Approach
Small experiments, honest write-ups.
How this lab works: small falsifiable questions, controlled experiments, first-party data, and write-ups that show the method — not just the conclusion.
Start from a falsifiable question
Every project begins as a question narrow enough to answer: not “how does AI search work” but “does adding schema markup change citation rates for these fifty pages”. If an experiment can’t fail, it isn’t run.
Collect first-party data
Server logs, controlled test pages, scripted model queries, repeated sampling over time. Secondhand screenshots and vendor claims are treated as hypotheses to test, never as evidence.
Publish the whole result
Each experiment ends in a write-up with the method, the numbers, and the parts that didn’t work. Where practical, code and data ship with the post so the result can be reproduced or torn apart.
Principles
- Measured effects over received wisdom — popular advice gets tested, not repeated.
- Conclusions stay inside what the data supports; open questions are labeled as open.
- This is independent, personal work — done on my own time and infrastructure, separate from my employer.