ce-optimize

To run controlled experiments for code optimization, provide a YAML optimization spec to automate metric-driven iterative batches against a repeatable measurement harness. The system subjects multiple plausible variants to common measurements and records durable results.

Metric-driven iterative optimization is systematically exploring code variants by generating hypotheses, testing them against a stable baseline, and gating improvements using hard metrics or LLM-based quality judgments to identify the best performing solution.

Yes, the optimization harness supports judge mode for prompt-based LLM evaluation alongside deterministic hard metrics. Judge mode uses LLM-based quality judgments to evaluate variants when hard metrics are unavailable or insufficient.

You can resume interrupted optimization batches using isolated git worktrees and a persistent experiment log. A strategy digest records baseline and best-so-far results, enabling resumable and auditable optimization across multiple sessions.

Yes, the measurement harness enforces gate checks during iterative optimization batches. Gate checks validate whether generated code variants meet predefined objective metrics before accepting them as improvements over the stable baseline.

Worktrees require a compatible git repository to safely modify code and reproduce results. If your software project lacks a structured version control environment, isolated worktree scaffolding for experimentation cannot be utilized effectively.