spirit answer) is one rhymed verse of around 47 characters produced from one question, not a 37,000-token continuous corpus, so the determinism claim and the corpus-generation claim are at different scales. Rosenthal's footnote (Muqaddimah trans. Rosenthal vol. 3, p. 197, fn. 880) describes the rationale of the letter-determination procedure and its relationship to the table as "by no means clear to me", and Link's reconstruction inherits that uncertainty for parts of the procedure.
I have also not found a primary-source attestation of the zairja as a pharmaceutical or medical device. Link's account, the standard scholarly reference, treats the device throughout as divinatory. If you are drawing on a medical-zairja tradition, a citation would help locate it.
Minor
The token count differs across your own materials. The forum post and the OpenSCAD comments cite 37,671; manuscript_data.js (generated 2026-04-07) gives 37,046. The C10 denominator does not match the data file. Worth reconciling.
None of this is a claim that the underlying structural intuitions are not worth pursuing. The four-slot decomposition in your wheel_spec (8 prefixes, 4 gallows, 9 named suffixes) tracks with what I've been working on independently - I'm curious how you landed on this configuration. The questions above are about the deterministic-generator claim and the za'irajah linkage specifically. If you can share the §8 rule and the source transliteration, I will run the same tests against your numbers and post the result.
Best,