FE-012A Threats To Validity

• Analyst contamination from prior Framework Engineering discussions
• Synonym normalization bias
• Artifact selection bias
• Over-abstraction
• Loss of domain meaning
• Reconstruction subjectivity
• Single-extractor bias
• LLM bias if LLMs are used as extractors
• Confirmation bias toward existing primitive set

Interpretation Note

This experiment is especially vulnerable to analysts seeing what they expect to see.

Any execution should preserve extraction notes, disagreements, and rejected merges so later reviewers can audit where interpretation drift occurred.