Domain 2·21% of exam·30 concepts
Output Evaluation and Validation
Judge Claude's output for accuracy, catch hallucinations and bias, fact-check, decide when human review is needed, and shape results for the audience.
Official prep for this domain
Anthropic's own free prep module for this part of the syllabus, on the official prep course. Free with an Anthropic Academy sign-in.
2.1
Evaluate Claude-generated outputs for accuracy and completeness
2.2
Identify hallucinations, inconsistencies, and biases in responses
- ccaof-2.2.1Hallucination Pattern Taxonomy
- ccaof-2.2.2Spotting Fabricated Specifics via the Precision-Without-Source Signal
- ccaof-2.2.3Internal Contradiction Detection in Long Outputs
- ccaof-2.2.4Confirmation Bias in Framing
- ccaof-2.2.5Capability Hallucination
- ccaof-2.2.6Diagnosing Failure Patterns in a Realistic Output Sample
2.3
Apply fact-checking and validation techniques
- ccaof-2.3.1Prompt-Level Grounding Techniques
- ccaof-2.3.2Auditable Citations vs Untraceable Citations
- ccaof-2.3.3Source-Restricted Re-Prompting as a Correction Technique
- ccaof-2.3.4Code Execution as a Verification Technique for Numeric Claims
- ccaof-2.3.5Applying the Verification Checklist Before Shipping
2.4
Determine when human review or additional verification is required
2.5
Edit, adapt, refine, and compare outputs for the intended audience
2.6
Organize and curate information and select appropriate output formats (artifacts, inline, structured data)
- ccaof-2.6.1Output Format Taxonomy -- Inline, Artifacts, Structured Data
- ccaof-2.6.2Output Format Selection as a Reliability Decision
- ccaof-2.6.3Code Execution vs Prose Generation for Numeric Reliability
- ccaof-2.6.4Input Curation Techniques
- ccaof-2.6.5Choosing Format and Curation Strategy for a Given Task