Working paper v1.2 (not peer reviewed). This paper proposes epistemic assurance as a minimal, assurance-facing diagnostic lens for one class of epistemic risk: cases where action-guiding representations stay coherent, fluent, and procedurally acceptable while becoming insufficiently governed by corrective contact with reality — the condition the companion paper names synthetic coherence. The lens targets one assurance object, representational answerability: whether action-guiding representations remain corrigible by evidence across the full chain that produces, verifies, challenges, corrects, authorises, and uses them. It uses three dimensions — grounding, contestability, and revision fidelity — as positions in a single corrective loop, with revision fidelity (whether corrected understanding reaches the artefacts that continue to guide action) receiving particular emphasis as the dimension most likely to reveal cross-interface failures. The lens is applied to three current governance and assurance pressure points — EU AI Act Article 14 (human oversight), ISO/IEC 42005 impact-assessment reassessment triggers, and SR 26-2 effective challenge — and includes an entry condition (the action-guiding representation under review must be specified) and a diagnostic-priority rule that avoids displacing more direct explanations such as fraud or negligence. The contribution is deliberately narrow: a cross-scale diagnostic hypothesis, most developed at AI-mediated and institutional-workflow scale, that tests correction mechanisms rather than truth directly and is intended to sharpen existing oversight, impact-assessment, and effective-challenge regimes rather than replace them. This paper forms the bounded, assurance-facing entry point into a broader epistemic-governance programme; the broader programme lies outside its scope. Changes from v1.1: the assurance object fixed explicitly as representational answerability, with epistemic assurance as the diagnostic lens; the first triad dimension renamed from evidential provenance to grounding, with evidential provenance retained as its operational sub-test; the representational chain extended to include authorisation (production, verification, challenge, correction, authorisation, and use); abstract rewritten as a single comprehensive paragraph; operational boundary test added distinguishing revision fidelity from ordinary document control; floor-outcome clarification added to the limitations and validation agenda; the lens’s ex-ante and ex-post uses distinguished explicitly; Figure 2 (minimal corrective loop) added; reference and citation corrections (Vaughan 1996 added; Ghareeb et al. author data aligned with the Nature record; SR 26-2 citation label; ISO/IEC 42005 wording precision; spelling and chain-definition consistency); redundancy trims and consistency edits throughout. Companion paper: Synthetic Coherence — DOI: 10.5281/zenodo.18916480.
Hermod Rypdal (Tue,) studied this question.