monitoring-script-debugging

Cross-validate the reported number with an independent shell pipeline, such as grep plus a Python counter for unique exception types. If the script and cross-check differ by more than 10%, the script has a parser or counting bug that must be patched before reporting any results.

Lenient parsers corrupt valid data when regex-based comment stripping matches /* or // inside JSON strings, such as URLs or glob paths. Fix it with string-aware character-by-character comment stripping and always try the standard json.loads parser first before falling back to lenient parsing.

Count unique error signatures rather than total matching lines, and count each traceback block once instead of every frame line. Use a rolling time window based on per-line timestamps so old failures age out, and normalize retry counters like (1/3)(2/3)(3/3) into a single signature.

A permanently red check usually scans the entire log history instead of a rolling window, so old failures never age out. Parse per-line timestamps, let untimestamped traceback lines inherit the previous line's timestamp, and count only entries within a window such as 7 days.

Mass identical TypeErrors like '<=' not supported between float and str indicate one config loader bug, not many check bugs. Probe the loader output directly; fallback YAML parsers often fail to strip inline comments or parse flow and block lists, returning strings where numbers are expected.

fnmatch patterns are anchored at the start of the path, so references/** only matches paths beginning with references/. To exclude a directory at any depth, prefix the pattern with **/, for example **/references/** or **/node_modules/**.