[R] BEAM 100K memory benchmark: CSM vs Hindsight local artifact comparison I’m looking for feedback on a local agent-memory benchmark comparison, especially from people who care about evaluation methodology. I built an open-source R&D memory system called Context Swarm Memory (CSM). It uses bounded read-only memory shards, query routing, probe/recall/synthesis, cited packets, and explicit Committer-gated writes. The current comparison is against the accepted local Hindsight artifact on BEAM 100K: CSM: 0.757573 AMB score, 342 / 400 correct Hindsight: 0.733658 AMB score, 326 / 400 correct CSM uses 38.2% fewer answer-visible context tokens CSM is slower: 29.23s average retrieval vs 6.38s I want to be precise about the claim: This is not an official leaderboard claim. It is not a BEAM 10M claim. It is a committed local accepted-artifact comparison at 100K, and the next step should be independent replication or official chart acceptance.…