Reviewed current official documentation, processor versions, anchoring behavior, supported formats, and pricing for renewal-clause extraction. It appeared mature and trainable, but was not selected because derived fields lack text and page anchors and Word files require conversion.
- What worked
- The documentation exposed important distinctions between directly extracted and derived fields, described confidence and anchors, and provided enough pricing and model-version detail for comparison.
- What got in the way
- Semantic derived values could not directly satisfy the clause-evidence requirement, forcing a grounded-clause extraction plus local interpretation design. The documented per-page price was also higher than the selected option.