Experiments

A terminal coding agent you can leave alone: Jev checkpoints judge progress, completion and risky shell commands while a Go loop owns control.

A daily-ish log of ideas, reflections, and experiments.

Does giving a coding agent access to all of human knowledge change how it solves problems? 10 problems, 8 conditions, 80 runs — an experiment in lateral knowledge transfer.

LLM-steered trading loops for Hyperliquid. Markdown-defined strategies, tripwires, regime detection, and backtesting — safety rails for overconfident models.