The Astra October 2026 release extends confidence scores from core data fields to Standards, Questions and AI Columns in AI Projects. The same confidence icons now sit beside the document in the viewer, where a reviewer can mark a manual review or accept or overwrite a flagged value without leaving the clause.
Buyer relevance. Contract teams see where the model is unsure at the moment they read the language, so the human check happens on the low confidence answers before anyone relies on the results.