Building a Micro School: The 2-Hour Core Model

Alpha School promises "top 2% test scores" and "2.6x faster learning" in just two hours daily using AI tutors—yet when Pennsylvania evaluated their charter application in January 2025, they rejected it unanimously as "untested." When journalists requested data, Alpha declined to share it. When a parent tested the homeschool version (same software), it showed only 1x baseline learning, not 2.6x. This episode examines the gap between marketing and reality: what "AI tutor" actually means (adaptive software, not generative AI), why AI tutoring design determines whether it doubles learning gains or reduces performance by 17% (Harvard vs Penn studies), and why the homeschool pilot reveals the in-person environment—not the software—may be the active ingredient. Read the full research report at https://research.yuda.me/podcast/episodes/building-a-micro-school/ep2-two-hour-core-model/report.md Key Sources: • Pennsylvania Department of Education Charter Rejection (Jan 2025): Called Alpha's AI model "untested" on all 5 statutory criteria • Harvard RCT - AI Tutoring Success: Scaffolded AI tutoring produced 2x learning gains (p < 10⁻⁸) • Penn/Wharton RCT - AI Tutoring Harm: Unrestricted ChatGPT access led to 17% worse test performance • Dan Meyer's Critical Investigation: Documents conflicts of interest and actual vs claimed model • ACX Parent Review: Notes homeschool version showed only 1x learning speed (not 2.6x)