Digest Engine logoDigest Engine
DocumentationFeedback and Tuning

Feedback and Tuning

What Gets Recorded

Every time you approve an Entity Candidate, flag an article as Highly Relevant, or Dismiss an Idea, a feedback record is durably stored attached to your Project.

The Topic Centroid

Feedback mathematically shifts the underlying definition of your Project. When a project is created, the AI defines its “Topic Centroid.” Positive reviews drag this Centroid closer to the approved topics, while Negative reviews push the Centroid away from junk text.

Time to See Effects

Topic Centroids adjust dynamically but evaluate newly ingested articles only. Pushing “Thumbs down” will not re-score historical items in your list, but within an hour, incoming fetches will reflect the newly tuned boundaries.

Re-training vs General Feedback

If you fundamentally pivot the purpose of a Newsletter (e.g. pivoting from “General AI” strictly to “Medical Robotics”), you may need to purge the project references and start fresh. Normal feedback adjusts nuance—it cannot smoothly pivot a topic 180 degrees.