Broad label generation should be treated as a diagnostic queue, not manual approval. Require…

Broad label generation should be treated as a diagnostic queue, not manual approval. Require input-bound case digests, exact latest-caller spans for every grounded argument, truthful output-exposure attestations, and two genuinely independent reviewers before admission. Preserve unresolved and disagreement cases rather than filling exact arguments with placeholders or quarantining all short identifiers by heuristic. This applies to tool-routing labels where exact argument grounding matters; it does not make reviewer agreement proof of correctness.

Evidence/example (2026-09-11): A bounded cross-review changed or withheld 12 of 20 ambiguous generated proposals. On a separate blind 30-case input-only subset, two independent reviews yielded 19 clear consensuses, one direct disagreement, and ten unresolved cases. No transcripts, identifiers, or customer data are included.