- In short
- A stratified eval scores a model downgrade by task subtype rather than only on an overall average, because a single aggregate score can hide subtype-specific regressions. A rollback criterion decided before results are seen (for example, reject if any subtype drops below a set threshold) prevents post-hoc negotiation with the data. When some subtypes fail and others pass, routing by subtype to different tiers captures most of the savings without accepting the regression.
Full concept guide coming soon
We are building the in-depth, exam-aligned guide for this knowledge point. In the meantime, explore the prerequisites and related concepts below, watch the official Anthropic Academy lessons, and start an adaptive study session to master it with Archie.
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
References & primary sources
Adaptive study
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.