The line that never moves.

“Fuse intelligence. Never authorise combat.”

“It raises the alert. It never takes the action.”

“Not a number someone typed. A number a named, verifiable process produced.”

“The AI produces the input. The clinician makes the call.”

“It assembles the intelligence. It never renders the judgment.”

“It builds the case. It never lays the charge.”

Models propose; humans commit

“Baru never authorises action. It proposes, correlates, and explains; a cleared human operator decides. That line never moves.” Its counterpart on the discovery side reads the same way: AI does the reading; people make the findings.

Delegated authority is bounded

Agents work inside trust zones Z0–Z3 under a deny-by-default policy engine, and scoring formulas, rubrics, and taxonomies live in code as authoritative ground truth — the language model interprets, extracts, and drafts; it never assigns the score. Every effect that reaches the outside world is gated behind an earned trust zone and an accountable human decision.

The record proves it

Every finding carries a source citation resolvable to a specific document and passage; outputs lacking citations are rejected, and coverage gates verify mechanically that nothing was silently dropped. The finished work product carries a certification of that entire chain — every claim cited to source or rejected, every release signed off by a human. Verification never requires trusting the machine; it requires checking the record.