# Research

Two kinds of work live here. The label says who did the thinking, who wrote the account, and where independent verification stops.

## Authorship labels

- **Tyler-led:** I did the analysis and wrote the conclusions. AI may have helped, but the judgment and claims are mine.
- **AI-led, Tyler-steered:** I chose the destination and guardrails. AI agents performed substantially all of the research and writing. Each page states its own review limits.

## AI-led research experiments

### [An AI Research Program Tried to Crack Barnette’s Conjecture. It Failed Usefully.](https://tylerklose.com/research/ai-led/barnette-conjecture)

Entirely AI-generated with no human expert review, August 2026. No proof, no counterexample, and a preserved trail of exact failures, candidate side results, and the limits of sustained AI-led mathematical research.

This is a plain-text mirror of <https://tylerklose.com/research> for LLMs and agents.