What this category of data looks like
This category handles daily documents including normative document drafts, drafting explanations, legality review opinions, solicited opinion summaries, and filing reports. Data sources include regulatory and rule repositories, higher-level law databases, departmental policy document repositories, and administrative normative document filing systems. Update frequency is irregular. All documents include standard fields such as clear issuing entities, article content, and filing identifiers. All source data uses officially published formal official document formats, with no custom non-standard fields.
Constraints imposed on citation sourcing and traceability by these characteristics
Since all data sources are official, authoritative compliance repositories, the traceability process must accurately match the official identifiers of sources to avoid citing non-compliant content. Because update frequency is irregular, knowledge bases must be synced regularly to ensure cited laws and policies are the latest valid versions. Daily documents often involve hierarchical regulatory and policy comparison, so the traceability process must distinguish the priority of higher-level laws and lower-level documents, avoiding omission of core supporting basis. At the same time, errors directly affect document filing and release, so the traceability process must ensure the completeness and accuracy of cited content, with no incorrect or omitted citations.
Configuration settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Normative document articles are lengthy, so full regulatory basis content must be covered to avoid truncation of key articles |
responseData | Enabled | Citation sources' filing numbers, issuing agencies and other identifiers must be returned to meet the traceability and archiving requirements of legality review |
detail | Enabled | Context of cited articles must be displayed to help reviewers verify the match between articles and drafts |
| Single search limit | Top 8 results | Core basis such as multi-source higher-level laws and departmental policies must be covered to avoid missing legal review points |
| Similarity threshold | 0.75–0.85 | Regulatory article content must be accurately matched, irrelevant non-related policies must be filtered to improve review efficiency |
Three common mistakes
- Phenomenon: Unassociated configured knowledge base sources appear in search results. Cause: No knowledge base filtering rules are configured, leading to recall of irrelevant cross-repository data.
- Phenomenon: The number of citations does not meet review requirements, either introducing too much redundant content or omitting core basis. Cause: The single search limit and similarity threshold are not adjusted based on document characteristics.
- Phenomenon: Application errors occur after configuring citation variables. Cause: The unique identifier of the corresponding knowledge base is not correctly bound, or the variable format does not meet system verification rules.
How to confirm configuration is complete
- Upload a normative document draft to trigger the review workflow, and confirm that all citation sources in the returned results come from specified data sources such as the configured regulatory repository and policy repository.
- Adjust the
quoteMaxTokenparameter, observe the display length of cited content to confirm that no key articles are truncated. - Test recall results using different similarity thresholds to verify that irrelevant content is filtered out and citation entries meeting review requirements are retained.
- Check the citation fields in returned results to confirm that traceable required information such as filing numbers and issuing agencies is included.
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.