research notebook

notebooks & run-reports across projects

AutoTreeSearch

Measuring whether autoresearch decisions (proposer model, prompt, methodology) statistically change optimization performance — on a local 3090 GPT benchmark and the WGR C++ speedup tasks.

notebook →

Breadth vs Depth in LLM Autoresearch: allocating a fixed candidate budget20260625-breadth-depth
J0 vs J1: does the orchestration-juice flag help a cheap proposer?20260626-j0-j1-deepseek