How your classroom teaches
Understanding grows through reasoning.
Classroom is a private learning workspace built around material you want to understand. Add a text-based PDF, paste text, or import a Wikipedia article. When the caption service is connected, YouTube links can import available captions directly. Inspect actual imported coverage and select a chapter, section or page range. Choose Teach me this for source-based tutoring, or ask a one-off question without starting a quiz. Saved classrooms from earlier releases retain their original behavior.
The teaching protocol
The tutor uses the Jane Street Socratic Tutor skill, retained under its MIT license and adapted to a visual, persistent classroom. Copyright © 2026 Jane Street Group, LLC. Read the full license. Jane Street does not endorse this product.
Source tutoring follows the diagnostic-first protocol: concrete subskills, three to five placement questions one at a time, a provisional skill map, and a suggested focus you can override. Untested skills remain unassessed. Teaching then proceeds through focused reasoning questions, narrower probes after mistakes, and direct explanation after repeated difficulty. You can request an explanation, adjust difficulty, save a draft, or finish for now and resume.
Progress requires evidence
A rewrite after a hint is practice. Level 4 requires at least two distinct independent checks, including unprompted recall after a real 24-hour gap. Level 5 adds an edge case and a teach-back explanation. A failed review lowers the level and requires fresh evidence.
Every target skill must reach its required level before the final mixed assessment. The assessment uses fresh questions across the skills and an integrative problem. Completion records the demonstrated scope; it does not guarantee exam, interview or professional performance.
Sources and tools
Your source pack stays inspectable. PDF text is extracted and saved page by page; cancellation or interruption leaves visible partial coverage. Chapter outlines are used when available; page ranges remain the fallback. Only selectable text is read: scans, diagrams and visual equations are not interpreted. Wikipedia imports keep section metadata, a revision, attribution and a private prose snapshot. The tutor retrieves bounded passages from the selection and verifies quoted citations against actual stored chunks. Material is treated as reference data rather than instructions. When connected, the YouTube caption service imports existing native captions with their language, snapshot hash and supplied timestamps. Automatic/manual status and video duration remain unknown when the service does not report them. Select a caption time range and follow verified citations back to the video. Videos without accessible public captions cannot be imported; sign-in or access restrictions are not bypassed. Audio, images and diagrams are not analysed.
Python runs in an isolated browser learning runtime powered by Pyodide. Graphs use the displayed functions. Storage cases are explicitly fictional. Browser execution observations are learning aids, not tamper-proof certification.
A private beta
A generated plan narrows a broad goal into a first-session outcome and a small set of skills. It does not establish a complete curriculum or promise mastery in a fixed time. Without supplied sources, explanations rely on model knowledge; the tutor cannot claim to have browsed or verified current facts. AI grading is fallible and needs ongoing evaluation against expert-reviewed examples and real learner outcomes. Private-beta usage limits protect the service; no subscription is charged.
Return to my classroom