test(quiz): first unit coverage for the answer submission flow

[tongshu2023]

Summary

Closes #32.

submit_answer handles grading, LLM error classification, wrong-answer records, LOOM mastery updates, and background diagnostics — and had zero tests. This adds the first unit coverage, calling the endpoint function directly against a real in-memory SQLite database (auth dependency bypassed by passing the user object; the LLM-backed classifier mocked for determinism — no real API calls).

Tests

tests/test_quiz_submission.py — 9 tests:

Happy path

• correct submission returns the proper grade with explanation, persists a correct PracticeResult, and creates no WrongAnswer record
• grading is case- and whitespace-insensitive

Wrong answers

• an incorrect answer triggers error classification; the category lands on both the PracticeResult and the WrongAnswer record
• classifier failure degrades gracefully with the error_classification_failed warning while still persisting the submission
• wrong answers schedule both background diagnostics (auto-derive + confusion detection); correct answers schedule the effective-review check

Error cases

• unknown problem id raises NotFoundError
• a problem in another user's course is invisible (ownership join enforced)

Mastery

• knowledge-pointed problems update LOOM ConceptMastery through the real BKT path (not a mock): practice/correct counters increment and the mastery score becomes positive

Results

pytest tests/test_quiz_submission.py -v: 9 passed locally.

Acceptance criteria

• At least 5 test cases covering happy path and error cases (9)
• LLM calls are mocked (no real API calls in tests)
• All tests pass with pytest tests/test_quiz_submission.py -v

🤖 Generated with Claude Code