Skip to content

Release 2.3.0 - #120

Merged
overwrite00 merged 8 commits into
mainfrom
develop
Sep 19, 2026
Merged

overwrite00 merged 8 commits into
mainfrom
develop

Conversation

@overwrite00

Copy link
Copy Markdown
Owner

Release 2.3.0

Promotes develop to main for the 2.3.0 stable release (validated through v2.3.0-beta.1 to beta.3; the stable build is promoted from v2.3.0-beta.3 without rebuilding).

Added

  • Dialog to choose which data types the AI scan looks for (remembered between runs).
  • Reconstruction of native (non-scanned) PDFs keeps the original font, size, colour and baseline; restore map version 3 (v1/v2 still load).

Fixed

  • AI scan no longer selects parts of words or spreads a single NER hit page-wide: offsets are mapped to word boxes and matching is whole-word.
  • AI scan no longer flags labels, headings, list items and identifier names ("Telefono:", "- Presente", "Configurazione", "Partita IVA"); names glued to addresses or role words are cut.
  • AI scan crash in packaged builds when an email was found (tldextract snapshot was not bundled).
  • A value wrapped over several boxes is restored per box.

Changed

  • mypy is now a blocking CI step; PyMuPDF is imported as pymupdf.

After merge

Tag v2.3.0 on main to run the stable promotion workflow. CHANGELOG.md already has the [2.3.0] - 2026-09-19 heading and an empty [Unreleased], as release.yml requires.

Offset-based detection mapped to word boxes, whole-word propagation, per-type thresholds and generic-word filtering, address splitting for names, entity selection dialog, pymupdf import and blocking mypy in CI.
…ve PDFs (#113)

Record the typography of each redacted value at pseudonymization export (restore map v3) and rewrite it after apply_redactions with the embedded font when usable, else the closest Base-14 font.
Document the new AI detection pipeline and restore map v3, and bump the version to 2.3.0-beta.1.
Collect tldextract in the PyInstaller recipe so Presidio's email recognizer finds its suffix snapshot in the frozen app.
Ship the tldextract packaging fix in a new beta.
…117)

Judge NER spans line by line and reject single-word PERSON/LOCATION hits that are labels, list items, headings, common nouns or identifier names; strip role words from names.
Ship the AI false-positive fixes in a new beta.
Update docs, move Unreleased under 2.3.0 and drop the beta suffix.
@overwrite00 overwrite00 self-assigned this Sep 19, 2026
@github-actions github-actions Bot added type:docs Improvements or additions to documentation type:ci CI/CD, GitHub Actions lang:python Python-related scope:core Core library/framework changes labels Sep 19, 2026
@overwrite00
overwrite00 merged commit 72bd3a9 into main Sep 19, 2026
8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lang:python Python-related scope:core Core library/framework changes type:ci CI/CD, GitHub Actions type:docs Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant