News Feed

1,061 Sources

13
31
181
359
107
70
15

Far Left13

Left31

Center Left181

Center359

Center Right107

Right70

Far Right15

Not Placed on This Axis285

How we score →

Scoring Methodology

Two Things, Six Lenses

Void News reads two things: what the outlet has published before, and what this specific article actually says. Both, across six axes, by rule-based NLP. No LLM calls. The same text always produces the same scores. On political lean the track record sets the anchor and the words move the score from there, within bounds: 10 points at most for an outlet we have rated, 24 for one we have not. An article under 150 words is not read for lean at all; it takes its outlet’s placement, and if the outlet has none it does not vote. Sensationalism, opinion, and framing are read from the words of a full article. Every score discloses how it was reached.

What that came to on the run of 2 October 2026: of 11,666 articles collected, 57% were under 150 words and were not read for lean. 4,628 of them came from the 236 sources that reach us through Google News, which delivers a headline and nothing more. Across 3,796 full articles from rated outlets, the words moved the score 1.4 points on average and left 63% of them exactly on the record.

Not every outlet is placed on a left-right axis. It describes American and British politics well and describes much of the world badly, so a national daily in Nairobi or Tokyo is marked not placed on this axis rather than assigned a position we cannot defend. Those articles are still read for their own words, and still count as coverage. What they do not do is vote on which way a story leans, because inventing a placement would put a measurement in front of you that we never made.

Live Autopsy

Pick any article. Watch the scores populate in real time. Every number traces back to a signal decomposition: keyword counts, entity sentiments, attribution density. Every score shows its work.

42 articles, scored blind, checked by hand.

42 ground-truth articles across 9 categories. Each scored by the engine, then verified by hand against editorial consensus. A cross-axis correlation gate ensures no two axes are secretly measuring the same thing.

42ground truth
9categories
r < 0.70cross-axis
1,061Hand-Curated Sources
US Major43
International420
Independent598

7 lean zones. L:R ratio 1.17:1. Every score is deterministic. Same text, same result. Every score includes a structured rationale. If a score looks wrong, the rationale tells you why.