Safe Revocation Without Complete Diagnosis: Exact Resource Frontiers under Coupled Faults for Subjecthood-Agnostic AI Coexistence

This theoretical paper derives exact integer backup-resource frontiers for subjecthood-agnostic safe revocation under a shared adversarial root budget. Diagnostic corruption and backup loss are coupled rather than independently worst-cased. General formulas apply to arbitrary channel counts, primary replacement counts and fault budgets; constructive allocations attain the lower bounds. A fixed architecture with f+1 diagnostic and f+1 backup roots reaches its bank-capacity minimum without unique diagnosis, while a specified decoupled rule can require an unbounded factor more reserve. Extensions handle arbitrary one-fault diagnostic maps and replicated co-located roots. Independent finite verification supplements written proofs. The paper does not assume machine personhood or claim a real ASI safety experiment. PR-v6.0.0. Not peer reviewed. Source and reproducibility data accompany the paper. No empirical ASI or consciousness claim.