AI coach literature strands
The strand-by-strand literature reviews behind the AI coach work. The meta-analysis (../META_ANALYSIS.md) draws on them. Each lists only sources downloaded and read, gives design rules, and flags where the spec conflicts with the evidence. Archived copies sit in the session scratchpad under sources/<strand>/, to be published to gs://avantifellows-research-sources/ai-coach/.
| File | One line |
|---|---|
| ai_tutors.md | RCTs of LLM tutors and coaches. Answer-giving lowers unassisted scores. Key-grounded, step-gated tutors remove the harm and sometimes add gains. Students rarely use a tutor they are not routed to. |
| learning_science.md | Retrieval, spacing, interleaving, worked examples, feedback, mastery and negative marking, checked for strong JEE students. Also corrections to the Learning Framework’s citations. |
| adaptive_platforms.md | Mindspark, Khan Academy, ASSISTments, Cognitive Tutor, Alcumus, Brilliant and Alpha School. Outcomes track a protected, supervised slot more than the software. |
| hybrid_human_ai.md | Who does what between teacher, student and AI: tutoring dosage, coaching, task goals, AI-assisted tutors, early-warning flags and teacher time. |
| india_context.md | Indian ed-tech RCTs, BFLPE and rank feedback, Indic-language accuracy, Anthropic’s age rules, DPDP Act and Rules, NTA dates, and what a 40-student pilot can detect. |