Getinside · Print Preflight Checker

A Streamlit app that validates print-bound PDFs and images (PNG/JPEG) for Getinside ad campaigns against:

• Technical print specs — dimensions, bleed, DPI, colour space, QR presence and minimum size, QR URL alignment with the configured campaign.
• Mandatory legal/commercial mentions — advertiser identification (company, legal form, address, share capital, RCS), commercial offer (end date, promo code), printer mention, and industry-specific regulatory phrases (alcohol, food, drugs, gambling, automotive, consumer credit, financial products, insurance, toys).

Architecture

preflight/                  # all business logic
  formats.py                # FormatSpec registry (A5, A6, Carte cadeau, 15×10, Custom)
  industries.py             # IndustryRule registry (11 sectors)
  text_normalize.py         # OCR-tolerant normalization + fuzzy matching
  document.py               # PDF / image abstraction (Document, Page)
  extract.py                # text-layer first, OCR fallback (lang='fra')
  qr.py                     # cv2.QRCodeDetector wrapper
  logos.py                  # phash matching against assets/logos/
  pipeline.py               # CheckContext + run_all_checks orchestrator
  checks/
    dimensions.py / colorspace.py / qrcode.py / logos.py
    advertiser.py / offer.py / printer.py / industry.py
assets/logos/<category>/    # reference logos (PNG/JPEG/SVG accepted)
app.py                      # Streamlit UI — pure plumbing
tests/                      # pytest suite (unit + integration)

Each check returns list[CheckResult] with one of three severities:

• ERROR — blocks print
• WARNING — needs human review
• INFO — advisory / detected value (e.g. extracted QR URL)

Local development

macOS

brew install tesseract tesseract-lang
python3.13 -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
pytest -q
streamlit run app.py

Debian/Ubuntu

sudo apt-get install tesseract-ocr tesseract-ocr-fra libgl1 libglib2.0-0
python3.13 -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
pytest -q
streamlit run app.py

Deployment — Streamlit Community Cloud

1. Push to GitHub.
2. Connect the repo at https://streamlit.io/cloud.
3. Set the entrypoint to app.py.
4. packages.txt installs Tesseract + French language data + GL libs; requirements.txt is installed in the Python environment automatically.
5. Default upload size is set to 50 MB via .streamlit/config.toml.

Logo library

assets/logos/ is the reference library used by the perceptual-hash (phash) check. Subdirectory names are the category, e.g.:

assets/logos/
  cartouche_info_tri/
    Logo triman + Info-tri.svg
  getinside/
    Logo getinside.svg
  imprim_vert/
    Logo Imprim_vert.svg
    Logo Imprim_vert_simple.svg

Accepted file types: .png, .jpg, .jpeg, .svg (SVGs are rasterized on load via PyMuPDF). Add a logo variant by dropping a new file into the appropriate category directory and restarting the app.

The phash check is intentionally honest about its limits:

• It compares the whole rendered page hash against each variant — this confirms similar imagery is present, not the precise size or position.
• The 8 mm minimum size for the Cartouche Info-tri is surfaced as an INFO-level reminder for visual confirmation.
• If the library is empty the check returns INFO and asks for manual verification, rather than silently passing.

Known limitations (v1)

• TAC (taux d'encrage ≤ 300 %) and ICC profile (FOGRA39) are advisory only: the app reminds the operator to verify them in the prepress tool. Per-pixel TAC computation is intentionally out of scope.
• Multi-page PDFs > 2 pages are rejected — the tool targets recto or recto-verso flyers / cards.
• Mixed PDF + image uploads are rejected.
• Vector-level overprint, trapping, transparency flattening are not checked.
• The phash logo check provides confirmation, not localization (see above).

Tests

pytest -q                       # full suite (~85 tests)
pytest tests/test_pipeline_integration.py -v

OCR-dependent tests are auto-skipped if tesseract is not on the host.