Methodology — how Provenance scores x402 endpoints

Plain-English description of every signal behind the numbers on this site, what they can and cannot prove, and how to dispute a rating.

Data sources

• Coinbase x402 Bazaar — the public discovery feed (23,338 listings at scan time), including each listing's own published quality.l30DaysTotalCalls and payer counts.
• GoPlausible x402 registry (Algorand) — 38 listings.
• Base mainnet (eip155:8453) — raw USDC (0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913) transfer logs to each merchant's payTo address over a 7-day window.

No private data, no scraping behind authentication, no payment interception. Everything used here is publicly readable by anyone.

The 8-signal organic-revenue-quality (ORQ) model

Each scored merchant gets a wash-risk score from 0 (fully organic-looking) to 100, combining eight signals over its observed on-chain receipts:

1. Distinct-payer count — how many unique wallets actually paid. One or two payers behind thousands of claimed calls is the strongest single wash indicator.
2. Payer revenue concentration (HHI) — Herfindahl index of revenue by payer cluster. “Top cluster holds 99% of revenue” means one actor funds nearly everything.
3. Funding-cluster analysis — payers funded from a common source wallet are collapsed into one cluster before concentration is computed.
4. Payment-timing regularity — coefficient of variation of inter-payment gaps and the share of identical-interval gaps. Metronomic timing is machine-generated load, not demand.
5. Wallet age & freshness — clusters of newly created wallets that only ever pay one endpoint.
6. Self-dealing loops — payout flows that return to wallets that fund the payers.
7. Claimed-vs-observed gap — registry-claimed 30d call volume versus on-chain receipts in the observation window (see the settlement caveat below).
8. Cross-endpoint payer overlap — the same tiny payer set boosting several endpoints at once.

Grades: A (<10), B (10–19), C (20–41), D (42–69), F (≥70). Organic-adjusted calls = claimed 30d calls × (1 − score/100). Wash-rank inflation = organic-adjusted rank − raw-volume rank: how many leaderboard positions the raw number buys beyond what organic-adjusted volume supports.

The settlement-visibility caveat (“n/v”)

x402 payments frequently settle through a facilitator, which may batch or defer transfers. In that architecture, few or no direct on-chain USDC transfers to the merchant's payTo address is normal and legitimate. Where a high score is driven mainly by missing on-chain receipts or a near-zero visible payer set — rather than by positive wash signals like metronomic timing or self-dealing loops — we display n/v (not verifiable): insufficient on-chain visibility (batch settlement possible) instead of a letter grade. The raw score is preserved in the open-data JSON for transparency, flagged with the same caveat.

Wrapper Watch: reseller signatures

Wrapper Watch flags endpoints whose public listing exhibits signatures consistent with reselling a first-party API from a domain the brand does not control. Signals, each contributing points to a 0–100 likelihood score:

1. Brand-term-off-domain — the listing references a first-party brand (in its description, tags, URL, or schema) while hosted on an unrelated domain.
2. Proxy language — self-describes as a proxy (“API proxy”, “powered by …”).
3. Generic passthrough schema — the input schema is a bare passthrough (a single prompt/query/url field), consistent with forwarding to an upstream API.
4. Cost-plus markup — per-call price is a near-multiple of the brand's public price.
5. Domain forensics — generic infrastructure hosting (e.g. *.railway.app, *.vercel.app) while referencing premium first-party data.
6. Live-402 probe — where noted, we confirmed the endpoint actually returns HTTP 402 Payment Required today.

Thresholds: flagged at ≥45; high confidence ≥70; medium ≥50. Every flag means “suspected — verification recommended”, and the resale-authorization status of every flagged endpoint is unknown to us. Brand pages exist so the one party who can verify authorization — the brand's own API/trust team — can review the evidence quickly.

Limitations

• Transfers to a payTo wallet may include non-x402 USDC receipts; we cannot always separate them.
• Low payer diversity can also reflect a young service or one large legitimate customer.
• Facilitator batch/deferred settlement hides receipts from direct on-chain observation (see n/v above).
• Brand-term matching can produce false positives (e.g. comparative mentions, tooling that merely integrates a brand). Comparative mentions are down-weighted but not perfectly.
• The observation window is 7 days for on-chain receipts vs the registry's 30-day claimed volume; the claimed-vs-observed gap signal accounts for this scaling, but short windows add noise.
• A flagged endpoint may hold a redistribution license we cannot see. We publish signatures, not conclusions.

Disputes & corrections

If you operate a listed endpoint or represent a named brand and believe a rating or flag is wrong — or you can demonstrate authorization — open an issue at github.com/apeirontrade/provenance-site/issues. Provide the endpoint URL and any evidence you can share. We correct errors and annotate verified authorizations in the next weekly build, and will publish the correction alongside the original.

Legal framing

This site publishes independent statistical analysis of public blockchain data and public API-registry listings, in the tradition of independent ratings and consumer-protection research. Statements about named endpoints are limited to (a) verifiable public facts (listings, prices, on-chain transfer patterns) and (b) clearly-labelled statistical opinions derived from them (“consistent with”, “suspected”, “verification recommended”). No statement on this site asserts, and none should be read as asserting, that any named party has committed fraud, breached a contract, or violated any law.