Blog

Notes on falsification methodology, retrieval evaluation, and computational philology. RSS.

From dossier prediction to validated lift: 8 of 10 items shipped, harness PASSes, and an UNDER finding worth the whole exercise 1 May 2026
A case study in adversarial calibration: how to use a research dossier so that disappointment becomes data.
The corpus drift trap: why your RAG's nDCG is probably lying to you 1 May 2026
A debugging story, and a 200-line tool that prevents the next one.
Wiring up dead code: how I found +1.9% nDCG sitting unused in my own repo 1 May 2026
Most production RAG pipelines have features that were "shipped" months ago and are silently doing nothing. Here's how to find yours.