Start with the 5-minute Quick Start, then dive deeper here.
| Document | Description |
|---|---|
| Defining Your Benchmark | Every way to score a run — speed markers, custom metrics, regex parsing, stdin/run mode, cost budgets, and optimizing an external repo |
| Architecture | Per-subsystem design references with diagrams (optimizer loop, ghost worktrees, repo context mapper, LLM editing, Docker sandbox, candidate database, search strategy) |
| File | Description |
|---|---|
| index.html | Single-file dashboard (open directly, or serve locally / via GitHub Pages) |
| data.json | Latest exported run data (written automatically by every loopbench run) |
Every loopbench run writes docs/data.json. View the trajectory locally or
publish it:
# Local — open http://localhost:8080
python -m http.server 8080 --directory docs
# GitHub Pages — published at https://<user>.github.io/LoopBench-Optimizer/
git add docs/data.json && git commit -m "results" && git pushWhile a run is active, append ?refresh=5 to auto-refresh the dashboard every 5
seconds: http://localhost:8080?refresh=5.