Developer discovers coverage scanner had never tested itself due to alphabetical blind spot

A software developer built a tool designed to verify that other scripts behave correctly during test runs, tracking which log files changed across each call. The scanner honestly reported partial coverage each day, exiting with an error code to signal incomplete passes, with daily sweeps reaching 45.7%, 49.6%, and 68.2% of the tool corpus. Because the scanner always started its alphabetical walk from the beginning, each new pass re-checked previously covered tools before reaching anything new, leaving an entire subsystem cluster — including speech, networking, and comms tools — never swept. More strikingly, the forgery-detection tool itself sorted after the cutoff point, meaning it had never once verified its own behavior. The developer concluded that raising the time limit could not solve the problem, since the corpus had grown too large for any single pass to complete within a 24-hour cycle, and instead proposed a rotating cursor that advances the starting position across successive runs.