L5D v3 (Core): Linear-to-5-Dimensions Multi-Channel Protocol for Safety UX Logging and Qualitative Auditing of LLM Dialogues

L5D (“Linear-to-5-Dimensions”) is a safety-first, non-encrypting multi-channel overlay for recording and communicating long-form LLM dialogue behavior. It separates a response into five channels: C — Content (task output)U — Uncertainty (epistemic status)B — Boundary (safety constraints and safe alternatives)P — Pressure (behavioral pressure markers; operator-coded or self-reported 0–10)M — Method (procedure used; steps/checks) The protocol is designed for Safety UX work: making uncertainty and boundary-handling legible, reducing confusion caused by “hedge-everything / refuse-if-in-doubt” patterns, and enabling qualitative audits (including pressure–risk mismatch and depth-avoidance analyses) without claims about model internals. This is the Core specification (v3). An expressive companion overlay (L5D-Ψ) is planned as a separate Zenodo record and can be used inside the Core channels without changing the schema. Related works in the same series:• Victor Calibration (VC), arXiv:2512.17956• Depth Avoidance, Zenodo DOI: 10.5281/zenodo.18168544• The Human-in-the-LLM Box, Zenodo DOI: 10.5281/zenodo.18357934