Benchmarks-as-Contracts: A ReceiptBench Spec Template for Regimes and Closure

Benchmarks for tool-using agents are converging away from pooled scalar scores and toward regime-explicit evaluation: admissible evidence, checkable closure, and verifier-bounded verification budgets. Our prior work argued this shift is best understood as operational confirmation of Large Language Fields (LLFs): a field-level contract layer above models where meaning, evidence, and “done” are governed by reusable, versioned rules. ReceiptBench made this convergence executable by replacing single scores with typed receipts — minimal, time-scoped claims emitted by evaluators/verifiers — so that scores become explicit policy projections over receipt-space. This paper introduces a practical missing layer: a specification template that turns any benchmark into a contract artifact Π_b with declared admissibility, budgets, and closure predicates, and with mandatory receiptboard reporting that preserves subfield heterogeneity. We further show that a benchmark spec template is naturally a Metafield: a design-time contract family that instantiates benchmark contracts across tasks and domains while enforcing migration semantics to prevent silent semantic drift. The result is governance-ready evaluation: “done” becomes auditable, tail risk becomes visible through worst-subfield and closure-integrity signals, and benchmark comparability becomes a property of declared regimes rather than of leaderboard folklore.