Every story gets one number, 0–10, and the reasoning that produced it. This is what the number is anchored to.
Every story is scored 0–10 by a calibrated AI prompt that benchmarks against real historical market events rather than against the other stories in that morning’s batch. The anchoring is the whole point: it is what keeps a 7.0 in June comparable to a 7.0 in December. A ranking that is only relative to one day’s news drifts with the news cycle; a fixed reference table does not.
Internally the model works on a 0–100 scale and the site displays 0–10 — dividing by ten is the entire conversion. A raw evaluation is normalised through a sigmoid curve so scores spread cleanly across the range instead of clustering in the middle, which is what makes a routine 4 look different from a must-read 8 at a glance.
These are the fixed reference events the scale is pinned to. They are published in full because an anchor table nobody can inspect is a claim rather than a methodology:
Scores fall into seven calibrated tiers. The reasoning published with each story names the tier it landed in, so the number and the words describing it can never say two different things:
Each story page carries the score, the reasoning behind it, and the outlets that reported it, by name. Where ideological framing applies to the subject it also shows the Left / Center / Right split of those outlets; where it does not — a product review, say — no split is shown at all, because a fabricated one would be worse than silence. Stories covering the same event are clustered and only the lead story of a cluster is published, so one event is never scored twice under two headlines.
What the score is not is a commercial signal. It is identical for free and paid readers, and paid analysis is layered on top of it rather than weighted into it. About Neat Digest covers the sources and the independence rules behind that; the FAQ answers the short questions.