ADRIAN SHARMAN
This paper applies Shannon entropy analysis and corpus-wide computational classification to the complete Open Khipu Repository (612 khipus, 39,172 pendant cords, 720 unique compound colours) and demonstrates that the Inca khipu colour system operates as a three-tier information hierarchy structurally analogous to the vowel–consonant distinction in natural language alphabets. The standard practice of collapsing compound colours to their primary component — followed by every computational khipu study since the Aschers — discards 1.366 bits, fully 27% of the total colour entropy. Corpus-wide classification identifies three co-existing administrative systems: System A (Logistical/Warehouse, 154 specimens, 25.2%), characterised by cotton fibre, polychrome colour matrices, and S-type decimal arithmetic; System B (Demographic/Governor, 28 specimens, 4.6%), characterised by camelid or human hair fibre, monochrome flat structure, L-type tactile knots, and Gaussian value distributions; and System C (Categorical/Codebook, 63 specimens, 10.3%), exhibiting Zipfian power-law distributions with a shared pan-Andean vocabulary of 11 categorical codes. The newly identified "hair khipu" (KH0289, Gold Museum Lima) extends the System B evolutionary lineage through Wari antecedents (~600–1000 CE) to a pre-Wari human-hair origin. Cross-site matched-pair analysis identifies 3,512 khipu pairs sharing sequential value matches, of which 1,747 span different archaeological sites. Monte Carlo simulation confirms this exceeds chance expectation by a factor of 8.1 (p < 0.001). The resulting network maps a continental bidirectional administrative pipeline spanning over 1,000 km, with three hub sites: Incahuasi, Pachacamac, and Leymebamba. Audits of the Medrano/Urton Santa Valley and Leymebamba tomb collections demonstrate that both contain exclusively System A state copies, resolving the long-standing puzzle of why most surviving khipus appear purely arithmetic: the archaeological record preserves the formatted receiver end of the postal pipeline, not the raw governor originals. The paper presents 30 computational findings and 9 formally falsified hypotheses. All results are validated through a three-layer "Glass Box" reproducibility framework: interactive discovery via KARP MCP tools, algorithmic extraction into four standalone Python scripts requiring only the Python 3.8+ standard library, and 8 raw SQL queries executable against the public OKR database. Research was conducted using the Multi-Agent Deliberation Distillation (MADD) methodology, orchestrating Claude, Gemini, and Grok through structured adversarial council sessions on the KARP platform. Extends MDP Paper III (DOI: 10.17613/c97d4-pgs78). Code and data: github.com/souldriver007/mdp-ancient-scripts