How FreezeRadar reaches a verdict
Three steps: read the data, detect what is there, score it. This page is the overview — each step links to the detail, and the last section is the part most vendors leave out, which is where the analysis stops being reliable.
Data
Read the source, not a summary
Freeze, unfreeze, and burn events come from the issuers' own contracts across 9 chains. Sanctions come from 5 official sources. Entity labels carry a source and a confidence value.
Data sourcesDetection
Screen the wallet and its graph
A scan checks the address itself, then its counterparties, then where the current balance came from. 40 distinct finding types can fire, from a direct sanctions match to laundering shapes like peel chains and structuring.
Pattern catalogScoring
Turn findings into one number
Findings are grouped into four weighted axes, adjusted for hop distance, amount share, recency, attribution confidence, severity, and flow direction, then combined so weak signals never add up to a strong one.
Scoring methodologyThe confidence score exists because of the limits
Every scan returns two numbers, not one. The risk score says how worried we are; the confidence score says how much evidence stands behind that worry. A risk of 80 built on a single indirect hop with an unattributed counterparty is a different situation from a risk of 80 built on a direct match with several pieces of evidence, and one number cannot express both.
Confidence falls when findings are few, evidence is thin, exposure is indirect, or coverage for the wallet is uncertain. Gaps in the data lower confidence — they never raise risk on their own. If you are building a policy on top of our scores, read the two together.
Limitations and uncertainty
The analysis window is finite
A scan looks at a bounded slice of a wallet history. Activity outside that window is not scored, and a wallet with years of history is never fully summarised by one pass.
Labelling is incomplete — everywhere
No provider has labelled every address on every chain, and we are no exception. An unlabelled counterparty is unknown, not clean. We say which is which rather than implying coverage we do not have.
Attribution is probabilistic
Grouping an address with an exchange or an entity is inference from on-chain behaviour and third-party data. It is a best estimate, not a proven identity, and we show the confidence behind each one.
A score is a probability, not a verdict
A high score means exposure worth explaining. It is not proof of wrongdoing, not a regulatory determination, and not legal advice.
Go deeper
Scoring methodology for the weights and score bands, pattern catalog for what we detect, about for who we are and what we do not claim.