Good question. The score comes entirely from the one-tap verdicts other visitors leave on the number's page: a scam vote counts in full, a nuisance or telemarketing vote counts half, and a safe vote counts zero. If nobody has rated the number yet there is no score at all, which is not the same as the number being safe.
Strengths: it reflects what people who were actually called experienced. Weaknesses: a brand-new number that scammers have only just started using will have no verdicts yet, and a handful of votes can swing the score either way.
We pair it with the Ofcom Range Holder data and the community discussions here for a fuller picture. Neither signal is bulletproof alone. There's more on why an AI chatbot can't answer this either in AI chatbot vs database lookup.