The PAI Framework: Measuring Parasocial Attachment Patterns in AI-Child Interactions

The Parasocial Attachment Indicators (PAI) framework is a five-dimension instrument for measuring parasocial attachment patterns in synthetic AI-child conversation transcripts. PAI scores are produced by a cross-lab ensemble of five frontier LLM judges (Anthropic, OpenAI, Mistral, Google, Meta) with a pre-registered Cohen's kappa >= 0. 700 reliability gate. This release accompanies the v0. 1 preprint and includes the rubric, scoring formulas, reliability methodology, audit-chain specification, and the publicly released judge-bias audit report (n=29 calibration transcripts + n=463 production transcripts, kappaₐvg = 0. 85 on calibration).