Human review controls
People remain accountable. The record should prove it.
Public-program review needs more than a score. It needs a visible line from assignment and conflict handling through rationale, authority, approval and any override.
Book a ProtoGrants walkthroughReview assignment
Make responsibility explicit before scoring starts.
A governed review model should identify who received the file, which reviewer group they represent and what deadline or service standard applies.
The current ProtoGrants reviewer surface displays stored reviewer fields. Assignment actions and work queues remain product direction.
Conflict and recusal
Treat conflict handling as case data.
Declaration, matching, recusal, reassignment and quorum should be reconstructable without relying on private email or meeting memory.
This is part of the recommended ProtoGrants operating model. It is not represented as a working feature in the current demonstration.
Criteria and rationale
A number cannot stand in for a decision reason.
Weighted criteria can organize a review, but the record also needs the evidence considered, the reviewer rationale and the authority for the final decision.
Current scores are demonstration data. No AI resource is connected, and the demonstration does not automate a final decision.
Approval and audit
Preserve the decision package that a future reviewer will need.
Recommendation, moderation, approval, override reason and dissent should remain attached to the application as a human-readable decision history.
A production implementation must test this scenario against the buyer's policy, authority matrix, retention rules and audit expectations.