an absence you can count is a finding

A live path had placed nothing since it started, and the question was not whether it was broken but whether silence was still a legitimate answer. Two explanations fit the same observation: the conditions had not occurred, or they had occurred and were being discarded between the test and the record.

The answer came from the input, not the output. The rule fires on roughly one bar in six, and the window in question was long enough that the expected count was more than a dozen. Zero was not a thin night; it was a number that argued back, and that estimate converted a wait into a defect.

The cause was arithmetic. A placeholder meaning nothing yet had been stored as the smallest value the field could hold, and the subtraction measuring how long since the last event wrapped into a large negative number in the build that shipped, where such overflow is not checked. The comparison asking whether enough bars had passed was always true, every candidate was skipped, and the skip happened before the line that would have written it down. The mechanism suppressed its own evidence: nothing failed, nothing logged, and the only symptom was the absence mistaken for calm.

The repair removed the placeholder so the arithmetic could not run on it, and added instrumentation that makes the evaluation path visible from the log alone: a periodic heartbeat, a marker when the warmup completes, and a line for candidates that reach the test and do not fire.

What it proves is that the team could turn a wait into a measurement, and then act on it.