STATE - AI Tutor Learning Guidance

Current truth

AI Tutor is a cross-module, on-demand learning guide. It provides contextual explanations and actionable guidance from platform knowledge sources, while preserving learning integrity during active attempts.

Rules

• Role baseline:
  • AI Tutor is teacher-like support for learning guidance, not an auto-solver.
• Invocation baseline:
  • on-demand only,
  • no auto-open panel behavior.
• Grounding baseline:
  • source priority is session context -> platform KB -> learner history -> general fallback.
  • response should expose provenance class (practice|course|blog|history|general_fallback).
• Active-attempt integrity:
  • no final answer/answer key/ready-to-submit rewrite during active attempt.
  • enforced hint ladder: orientation -> concept hint -> scaffolded steps -> self-check.
• Output artifact baseline:
  • concise answer,
  • guided checklist,
  • mini exercise,
  • diagram/mindmap-ready structure,
  • progress insight on demand.
• Action continuity:
  • non-trivial responses should include do now and/or deep-linkable next action when available.
• Existing high-attention plan contracts remain active:
  • 7-day plan structure,
  • reminder constraints,
  • required-day semantics.
• Smart Search boundary/handoff dependencies:
  • AI Inline (Layer 1) and AI Tutor (Layer 2) are explicitly separated.
  • Inline escalation opens Tutor with export packet or minimum fallback seed.
• Handoff payload performance baseline:
  • Inline-to-Tutor packet dùng tiered compression (full|balanced|lite) theo device/network profile.
  • Compression không được phá vỡ required minimum + fallback-open invariant.
• AI full-test mapping baseline:
  • AI Thi thu Full Test dùng canonical intent ACT_TEST với variant AIF_MOCK_FULL_TEST (không tách intent mới ở v1).
• Diagram/mindmap rendering baseline:
  • render 3 tầng: visual_native -> structured_graph_text -> guided_checklist_fallback.
• Provenance visibility baseline:
  • depth 3 mức (compact|standard|detailed) với segment default và detailed on-demand.
  • general_fallback luôn có disclosure giới hạn dữ liệu.
• Sandbox exception baseline:
  • chỉ cho phép sample/reference answer trong context non_graded_sandbox và không ảnh hưởng score/streak.
  • graded attempts vẫn giữ hard no-final-answer rule.

Boundaries

• This state does not define model/provider infrastructure choices.
• This state does not replace teacher authority for important course tests.
• This state does not change existing 7-day reminder delivery channel contracts.

Dependencies

• Upstream:
  • Smart Search handoff payload and module context adapters.
  • Practice active-attempt status and submission lifecycle.
  • Learning metrics/progress signals.
• Downstream:
  • AI Tutor platform UX specs,
  • module-specific handoff UX,
  • AI Tutor quality and release gates.

Decision trace

• DEC-0038: on-demand behavior baseline is retained for Home-level invocation.
• DEC-0024..DEC-0031: 7-day plan and reminder policy cluster remains active.
• DEC-0082: AI Tutor role/capability baseline.
• DEC-0083: grounded knowledge + recommendation contract.
• DEC-0084: no-final-answer active-attempt policy.
• DEC-0085: output artifact and progress insight taxonomy.
• DEC-0086: Smart Search AI Inline vs AI Tutor boundary.
• DEC-0087: inline-to-tutor export-context and fallback handshake.
• DEC-0088: mobile payload compression tiers for inline-to-tutor handoff.
• DEC-0089: AI Thi thu Full Test mapping to ACT_TEST + AIF_MOCK_FULL_TEST variant.
• DEC-0090: diagram/mindmap layered rendering fallback contract.
• DEC-0091: provenance visibility depth by segment with on-demand detailed mode.
• DEC-0092: sandbox-only exception boundaries for non-graded practice.

Open questions

• No blocker-level open questions for AI Tutor core logic at this stage.