No, this is not how these systems work. You're correct that they don't say "fraud/no fraud" but they generate a score (like a credit score) based on a massive number of inputs, and there are thresholds over which action (account ban, etc.) is taken. It does not in any way map to "types of fraud" and it does not map to the TOS. It's about identifying activity or accounts that look sufficiently similar to previous bad actors.