Answers are grounded in this paper and cite their sources.
Test of what carries the gain of a frozen annotation harness for LLM coding of teacher talk moves. Nothing is optimized. Claude Sonnet 5.5 codes 104 new passages from 15 teachers of TalkMoves Subset 2 with twelve instruments: the harness, the manual, the manual plus one class of rules, and the harness without one class of rules. Classes are rules already in the manual, operationalizations, added conventions and rules in tension with the manual. Confirmatory hypotheses compare added conventions with repetitions of the manual in both directions, with a teacher-level cluster bootstrap and Holm correction. Registered before the passages were drawn. The deposit includes all instrument texts except the manual.
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: