LongHarness tests evidence management for long-context agents
The framework supports agent answers by updating evidence and conclusions together across long documents.
Event / data period: 2026-10-09
Published: · Updated:
Researchers affiliated with Fudan University, Shanghai AI Laboratory and Peking University in China introduced LongHarness. The system manages source-linked evidence across long document collections; as new evidence arrives, it updates both the evidence pool and provisional conclusions.
In the evaluation on the first version of OfficeQA Pro, the paper reports a 60.2% exact-match rate with GLM-5.3-Flash. The authors describe a substantial increase over direct model use, but the result is limited to the constructed benchmark and the tested model and retrieval setup. It does not establish the same performance in every real long-context workflow.
Data
Research result
OfficeQA Pro V1; benchmark of about 100 million tokens; exact match
| GLM-5.3-Flash exact-match rate on OfficeQA Pro V1 | 60.2 | OfficeQA Pro V1; benchmark of about 100 million tokens; exact match |
Methodology
The value is limited to the benchmark experiment reported in the paper; it is not a field result or general performance guarantee.
Updated:
Subjects and places
Subject proposals
- Science and technologySuggested
- Preserved editorial category/topic; broad candidate onlyarxiv.org ↗
Geographic references
Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.
- China · Place role: Coverage · Place type: Country or area · Location precision: RegionalEvidence
- Explicit preserved editorial country/region; no city or national centroid inferredarxiv.org ↗