Posts
All the articles I've posted.
Seven Checks for Systems That Look Correct
Published: at 06:00 AMA May–September retrospective reading path through data correctness, Kafka, RAG, caching, MCP authorization and evaluation, with refreshed series chapters.
95% Agreement, Zero Failures Caught: Testing Your LLM Judge
Published: at 06:00 AMAn executable confusion-matrix example shows why agreement can hide a useless judge. Evaluate failure detection and actual task outcomes separately.
Your MCP Server Is an Authorization Boundary
Published: at 06:00 AMA tool schema is not permission. Separate token audience, client consent, session routing and object ownership in remote MCP services.