Baton Benchmarks

This directory contains reproducible protocols, not marketing estimates.

The first case measures a shared-contract, multi-surface refactor because it exposes the behaviors Baton is designed to govern: pre-fan-out convergence, artifact ownership, context reuse, centralized verification, and visible coverage gaps.

Current status: protocol published; paired trials not yet run.

Do not convert illustrative scenarios from the main README into benchmark claims. Publish results only with raw evidence, environment details, repeated paired trials, and explicit limitations.