Methodology
How this service separates detected risk, evidence quality and the ability to make a full assessment.
Deterministic analysis
Risk scores, signal severity, Confidence and assessment state are calculated by versioned deterministic code. Language models do not assign scores, change severity or invent evidence.
Risk is not Confidence
Risk summarizes supported adverse indicators. Confidence describes how strongly the available evidence supports the assessment. Low detected risk with weak evidence is not presented as a low-risk assessment.
Evidence domains
Coverage is evaluated across contract structure, administrative controls, supply controls, trading/sellability, holder concentration and source transparency. Critical blind spots can force PARTIAL_EVIDENCE or INSUFFICIENT_EVIDENCE.
Snapshots and provenance
On-chain observations preserve chain ID, block number, block hash, block timestamp and snapshot finality where supported. Ordinary recent head reads are labeled HEAD_PINNED rather than falsely described as finalized.
External intelligence
External providers are treated as untrusted, replaceable enrichment. Provider outage, malformed data or rate limiting lowers evidence quality; it never becomes token risk by itself.
Limitations
Smart contracts, ownership, liquidity and external intelligence can change after a snapshot. No automated analysis can guarantee safety or detect every failure mode.