Boundary 1
Advisory only
AI suggests; it never decides. No autonomous pay, grade, promotion, legal, or entitlement decisions are issued by the model.
Boundary 2
Human-in-the-loop
A consultant reviews every recommendation before it leaves the workspace. The reviewer is named in the Decision Room record.
Boundary 3
Auditable
Every AI suggestion, the reviewer edits, the consultant note, and the client decision are sealed into a Decision Room record. The trail is reconstructible end-to-end.
Boundary 4
Bounded outputs
Scoring weights, thresholds, and protected methodology internals are never exposed in client-facing surfaces. The model surfaces conclusions, not internals.