AT-301c · Module 2

Multi-Critic Patterns

4 min read

A single critic has blind spots. Two critics with different evaluation priorities surface issues that neither catches alone. The pattern: Critic A evaluates on technical correctness and structural integrity. Critic B evaluates on audience impact, tone, and engagement. Their scores are weighted by the artifact type — technical documentation weights Critic A at 0.70 and Critic B at 0.30; marketing content inverts the ratio.

The multi-critic approach adds cost — roughly 1.8x the single-critic model — but reduces the false-pass rate from 14.23% to 3.67%. A false pass is an artifact that clears the quality gate but fails in production. For customer-facing deliverables, the 1.8x cost is a bargain against the reputational cost of a false pass. For internal artifacts, a single well-calibrated critic is sufficient.

Do This

  • Use multi-critic for customer-facing deliverables — false-pass rate drops from 14.23% to 3.67%
  • Weight critics by artifact type — technical work weights the technical critic higher
  • Require critics to independently score before comparing — prevent anchoring bias

Avoid This

  • Use multi-critic for internal drafts — the cost-to-quality ratio does not justify it
  • Average critic scores — that hides disagreements that deserve investigation
  • Let critics see each other's scores first — the second critic anchors to the first