Start with the product.
End with the falsifiable claim.
The full Citadel case in one short pass: progressive first use, operation evidence, three frozen local studies, and the exact result Sentient funding would test.
Watch the evidence reproduce.
This 42-second supplement is rendered from commands actually executed in the release checkout. The committed JSON retains every output line, exit code, output digest, and source revision.
Citadel starts with one command: /do. You describe the engineering outcome, and Citadel chooses the smallest operating lane that can carry it. When work lasts longer, Citadel adds repository state, recovery, coordination, bounded execution, and a concrete next action around Claude Code or Codex.
The research question begins where ordinary orchestration ends. A smaller model is not cheaper if its answer fails, the verifier forces a retry, or the operation hides part of its cost. Citadel binds the declared route to observed execution, a verdict outside the routed model, cost lenses, and a signed receipt.
V1 recorded more verified cells, but its apparent savings reverse when one same-route timeout pair is excluded. V2 matched baseline cell completion, but 12 small-model failures escalated to 7B and made the policy use 15.7% more GPU energy. V3 moved to six repository fixtures: integrity held, yet the 7.1% energy reduction missed its frozen 20% gate. Citadel published all three boundaries.
Sentient funding would test a learned operation-value controller across agent frameworks, repositories, models, tools, and hardware. The public target remains at least 80% absolute verified completion, at least 95% of a valid frontier baseline, and at least 30% lower measured end-to-end cost. Frontier must first clear 80% overall and 70% in every frozen task stratum.