optimize

Run a metric-driven Infiquetra optimization loop — define a measurable goal, baseline current behavior, generate a hypothesis backlog, run bounded one-variable experiments, measure each against hard degenerate gates before any LLM-judge, keep the best real improvement and revert the rest, and converge on the measurably-best version. Eight metric classes (perf, cost, reliability, agent-usability, security, quality/accuracy, developer-experience, maintainability). Off-chain and saga-untouched; it finds the winner, then routes it to /work to ship — it never commits, merges, or deploys. Triggers on "optimize this", "make it faster/cheaper", "tune this", "baseline and improve", "drive this metric to a target", or a perf finding handed in from /qa or /loop.