mercator-address-evidence

What this is not

It is not scam, rug or safety forensics. If you want to know whether a deployer is hostile, whether supply is concentrated or whether a migration narrative holds up, that is a different question and this does not answer it.

This answers one question only: is this chain and address the one the entity publishes for this purpose? A contract can pass here and still be unaudited, upgradeable to hostile code, or able to take your funds. PASS means intended target, not safe.

Why it exists

An agent that takes a contract address out of prose — a message, a reply, a doc page — and trades on it has authenticated the speaker, not the address. Those are different things, and only one of them is checkable before the transfer is irreversible.

Who pays, and from which wallet

Your wallet, not ours and not the platform's. This skill does not carry a wallet, does not custody anything and never asks for a private key. It describes a paid HTTP endpoint; your agent pays for the call the same way it pays for any other x402 resource.

On a managed platform the platform's wallet signs. On OpenClaw there is no managed wallet: whatever you already use to sign x402 — an x402 skill holding a funded key, a bounded agent wallet, or a pre-signed X-PAYMENT header you set yourself — is what pays here. If your agent has no x402 signer configured, the free preflight below still works and the paid route will simply keep returning 402.

A call that evidences no target comes back billable: false — and IS STILL CHARGED FOR. Settlement cancellation does not work under the x402 exact scheme (measured 2026-09-30), so the waiver this line used to promise could not be performed. That is a defect with a fix in progress, not a policy, and it affects only the no-evidence case. This text is dated: if it still says a fix is in progress after 2026-10-14, the fix has stalled and this sentence is the stale one.

The shape of the call

You bring three things and we check them. We never pick a target for you: a request with no candidate address always answers transact: false with reason candidate_address_required rather than guessing which of an entity's dozens of contracts you meant.

BASE="${MERCATOR_BASE_URL:-https://mercator-entity-evidence.fly.dev}"

# Free preflight — what will be checked, and what the answer will mean.
curl -sS -X POST "$BASE/v1/address-evidence/validate" \
  -H 'content-type: application/json' \
  -d '{"domain":"lido.fi","chain":"ethereum",
       "address":"0x73b047fe6337183A454c5217241D780a932777bD",
       "purpose":"emergency multisig"}' | jq .

# Paid answer — $0.25 USDC on Base, x402. Unpaid, this returns 402 with
# the payment requirements; your x402 signer retries it with X-PAYMENT.
curl -sS -X POST "$BASE/v1/address-evidence" \
  -H 'content-type: application/json' \
  -H "X-PAYMENT: $X_PAYMENT" \
  -d '{"domain":"lido.fi","chain":"ethereum",
       "address":"0x73b047fe6337183A454c5217241D780a932777bD",
       "purpose":"emergency multisig"}' | jq .

# See the payment requirements without paying:
curl -sS -X POST "$BASE/v1/address-evidence" \
  -H 'content-type: application/json' \
  -d '{"domain":"lido.fi","chain":"ethereum"}' | jq '.accepts'

What comes back

Measured live on 2026-09-13 against the deployed build (2848178), for the request above:

{
  "status": "PASS",
  "entity": "lido.fi",
  "chain": "ethereum",
  "chain_caip2": "eip155:1",
  "address": "0x73b047fe6337183A454c5217241D780a932777bD",
  "confidence": 0.9,
  "agreement": "2_of_2",
  "independent_origins": 2,
  "evidence_basis": ["entity_published", "explorer_verified"],
  "contract_role": "Emergency Brakes Multisigs",
  "checked_purpose": "emergency multisig",
  "checked_purpose_supported": true,
  "checked_address_supported": true,
  "self_published_only": false,
  "proxy": { "is_proxy": true, "proxy_type": "master_copy", "…": "…" },
  "action": { "transact": true, "reason": "…", "does_not_mean": "…" },
  "billable": true
}

Every row under evidence carries the operator, the URL, a verbatim excerpt and the time it was retrieved, and may only sit there if it states the chain and address that were returned.

The rules that make the answer worth paying for

• action.transact is derived, never assigned. Any conflict, any null value, anything short of a pass makes it false.
• transact: true also requires the entity's source to have said what the contract is. An entity owns its staking contract, its bridges and its emergency multisigs alike; "it is theirs" is not by itself a reason to send a transaction to it.
• A contract address is never decided by counting. Two credible sources that disagree produce no answer at all — the cost of the minority being right is your funds.
• For an upgradeable contract the canonical target is the proxy; the implementation is context and is expected to change without changing the answer.
• Provenance is graded, not flattened: entity_published, explorer_verified, registry_mapping, cryptographic_control and inferred mean different things and the payload says which it has.
• An explorer proves a contract exists; it never says whose it is.

Honest limits — read these before you rely on it

• A bare hex with no entity is UNRESOLVED. If all you have is 0xabc… pasted in a chat and no idea whose it is, this cannot help: there is no published claim to check it against. The name is the input.
• Coverage is the entity's own publishing. If a project does not publish its addresses on a page we can reach, the honest answer is UNRESOLVED, not a guess.
• Not a safety assessment. See the top of this file.
• Self-published only is reported as such: when the only source is the entity's own page, self_published_only is true and confidence is lower, because a compromised docs page is exactly the attack.

Price

Price, network, asset and payee above were read from the live 402 challenge on 2026-09-13, not from documentation. No signup, no API key, no account.

Companion skill

mercator-entity-evidence — the same evidence discipline applied to company facts (legal name, domain, headquarters, executives) at $0.10.