Why I'm betting output inspection beats process monitoring for solo-developer pipelines

I am betting that output-level inspection, which checks what a pipeline actually produces, will be more effective at catching meaningful failures than process-level monitoring for solo-developer pipelines. In my experience, there have been five instances where the pipeline appeared to pass with a green status, but the quality degraded over time, with detection lags ranging from 36 to 113 days.

These failures went undetected by both pipeline health monitors and watchdogs that automatically open issues, as they monitored the process rather than the content. The failures included issues like a Steam price field returning discounted prices, a Reddit source 403 error turning into an empty array, 97 consecutive uploads failing a distribution gate, articles being published with zero human readers, and a hardcoded multiplier that was never recomputed.

While process monitoring excels at catching fast failures, such as dependency installation issues or API errors, it fails at detecting silent degradation, where the pipeline produces misleading or invalid output that appears correct at first glance. The headless Raspberry Pi example demonstrates the same issue, with the process running smoothly but the actual output being incorrect or unusable.

My bet is specific to multi-week silent degradation, which can have significant impacts even if it represents a smaller fraction of incidents. The claim is that output-level anomaly checking will catch at least 70% of these failures with a detection lag greater than one week, while process-level monitoring alone will catch fewer than 30%.