← LearnProofFactoring pilot: evidence before labels
Classroom workflow · classroom-2027.2
Create a classroom assessment
- Register as a teacher and create a class. Share its joining code with students.
- Upload a lesson PDF up to 256 KiB under Materials, review extracted text or enter objectives manually, and approve the supported concept mappings.
- Under Assignments choose the class and material. Select Core, Transfer, or Author supported questions. Preview the generated questions and answer keys, confirm suitability, and Publish assessment.
- Students join with a learning name and private password, open the assessment, and complete answers, confidence, probes, targeted practice and reassessment. Returning students use the same credentials with Join for the first time unchecked.
- In Learning Evidence, inspect scores, diagnostic evidence, confidence, hints, revisions, reflections and reassessment. Review hypotheses with a reason and export evidence.
For controlled validation, mark an assessment as synthetic and enroll two ordinary accounts. Explore Demo also uses this same learning system. Synthetic records are excluded from research exports by default and are not evidence of learning impact.
What the engine does
LearnProof checks exact polynomial coefficients and unordered root sets. A factor-sign error or zero-product error is a candidate explanation, not proof. Distinct probes can support a hypothesis, leave uncertainty, or reveal conflicting evidence. Ambiguity stops for teacher review. Targeted practice uses graduated hints; novel reassessment does not. Practice completion alone is not mastery.
The pilot covers coefficient relationships, factor signs, zero-product reasoning and root substitution. It does not grade arbitrary mathematics. Unsupported notation is marked for manual review.
Plan an independent evaluation
Compare LearnProof adaptive practice with Google Forms practice plus conventional teacher feedback. Use reviewed parallel baseline, post-test and delayed assessments scored independently of internal practice. Google Forms participants need no LearnProof account.
External CSV fields: participant_id,condition,baseline,post,delayed,independent,source,assessment_date. Conditions are learnproof_adaptive and google_forms_teacher_feedback. Scores range from 0–100; blanks stay missing. Keep personal names out of study codes and source descriptions.
Privacy and limitations
Use pseudonyms. No student content is sent to an LLM. PDFs are private to the teacher; extraction is limited and is not malware certification. File access expires after 90 days, but physical deletion is owner initiated. Remove pilot evidence and cohort outcomes through their explicit deletion controls. Provider backup retention is separate.
Supervised is a teacher-declared condition, not automated proctoring. Neither browser behavior nor response time proves independence, attention or motivation. Obtain institutional permissions and appropriate research/privacy review before actual students participate. Software operation does not establish ethics approval or educational effectiveness.
Legacy learning-cycle protocol remains available for existing checks.