What This Category's Data Looks Like
Daily documents for this category include basic government service item catalogs, implementation lists, service guides, standard material catalogs, flowcharts, and frequently asked questions. Data sources are the national basic government service item catalog, administrative licensing item list, power and responsibility list, and local government service item databases. The update frequency is usually once per month. Document fields cover item names, acceptance conditions, application material lists, service processes, etc. There is no fixed unified unit, and field integrity must comply with government service standardization specifications.
Constraints Imposed by These Characteristics on Citation Sources and Traceability
The data sources for this category cover multi-level government catalogs from national to local levels. Citation traceability requires clearly marking the data's belonging level and issuing unit. The monthly update feature requires regular synchronization of knowledge base source data to ensure that Q&A content aligns with the latest policies. The multi-field document structure requires precise matching of specific document fragments corresponding to the item when citing, to avoid confusion across fields. As a rule maintenance work, citations must strictly correspond to policy provisions and item entries, and cannot be associated with individual case handling data.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Adapt to the length of single fragments from government service documents, to avoid truncating key policy statements |
responseData | Return complete citation metadata | Meet compliance requirements for government traceability, which requires displaying issuing units and update times |
detail | Enable detailed traceability | Clearly point to the chapter position of specific documents, making it easier to verify citation accuracy |
| Maximum recall results | Top 3 entries | Avoid redundant content from excessive citations, focusing on core policy basis |
| Knowledge base synchronization cycle | Once per month | Match the monthly update rhythm of this category's data to ensure content timeliness |
| API timeout | 600 seconds | Adapt to the loading duration of government service documents, avoiding request interruptions due to overly long content |
Three Common Misconfigurations
- Unable to output citation content after connecting the knowledge base via the tool call module. The cause is that the
responseDataconfiguration is not enabled, and the setting to return citation metadata is not activated, resulting in the interface not returning citation identifiers. - Garbled quotation marks appear at the end of dialogue output, which then revert to standard quotation marks after output. The cause is that the format parsing of citation metadata does not adapt to the multi-field structure of government service documents, and special characters are not properly escaped.
- Returns citations from non-target files even when the knowledge base has no matching content. The cause is unreasonable recall threshold settings, failure to filter irrelevant document fragments, and incorrect association of content not belonging to the current government service item.
How to Verify Successful Configuration
- A test query containing a clear government service item name is initiated, and output content is checked for inclusion of metadata such as the issuing unit and update time of the source document.
- The interface return results of dialogue requests are reviewed, citation-related field information is confirmed to be included, and whether the
responseDataanddetailconfigurations are effective is verified. - A knowledge base synchronization task is triggered, synchronization logs are reviewed, the synchronization cycle is confirmed to meet the requirement of once per month, and alignment with the data update rhythm of the category is verified.
- A query for an uncovered non-target item is input, the output is confirmed to not include irrelevant citation content, and the accuracy of the recall logic is verified.
The values provided are common starting points and should be measured against local deployment 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.