Socragate
An embeddable pedagogical-integrity layer for AI tutoring products that classifies the learning task, selects an educator-approved assistance policy, checks candidate responses for direct-answer leakage and safety, offers graded hints or worked examples when appropriate, and records a parent- or teacher-visible learning-event capsule while keeping learner request, skill and assignment context, assistance policy, model candidate, safety flag, pedagogical decision, delivered response, learner work, assessment, teacher judgment, parent visibility, correction, and learning outcome separate.
AI tutors need to balance productive struggle with clear explanation, accessibility, safety, and legitimate requests for direct instruction. Socragate cannot promise to block every answer leak, and “never give answers” is not universally sound pedagogy. It must distinguish homework assessment, practice, formative hints, worked examples, accessibility accommodation, urgent safety, and teacher-authorized instruction. The learning-event capsule records observable prompts, hint levels, student responses, sources, and policy decisions; it must not expose hidden model chain-of-thought or create invasive child surveillance. Parent and teacher access depends on age, consent, school role, privacy policy, and learner rights.
An edtech product leader, tutor builder, curriculum organization, school technology team, or educational creator shipping an AI tutor and needing configurable assistance plus safety controls.
Tutor builders and edtech product teams have a named integrity and parent-trust problem.
Policies, classifiers, hint ladders, safety checks, event capsules, tests, and reports scale in software.
Two cross-references, three inbound connections, and two direct connections support the trust-layer thesis.
The source verifies open guardrail and PII primitives, a funded tutoring product built around not giving direct answers, a defined edtech-builder buyer, and no supplied horizontal pedagogical-integrity layer.
No structural copying cost is proven, pedagogical policy varies by age and task, direct-answer detection is fallible, child privacy and safety require operations, and the horizontal layer can be absorbed by tutor platforms.
Discussion
No comments yet — be the first to weigh in.