A multi-turn tutoring application accumulates conversation state over 40+ exchanges. By turn 30, the messages array is consuming 60,000 tokens and the student's current problem is only 5 turns old. The architect wants to compress old conversation state without losing pedagogically important context (student mistakes, misconceptions corrected, concepts mastered). What is the optimal compression strategy?

C is correct because semantic compression via a 'student model' summary preserves the pedagogically critical information — what the student has learned, what errors they persistently make, and where they currently are — in a compact structured format. Replacing raw message history with this structured summary after every 10 turns keeps the messages array bounded while retaining actionable pedagogical context. A is wrong because removing every other message breaks the logical flow of multi-turn reasoning — questions lose their answers, explanations lose their context — and creates incoherent conversation history. B is wrong because for tutoring, earlier turns often contain the origin of a student's misconception or proof that they mastered a prerequisite concept — discarding this loses valuable diagnostic information. D is wrong because temperature controls output randomness and has no relationship to context dependency or memory compression.