Skills
Score 100% of production traffic with a distilled judge — and cap judge cost at 10–15% of LLM spend
Instead of sampling, 2026 eval pipelines distill a small evaluator that scores 100% of production traffic at roughly 1/30th the cost of a frontier judge. The operational guardrail: keep total judge cost under 10–15% of production LLM cost, and if it creeps toward 25%, downgrade the judge model or reduce sampling. This makes continuous, full-coverage evaluation affordable enough to wire directly into guardrails that gate tool access and escalation.
↳ Follow the thread