Morphotactic Structure, Archaeological Ground-Truth Locks, and Logo-Syllabic Proto-Dravidian Modelling of the Indus Script: Cross-Validation Across 5,680 Inscriptions

This monograph and accompanying dataset establish a two-tier computational and epigraphic framework across the complete digital corpus of N = 5,680 Indus Valley (Harappan) inscriptions (16,820 sign tokens). In Part I (Model-Independent Empirical Proofs), we demonstrate: (1) Minimum Description Length (MDL) Markov compression of 41,755 bits; (2) Statistically undeniable physical archaeological ground-truth locks (Sign W048 enriched 27.9x with Elephant iconography, 16.7x at Lothal port, and 12.7x on clay cargo bale tags; Sign W840 enriched 8.7x on stoneware bangles), achieving 66.2% out-of-sample blind test accuracy across 385 unseen physical artifacts vs. 25.0% random baseline (p < 10^-45); (3) Exact 24/24-sign minimal-pair structural alignment of the twin 3-sided tablets (M-494 / M-495) and the 7-sided prism tablet (ID 2844); and (4) Zero-leakage Split-Half and Leave-One-Out (LOO) cross-validation achieving Benjamini-Hochberg FDR q < 0.05 on 74.8% of core signs and 99.12% total coverage (224/226 signs) when combined with 29-dimensional cosine allograph resolution. In Part II, we map this structural skeleton onto a 1-to-1 Proto-Dravidian (*PDr) lexicon constrained by five regular DEDR sound laws and Mesopotamian cuneiform synchronisms (`[...]-ib-ra, lú Me-luh-ha`, `el-lu / i-li`, `ma-gi-lum`), alongside the multi-script benchmark of the ArchaeoScript-AI v1.0 engine on Minoan Linear A and Proto-Elamite.