Max Woolf published the prompts behind a 7.5-32x Rust speedup loop.

His agentic iteration post sets a criterion baseline, requires at least 1.2x, and iterates until benchmarks stop improving. Forbid unsafe, benchmark manipulation and target-cpu=native. Explicitly allow radical low-level changes. Fan out 7-12 cheap Luna subagents, run a 20% line-cut refactor pass that unexpectedly gave double-digit speedups, finish with a "try a breakthrough, don't give up easily" pass. His UMAP port runs 4-15x faster than umap-learn. Single practitioner, unaudited numbers, but every step is reproducible.