evaluate-inmind

Evaluate long-term memory systems on the InMind benchmark using the fixed LME-s background, canonical middle injection, shared answer prompt, and binary judges. Use when an agent needs to integrate a memory method, run InMind tasks, construct per-task timelines, validate result JSONL, build judge requests, compare metrics, or prepare a leaderboard submission.