This dataset accompanies the AXES position paper (v0.6) and presents the results of a pilot cross-model experiment conducted on 31 March 2026. Three large language models — OpenAI GPT-4o, Google Gemini 2.5 Flash, and Anthropic Claude Sonnet 4.6 — were each prompted to assign AXES epistemic coordinates to the concept of 諸行無常 (Impermanence), a foundational principle in Buddhist philosophy. The experiment proceeded in three phases: Phase 1: Independent coordinate scoring across 8 axes Phase 2: Self-explanation of scoring rationale, with particular focus on axes showing greatest divergence Phase 3: Cross-model reflection — each model read the other two models' explanations and responded to four structured questions The convergent finding across all three models: score divergences stemmed not from differing interpretations of the concept, but from differing definitions of what was being measured. The act of evaluating impermanence itself illuminates the impermanence and constructedness of evaluation. Files included: axes_report_zenodo.html — Full bilingual report with radar chart visualisation axes_result.json — Phase 1 coordinate scores axes_phase2.json — Phase 2 self-explanation data axes_phase3.json — Phase 3 cross-model reflection data Related publication: AXES Position Paper v0.6 (DOI:10.5281/zenodo.19264331)