a gate the system could not physically reach

A live path produced nothing for a stretch of hours. The obvious readings were that there were no qualifying candidates, or that the feed was broken. Both were wrong, and the way the truth was found is the useful part: rather than guessing, the team wrote down what each hypothesis would predict and went looking for observations that could separate them.

The decisions were in fact being produced, and on schedule. The gate that admits them required the decision to reach the execution side inside a window measured in minutes. Delivery of the underlying data arrived with a lag that varied from a few minutes to a quarter of an hour. The window was therefore not tight, it was unreachable — and it had been unreachable the whole time, written into the configuration in the grammar of a tolerance.

The same investigation found a second, unrelated cause of silence elsewhere: a pinned specification older than the binary actually running, which the harness refused exactly as designed. The refusals were the system working. The appearance of a broken system came from the absence of any record saying why.

The fix treated lateness as a first-class input: an incremental engine that reprocesses only the new data instead of replaying history each cycle, so the decision exists before the window opens rather than after it closes.

It proves the difference between silence and refusal, and the value of writing that difference down.