monitoring

Noticing that something got worse — and why none of the signals a normal deploy is watched by go off when an LLM starts answering badly.

2 posts

• The silent regression: why LLM quality drops and nobody notices

  4 August 2026 · 5 min read

  No alert fires, because nothing breaks — the answers are still fluent, confident, and worse. Six causes with the detection method for each, and why a site-wide average is where regressions hide.

  • production
  • evaluation
  • monitoring

• Measuring what your LLM does under attack

  25 July 2026 · 5 min read

  Injection has no clean fix, so the answerable question is what fraction gets through and whether that moved. Blocked, deflected, complied, over-refused — and why the outcome nobody counts is the one that quietly costs you real users.

  • security
  • production
  • monitoring