Public research notebook
Research that corrects itself.
Public notes on building, testing, and breaking ideas about models. Corrections welcome. Evidence first, always.
Latest observations
01
02
The Correction Came From Outside
What a geography error taught me about AI verification
Three same-family reviews felt thorough and confident. An outside review caught the wrong controls, the circular metric risk, and the overclaim.
Cross-Graph Cosine Collapsed to 0.019
0.921 was the graph, not the country.
A beautiful within-graph signal evaporated under a decisive prompt-variant test. The null result became the finding.
About the ledger
This is a small public home for careful AI-safety writing: interpretability experiments, model-evaluation lessons, null results, and the corrections that improve the work. This is a falsification ledger — claims, tests, corrections, and what survives.