Claim tiers
Every score falls into one of five bands. The numeric range is fixed; the band label is what we publish next to a score on every tile and verdict card.
How we score
Each article URL runs through a deterministic heuristic. The model starts at a neutral baseline of 5 — not zero — so the absence of red flags is not itself a pass. Five signals sum onto that baseline; the final 0 — 10 result is rounded to one decimal place. The heuristic itself is pure: given the same body text, it returns the same breakdown every time.
- Length — −0.6 … 0 (penalises thin pages)
- Sensationalism — −3 … 0 (caps + exclamations + sensational words)
- Hedging — −2 … +2 (balanced reporting rewarded)
- Attribution — 0 … +3 (named sources + institutions)
- Structure — 0 … +2 (byline, publish date, author marker)
The breakdown is what you see on the article page — every signal is listed with its raw value and its contribution to the headline score, so anyone can retrace how we got there.
Source disclosure
The News Wall lists every outlet we score. Outlets fall into three kinds, and the kind is published on the tile so readers see how a score was set.
sourceId against the trending RSS corpus. Their score is recomputed each day from observed coverage; no hand-set baseline.sourceId. The engine runs first; if no trending rows land for the outlet on a given day, the curated houseScore surfaces.