document_sorter/README_V1_4.md

Quick Sorter v1.4 – focused content-reading upgrade

Changes from v1.3:

• Document extraction failures are no longer silent. The GUI shows a Content status such as docx:1834, missing:python-docx, or pdf:no-text-or-scan.
• DOCX extraction reads document metadata, normal paragraphs, tables, headers, and footers. This matters for Ablaufpläne and forms whose useful text often lives in tables.
• PDF: metadata + first 3 text pages.
• XLSX: sheet names + first rows of first 2 sheets.
• PPTX support via python-pptx.
• ODT support via odfpy.
• The LLM now extracts structured semantic fields: topic, subtopic, document type, detail, version. Python builds the final filename deterministically from those fields.
• Generic filenames with no readable content are capped below auto-confirm confidence instead of guessed confidently.

Recommended packages inside your pyenv virtualenv

python -m pip install python-docx pypdf openpyxl python-pptx odfpy pillow

No --break-system-packages is needed inside a proper virtualenv.

Check what this exact Python environment can read:

python quick_sorter.py --check-deps

(dummy is ignored when --check-deps is used, but the positional argument is still required by argparse in v1.4.)

Best first test

Start the app against the test copy and select only 2026-08-21_Ablaufplan.docx. After Analyse selected, inspect the new Content and Semantic columns. If the content reader works and the document actually contains the event name, you should see something like:

• Content: docx:...
• Semantic: cafe-bach / bach-voices / ablaufplan
• Destination: Cafe_Bach/Veranstaltungen
• Name: 2026-08-21_cafe-bach_bach-voices_ablaufplan.docx

If Content says missing:python-docx, install python-docx in the active virtualenv. If Content says docx:no-text, the file may use unsupported drawing/textbox content; that is a separate extraction case to add only if needed.