A Framework for Strong Context Design
This lesson introduces a systematic classification framework for context inclusion decisions. The framework replaces intuitive relevance judgments with a repeatable, category-based filter.
Three-Tier Classification
Each candidate piece of context is classified into one of three categories:
Essential
Required for task correctness. Without this information, the model cannot produce a correct result. Always include.
Helpful
Not required for correctness but improves output quality or efficiency. Include conditionally, when context capacity permits.
Noise
Irrelevant or distracting to the task. Inclusion degrades output quality. Exclude unconditionally.
Application Procedure
- Begin with Essential context. Include only what is required for correctness.
- Add Helpful context incrementally. Introduce supplementary information only after the essential baseline produces correct output.
- Exclude Noise unconditionally. Information that does not contribute to the task actively degrades performance.
Key Takeaways
- The Essential / Helpful / Noise classification provides a repeatable decision filter
- Context should be built incrementally from the Essential baseline
- Noise exclusion is unconditional — its inclusion always degrades output