Most scoring systems hide disagreement. Average three scores into one number and a confident 9 plus a worried 4 becomes a bland 6.5 that nobody actually believes. Fusion does the opposite: disagreement is recorded, labeled, and shown to you.
What a split means
When a jury votes 2–1, the majority carries the decision, but the dissent is preserved verbatim in the run record. The dissenting judge's criterion-level objections stay attached to the receipt.
This matters because a dissent is usually specific. The losing vote is rarely "this is all bad"; it is "criterion four is not actually satisfied because the migration path loses data on the legacy table." That sentence is the single most valuable line in the whole run for the human who has to merge it.
Contested runs fail closed
With a smaller jury, a clean split has no majority at all. Fusion marks that run contested rather than manufacturing a winner. A contested run is not a verified run; it is an explicit signal that the evidence did not settle the question, and a human decides what happens next.
Failing closed is the point. A verification layer that always produces a confident answer is indistinguishable from no verification layer.
Where you see it
The receipt shows the vote as it happened: which criteria were unanimous, which were carried on a split, and what the dissent said. Two runs can both end in "pass" while carrying very different risk profiles, and the receipt is honest about the difference.
When you review a Fusion result, read the dissent first. It is the cheapest risk report you will ever get.