How we rate claims
Every rating answers two separate questions — because mixing them up is how this subject usually goes wrong.
1 · Does it break the laws of nature?
If the facts are true, could nature do this? Read assuming every documented fact holds — even for hoaxes. Real fairy photographs would sit high here; whether the photographs are genuine is the next dial’s job. For timing stories the question becomes: was this more than coincidence?
2 · Is there evidence it’s true?
Did it happen as reported? Records, adversarial scrutiny, independent sources, named witnesses, contemporaneity — and direction matters. A confession or exposure drives this dial to the floor no matter how thick the file is. And positive proof of falsity outranks mere silence: a refuted claim always sits below one that is merely undocumented. Each grade names its kernel — the minimal disputed facts this dial actually scores.
How the two come together
We don’t blend the two dials into a single score. A claim only rises when both are high — the event would have to break nature andthe record has to say it really happened. Where it lands on the two dials sorts it into one of the six tiers below, with one plain sentence reconciling them. No bonus points, no benefit of the doubt, and never “certain” in either direction.
The six tiers
A quick read of the two meters at each tier — the top bar is how miraculous (brown → gold), the bottom is how strong the evidence (red → green). Thresholds are published so you can check our work — and argue with it.
See the Gold-standard cases and both Top-10 boards →
Meter ≥ 70 and Evidence ≥ 80
Clearly miraculous if true, and strongly evidenced that it happened. The whole mission is making this list longer — honestly.
Meter ≥ 60 and Evidence ≥ 40
Clearly miraculous if true, but the record is thinner. Better documentation could promote it — these are our research priorities.
Meter 41–69 and Evidence ≥ 40
A decent record and a genuinely contested mechanism — the honest middle where serious people can disagree.
Meter ≤ 40 and Evidence ≥ 40
It happened — and nature accounts for it. These are not failures; many are the best good-news stories in the catalog.
Evidence 10–39
The record simply can't carry the claim in either direction. Most old legends live here, honestly labeled.
Positively refuted — not merely undocumented
An override, not just a low score — shown on the claim as “Proven False.” This is a different state from Unproven: there the record is simply silent; here it actively says the claim is false. The label names which: confessed means those responsible admitted it; refuted means positive evidence shows the claimed facts are false. Publishing these is what makes the rest credible.
Worked example: a Gold-standard case
Antonietta Raco — primary lateral sclerosis resolved after a 2009 Lourdes pilgrimage, recognized as the 72nd Lourdes miracle in 2025. The first dial sits high — hard to explain (PLS has no documented spontaneous-remission mechanism, though rare PLS-mimicking conditions keep it short of certainty) — and so does the second: strongly attested (a 16-year inquiry, a 17-of-21 vote by an international medical committee, the strongest documentation in the modern record). Both dials high is rare — only four claims in the catalog manage it — and that is what earns the Gold standard.
Worked example: the Cottingley fairies
The 1917 fairy photographs sit very high on the first dial — genuine photographs of fairies would be extraordinary by any standard. But the second dial is on the floor: the photographers confessed, the fairies were paper cutouts — so the verdict is Proven False. A high first dial on a disproven claim isn’t a contradiction — it’s the system refusing to pretend fairies would be boring.
Taking the natural explanation seriously
Before the dials, there is one question: could nature already account for this? Almost every claim has one of six ordinary rivals doing the heavy lifting — spontaneous remission, coincidence, expectation, misdiagnosis, deception, or misperception. We treat each one on its own terms, and we say where it stops being enough. The standing Natural Explanation library sets out each mechanism and lists the cases it best accounts for.
The full working rubric — sub-signals, anchors, version history — is maintained openly and revised as we calibrate. Spot an error in a rating? The evidence ledger on every claim shows exactly what the number rests on. See how the rubric audits itself on the calibration page, or take the whole corpus home from data & badges. More about the project →