The quality documentation is organized by purpose:

• benchmark methodology defines the corpus, extraction protocol, tokenization, and metrics;
• 0.9 quality report records the release benchmark and security checks;
• 0.8.4.1 baseline preserves the previous-version benchmark;
• engine comparison summarizes the ten-engine comparison;
• detailed extraction results contains complete corpus, failure, diagnostic, and per-page tables;
• user issue corpus records the issue-derived fixtures.

Raw machine-readable comparison results are stored in benchmarks/results/. Saved page fixtures and their provenance are stored under pages/.