Why We Removed Composite Content Scores
Why similarity, readability, and trust grades should not decide whether an article can publish — and what replaces them.
A Mathematical Value Is Not Automatically a Quality Measure
Embedding similarity, readability formulas, keyword density, and model confidence can all produce repeatable numbers. That does not prove that the number measures usefulness, trust, factual accuracy, or ranking potential.
SIA previously displayed semantic similarity as a percentage. The calculation was reproducible, but the percentage was not calibrated against a reliable publishing outcome. Treating it as an article grade created false precision.
What We Removed
We removed composite quality scores, trust and readability grades, arbitrary pass thresholds, and similarity-based publishing signals. They no longer appear in the pipeline, calendar, dashboard, or publishing decision.
What Replaces Them
Publishing now uses a short delivery contract recomputed from the current saved article:
- -The article content is present.
- -Generated-media placeholders are resolved.
- -At least one approved research source is cited when approved sources were included in the article plan.
- -Generated image integrity is checked separately by the publisher.
Each check has a literal pass or fail result and an exact cause. It does not pretend to predict editorial quality or search performance.
Where Judgment Belongs
Editorial quality still matters, but it requires evidence and human judgment. Search outcomes should be evaluated from observed impressions, rankings, clicks, citations, and conversions after publication—not inferred from an invented pre-publish score.
SIA reports delivery facts as facts and leaves editorial judgment to editors.