Papers, reports, essays and evidence records by Margaret Stokes concerning Aurora-Lens, epistemic legitimacy, runtime AI governance, persistent state and intellectual provenance.
The featured publication documents the scope, sequence and attribution of Aurora-PEF and Aurora-Lens in later discussions of runtime AI governance.
A documented correction of the chronology, scope and attribution of Aurora-PEF and Aurora-Lens in later discussions of runtime AI governance.
Canonical record: Zenodo DOI 10.5281/zenodo.22867653
OSF archival copy: OSF DOI 10.17605/OSF.IO/RH8GY
SSRN: Pending review
Foundational papers and reports on Aurora-Lens, the consequence boundary, persistent referential grounding, and epistemic legitimacy as a governance layer.
Aurora-Lens is a deterministic commitment-governance architecture that determines whether candidate information, interpretations, determinations, generated outputs, releases or actions may become operational consequence.
In July 2026 Anthropic identified the J-space: a sparse, privileged subspace of a language model's activations in which some representations become available for report, modulation and reasoning. The paper examines what may enter governed runtime clearing under persistent-state architecture.
The World Before the Word argues that language and AI systems become answerable only by referring to a world that persists independently of their representations.
Proposes that cognition is activation-bound and present-centred, that apparent continuity is reconstructive, and that coherent cognition requires persistent referential grounding in the world. It considers implications for human cognition, AI memory, identity and inference-time governance.
Introduces epistemic legitimacy as a governance requirement for AI systems and presents Aurora-Lens as a provider-agnostic runtime governance layer that controls whether a conclusion may be admitted, clarified, refused or stopped. The SSRN public record was created on 16 February 2026; its public document was updated on 6 March 2026.
Version 1: 10.5281/zenodo.18653120 · Later version: 10.5281/zenodo.19504665 · SSRN Abstract 6244239
Applied governance work mapping Aurora-Lens to OECD guidance, ISO/IEC 42001 evidence requirements, AuthZEN analysis, and cross-architecture terminology.
This independent technical note analyses the AuthZEN Access Request and Approval Profile — Draft 1, published 27 July 2026, and whether portable approval state conveys approver-eligibility basis.
Examines how Aurora-Lens can provide decision-level runtime evidence within an ISO/IEC 42001 artificial intelligence management system (AIMS).
AI governance discourse is increasingly fragmented by inconsistent terminology. Concepts such as admissibility, authority, continuity, legitimacy, runtime governance, auditability, and consequence are mapped across architectures.
Maps Aurora-Lens to OECD due-diligence guidance through admissibility gating, deterministic interventions and tamper-evident audit records.
Evidence records preserving development sequence, prior art, maintained unresolved state, and repository provenance.
Principal readable record and accompanying evidence files preserving the communications and chronology concerning Aurora-Lens, VALO and REHT; Margaret Stokes's authored contribution to VALO; access to related repositories; and the documented provenance of those materials.
Records priority concerning the maintained unresolved state and documents its subsequent appearance in the PEACE Protocol, including the documented chronology and supporting evidence.
A dated record of the development, disclosure and publication of Aurora-Lens and related architecture. Public SSRN record created 29 April 2026; updated 8 May 2026.
Essays published outside the formal preprint record.
Published in Build With Her, Issue 08. I came to AI through curiosity, then became increasingly troubled by hallucinations, collapsed ambiguity and the difficulty of trusting systems that could produce confident but unsupported answers. That led me to build Aurora-Lens, a system designed to maintain a persistent world, preserve uncertainty when it cannot be resolved, and stop actions before unsupported claims become consequences.
ORCID, SSRN, and the public PyPI package record for Aurora-Lens.
The OSF project for Epistemic Legitimacy as a Governance Layer for Large Language Models: Architecture and Implementation is at osf.io/86bxj.
For patent filings, dated disclosures and the development chronology, see Proof & Precedence. For software packages and implementation resources, see Resources.