seabench

Overview ¶

Command seabench is the SeaPortal benchmark / evaluation harness.

The first subcommand is `eval`: run the corpus at tests/eval/corpus.yaml through four in-process extractors (seaportal, go-readability standalone, html-to-markdown standalone, strip-tags baseline) and emit a Markdown report with per-extractor precision / recall / F1 plus machine-independent time ratios relative to the strip-tags baseline.

All extractors run in-process — no subprocesses, no network, no LLM. The scoring signal comes entirely from must_include / must_exclude substring lists declared in the corpus YAML.