More Accurate Topic Assignment for Blocked Articles

This release focuses on making topic and policy evaluation more consistent, especially for articles that are not yet eligible for publication. It also includes user-noticeable performance and reliability updates in admin automation checks.

New Capabilities

Evidence-backed AI topic assignments are now persisted even when an article remains blocked for insufficient publication evidence. This means relevant articles with strong topic matches can still receive a deterministic primary topic for editorial policy evaluation.

Publication safety behavior stays the same. Articles are still blocked when evidence is insufficient, unrelated, comparison-only, or incidental-only.

Fixes and Performance

Admin automation text checks now classify article text more accurately across non-English scripts. This reduces false “unreadable” classifications for valid non-ASCII content.

Several checks were also adjusted to do less unnecessary work on long article bodies. In practice, this helps the admin dashboard stay more responsive when processing large amounts of content.

Reader parity updates were validated across the web app, the Chrome extension, and the Edge extension. This helps keep reading interactions consistent across supported surfaces.