What this category of data looks like
Data for this category comes from databases, data dictionaries, masking tools, and access approval systems. Update frequency is per batch or per project. Daily generated documents include masking plans, field mapping tables, masking verification reports, and approval records. Masking plans contain sensitive field identification rules, execution workflows, and verification requirements. Field mapping tables mark original field names, masked field names, and corresponding masking rule types. Masking verification reports record sensitive information processing results and compliance check status. Data structure is mostly structured fields, with no complex nested levels. Single document size fluctuates with project scale.
What constraints do these characteristics impose on knowledge base retrieval and recall
This category of data is updated per project batch and covers multiple types of business documents, which imposes three types of constraints on retrieval and recall. First, limit the retrieval scope to the exclusive document set of the corresponding operations project. This avoids retrieving masking configuration data from other projects. Second, distinguish retrieval weights for different document types. Structured fields in field mapping tables require priority matching of precise keywords. Masking verification reports require priority recall of the latest versions. Third, since data sources include tools and approval systems, regularly sync the knowledge base to ensure retrieved content matches the latest configurations.
How to configure settings
In FastGPT v4.17.0, adjust the knowledge base retrieval and recall logic using the following configuration items:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Similarity threshold | 0.75–0.82 | Masking-related documents are mostly structured fields. Balance precise matching and recall coverage, calibrated based on actual testing |
| Single search limit | Top 6–10 results | Single masking document size is moderate. Excessive recall increases context processing load and slows response speed |
| Retrieval mode | Precise matching | Structured documents such as field mapping tables require precise matching of field names to avoid fuzzy recall of configuration content from unrelated projects |
| Reranking toggle | Enabled | When retrieving mixed document types, reranking improves the sorting priority of compliance-related documents such as masking verification reports and approval records |
| Query optimization | Enabled | Operations tickets often include vague descriptions. Expanded queries can match workflow keywords and rule descriptions in masking plans |
| Specified retrieval set | Specify the set ID for the current operations project | Limit retrieval scope to the exclusive masking document set of the corresponding project to prevent cross-project sensitive data leakage |
| Context length | 1000–2000 characters | Must accommodate the core content of a single masking document to avoid truncation of key configuration information |
Three common misconfigurations
- Retrieval results include masking document content from non-target projects, and cannot accurately recall documents from the specified set. The cause is failure to configure the specified retrieval set parameter to limit scope, resulting in recall of all masking data in the knowledge base.
- When consecutively submitting operations ticket queries, subsequent questions cannot associate previous masking configuration information, leading to irrelevant answers. The cause is failure to enable context-aware retrieval logic, where documents are only recalled independently for each single query without retaining historical interaction context.
- Retrieved masking rules do not match actual configurations, and sensitive field matching errors occur. The cause is failure to set a reasonable similarity threshold, causing low-match irrelevant documents to be recalled first.
How to confirm configuration is valid
- Submit a retrieval request targeting the masked fields of a specified project. Verify returned results only include documents from the corresponding project to confirm the specified retrieval set configuration takes effect.
- Submit two consecutive related operations ticket queries. Verify the second query can associate context information from the first query to confirm the context-aware retrieval logic works correctly.
- Adjust the value range of the similarity threshold. Verify the matching degree of retrieval results changes as expected to confirm the threshold configuration is reasonable.
- Submit a ticket query with a vague description. Verify workflow descriptions in masking plans are recalled to confirm the query optimization configuration takes effect.
- Verify the number of returned retrieval results falls within the configured range of the single search limit, and confirm the context length configuration can accommodate core retrieval content.
The values provided 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.