Open Standard · v0.3 · MIT
Disclose Disclose

The open standard for merchant performance

Let any agent see how a merchant actually performs before it buys.

Merchants publish operational data at one open endpoint. Signatories with something to lose attest to it. Agents draw their own conclusions.

Agents can already read price, availability, reviews, and policy text. They still cannot reliably see refund speed, on-time shipment, chargeback ratios, or return rates. Disclose gives merchants one place to publish those performance signals: /.well-known/disclose.

agent · tool calls illustrative
About to buy from shop.example. Check the policy, UCP capability, and any performance disclosure before I confirm.
fetching policy + endpoints
GET /.well-known/ucp
UCP present. Checkout and fulfillment capabilities confirmed.
GET /policies/refund-policy
30-day returns, customer pays return shipping, refunds "within 10 business days."
Policy states the worst case. It says nothing about what actually happens. Checking Disclose.
Disclose MCP · get_merchant_disclosure
GET /.well-known/disclose
SignalValueAttestation
refund_processing_time_median_days2.4computed
on_time_shipment_rate96.2%signatory
chargeback_rate0.31%none
dispute_win_rate61%none
product_return_rate4.8%computed
platform_seller_tenure_days1460computed
Policy says up to 10 business days. Observed median is 2.4. The agent decides how much weight each attestation level earns.
/.well-known/disclose
Canonical discovery path. Same convention as robots.txt. No merchant registry required.
Three attestation levels
None (self-reported), computed, or signatory. Agents see the difference.
Open
MIT licensed. Published on the merchant's own domain. No platform lock-in.
01 · Discovery

One file on the merchant's domain.

Merchants publish a structured disclosure document at /.well-known/disclose. Agents that know the convention fetch it directly. Fallbacks exist for platforms that do not support the well-known path. No central registry of merchants is required.

02 · Attestation

Source and strength are explicit.

Every signal carries an attestation level: none (self-reported), computed (calculated from source system data by a third-party tool), or signatory (cryptographically signed by an authorized third party). Agents decide how much weight each level earns.

03 · Scope

Merchant, offer, or item.

Signals can be published at merchant, offer, or item scope. An agent comparing the same product across two sellers can read entity-level performance and, when available, SKU or item-level attributes from the same endpoint.

Sits below discovery and above the transaction.

UCP and ACP handle product discovery and checkout. MCP is the tool layer most agents already use. AP2 covers payment authorization. None of them standardize merchant performance signals: refund processing time, on-time shipment, chargeback rates, dispute outcomes, return rates, or platform tenure.

UCP answers how to buy. Disclose answers whether to buy, and from whom. It supplies machine-readable evidence and nothing else. No scores, badges, rankings, or recommendations. Agents weight the signals by their own risk models and the buyer's context.

Seven core signals to start. A full library behind them.

The Core Commerce Profile is the first standard signal set for agentic commerce. Each core signal uses a trailing 90-day window, a published methodology, and declared freshness, so the same number means the same thing at every merchant.

  • product_return_rateReturned units over shipped units
  • on_time_shipment_rateOrders shipped within the stated window
  • order_accuracy_rateOrders without wrong or damaged items
  • refund_processing_time_median_daysReturn receipt to refund, median
  • chargeback_rateChargebacks over completed transactions
  • dispute_win_rateDisputes resolved in the merchant's favour
  • platform_seller_tenure_daysDays active on the primary platform

The broader attribute library

  • Product quality
  • Returns and refunds
  • Fulfillment
  • Inventory and availability
  • Shipping and delivery
  • Financial risk
  • Customer support
  • Pricing integrity
  • Subscriptions
  • Sustainability and ethics
  • Identity and legitimacy
  • Review signals

Participation is voluntary. A merchant can publish one signal or all seven and add more over time. A missing signal has no defined meaning, and agents are told not to treat absence as poor performance or concealment.

A number is a claim. A Signatory makes it an argument.

The schema and the endpoint are easy to copy. What matters is who stands behind the data. A Signatory is a named third party that already holds a merchant's operational data and cryptographically signs specific signals. If the number is wrong, the Signatory's own reputation takes the hit.

none
Self-reported by the merchant. Useful, but unverified.
computed
Calculated from platform data by a third-party tool. Harder to fake.
signatory
Signed by an authorized third party with something to lose.

Signatories are listed in a public registry, authorized one signal at a time, and can be suspended or revoked in the open. A Signatory vouches for specific numbers, never for the merchant as a whole.

Returns platforms, logistics providers, and payment processors are natural Signatories. BNPL providers are especially strong candidates. They carry direct financial exposure when a merchant fails to deliver, so their attestations are backed by capital, not just reputation.

Platforms attesting to their own merchants have a conflict of interest. Independent Signatories are the design choice that keeps the evaluation layer from belonging to any one platform.

One endpoint. Provenance attached to every value.

An agent fetches a single disclosure document. Each signal arrives with its attestation level, observation window, methodology, and freshness metadata, so the agent knows what it is looking at and how recent it is.

"disclose:on_time_shipment_rate": {
  "value": 0.962,
  "unit": "ratio",
  "observation_window_days": 90,
  "window_start": "2026-07-03T00:00:00Z",
  "window_end": "2026-09-30T23:59:59Z",
  "generated_at": "2026-10-01T09:00:00Z",
  "reported_by": "example-logistics.com",
  "source_of_record": "example_logistics_api",
  "attestation_level": "signatory",
  "methodology_version": "core-commerce-v0.1",
  "attestation": {
    "signatory_id": "example-logistics.com",
    "signing_key_id": "exlog-2026",
    "signature": "eyJhbGciOiJFUzI1NiJ9..."
  }
}

Simplified and illustrative. See the spec for the full disclosure document. Agents interpret; the standard does not.

MIT licensed. Maintainer-led while the profile is validated.

Disclose is published as a framework, not a protocol. The distinction is intentional. The Core Commerce Profile, the merchant publishing model, the agent consumption model, and the Signatory registry are still being validated through implementation.

The specification is designed to be owned by no single company. The spec defines public triggers for moving to working group governance, such as live merchant publishers, independent agent consumers, and Signatory applications. Every material change happens in public on GitHub.

Commercial implementations, including Disclose Labs products like Sure Signal, follow the same rules as everyone else and get no special status under the spec.

The spec is live. The first cohort is forming.

The schema and reference tooling exist, including an MCP server agents can call to fetch disclosures. The next step is testing the model against real merchant data and real agent workflows.

We are looking for merchants willing to publish, potential Signatories willing to attest signals they already compute, and agent teams willing to consume the data and tell us what is missing.

We only use your email to follow up about the working group.

Agents should not have to guess.