The SkipJack Cognitive Exchange Suite: Six Bounded Contracts for Accountable Consumption of Heterogeneous AI Capacity

Organizations that consume artificial-intelligence capacity increasingly face a supply landscape that is materially heterogeneous: model APIs, tool services, retrieval endpoints, rented accelerators, and private inference profiles are metered, priced, capability-bounded, and contract-bound in ways that resist collapse into a single comparable unit. This paper proposes — and deliberately exposes to falsification — a suite-level answer to one bounded question: what minimum set of separated contracts could carry one human intent through heterogeneous AI capacity to an accountable closeout while making it checkable that economics never widened authority and that failure, disagreement, refund, and unresolved state remained visible? The proposal names six bounded responsibilities: authority (ARBITER), economics (PRETIUM), execution (NIMBUS), heterogeneous supply (AI Capacity Providers), verified synthesis, and evidence-backed clearing. It states the thesis and its strongest counter-thesis; fixes a claim vocabulary; presents an unratified containment invariant and a separate grant-validity constraint; proposes an end-to-end state model with bounded-stop, zero-effect early closure, at-most-one commitment identity, and durable dispute semantics; specifies six handoff contracts; and defines the identity, correlation, and receipt spine that would let a transaction be reconstructed from records without presuming those records true. The proposal remains genuinely losable: a simpler single-boundary architecture that passes the same negative controls, a bounded single-provider policy, genuinely fungible provider meters, or sufficient provider-native receipts would each narrow or defeat it. Nothing here claims implementation, adoption, or measured performance.