PRODUCTION SCENARIO
A Python client receives invoice fields from a Content Understanding analyzer and posts the totals straight into an accounting system.
What should the client read to decide which results need human review?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: The confidence score returned for each extracted field
Each field can carry a confidence score between 0 and 1, enabled with estimateFieldSourceAndConfidence, and the documented pattern is to automate high-confidence results and route low-confidence ones to a person. An analyzer's overall accuracy is an aggregate from training and cannot tell you which invoice went wrong. The content_filters array reports safety flags, a different concern. An HTTP 200 only says the call succeeded.