Backfill Text Geometry

🤖 AI Drafted (Not reviewed)

Use this when: pages already have a transcription but no word boxes. Apple Vision detects regions on the page (free, on-device), then the existing text is aligned onto those boxes, so every word in the transcript gets a position — measured where Vision read the same word, interpolated between anchors where it did not. Boxes are saved as provider aligned:apple_vision, never as a measured OCR pass, and a page whose alignment cannot be trusted is refused rather than given a wrong overlay.

Folder	/Detect Regions
Steps	3
Tags	preset, regions, bboxes, backfill, alignment

Steps, in run order

1. Files

Tool: Files — Pass through input files from workflow context

2. Detect Regions

Tool: Detect Regions — Finds WHERE the words are, on-device and free. Apple Vision reads the page locally to locate line and word boxes, so it does produce text — that text is a by-product of finding the boxes, not a transcription: it never replaces the page’s transcript, and no model is called. Runs before a transcriber so every box exists up front.

Settings this step uses:

Option	Value
language	en

3. Merge Geometry

Tool: Merge Geometry — Place a reviewed transcription’s words on the measured word boxes from an OCR pass, so corrected text becomes clickable on the page. Every word records whether its box was measured or interpolated, and a page whose line structure cannot be trusted is refused rather than given a confident wrong overlay.

Settings this step uses:

Option	Value
geometry_artifact_type	regions
text_artifact_type	transcription