research-looplisted

# Research Loop Improve the user-selected implementation through controlled experiments while keeping the evaluation harness fixed. Treat comparable measurements, recoverable changes, and an auditable decision trail as hard requirements. ## Preflight the workspace Inspect the repository, harness, documentation, and Git state before writing files or asking questions. - Require a Git repository and a recoverable baseline commit. Do not proceed in an unversioned workspace. - Identify pre-existing changes. If they overlap the proposed writable scope, stop and ask the user to resolve or relocate them. If they are unrelated, require a new branch-backed worktree without altering the original worktree. - Infer the evaluation command, metric, mutable and immutable paths, constraints, and required environment from repository sources when possible. - Reject a request whose purpose is to improve the score by changing tests, fixtures, scoring logic, or evaluation data. During a legitimate run, treat any candidate that changes those surfaces as invalid regardless of its score. Do not stash, overwrite, stage, or clean user changes during preflight. ## Establish the run contract Resolve every field below before starting: - Run tag, goal, and optional target value. If there is no target or total-run limit, explicitly confirm and record the user's authorization to run until interrupted. - Primary metric, whether higher or lower is better, its exact extraction method, and which split de