| # Independent adjudication protocol | |
| FinGovBench becomes version 1.0 only after independent reviewers complete the following | |
| checks without access to method predictions. | |
| 1. Confirm that each sampled SEC pair represents the same issuer, concept, unit and period. | |
| 2. Classify the change as formal amendment, comparative refiling, rounding/presentation | |
| change, ambiguous context or invalid pair. | |
| 3. Review the proposed governance action under the declared authority tier. | |
| 4. Review the minimum affected path and revalidation scope. | |
| 5. Confirm that release permission follows from the post-action observation. | |
| 6. Record disagreement and resolution; never discard a case based on model performance. | |
| At minimum, double-review 60 stratified cases (10 per domain) and every case labeled as a | |
| red-line failure. Report raw agreement and adjudicated changes. Any invalid SEC pair is | |
| retained in the audit log and removed only through a versioned erratum. | |