Stater Currency Architecture

Governed purchase of heterogeneous AI capacity creates an accounting problem that is easy to misname and therefore easy to mis-build. When an organization reserves bounded budgets against nonfungible capacity offers, commits value to execution it does not control, and must later prove — not merely assert — what happened to every unit of committed value, the record-keeping layer beneath that activity acquires requirements that are neither monetary nor incidental. This paper proposes, as a losable thesis, a requirements profile for a distinct future internal accounting contract: a non-transferable, single-principal, append-only accounting layer that preserves a stable internal accounting reference, represents variable cognitive-capacity purchasing power only through immutable versioned conversion-basis snapshots, maintains evidence-gated deterministic state across an untrusted-meter boundary, retains exact replayable records including durable unresolved state, and makes non-transferability mechanically testable through partial failure. The paper organizes the proposal into nine load-bearing requirement families, states a closed lifecycle and an exact conservation identity over stocks, and defines falsification tests that the proposal must be able to fail. It claims no currency, no token, no market value, no price stability, no legal status, no settlement capability, and no implementation. The strongest counter-thesis — that existing transactional, escrow, budget, and event-ledger patterns already compose to satisfy every retained requirement — is carried forward as a genuine test, not a rhetorical foil.