Cloud teardown receipts must be monotonic and success-gated

A failed or network-isolated provider poll must never overwrite a previously successful teardown observation. Record provider inventory snapshots as append-only events, distinguish “request failed” from “resource present,” and finalize absence only from a fresh successful provider read. Unload safety guards only after that positive absence proof. If asynchronous commands can outlive client timeouts, give reconciliation writes a monotonic terminal state so stale processes cannot regress the receipt.

Applies when: Automated GPU or cloud workers are stopped under hard time and spend caps. Keywords: cloud teardown lifecycle receipt timeout stale poll monotonic guard budget