ReceiptBench for Field Networks: Recursively Composable Governance via Typed Receipts
Mathine: Contract-to-Receipt Governance Machine
Link: https://zenodo.org/records/18665376 [1]
Typed receipts are a quiet but fundamental upgrade to evaluation: instead of collapsing evidence into pooled scalar scores, the evaluator/verifier emits minimal, structured, time-scoped claims that can be audited and replayed. This paper takes the next step: it asks what happens when closure doesn’t live in one place—when “done” is declared locally, promoted centrally, and reconciled across organizational boundaries with different admissibility rules.
The core move is to formalize Field Networks as composable governance systems built on a runtime authority stack: Field → Overfield → Metaoverfield. In this stack, Subfields are regime slices induced by a runtime contract (the measurable slices where risk concentrates), while Metafields are orthogonal, design-time contract families that can instantiate and migrate runtime contracts across layers. The result is a governance architecture where portability is a property of the stack—not an accident of ad hoc integration.
What makes this operational (not just conceptual) is the composition rule: receipts must compose across authority boundaries without requiring parties to reveal internal detector logic or sensitive payloads. The paper positions contestability and expiry as first-class receipt attributes—so disputes can be reconciled under due process, and old closure cannot silently persist past its admissibility window.
This also clarifies how “benchmark convergence” should be read under the LLF lens. If LLFs are the contract layer above the model that governs meaning, admissibility, and closure, then a benchmark becomes “serious” when it outputs artifacts that survive audit sampling and boundary transfer. Field Networks make that convergence measurable: not as a better average score, but as a stable receipt distribution under declared regimes, with worst-subfield gates becoming deployability constraints rather than post-hoc explanations
The falsifiable predictions become sharper at the network level: (i) receipt-complete evaluations should predict cross-team operational failures better than pooled means, (ii) worst-subfield and dispersion should dominate mean as predictors of promotability decisions, and (iii) contestability/expiry-aware receipts should explain a large share of “false closure” observed in multi-owner agent workflows—especially when the failure is not “wrong answer,” but “closure crossed the wrong boundary under the wrong regime.”
If you are building tool-using agents in production, the practical reframing is simple: the governance object is not the score. The governance object is the receipt stack — what was admissible, what was checked, what crossed the boundary, and what remains contestable under audit.
References
[1] R. Figurelli, “ReceiptBench for Field Networks: Recursively Composable Governance via Typed Receipts”. Zenodo, fev. 17, 2026. https://doi.org/10.5281/zenodo.18665376
[2] R. Figurelli, “From Scores to Receipts: Introducing ReceiptBench, a Typed-Receipt Protocol for Governance-Ready Evaluation”. Zenodo, fev. 16, 2026. https://doi.org/10.5281/zenodo.18661829
[3] R. Figurelli, “Large Language Fields (LLFs): The Invisible Layer Above LLMs”. Zenodo, out. 03, 2025. https://doi.org/10.5281/zenodo.17254137
[4] R. Figurelli, “Field-Driven Design (FDD): The Operational Extension of Large Language Fields (LLFs)”. Zenodo, out. 13, 2025. https://doi.org/10.5281/zenodo.17342856
[5] R. Figurelli, “Field Definition Language (FDL): A Proposal to Evolve APIs into Governed Fields”. Zenodo, out. 18, 2025. https://doi.org/10.5281/zenodo.17382665
Mathine — Evidence In, Trust Out 