Summary

• require Guided Learning choice quizzes to register complete option bodies
• normalize textual or prefixed expected answers to stable option labels before grading
• recover option descriptions from matching legacy ask_user turn events when possible
• persist normalized choice data to the Question Bank and add regression coverage
• update English and Chinese mastery prompts to describe the option contract explicitly

Root cause

mastery_quiz previously described its options argument as choice labels, so
the model could register only ["A", "B", "C", "D"] while ask_user held the
actual descriptions. The Question Bank then persisted the label-only array.

The same boundary also allowed expected_answer values such as Step 6, while
the interactive card returned C. Deterministic grading compared those values
directly and marked a correct selection as wrong.

User impact

Choice questions now retain the same full option text shown to learners, and
correct card selections are graded against normalized labels. Bare label-only
registrations fail fast so the agent can retry with complete data.

Validation

• ruff check deeptutor/capabilities/mastery/tools.py deeptutor/learning/tests/test_mastery_tools.py
• ruff format --check deeptutor/capabilities/mastery/tools.py deeptutor/learning/tests/test_mastery_tools.py
• python -m pytest -q deeptutor/learning/tests/test_mastery_tools.py (16 passed)
• manual Docker smoke test for full option persistence and Step 6 → C normalization
• browser verification of Question Bank rendering and Wrong Only filtering

Fixes #603