ce-optimize

Automate metric-driven optimization by running iterative experiments to discover the best code variants using a repeatable measurement harness that evaluates hard metrics or LLM-based judge scores.

You can use an LLM-based judge score to guide iterative code optimization by configuring the measurement harness to evaluate and compare configuration variants automatically across controlled batches.

Manage worktrees for parallel code optimization experiments by using the automated harness to handle dependencies, run variants in controlled batches, and maintain logged state with per-experiment measurements.

Metric-driven optimization supports evaluating clustering quality and search relevance by targeting problems where a scalar hard metric can guide iterative improvements and compare variant performance.

Limitations of using a measurement harness include requiring a repeatable scalar metric or judge score to guide improvements, meaning problems without quantifiable baseline measurements cannot be evaluated effectively.