Ensemble Review

Overview

Run the same review task across multiple Claude instances and aggregate results. This catches findings that a single pass misses (due to attention variability) and builds confidence through consensus. Particularly valuable for high-stakes reviews where false negatives are costly.

Core Pattern: N-Instance Review

import asyncio
from anthropic import AsyncAnthropic

client = AsyncAnthropic()
N_INSTANCES = 3