grobidOrg / grobid / 37747549986

44%

master: 44%

push

github

training(data): header medium-confidence fixes from the second worklist

Works through the 176 medium-confidence rows of the rebuilt header
worklist. 114 rows fixed, 16 with a <!-- check: ... --> comment next to
the edit where the decision is debatable; 16 rows left untouched with a
<!-- check --> comment; the rest rejected or accepted as layout-forced.

Main changes:
- trailing footnote numbers, markers and page numbers moved out of
  addresses, keywords, copyright notes (first-page numbers become <page>)
- PDF production dates of preprints labelled as submission like the
  '(Dated: ...)' lines; 'Preprint submitted to ...' dates merged into the
  submission note; template placeholders untagged
- isolated journal names, page numbers, ISSN/PII, doctypes, conflict,
  funding, availability and copyright statements tagged where untagged
- citations of other works, download banners, related-article URLs,
  report numbers untagged
- affiliation up to the last organisation name (E.R.A. 557, Swiss Branch,
  CAS, Centre for Health Science), trailing footnote numbers out of the
  1609_07374 addresses
- line-based splits of affiliation, author and title fields; same-line
  marked affiliations kept in one field per the guideline
- guideline: a footnote marker at the end of the title stays inside the
  title (corpus practice, 19 vs 0)

10154 of 25710 branches covered (39.49%)

Branch coverage included in aggregate %.

21271 of 46382 relevant lines covered (45.86%)

1.86 hits per line