Mathine — Evidence In, Trust Out

Zero-Trust Science: A New Architecture for Scientific Closure (Beyond Peer Review)

Mathine: Scientific Closure Tribunal Machine
Link: https://doi.org/10.5281/zenodo.18509038 [1]

Scientific claims increasingly depend on toolchains—datasets, lab protocols, model weights, preprocessing pipelines, numerical solvers, proof checkers, and fast-changing software environments. In that regime, “trust” often fails to travel: a claim can be locally persuasive yet globally fragile because downstream users cannot cheaply verify the decisive steps. [1]

This paper proposes a zero-trust architecture for science: never trust by default, always verify by portable artifacts. The core shift is to treat correctness closure as a protocol outcome rather than a social outcome—deterministic, portable, and repeatable under declared admissibility rules. [1]

Three building blocks anchor the architecture. First, Tribunals: explicit closure policies that define what counts as admissible evidence and what “closed” means for a claim. Second, Receipts: minimal verifiable artifacts that let an independent verifier check the decisive steps without replaying the entire upstream world. Third, typed HOLD states: first-class outputs that localize what failed to close (and why) instead of forcing false certainty. [1], [2]

A fourth layer makes this scale under drift: canonicalization and identity control so verification remains meaningful as environments, libraries, and dependencies change. Without an identity layer, even “good receipts” decay into uncheckable narratives when toolchains move. [1], [5]

Crucially, the goal is not to eliminate peer review but to re-scope it. Peer review stays focused on interpretation, significance, and assumption critique, while correctness closure becomes increasingly receipt-driven and machine-checkable—an unavoidable requirement under AI-driven claim volume. This makes scientific trust portable in the same sense that proof logs and certificate checking make mathematical trust portable. [1], [3], [4]

References
[1] R. Figurelli, “Zero-Trust Science: A New Architecture for Scientific Closure (Beyond Peer Review)”. Zenodo, Feb. 6, 2026. https://doi.org/10.5281/zenodo.18509038
[2] R. Figurelli, “Math Machines: The Systems Architecture of Mathematical Trust”. Zenodo, Feb. 6, 2026. https://doi.org/10.5281/zenodo.18511647
[3] R. Figurelli, “The End of Exhaustive Search: A General Theory of Solving Hard Problems by Certified Transport”. Zenodo, Feb. 2, 2026. https://doi.org/10.5281/zenodo.18460311
[4] R. Figurelli, “From Scores to Receipts: Introducing ReceiptBench, a Typed-Receipt Protocol for Governance-Ready Evaluation”. Zenodo, Feb. 16, 2026. https://doi.org/10.5281/zenodo.18661829
[5] R. Figurelli, “Benchmarks-as-Contracts: A ReceiptBench Spec Template for Regimes and Closure”. Zenodo, Feb. 17, 2026. https://doi.org/10.5281/zenodo.18675035