aps-practice-recommendation

Start from the core question of why this problem for this solver at this moment, factoring in user goal, skill state, attempt history, hint usage, and motivation. The skill provides requirement tables, recommendation types, and MVP logic templates to structure that decision.

Difficulty labels are contextual and inconsistent across platforms, and they ignore concept knowledge, insight difficulty, implementation burden, and the user's actual mastery. The skill recommends targeting a predicted solve-probability band using rating models like Elo or Glicko instead.

Fixed curricula suit beginners and fundamentals because they are clear and explainable, while rule-based recommendation is a transparent early MVP choice. Personalized ranking and hybrid approaches fit later stages once sufficient interaction data exists.

Use adaptive placement testing that selects maximally informative problems near a 50 percent predicted solve probability, converging on an ability estimate in roughly five to ten problems. Glicko rating deviation or IRT standard error provides a natural stopping rule.

Hide tags until an attempt when practicing pattern discovery or interview preparation, since early tag reveal reduces insight development. Tags can be revealed as a hint level with logging, or shown upfront only for beginner concept drills.

Collaborative filtering suffers cold-start problems for new users and problems, and it reproduces filter bubbles by recommending patterns the user already over-practices. Content-based matching on concept tags and difficulty is the stronger default for this domain.