Back to Academy
Relevance 7/10Prompting and EvaluationAdvanced7 min read
Long-Context Evaluation
Long-context evaluation tests whether models use and retain relevant information across large context windows.
Why it matters for annotators
Long-context reliability is increasingly important for enterprise assistants.
Visual mental model
Long input context -> response fidelity and retrieval checks.
Examples (bad vs good)
Scenario: Real annotation scenario involving Long-Context Evaluation
Bad: Labeling quickly without applying project rubric.
Good: Applying rubric criteria, documenting rationale, and escalating uncertainty.
Common mistakes
- Skipping guideline details for edge cases.
- Applying inconsistent criteria across similar samples.
- Avoiding escalation even when uncertain.
Submission checklist
- Read the latest guideline update before each batch.
- Apply rubric dimensions explicitly in each decision.
- Escalate ambiguous items with concise rationale.