"""Benchmark harness: corpus, annotation sets, metrics, runners."""