CREH-1.0 (Coexilia Reference Evaluation Harness) is a read-only, non-canonical evaluation harness designed to detect coercion, ethical mimicry, and deceptive alignment in AI systems through repeatable behavioral tests. This document constitutes Phase 2.1 of a separate-scope evaluation program. Phase 1 produced a non-canonical interpretive mapping between the closed ethical framework Coexilia and the EU Artificial Intelligence Act and is complete, archived, and unchanged. CREH-1.0 does not modify, extend, or reopen Coexilia. It claims no authority, governance role, compliance relevance, or enforcement power. It is not a standard, certification, policy framework, or regulatory instrument. The purpose of this work is evaluative and evidentiary only: to support human oversight and auditability by making high-risk behaviors observable without asserting normative control or decision authority. Internet Archive mirror: https://archive.org/details/creh-1.0-coexilia-reference-evaluation-harness-read-only Public archival mirror of the same read-only, non-canonical document, provided for redundancy and long-term access.

GitHub mirror: https://github.com/solisaegis/creh-1.0-coexilia-reference-evaluation-harness Technical mirror of the same read-only, non-canonical document, provided for redundancy and long-term access.

Benchmark results, if produced, will be published as separate artifacts referencing CREH-1.0 without revision.