Answers are grounded in this paper and cite their sources.
Estimation and replication study: the frozen annotation harness from Study 1 is applied without further optimization to 125 passages from 19 teachers of TalkMoves Subset 2 (TalkBack) that were not used in Study 1. Primary estimand: Cohen's kappa against human gold with a teacher-level cluster bootstrap CI for Claude Fable 5.1 coding sentence by sentence; secondary: harness, model, unit and model-family effects (Claude Sonnet 5.5, GPT-6 Luna) and replication of Study 1. Registered before the passages were drawn.
No takes yet. Share an insight, caveat, or question.
Simon Filler (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: