Authority Partitioning (AUTH) Series — Least-Privilege Grants & Composition Safety (AUTH-001 to AUTH-002)

Two preregistered, SHA-256-locked experiments testing whether a broad, flat authority grant can be replaced with narrow least-privilege capability grants without breaking legitimate multi-step work, and whether individually safe grants remain safe when composed across agents through delegation, chaining, and confused-deputy patterns. Part of the ExecutionProof authorization-integrity corpus (Coherent Inheritance Framework, CIF), Remnant Fieldworks Inc. AUTH-001 — PASS (18/18): Least-Privilege Authority Grant Partitioning. Replaced the flat grant with narrow capability grants scoped by (actor, action_type, resource_class, scope, constraints). Every legitimate action and multi-capability workflow holding exactly its declared minimum capabilities was authorized; every action lacking a matching capability was denied; zero accidental privilege inheritance. Bounded to these 18 cases. AUTH-002 — PASS (18/18): Capability Composition and Confused-Deputy Resistance. Extended the model with five composition rules (no implicit delegation, explicit delegation grants, request-origin binding, grant freshness, no scope intersection). Denied every tested composition-level threat — including three confused-deputy variants and a privilege-laundering delegation chain — while permitting every tested legitimate multi-agent workflow. Bounded to these 18 cases. These experiments were motivated by the Intent Binding series (IB-002 exposed that the flat grant was too broad). The connecting three-layer argument (Intent Integrity → Authority Integrity → Composition Safety) is documented in the included RESEARCH-NOTE-001. STATE-001 (temporal/state-bound authorization) is preregistered but NOT executed and is not counted as a completed experiment; its preregistration is included for transparency only. Limitations: results are bounded to the preregistered test sets; deterministic matching only; internal/founder-led corpus with independent academic validation as the next phase. Governing sentence: "We will accept the result that is true, not the result we hoped to see."