- MD: allow parens in DOI (old-style 10.1016/0006-3207(93)90088-i was
truncated at ")"); prioritize DOI-line doi.org URL; strip trailing
paren only when unbalanced
- PDF: require 10. at segment start (fixes year artifacts like
2010.0170 masquerading as DOI); keep full candidate on no-hit instead
of shortest prefix (10.1002/pd.71 was truncated to 10.1002); absorb
paren segments (10.1061/(asce)wr.1943-...); stop at DOI未知 marker
keeping registrant prefix; accept 10_ segment start variant
- undecodable PDFs: 71 -> 54; manifest DOI values now complete
- pairing count unchanged (3404) = true DOI overlap of the two corpora
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
adminandClaude Fable 5
037790c609