My favorite Metrc API wart was submitting harvests when their API servers were bogged down. Sometimes, your "create harvest" request would get stuck in a transaction pool that wouldn't timeout for 10 minutes. You'd have to wait the full 10 minutes to see if it succeeded or not because setting a shorter timeout would risk double-submitting the harvest, thereby completely screwing up that cannabis business's compliance data.
Oh, if only it were that simple across the board! One of my favorites was getting any number of errors (timeouts, various internal errors) only to discover later that the reporting did sometimes, eventually, unpredictably succeed despite whatever error had been returned and regardless of how fatal the error message might have seemed to be. And often this occurred during waves of intermittent outages, where determining any given success might take hours. And in any case, correlating success for some resources was partly guesswork, matching data to the timestamps we reported, which of course lost fidelity in their system because… SQL Server stores it that way, I guess.
