[DFL-AUDIT-153] GPT-5.6 usage evidence

## Finding

**Original score:** 1/10  
**Owner:** Human  
**Priority:** P2  
**Workstream:** `codex-narrative`

## Evidence

No repository proof; tokenizer use is not GPT-5.6 use.

## Acceptance

Resolve the finding with reproducible evidence, or record an explicit accepted-risk decision.

Dependencies: none.