A researcher replayed real Claude Code and Mooncake requests through a prefix-cache simulator to test whether alternative eviction policies could beat LRU, but failed across three approaches. The analysis reveals that under capacity pressure, most recomputation stems from tool-calling loops seconds apart rather than idle sessions, and the production LRU baseline proves surprisingly difficult to improve upon.
A token leak in Codex's goal mode causes users to burn through $200 weekly subscription limits in a day. The issue stems from the orchestrator restarting the model every 0.03 seconds while waiting for child agents, forcing it to reread 120–470k tokens per continuation until hitting rate limits. Proposed fixes include using models with native sleep support or implementing pauses between continuations.
LatentMathBench is a benchmark designed to test whether large language models like OpenAI's GPT-6 Astra can perform long chains of sequential reasoning in latent space rather than through visible chain-of-thought. The benchmark uses recurrent depth architecture with potential KV-cache sharing, which could allow reasoning to happen opaquely in cached states while producing arbitrary filler text, making the model difficult to monitor.
An open-source protocol for migrating Large Language Model KV-Cache states across datacenters, achieving 81.73ms latency and reducing GPU compute overhead by 95% compared to standard re-computation methods. The zero-copy transport eliminates prompt re-computation during session handoffs, enabling faster Time-To-First-Token latency.