Interference in LLM Interaction

This research note proposes an observation-level framework for identifying interference patterns in large language model interaction. It introduces a minimal diagnostic protocol to distinguish task-contingent effects of cultural default framing, decoding variance, and agreement shift, without making claims about internal model mechanisms.