interrogatelisted

# Interrogate Spawn several parallel reviewers over the same diff, same intent, same rubric. Deduplicate their findings, then deliver a single bucketed verdict. **The deliverable is a verdict, not a patch. Do NOT auto-apply changes.** ## Read this before you trust consensus The original method assumed reviewers from different vendors — different training data, different blind spots — so agreement across models was strong independent evidence. **That premise does not hold here. Claude Code is Anthropic-only. Every reviewer is a Claude model.** They share pretraining lineage, post-training, and safety tuning, so their blind spots are correlated. Two Claude models agreeing is weaker evidence than two vendors agreeing, and — more dangerously — *four Claude models missing the same thing tells you almost nothing about whether it's there.* Two consequences, and they are load-bearing: 1. **Manufacture the diversity in the prompt, not the model picker.** Each reviewer gets a different rubric emphasis (its lens) so the reviewers are actually looking at different things. Model choice varies reasoning depth and cost; the lens is what varies coverage. If you collapse all reviewers onto the same lens, you are paying 4x for roughly one review. 2. **Silence is not clearance.** "No reviewer flagged the auth path" does not mean the auth path is clean. Report unanimous silence on a high-risk lens as an open question in the verdict, never as an all-clear. Do not describe this setup to th