Existing Open-Ended Deep Research (OEDR) systems typically generate reports from scratch, which the authors argue is inefficient when reports must be continuously maintained as new information appears. To address this, a new paper on Hugging Face introduces Incremental Open-Ended Deep Research (Incremental-OEDR), a setting that treats a report as an evolving research state.
Under Incremental-OEDR, updates preserve valid knowledge, revise outdated or incomplete content, and incorporate newly available information. The proposed Structured Harness represents reports as structured collections of outlines, sections, and supporting evidence, and provides structured retrieval, a persistent structured evidence pool, and structured generation for selective report updating and evidence reuse.
The authors also establish a temporal evaluation framework spanning ten years, with a Single-Step Task and a Long-Chain Task to assess incremental updates over individual transitions and long-term update chains.
Experiments on DeepResearch Bench and DeepConsult, under both Open-source Configuration (OC) and Proprietary Configuration (PC), reportedly show that Incremental-OEDR maintains competitive report quality while improving report continuity and reducing research costs. On DeepResearch Bench, the approach achieves up to 0.51 higher content-level ROUGE-L F1, 0.63 higher outline-level EM F1, 33% lower token consumption, and 61% fewer search calls than OEDR.
Further details are available on the project page: https://ioedr-project.github.io/.