Three software bots each pull the next task from a shared queue and run it through a fixed, rule-based workflow: a predefined sequence of steps triggered the moment a task appears — the same UI-driven actions a human would click through, just executed automatically. Some tasks hit an exception (a form field the workflow doesn't recognize, a system timeout). A bot retries an exception up to the retry limit before giving up; retried tasks re-enter the queue rather than being lost.
Workflow = Sequence of Automated Tasks + Trigger Conditions
per task: P(exception) = exception rate
if exception and retries < retry limit → retries++, requeue
if exception and retries ≥ retry limit → failed
else → completed
throughput = completed tasks / elapsed time (tasks/min)
success rate = completed / (completed + failed)
queue grows when arrival rate > bots × processing speed
- Task arrival rate — how fast new work enters the queue; the article's "trigger conditions" that spawn a workflow run.
- Bot processing speed — how many tasks each of the 3 bots completes per minute once it starts one.
- Exception rate — the share of tasks that don't fit the rule-based path cleanly and need a retry.
- Retry limit — how many times a bot re-attempts an exception before routing it to the failed pile; raising it trades throughput for a higher success rate.