test: AI Features — Aristotle tutor, AI grading pipeline, voice & video exercises

[guillermoscript]

Scope

All AI-powered features: Aristotle course tutor, AI auto-grading for exercises/exams, voice exercises, and video exercises. These require an OpenAI API key with credits in .env.local.

Prerequisites

• OPENAI_API_KEY set and funded in .env.local
• Course with course_ai_tutors row: enabled = true
• Course must have lessons, exercises, exams

Personas

Persona	Email	Role
Student	student@e2etest.com	student (Default School)
Teacher	owner@e2etest.com	admin/teacher (Default School)

AI Components

Aristotle Tutor (/dashboard/student/courses/[courseId]/lessons/[lessonId])

• Floating chat panel visible when course_ai_tutors.enabled = true
• Sends first message → receives streaming response
• Tutor context includes correct lesson title and course structure
• Tutor knows student's completion status (which lessons done, exam scores)
• Pass/fail check uses 70% threshold (not passing_score column — that doesn't exist)
• Session created in aristotle_sessions table
• Messages saved to aristotle_messages table (both user + assistant)
• Past session summaries included in subsequent sessions
• Tutor stays on-topic — off-topic question redirected to course material
• No chat panel when course_ai_tutors.enabled = false

AI Exercise Grading (/dashboard/student/courses/[courseId]/exercises/[exerciseId])

Exercise types that use AI grading:

• free_text — open-ended text answer

• voice — audio transcription + evaluation

• video — camera recording + evaluation

• Submit free-text answer → AI evaluates, returns score + feedback

• Feedback rendered as markdown in results view

• Score saved to exercise_completions table

• XP awarded after submission (check xp_transactions row created)

• Resubmission: new exercise_completions row created (not overwrite)

• Teacher can view AI feedback in teacher submissions view

Voice Exercises (plan-gated: pro+)

• Plan gate: voice exercise shows upgrade prompt on free / starter plan
• pro+ plan: microphone permission requested
• Record audio → transcribed → AI evaluates
• Playback of recorded audio before submit
• Submit → score + feedback returned

Video Exercises (new feature — VideoExercise component)

• Camera permission requested
• Record video → saved
• Submit → AI evaluates
• Plan gate: same as voice (requires pro+)

AI Exam Grading (/dashboard/teacher/courses/[courseId]/exams/[examId]/submissions/[submissionId])

• After student submits exam, AI-graded submissions appear in teacher view
• Teacher can override individual question score
• Teacher can override overall score
• Save override → updates exam_submissions record
• Student score in progress page reflects teacher override

Known Issues (pre-confirmed fixed)

• exams.passing_score column does not exist — all pass logic uses hardcoded 70% threshold
• lesson_completions has no tenant_id — removed from Aristotle route query
• exercise_completions has no tenant_id — removed from Aristotle route query

Definition of Done

• Aristotle streams responses with correct course context
• AI grading returns score + feedback for all exercise types
• XP awarded on completion
• Teacher override saves and reflects to student
• All AI calls use model from AI_MODELS config (not hardcoded)

[guillermoscript]

QA — ⛔ BLOCKED on OpenAI quota; read-paths verified

Blocker (prerequisite not met)

The OPENAI_API_KEY in .env.local is present (164 chars) but unfunded. Direct probe:

POST https://api.openai.com/v1/chat/completions
→ HTTP 429  { "type": "insufficient_quota",
              "message": "You exceeded your current quota..." }

So all live-AI test cases are untestable right now: Aristotle streaming responses, AI grading of free_text/voice/video exercises, and AI exam-grade generation. These need a funded key (see the prerequisite in this issue).

Verified without live AI ✅

Check	Result
Aristotle panel visible when course_ai_tutors.enabled = true	✅ course 9999 ("Quick Test Course") as alice@student.com: floating "φAristotle" launcher (aria "Open Aristotle"), dialog labeled "Aristotle" / "Course tutor", quick-prompts ("Explain this to me", "Practice problem on this", "How does this fit the bigger picture"), Submit.
Lesson page renders with tutor context	✅ lesson content + 2/2 course progress + lesson nav
AI Exam Grading → teacher override	✅ already DB-verified in #287: per-question override → overall score recomputed (70→68) → review_status='teacher_reviewed' → reflects to student.
Pass logic uses 70% threshold (no passing_score col)	✅ code-confirmed (column doesn't exist)

Minor UI note

The Aristotle floating launcher (bottom-right) is visually overlapped by the Marker.io "Report a Bug" dev widget — they share the bottom-right corner. Dev/staging-only tool, won't ship to prod, but on local it covers the tutor button. Not filing as a bug.

Not testable until key funded

Streaming responses, session/message persistence (aristotle_sessions/aristotle_messages), AI exercise grading + XP award, voice/video record→transcribe→evaluate, plan-gating live behavior. No code changes — the AI pipeline code was corrected in prior PRs; this is purely an environment/credits blocker.