Hotfix: shipped PDF identity and presentation overflow checks could report a match or clean result without measuring the relevant evidence.
Fixed
- Inspect the actual npm tarball and classroom ZIPs against the selected release
source, including file hashes, existing private-identifier/credential patterns,
PDF text and binary/Office metadata. Publish the inspected tarball and download
both channels to compare the delivered payload. Recovery runs record the selected
tag commit and also verify npm versions that are already published. - Replace a contextual comment in an npm-shipped synthetic R fixture with neutral
wording. The test computation and expected result are unchanged. - Presentation overflow checks now request line coordinates from poppler instead of
silently treating word-only output as a clean measurement. Real-PDF regression
tests run in the existing CI step. The Nature/Lancet builder uses readable
outline/glossary text and native bullets, reserves space for wrapped headings
and captions, and supports choosing installed fonts without rewriting content.
Font-check success no longer implies the fonts are installed at the venue. - Consistency linting no longer reports a single COVID-19 spelling as mixed, splits
numeric/hyphenated abbreviations into partial tokens, or treats grammatical
follow-up/follow up and long-term/in the long term pairs as spelling drift.
Synthetic US/UK normal controls supplement the existing seeded-error challenge. - PDF rendering now honors frontmatter before wrapper/OS defaults for fonts,
page geometry, font size, line spacing and link colors. Explicit pandoc overrides
remain available. Glyph scans support an explicitly selected TTC/OTC face and
report why font coverage could not be checked; different faces are never combined.
Synthetic font and pandoc regressions extend the existing render test step. - Full-text retrieval no longer treats scattered title words or body/reference mentions
as a title match. Report schema 2 separates download results from advisory first-page
title/identifier evidence and records the PDF hash; related versions and incomplete
evidence remain visible for review. Existing retrieval counts and downloaded files are
preserved. Search and library-sync guidance carries the identity evidence downstream.