Reflective Light Cone V1.3 Local Validation Supplement: Companion Data and Code for "MUSE and Bounded Reflection"

Companion data and code for Toeda (2026), MUSE and Bounded Reflection: Mathematical Foundations of the Reflective Light Cone (10.5281/zenodo.20732746), section 21.5 and Appendix E. The monograph carries this V1.3 local validation supplement unchanged and states that it is deposited as a separate companion record; this is that record. Test A: a deterministic simulation of a finite perturbation budget for the reflective light cone (three scenarios; byte-reproducible under Python 3.10, 3.12 and 3.13). Test B: six arms of the OPERATE-FR Smoke-100 candidate operational track (100 items, one run per arm, temperature 0): Qwen 3.5 9B raw and MMV-engine governed, gpt-oss-120b (Groq) raw and route-transformer + post-validator, Gemma 4 12B raw and route-transformer + post-validator. Raw responses, scored summaries with per-task detail, the report, the aggregate summary and the scripts are included; the Smoke-100 item texts and labels are withheld because they seed the OPERATE-FR item pool. The 25 June 2026 files are byte-identical to the supplement MUSE describes (SHA256SUMS). A README added on 1 October 2026 gives reproduction steps, the public harness pointer and ten reading notes, including: in both post-validator arms a rule-based rewrite produced 14 of the 15 stale-premise successes; the overall gains of +0.22 hold without that item family (exact McNemar p = 0.0013 and 0.00014); the design cannot separate a capacity effect from an intervention effect; and the simulation's violation check cannot fail by construction. Supplemental evidence only: not an independent benchmark and not a proof of the Reflective Light Cone conjecture. Code AGPL-3.0-or-later; data and documentation CC BY-NC-SA 4.0. Disclosure: the author created the framework evaluated here. The consistency check and packaging were done with Claude Opus 5.5 (Anthropic) under the author's direction; the registered author is the human author alone.