ReceiptBench for Field Networks: Recursively Composable Governance via Typed Receipts

Typed receipts make evaluation outcomes auditable by emitting minimal, structured, time-scoped claims issued by evaluators/verifiers — rather than collapsing evidence into pooled scalar scores. In operational settings, however, closure decisions traverse organizational and contractual boundaries: local teams declare “done” under local admissibility rules; enterprises gate promotability under unified closure predicates; and cross-organization workflows require reconciliation under multi-owner due process. This paper formalizes Field Networks as composable governance systems built on a runtime authority stack — Field → Overfield → Metaoverfield — with Subfields as regime slices inside any runtime contract and Metafields as orthogonal design-time contract families that instantiate and migrate runtime contracts across layers. Building on the LLF lens that interprets benchmark evolution as convergence toward field-level contracts, we define how receipts compose across authority boundaries, how worst-subfield gates become deployability constraints, and how contestability and expiry enable reconciliation without sharing internal detector logic or sensitive payloads. The resulting architecture makes portability and auditability properties of the stack rather than artifacts of ad hoc integrations.