Texts in unknown scripts pose a methodological challenge: how can structural properties be rigorously characterized without presupposing a specific language or encoding? We present RESON v3.0, a self-correcting multi-axis framework combining six analytical layers (temporal fingerprinting, morphological analysis, grammar extraction, morpheme-section correlation, prosodic analysis, phonetic reconstruction) with built-in falsifiability testing. Applied to the Voynich Manuscript (Beinecke MS 408) as case study. Key results:— Seven orthogonal validation axes: 80.7/100 (4 pass, 3 partial, 0 fail)— Out-of-sample generalization: 79.9/100 (stable across 10 seeds)— Semantic consistency gate: 21.9% at freq≥10 (z=59.6, p<0.0001)— Cross-perplexity: folio classification 61.6% vs 20% chance (3.08x lift)— Cipher rejection: DET z=11.74 Negative results reported honestly:— Vowel harmony claim withdrawn after falsifiability testing— Morpheme decomposition: 37/100 (worse than whole-word baseline)— Position-filtered cross-reference: 34/100 (no enrichment) The framework does not claim decipherment or language identification. It demonstrates what can and cannot be rigorously said about the manuscript's structure, and where the analysis fails. Open-source: all code, 51 control texts, and intermediate outputs included.
Felipe José França da Silva (Wed,) studied this question.