Fusion bills on one event: the run met its frozen Definition of Good. No pass, no charge.
Why charge on proof
Most AI tooling charges for effort: tokens in, tokens out, compute time. That pricing quietly makes failure profitable for the vendor. Every hallucinated retry is revenue.
We sell verified outcomes, so the meter runs on outcomes. The internal ledger, not the payment processor, is the source of truth: a run's receipt decides whether it was billable, and billing is a downstream projection of that decision. If the gate failed, if hard criteria were not met, if the jury marked the run contested, the ledger records a non-billable run and you pay nothing for it.
A verification company should only get paid when the verification contract is met.
That sentence is the business model, not a slogan.
What this does to our incentives
Charging on proof puts us on the correct side of the table:
- A sloppy contract that nothing can pass produces free runs, so we are motivated to help you write criteria a machine can grade.
- A flaky gate that fails good work produces free runs, so gate quality is our cost problem, not your gamble.
- Verification standards cannot quietly erode, because eroding them is the only way to charge for junk, and the receipt makes erosion visible.
The honest fine print
"Free" means the failed run does not consume your credits. If you bring your own model keys, the model vendor still charged you for the tokens; Fusion's default local and BYO modes keep that spend inside your own accounts where you can see it. Every run's usage is counted in tokens, per role, on the receipt, pass or fail, so you can check what a run actually consumed. What we will not do is take a fee for work that did not meet the contract you froze.
The receipt says what passed. The invoice agrees with the receipt. That alignment is the pricing model.