Two Governed-Refusal Mechanisms: Graduated Mount-Time Comparability Gating, and Interrupt-Driven Revocation Reaching Suspended Agents

Defensive publication of two mechanisms, placed in the public record as prior art. (A) Graduated mount-time comparability gating for a distributable semantic index: a plaintext identity header naming an embedding model and dimension is read without decrypting the payload; a three-valued comparability state (match, dimension-only, mismatch) is determined against the host's active provider; the mount completes in all three states; similarity search is permitted, permitted-with-a-stored-unverified-indication, or refused accordingly, and under refusal the artifact's non-similarity surfaces continue to be served. (B) Interrupt-driven revocation reaching suspended agents: a revocation-class policy change identifies capabilities by scope, the affected agent population is resolved from a capability registry without interrogating any agent, and prioritised interrupt events are delivered at inter-transition boundaries irrespective of interrupt masking or suspension, executing a compiled corrective action before the agent resumes prior work. Measured: a request-time policy model reaches 0 of 10,000 suspended agents on revocation; this mechanism reaches 10,000 of 10,000 in one tick, with reaction latency constant rather than linear in the agent duty cycle. Each mechanism is stated with the prior art it was searched against, including what defeated adjacent claims, and with an explicit statement of what the searches did not cover. No patent has been or will be sought on either.