First citable release.
Contents: full draft manuscript (MANUSKRIPT_DRAFT_EN.md, v0.2), all analysis scripts (plain Python 3, stdlib only), audited corpus extract (data/), GORILA I apparatus coding with page references (data/gorila_apparatus.json), six pre-registered falsifiable predictions (PREDICTIONS.md / PREDICTIONS_EN.md, git-timestamped), literature review with prior-art documentation (LITERATUR.md).
Key findings:
- Arithmetic discrepancies on the Haghia Triada tablets correlate with erasures in numeral lines (5/6 deviators vs 0/8 exact tablets, Fisher's exact p = 0.003; independently supported by Montecchi's 2019 palimpsest list) — Linear A "calculation errors" are residues of document updates, not scribal incompetence.
- HT 13's famous discrepancy is value-independent: it reduces to 2J = J.
- HT 123 column constraints (H = entry-sign + ¼, hence H ≥ ¼) bipartition the fraction-value literature; the machine-validated ledger supports the constraint-respecting camp (Bennett 1950; Cash & Cash 2011) against H = 1/6 (Was/Younger) and H = 1/16 (Corazza et al. 2021).
- 53 data-quality corrections contributed upstream to mwenge/lineara.xyz (PRs #7, #8).
Analyses performed with AI assistance (Anthropic's Claude Fable 5), directed and reviewed by the author; all critical readings verified against the GORILA I facsimiles.