What the data for this category looks like
This category covers information related to the full customer lifecycle renewal process. Core daily documents include four types: customer health reports, renewal forecast tables, high-risk customer lists, and communication records. Data sources include CRM systems, product usage data, ticket statistics, and contract ledgers. Updates are performed weekly or monthly. Data fields include customer association identifiers, contract cycle nodes, usage behavior records, service interaction records, and more. The field structure varies across different document types, with no unified single-field format.
What constraints do these characteristics impose on the citation sources and traceability link?
Data comes from multiple heterogeneous data sources, so each cited fragment must be accurately linked to the original data source’s specific entry to avoid traceability confusion. Data is updated weekly or monthly, so the citation time range must match the update cycle to prevent expired citations from affecting health assessment judgments. The fields of different document types vary significantly, so unique identifier mapping rules must be configured for each document type. For example, contract ledgers must display contract numbers, and tickets must display ticket numbers, to ensure a clear traceability chain. The data interpretation process takes a long time, and a clear traceability chain can shorten troubleshooting time. This requires that citation configurations include clear source identifiers and hierarchies.
Configuration settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
quoteMaxToken | 8000–12000 token | Matches the average length of renewal health management documents, avoiding exceeding large model context limits |
responseData | Enabled | Meets the requirement to directly output original knowledge base content without additional formatting |
detail | Display data source type and unique identifier | Quickly locate original data entries and shorten troubleshooting time |
| Multi-knowledge base association configuration | Associate four data sources: CRM, product usage data, ticket statistics, contract ledgers | Covers all data sources for this category, supporting cross-source queries |
| Citation category display | Display document type | Helps quickly identify data categories to assist health assessment analysis |
Three common misconfigurations
- Symptom: Citation results only display general source names, without specific unique identifiers such as contract numbers or ticket numbers. Cause: No unique identifier mapping rules are configured for the corresponding document type, resulting in an unclear traceability chain.
- Symptom: Query responses time out each time, or returned results are truncated. Cause: The
quoteMaxTokenparameter is not adjusted based on the data scale of renewal health management, and the set value exceeds the maximum processing limit of the large model. - Symptom: Only content from a single data source can be cited, and renewal-related data from multiple knowledge bases cannot be obtained at the same time. Cause: Multi-knowledge base association configuration is not enabled, and only a single data source is bound, failing to cover the full data requirements of this category.
How to verify correct configuration
- Initiate a query involving renewal health-related data, and check whether each cited fragment in the returned results is accompanied by a data source type and unique identifier.
- Check the token usage of cited fragments to confirm they do not exceed the preset
quoteMaxTokenlimit. - After enabling the
responseDataswitch, check whether the original content of the corresponding document can be directly output without additional formatting. - Test knowledge bases associated with multiple data sources, and confirm that query results can cover different types of renewal-related documents.
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 2026-10-19.