< Articles />
Long-form writing by topic, newest first in each.
30 articles in 8 topics, across 8 pages.
Search
Matching articles from every page
Validation and testing, continued
Validation, data integrity and tests that cover the fix
← Earlier in this topic, on page 2-
Testing Against State Drift: Guarding a Config Tool Whose Source of Truth Lives Somewhere Else
My config tool for ESA's Pyxel has to stay in step with a project I don't control. How I test for drift at three layers, and the principle behind them: test each invariant at the seam where it lives.
AI reliability
AI systems that check their own work
-
A Rule It Cannot Quietly Break: Building an Assistant That Checks Its Own Answers
A model cited a real rule and then broke it. Every citation verified, and the answer was still wrong. Why I'm building Nexa — an offline validation layer that checks its own output against a schema-validated rule corpus — and the design choices behind its first finished version.
-
I Instrumented a System the OTel GenAI Conventions Weren't Built For - Here's Where They Broke
I instrumented a deterministic, content-blind AI workflow enforcer with the OpenTelemetry GenAI semantic conventions. It worked, but pinched in six places, each exposing one assumption: that a GenAI system is, by definition, a model invoker.
-
Validate Your LLM Workflow Before It Runs (Python + YAML Topology Checks)
Workflow graph highlighting an invalid execution path caught by topology validation before execution.