// Neutral clearing protocol

The clearing layer for AI training data.

Anonymous, request-driven matching between AI labs and vetted data & RL-environment suppliers. No brokers. No identities. Just the match.

YBacked by Y Combinator
// How it clears

Three steps. Both sides stay blind until settlement.

001

Request

A lab posts an RFD: task, eval harness, budget, license, constraints. The request goes live; the lab's name and roadmap never do.

002

Match

Vetted suppliers are matched to the spec by capability, never by identity. Demand finds supply without either side exposed.

003

Settle

Every sample is scored against the lab's own eval. On match, the deal clears through escrow, and only then do identities reveal.

FOR LABS · DEMAND

Source data you can't buy off a shelf.

POST
Define an RFD: task, eval, budget, license terms.
STAY ANONYMOUS
Suppliers see the spec, never your name or roadmap.
VERIFY
Every sample is scored against your eval before you pay.
RL ENVIRONMENTS
Commission custom environments, not just static sets.
Post an RFD
FOR SUPPLIERS · SUPPLY

Get matched to demand that already exists.

GET VETTED
One capability review, then you're a vetted supplier.
SEE REQUESTS
Browse anonymized requests by capability, never by who's asking.
PROVE
Submit against the eval; the protocol scores you objectively.
CLEAR & GET PAID
Settle on match, identity revealed only at clearing.
Apply as supplier
// Neutrality

United Data takes no side, holds no inventory, brands no supplier. It is a protocol, not a broker, just the rails the trade clears on.

Anonymous

No identities until clearing. Demand can't leak; suppliers can't be poached around the protocol.

Neutral

No markup on either side, no owned supply we'd rather sell. We never compete with the people on our rails.

Verified

Every match is scored against the demand-side eval. Provenance, licensing and indemnity clear with the deal.