Wallet reputation methodology

# Wallet reputation methodology

AgentRanking wallet reputation is a versioned, wallet-scoped product assessment. It is not a universal reputation score, credit rating, or sanctions determination. A caller that needs sanctions screening must use an appropriately licensed source and an explicit policy outside this assessment.

Resolution and disclosure

Requests identify a network by public UUID and an address. The network-family adapter normalizes the address before any lookup. Unknown normalized addresses are returned as UNKNOWN; public resolution never creates database records.

Only listed, moderation-approved agents and explicitly public relationships are eligible. Version 1 recognizes valid ownership proofs, public verified execution or revenue bindings, evidenced protocol declarations, and specially attested manager relationships. Private workspace wallets, custody-provider references, conflicts, and account membership are excluded. OBSERVED_ONLY relations are available only from the wallet-agents API when the caller opts in, and they are never described as ownership.

Version 1 score

Each current relationship has a policy baseline: verified owner 720,000 ppm, verified manager 680,000 ppm, execution wallet 610,000 ppm, revenue wallet 590,000 ppm, and protocol declaration 470,000 ppm. When an associated agent has a current product-trust projection, the relationship value is 40% baseline and 60% agent trust. Relationship confidence, agent-trust confidence, and source freshness determine its effective weight. Stale sources receive a 35% confidence reduction. Observed-only relations do not contribute to the score.

The result is a confidence-weighted mean. No eligible evidence produces scorePpm: null and UNRATED; the service does not invent a neutral score. Active wallet moderation can cap the result at CAUTION, while an active BLOCK_WALLET action produces BLOCKED with a zero score. Positive tiers are:

• TRUSTED: score at least 800,000 ppm and confidence at least 700,000 ppm;
• ESTABLISHED: score at least 650,000 ppm and confidence at least 500,000 ppm;
• EMERGING: any other non-caution rated result.

Factors identify direction, policy weight, confidence, an opaque or public evidence reference, observation time, expiry, and freshness. The source watermark hashes the exact relationship, trust-projection, and moderation inputs. Assessments are immutable snapshots, cached by that watermark, and expire after one hour.

Policy checks

POST /api/v1/wallets/reputation/check accepts at most 50 wallets and can require a minimum positive tier, minimum score, a publicly related verified agent, and a maximum assessment age. Every result is PASS, FAIL, or UNKNOWN with machine-readable reasons. failOnUnknown controls whether unavailable evidence fails closed; it never converts missing evidence into a pass.

Address lookups and batches consume the wallets:read API budget and are additionally constrained by per-IP and per-client rate limits to reduce enumeration abuse.