Answers are grounded in this paper and cite their sources.
In China's national variety approvals, each cultivar's growth duration and disease grades are reported under the ecological type zone of its trials. A cultivar approved for several zones carries one set of values for each zone, and these values often disagree between zones. We describe a multimedia knowledge pipeline that keeps the type zone, assessment method and source with every value. Using the official approval descriptions of 5,885 rice and maize registrations and the 2025 announcement, we kept every grade and duration under the type zone that reports it. We then counted cultivar--disease pairs whose grades differ between zones, compared pooled and conservative shortlists with the answer for the intended zone, compared the pipeline's determinacy test with published retrieval policies that decide when to read the source passage, and predicted withheld grades from the cultivar's other zones. For more than a third of the cultivar--disease pairs documented in several zones, the official grades differ between zones, and the documented durations of one cultivar can differ by weeks from zone to zone. Pooling across zones adds unsupported cultivars to the 2025 shortlists, whereas requiring every profile to pass leaves out supported ones. The pipeline answers from stored profiles when the determinacy bounds agree and reads the source passage otherwise, and this rule reproduced every documentary answer in the 2025 workload. On sampled requests with a shared language-model reader, the rule made no errors and read the source passage for only 1.2% of requests. It also used far fewer generation tokens than the adapted retrieval policies we evaluated. In predicting grades across zones, a random forest beat the zone--disease majority grade only slightly, and the cultivar-level sign test did not settle the difference. Keeping the type zone with each value avoids the errors of pooled summaries and ties every answer to the record that supports it.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: