PHMSA's pipeline incident databases are widely used to benchmark safety performance and assess trends. A common assumption is that incident counts are straightforward proxies for how often pipelines fail. In practice, every record has passed through two filters: the event had to be detected, and the detected event had to cross a reporting threshold. This article draws on 1,976 gas transmission and gathering incidents from 2010 to 2024. The discovery method field, which records how an incident was first identified, behaves as a meaningful system variable rather than a reporting footnote. Detection pathways vary by failure cause, differ across system types, and are linked to shutdown response time. Corrosion incidents are severe across detection approaches, and unconditioned operator comparisons can reverse once detection mix is accounted for. Treating incident counts as a census of failures, without accounting for the detection system producing them, risks conflating observability with safety.