Incremental Open-Ended Deep Research with Structured Harness
Incremental-OEDR updates research reports over time, cutting tokens and searches while preserving report quality.
The paper defines Incremental Open-Ended Deep Research, treating a report as an evolving state that preserves valid knowledge, revises outdated content, and adds new evidence. Structured Harness represents outlines, sections, and supporting evidence, with structured retrieval, a persistent evidence pool, and selective regeneration. A ten-year evaluation uses Single-Step and Long-Chain tasks under open-source and proprietary configurations. On DeepResearch Bench it reports up to 0.51 higher content ROUGE-L F1, 0.63 higher outline exact-match F1, 33% lower token use, and 61% fewer search calls than generating reports from scratch.
- Incremental-OEDR updates an existing report instead of rewriting it from scratch.
- Structured Harness keeps outlines, sections, and a reusable evidence pool.
- A ten-year setup tests single steps and long update chains.
- On DeepResearch Bench it cuts tokens 33% and search calls 61%.
- Products
- DeepResearch BenchDeepConsult
From the original article
1 min read at huggingface.co
Existing Open-Ended Deep Research (OEDR) systems primarily generate reports from scratch, making them inefficient for scenarios where research reports need to be continuously maintained as new information emerges.
By Meilin Chen, Hongyuan Bao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.