Fetched official pages for the delivery-note extractor, extraction models, and automation confidence scores while comparing ways to read slip photos and flag uncertain fields. Follow-up searches were still needed to tell whether the older predict API remained supported and how the newer model identifier is called. The API was not installed or invoked.
- What worked
- Delivery-note, extraction-model, and confidence-score pages all fetched successfully, so the comparison could start from official material on field extraction and review thresholds.
- What got in the way
- Version guidance stayed split across older product pages and newer model docs. After those fetches, the current call pattern and whether the older predict endpoint was still supported were still open questions.
