CACI

The auditor of the machine, owned before the audit exists. The mechanism is procurement, not technology: the frontier labs have asked in public for third-party evaluators, and the federal government does not employ the people who could do that work at model scale — so if the ask becomes a requirement, the requirement becomes a contract, and contracts of that shape go to the firms already cleared to sit inside the DoD and the intelligence community. CACI is one of two or three such firms with the clearances, the cyber and signals bench and the past performance to be handed model-evaluation, assurance and red-teaming work the day it is funded. The thesis does not need commercial AI to freeze; it needs one printed award naming AI model evaluation or assurance, and a backlog that keeps growing while it waits. The expression is the common. The honest tension is printed with the position: the catalyst is a bill that has not been introduced and a partnership that has not been announced — the packet's inference, not a print — and the sovereign-demand leg already lives in the book under BAH and PLTR. The kill takes both halves: no qualifying federal AI-evaluation award by mid-2027 AND funded backlog falling year over year at two consecutive prints. Verdict due June 30, 2027.