TUS OS — Making AI Answers Answerable

TUS OS is a governed claim-custody and supportability architecture for AI-assisted judgement. It addresses the movement from admitted evidence to decision-facing claim by making that movement inspectable before final report language is permitted to form. The architecture fixes the active question, evidence boundary, concern frame, Domain Profile and Binding before execution. CAST establishes the anchor state, after which the governed Triad is deterministically constituted across a fixed 64-state supportability field. A domain-blind kernel produces canonical receipt surfaces, while the pre-authorised Binding governs how emitted and allowlisted structures may later be interpreted in domain terms. Receipts provide local structural witnesses. The Triad extends inspection across separated starting positions. Pair Technical Reports determine what each LANDING/GUIDANCE relation materially supports at local level, while Technical Synthesis uses the completed pair reports together with the authoritative set-level surfaces to determine what carries across the whole run, what remains local or unresolved, and where the final claim boundary lies. The resulting custody chain governs claim strength in both directions: it blocks evidential overreach while requiring materially supported findings to be preserved at the strongest lawful bounded level. TUS OS therefore makes the final answer answerable to a visible chain of question, evidence, meaning, structural witness, synthesis and claim permission. Its governing question is: What can this admitted record responsibly carry? File: TUS_OS_02_Full_Public_Architecture_Paper_v1.0.pdfVersion: v1.0Release date: 12 August 2026SHA-256: a9c5eaf70e888633ca248a4accf65ba0f824069dce4ad392b01300ed3a899bc1File size: 219,045 bytes © 2026 Mark Whitlock. All rights reserved.TUS OS® is a registered trademark of Mark Whitlock.