Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Business data for social assistance and minimum living security review includes multiple document types: minimum living security application forms, family economic status verification reports, household survey forms, democratic review records, public notice materials, assistance approval forms, and more. Data sources include low-income population dynamic monitoring platforms, resident family economic status verification systems, and cross-department shared data from public security, human resources and social security, medical insurance, housing and urban-rural development, tax authorities, and other relevant departments. Data update frequency is monthly. Core fields include family income and property details, family member relationships, and assistance application and approval information. There is no unified fixed unit. The field format of each source system must be matched.
What constraints do these characteristics impose on the "source citation and traceability" link
The monthly update rhythm of cross-department shared data requires the traceability link to synchronously match the latest data source version. This avoids the risk of incorrect or missed security assistance caused by citing expired information. The mixed structure of multiple paper and system document types requires distinguishing recall rules for structured fields and unstructured text when associating knowledge bases. This ensures core information is recalled first. Traceability of sensitive fields such as family income and property requires clearly marking the data source department and acquisition channel. This meets the compliance requirements of government services. The multi-source nature of cross-department data requires integrating supporting content from each system during citation. This ensures the completeness and accuracy of question and answer basis.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum number of recall results | Top 15 entries | Adapt to the integration needs of multi-source cross-department data, avoid context overload that affects response accuracy |
quoteMaxToken | 800–1200 characters | Match the content length of long documents such as family economic status verification reports, ensure complete citation of business information |
| Citation source display rule | Classify by document type | Distinguish cross-department shared data from internal business documents, enable auditors to quickly locate supporting materials |
detail | Enabled | Display snippet content of original documents, meet the needs of detail verification during traceability |
responseData | Include source path and update time | Match the characteristics of monthly updated data sources, clarify the timeliness and compliance of cited content |
Three common mistakes
- Phenomenon: In non-tool call mode, the large model does not cite content from web search nodes. Cause: The linkage configuration between the associated knowledge base and web search is not enabled, or the recall threshold is set too high, causing business data to take priority over search results.
- Phenomenon: The citation list only displays the bottommost entry, and does not display content classified by source. Cause: The display rule grouped by document type is not configured, making it impossible to quickly distinguish internal business documents from cross-department shared data during traceability.
- Phenomenon: When the total word count of the associated knowledge base exceeds the limit, the system does not provide a clear prompt. Cause: The quantity upper limit of the associated knowledge base is not adjusted according to the average length of category documents, and the capacity threshold of single database and total database is not calibrated through actual testing.
How to confirm the configuration is complete
- Launch test questions involving family income queries or minimum living security qualification judgments, check whether the returned results mark the department and update time of the data source.
- View the citation list, confirm that each citation entry is attached with snippet content and storage path of the original document.
- Adjust the quantity of the associated knowledge base, verify whether the system provides a clear prompt when the preset threshold is exceeded.
- Import simulated monthly updated data, confirm that the new data can be normally recalled and displayed in the citation list.
The values provided here are common starting points, and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.