experiment

Automated code optimization requires defining a scope, a shell metric command outputting a single number, and an iteration budget. The system then runs iterative experiments in isolated git worktrees to discover changes that measurably improve the defined objective.

Iterative code experiments run safely by using git worktree isolation. This keeps proposed changes separate from your main branch, allowing the system to propose, implement, and evaluate modifications without risking your primary codebase during the optimization loop.

Setting up a metric-driven optimization loop involves providing a scope, a shell command that outputs a single numeric metric, and an iteration budget. The system measures the baseline, runs experiments in worktrees, and tracks progress against that baseline.

Automated code optimization works with any shell command that outputs a single number. This scalar fitness function guides the iterative process, enabling the system to identify code changes that improve the measurable objective and discard failures.

Git worktree isolation requires available disk space for each experiment and depends on shell metric commands providing consistent numeric output. The system enforces type checks and logging, but convergence detection and diminishing returns checks will halt the loop when improvements stop.

Iterative optimization loops stop early due to convergence detection and diminishing returns checks. When the system determines that continued experiments are no longer yielding measurable improvements to the numeric metric, it halts and provides structured reporting on the results.