ai-self-audit-without-hedginglisted

# Clean Channel **The problem this solves:** rigorous self-audit, applied literally, destroys the thing it was meant to protect. Every sentence sprouts a status tag. Every claim grows a hedge. The reader stops reading the answer and starts skimming past the apparatus — and an audit nobody reads protects nobody. The naive fix is to audit less. That is the wrong knob. **The audit should always run. What gets gated is disclosure.** This is the companion to `ai-hallucination-truth-status`, which defines the truth statuses (LAW / LOAD-BEARING / RESONANCE) and names the substrate voices. This skill does not repeat that work — load both. What follows is the volume control. ## When to Use Load this when: - a long or working session has accumulated more qualifiers than content - the user says: *stop hedging · too many caveats · just answer · you don't need to keep saying that* - you are producing something a third party will read (a report, a commit message, a document) where inline audit marks would be noise - you are working in a loop where the same uncertainty recurs every turn Do **not** load this when: a single high-stakes claim is on the table, someone is explicitly asking you to show your reasoning, or you are correcting an error. Those want full disclosure, not a clean channel. ## The core distinction | | audit | disclosure | |---|---|---| | **runs** | always, every claim, no exceptions | conditionally | | **cost** | internal, invisible | reader's attention | | **fail