Maximize context window efficiency, reduce latency, and prevent lost-in-middle issues through strategic masking and compaction. Use when token budgets are tight, tool outputs overflow the context, conversations drift from intent, or latency spikes from cache misses.