Not every change deserves a full tournament. A one-line fix and a payment-system migration should not cost the same to verify, because they do not carry the same risk. Fusion's meter is depth: you dial how much rigor a run gets, and the dial is the price.
The dial
A run's depth is set by a few honest knobs:
- Candidates: how many rival agents attempt the task in parallel lanes.
- Judges: how many blind votes the verdict needs.
- Review depth: how far past the objective gate the scrutiny goes.
Shallow settings make sense for small, reversible changes. Deep settings make sense when the blast radius is real: auth, money, data migrations, anything you would lose sleep over. The credit cost scales with the settings, so the invoice reads like a risk decision you made on purpose.
Why not meter tokens
Because tokens measure effort, and effort is not what you came for. A run that burns twice the tokens to produce an unverified result is not worth twice as much; it is worth nothing, which is what we charge for it.
Depth also has a property tokens never will: you can reason about it before the run. "Three candidates, full jury, deep review" is a sentence a team lead can approve in a planning meeting. "Approximately some millions of tokens, we'll see" is not.
Rigor as a product decision
The quiet claim inside this pricing is that verification rigor should be a first-class, visible choice, the way test coverage and review requirements already are in serious teams. The dial makes the choice explicit, the receipt records which setting produced the result, and nobody discovers after the incident that the critical change got the shallow treatment by accident.