Design question: release triggers, concurrent-checkout locking, cross-protocol abstraction #7

Hi — I'm working on OM World, a protocol for a decentralized intent economy. Cycles' reserve/commit/release pattern over AP2 mandates is the most thoughtful authority-lifecycle implementation I've seen for the payment-mandate space, and it maps directly onto our Agent Mandate primitive.

Three design questions where your implementation experience would be valuable:

1. Release trigger semantics

The reserve→commit→release lifecycle is classic, but "release" is the tricky one. When does a reservation get released — explicit "no thanks" signal from the merchant/user, timeout (and what window), agent-side abort, or both? And: if release fails (network issue, etc.), is the reservation auto-released after some grace period, or held indefinitely as a debt to clean up out-of-band?

We're designing a similar reserve-style lifecycle for compute mandates and the release semantics are where everyone we've talked to has different intuitions.

2. Concurrent-checkout locking strategy

Three classic approaches: pessimistic (one checkout at a time per mandate), optimistic (allow concurrent, detect conflict at commit), or queued (serialize, no conflict). Pessimistic kills agent throughput; optimistic surprises the agent when a commit fails; queued is fair but introduces head-of-line blocking. Which did you land on, and did you find any operational pattern that made the choice less painful?

3. Cross-protocol abstraction above AP2

AP2 isn't the only mandate-bearing protocol — ACP has Allowances, UCP inherits AP2 mandates, x402 is settlement-only but has receipt semantics. Does Cycles plan to abstract the reserve/commit/release pattern above the AP2-specific shape, or stay AP2-pure? If you've thought about the abstraction layer, what's the minimum primitive that survives the abstraction?

This is exactly the abstraction layer we're trying to specify in OM World — a mandate concept that's neutral across payment protocols. Happy to share what we have.