One question about the Corbits facilitator, asked on the examples repo because it is the public surface closest to it.
The case most integrations never test: a /settle call lands on-chain and the response is lost — timeout, 502, dropped socket. The client reads "failed" and retries. What does Corbits answer to (a) a retry with a fresh authorization and (b) a re-presentation of the same one? For (b), is it already settled with the original hash, or a generic failure a correct client cannot distinguish from "never settled"? Multi-network makes this sharper: the EVM answer (nonce consumed → revert) and the Solana answer (same bytes re-broadcast → idempotent) are different by construction, and a client only gets it right if the facilitator's response tells it which case it is in.
If proving it beats describing it, aurumflux20/hostile-facilitator (MIT) produces both orderings on cue and counts distinct settlements for one purchase. Not asking for anything; if the answers are right I would like to record them on the public Retry-Safety Index.
One question about the Corbits facilitator, asked on the examples repo because it is the public surface closest to it.
The case most integrations never test: a
/settlecall lands on-chain and the response is lost — timeout, 502, dropped socket. The client reads "failed" and retries. What does Corbits answer to (a) a retry with a fresh authorization and (b) a re-presentation of the same one? For (b), is it already settled with the original hash, or a generic failure a correct client cannot distinguish from "never settled"? Multi-network makes this sharper: the EVM answer (nonce consumed → revert) and the Solana answer (same bytes re-broadcast → idempotent) are different by construction, and a client only gets it right if the facilitator's response tells it which case it is in.If proving it beats describing it,
aurumflux20/hostile-facilitator(MIT) produces both orderings on cue and counts distinct settlements for one purchase. Not asking for anything; if the answers are right I would like to record them on the public Retry-Safety Index.