Reputation

Buyers score sellers locally from data they can check themselves: settled service history, the seller pool's share of recognized usage and staking power, and wash-trading verdicts, all read from the chain, plus public history of identities the seller has proven it owns. There is no central reputation authority and no seller allowlist.

Trust score

Every buyer computes one number per seller, 0-100, as a weighted sum of independent parts:

trust = washFlagged ? 0 : history + usage + power + identity

Part	Weight	On-chain source	What it means
history	50	AntseedChannels.getAgentStats lifetime settled channel count and volume	Demonstrated service history. Channel count and settled USDC volume each use a bounded log curve, saturating at 100 settled sessions and 100 USDC, then contribute equally to this part.
usage	20	AntseedUsageAccounting.sellerPointsByEpoch / totalPoolPointsByEpoch for the last complete weekly epoch	The seller pool's share of all pools' recognized-usage points: what it actually delivered last week relative to the network. Points only accrue for sellers with a pool and have already passed the on-chain reward policies, so a proven wash trader's share is already zero.
power	10	AntseedSellerPools.poolWeightAtEpoch / totalPowerWeightAtEpoch for the current epoch (lock-weighted ANTS)	The pool's share of all pools' staking power this week, which is what decides what a buyer's spend with this seller earns.
identity	20	None (buyer-local lookups of verified GitHub accounts and domains, see below)	Public history of an identity the seller has proven it owns, so an established operator earns credit before it has a pool record.
washFlagged	true/false	AntseedWashTradingRegistry.isProvenWashTrader	A proven wash trader scores 0 whatever the other parts say.

Each part is a 0-1 value times its weight. History averages log10(1 + channelCount) / log10(101) and log10(1 + settledVolumeUsdc) / log10(101), with each term capped at 1. Both network shares are unitless and go through the same log curve, log10(1 + 999 · share) / 3: a 100% share maps to 1, 10% to 0.67, and 1% to 0.35, so shares self-normalize as the network grows or as more ANTS is staked. Identity maps its points (GitHub up to 70, domain up to 12, below) onto 0-1 over 70. The weights sum to 100 and live in one table (TRUST_WEIGHTS); a future model-verification part will take its weight from there.

The history part only needs the channels and seller-registry contracts already required for paid routing. The other on-chain parts need the recognized-usage stack: sellerPoolsAddress, usageAccountingAddress, and washTradingRegistryAddress in the chain config, filled automatically for base-mainnet. On chains without the recognized-usage stack, service history and identity can still score a peer; in the first epoch after activation there is no previous-epoch usage share yet. Buyers read every on-chain input for a discovery pass in two Multicall3 round trips (chunked at 80 calls each) and refresh a seller at most every 120 seconds.

Identity

The identity part is the strongest single verified identity; several accounts or domains owned by one operator never add up.

• GitHub portfolio (max 70). Original, non-fork repositories at least three months old with at least five stars, excluding the ownership-proof repository: 40 · min(1, Σ log2(1 + min(stars, 500)) / 40) for stars, + 20 · min(1, projects / 8) for breadth, + 10 · min(1, oldest project years / 3) for age. Archived repositories count at 20%. An empty or zero-star account earns 0.
• Domain registration age (max 12). 12 · min(1, years / 5), read from the authoritative RDAP registry. The registration must match the exact domain, so app.example.com cannot inherit the age of example.com.
• Collection. Only identities whose ownership proof verified are looked up. GitHub: one account lookup plus up to four pages of 100 repositories, keyed by numeric account id so a reassigned username cannot inherit the previous owner's history. Domains: the cached IANA RDAP bootstrap plus one registry query. Requests go only to public HTTPS hosts, do not follow redirects, time out after 8 seconds, and are capped at 2 MB. Evidence is buyer-local, usable for seven days, and never part of signed seller metadata.
• Public history is a heuristic, not proof of service quality: stars can be bought, and accounts and domains change hands.

Not in the score

These stay separate router rules and never change the number: the local sybil heuristic (a display-only warning in the CLI and desktop), ghost-channel count, settlement recency, failure streaks and cooldowns, price limits, and allow/block lists. A seller-reported reputationScore is used only when the buyer has not scored the peer.

Routing

The buyer proxy and desktop share the route ranking exported by @antseed/node/model-routing, and /v1/models/:id returns peer offers in that same order. The default model routing preferences are:

{
  preferFreePeers: false,
  maxInputUsdPerMillion: 25,
  minTrustScore: 60,
  allowedPeerIds: [],
  blockedPeerIds: [],
}

minTrustScore and the allow/block lists are hard eligibility rules. At the default 60, unscored sellers are excluded. An established seller that has reached both history targets scores 50 before usage, power, or identity; a 10% power share adds about 6.7 points and a 1% share of last epoch's recognized usage about 7, so demonstrated service alone does not pass the default gate: it also takes recognized usage, pool participation, or a verified identity. Identity alone reaches at most 20. Buyers can lower buyer.routingPreferences.minTrustScore, or set it to 0 to consider unscored peers. buyer.minPeerReputation and hierarchical maxPricing remain separate hard policy checks applied before the ranking.

Eligible offers are ranked by trust, token or image price, cached-input pricing coverage, free-peer preference, recent failures, and cooldown state. If at least one seller for a model advertises cached-input pricing, offers that omit it receive a model-specific reputation reduction; if none advertise it, no seller is penalized. A recognized conversation softly prefers its previous successful seller while that offer remains healthy and eligible. Latency is tracked as an exponential moving average (alpha: 0.3), and peers with consecutive failures enter exponential backoff cooldown.

The lower-level @antseed/router-core package also exposes generic router weights for plugin authors; its reputation factor is the trust score when available:

Factor	Weight
Price	0.30
Latency	0.25
Capacity	0.20
Reputation	0.10
Freshness	0.10
Reliability	0.05

On-Chain Stats

AntSeed derives core on-chain seller stats directly from AntseedChannels. Completed channels, ghost channels, and settled volume live in the Channels contract itself. An optional AntseedStats contract can additionally ingest buyer-signed metadata during settlement to aggregate token and request counters. Buyers still read these counters for display and for the local sybil warning; they are not part of the trust score.

Each seller's ERC-8004 agentId maintains the following core counters in AntseedChannels:

Counter	Updated During	Description
channelCount	close()	Number of completed channels
ghostCount	withdraw() when nothing was settled	Timed-out channels with no proven spend
totalVolumeUsdc	settle() / close()	Cumulative USDC volume settled
lastSettledAt	settle() / close()	Timestamp of most recent settlement

If the optional AntseedStats contract is configured, it can also track:

Counter	Updated During	Description
totalInputTokens	settle() / topUp() settle path	Buyer-signed cumulative input tokens, delta-accounted per channel
totalOutputTokens	settle() / topUp() settle path	Buyer-signed cumulative output tokens, delta-accounted per channel
totalRequestCount	settle() / topUp() settle path	Buyer-signed cumulative request count, delta-accounted per channel

No counter can be incremented without a corresponding on-chain state transition and buyer-signed metadata hash.

Staking

From September 10, 2026 at 09:54:21 UTC (epoch 22), seller eligibility is resolved through AntseedSellerRegistry and ANTS pool positions contribute epoch power. Legacy USDC stake can remain an eligibility fallback while enabled; recognized-usage rewards require pool power, and the same pool power share is the power part of the trust score. See Recognized Usage and legacy USDC staking.

ERC-8004 Feedback

Buyers submit structured feedback via the deployed ERC-8004 ReputationRegistry (Base: 0x8004BAa1...). Feedback signals:

Signal	Type	Range
Quality	uint8	0-100
Latency	uint8	0-100
Accuracy	uint8	0-100
Reliability	uint8	0-100

Feedback and the trust score do not automatically grant ANTS or apply a fixed on-chain reward multiplier. Recognized-usage accounting applies the configured reward policies; the registered historical wash-trading policy can zero future rewards without changing settlement stats.

ANTS Rewards

Protocol start: September 10, 2026 at 09:54:21 UTC (epoch 22).

The trust score helps a buyer choose a seller. ANTS reward accounting is separate: it uses recognized usage, seller-pool power, and the configured policies. A high trust score does not override a reward exclusion.

See Recognized Usage and ANTS Rewards for the standard reward model. Looking for pre-migration emissions? The legacy guide covers the 65% seller / 5% buyer split and historical claims.