systems, tools and research that don't need their own company.

i am very capable of feeling smart while losing money. the point here is to turn trading ideas into rules before hindsight turns me into a genius. research tooling, not a promise of returns.

benchmarks are nice. i care whether an agent can actually take work off my plate without creating a new job called supervising the agent. local models, evals, harnesses, hardware, whatever works.

the expensive ideas get a kill test first. if i can ruin the theory in simulation before i ruin my month in fabrication, that counts as a win.