So I ran into a pretty annoying problem while building RoundWatch, and I m curious how others deal with this.
Catching the moment a blockchain payment happens? Honestly, that s the easy part.
The real headache starts when something breaks mid-flight: Your agent finishes the job -> the payment goes through on-chain -> but the system crashes right before updating the local state.
When everything boots back up, how does your system know what actually went down? If you don't handle this right, you risk double-spending, losing track of progress, or getting stuck trying to resolve a payment that s already settled.
RoundWatch gives autonomous workflows a durable boundary between completed work and what happens next. It observes x402 payments, advances state only after verified settlement evidence, and preserves progress across restarts so agents can continue safely.
