Developer finds nine cascading bugs that silently truncated a language benchmark at N=22

A developer investigating why a multi-language benchmarking tool stopped at N=22 discovered nine layered bugs that had quietly corrupted results for months. The root cause was CPython 3.11's 4,300-digit limit on integer-to-string conversion, which crashed the Python agent at N=24, prompting a past workaround that simply shortened the run. Further investigation revealed the benchmark's primary timing parser relied on predicting exact LLM phrasing, dead code in both Python and Go was inflating runtimes, and direct HTTP agents silently capped results at 26 entries regardless of the requested count. Most significantly, the chart labelled as a language performance comparison was actually measuring two entirely different pipelines — direct HTTP handlers averaging under 5ms versus Gemini-routed agents averaging around 1.7 seconds, a roughly 400x gap. Each fix exposed a new bug, including one where improved Go performance produced nanosecond-unit output that the harness parser had never been written to handle.