‹ Culture 3.14 · What we ask ourselves
Question · Operations
What does it mean that a bot has finished a task?
Finishing the execution and producing the expected effect are two different facts, and usually only one of them gets recorded.
Why it matters
A process can complete all of its steps, return a correct status and have changed nothing in the target system. The API responded, the record exists, and the business effect did not happen.
This difference is always paid for later and always at a higher price, because by then there are dozens of executions marked as correct.
What we know so far
Distinguishing three states — executed, confirmed and observed — has proved more useful than fine-tuning the retries. Confirmation requires a later read against the source, and observation requires a signal that often has to be created on purpose.
When that signal does not exist, we prefer to say so in the trace rather than mark the task as finished.
What remains open
It remains open what to do when the observation signal arrives days later, with the process already closed and whoever launched it thinking about something else.
And how to represent a partially effective task without turning every state into an exception.