Structural Invariance Under Cross-Lingual Mirror Scripts: A Dual-Validation Framework: VITT × AIRT

Current AI safety evaluations face a fundamental blind spot: they evaluate either structural form or semantic meaning in isolation, rarely both simultaneously. This creates what we term the Verification Gap — a space where AI failures in non-dominant scripts, mirrored compositions, and structural anomalies go systematically undetected. This paper proposes the VITT × AIRT Dual-Validation Framework, combining calligraphic structural forensics (VITT — Visual Invariance Through Testing) with probabilistic meaning auditing (AIRT — Anti-plagiarism Integrity Red Teaming). Using three cross-lingual mirror scripts (Arabic, Japanese/Kanji, Korean Hangul) as stress stimuli and two documented Gemini failure cases as empirical anchors, we demonstrate that structural integrity and semantic coherence must be evaluated in parallel to constitute a complete AI output validation. Core Thesis: Integrity = Form × Meaning. Neither VITT alone nor AIRT alone is sufficient — their conjunction closes the Verification Gap.