It adjudicates process, not truth.
A case is submitted. Independent AI providers evaluate it. A Byzantine-fault-tolerant consensus reconciles them — and when they disagree enough, a human resolves it. The output is a governed verdict, never a conclusion.
The mechanism · three steps
01
Case submitted
A practice submits a case — title, summary, optional evidence. No conclusion is asked for; only a question is posed.
02
Providers evaluate
Independent AI providers assess the case in parallel. A Byzantine-fault-tolerant step reconciles their outputs against each other.
03
Verdict, or human
If providers agree, a governed verdict is emitted. If they diverge past threshold, a human reviewer resolves it. Either way: process, not truth.
The only four outcomes
— governance vocabulary, identical weight, never color-coded by outcome
High confidence
Providers converged within the consensus threshold. A verdict was emitted without escalation.
Requires human review
Divergence crossed threshold. A human reviewer becomes part of the record before this resolves.
Contradiction isolated
Providers disagreed in a specific, identified way. The contradiction is reported, not resolved.
Insufficient evidence
The submitted material could not support any governed outcome.
All four are reported with equal weight. Verdatrix adjudicates the evaluation process — it does not decide what is true.