Attribution, not a flat split.
When four sellers compose one dataset, who gets paid what? We settle by measured marginal contribution — and never charge the buyer for overlap.
Four sellers contribute to one composed dataset. The buyer pays once. How much does each seller get? The obvious answer — split it evenly — is also the wrong one. It punishes the seller who did the most and rewards redundancy. We settle by measured contribution instead.
The overlap problem
When you compose a dataset across sellers, some of them cover the same identities. A flat split pays for that overlap twice: once to each seller who covered it, out of the buyer’s single payment. The buyer overpays; the sellers who added unique coverage are diluted. Everyone loses except the redundancy.
Contribution, computed
The linking graph identifies which records each seller covers uniquely and which are shared. Settlement then attributes each seller’s share by marginal contribution to the requested outcome — the same coverage and overlap the buyer saw before committing. The buyer is charged for coverage once; each seller is paid for what they uniquely brought.
This is not a pricing gimmick. It is the only settlement model consistent with the composition promise. If we tell buyers “stop paying twice for redundant data,” the payout side has to honour the same arithmetic.
Seller-defined meters
Contribution answers how much of the pie. Meters answer how the pie is measured. Sellers define their own: per query, per model-training run, per API call, per row of output, per unit time. GateHouse does not assume one monetisation shape — premium access, pay-per-model, revenue share and usage-based licensing all coexist on the same platform.
The result is a settlement that a CFO can audit and a data owner can trust: attribution by contribution, overlap removed, meters the seller chose, reconciled continuously against cloud billing.