The Substrate Floor: Nested Noise Floors for Auditable Abstention Across Heterogeneous Instruments
Year: 2026
Type: preprint
Abstract: An instrument abstains honestly when its signal falls below a noise floor it computes for itself. This paper generalizes that discipline to an ensemble of heterogeneous instruments: a finding survives only if its signal clears the outermost noise floor in play, where floors nest – operator within artifact within substrate. The same move lifts up a level: a cross-instrument ensemble abstains below a substrate floor, the dispersion of instruments’ verdicts on one aligned claim. Reconciliation returns a typed verdict – corroborated, contested, substrate-conditional, or jointly-unresolved – never a forced point estimate; agreement is triangulation, disagreement is the floor itself. A no-rescue lemma forbids agreement among individually-noisy instruments from manufacturing a finding. The rule composes instruments of different kinds that share no output space, reconciling an eight-dimensional measurement vector and a categorical specification audit on an aligned ordinal verdict, and below a specification instrument’s coherence floor abstention is mandated by a coverage-impossibility result. The contribution is an open-source computational framework – already shipped, with worked cases reproducing all four verdicts and a seeded Monte-Carlo map of where it diverges from meta-analysis, sensor fusion, and selective prediction. It is a typed-verdict lattice over nested floors, not a relabeling of heterogeneity statistics. Includes zharnikov-2026ay-substrate-floor.yaml (Paper Spec v0.1.0) – a machine-readable specification of the paper’s claims, assumptions, and dependencies. The paper’s full machine-first bundle (the SPINE claim/dependency graph and the ONTOLOGY term module) lives in the public repository; see github.com/spectralbranding/paper-spec for the standard. This PDF is generated programmatically from that machine-first source under a research-as-repository model.
Author: Zharnikov, Dmitry
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7165882367
Cites:
Cited by:
Related to: