source-quality.md

Source quality and representative works

Source review is profile-specific. A source can be authoritative for one type of claim and unhelpful for another, so the system does not treat popularity as a global truth score.

Run a review for the active profile:

bin/iread sources-review --output data/source-review.json

Each source receives four separate outputs:

• domain_fit: relevance to the active profile.json and topic taxonomy.
• content_quality: evidence, credibility, originality, and clickbait risk.
• collection_quality: connector availability, recency, body coverage, and archive depth.
• confidence: the number of analyzed articles supporting the observed score.

Configured source profiles are priors. Article-level observations gradually replace those priors as the analyzed sample grows. The default policy treats 5 analyzed articles as medium confidence and 20 as high confidence. A high grade with low confidence therefore remains provisional.

The output also selects representative works. It includes a high-quality sample, a recent sample, and an originality-oriented sample when distinct articles are available. Users should review these works before accepting a recommended source.

Rating weights, confidence thresholds, grades, and target source roles are configured in config/source_policy.json. The source portfolio should cover a mix of primary sources, expert voices, independent reporting, specialist analysis, and discovery signals instead of selecting sources by score alone.