What data for this category looks like
Data sources for this category include fault timelines, system logs, monitoring metrics, and code commit records. Update frequency is triggered by major incidents or scheduled monthly synchronization. Daily deliverables include RCA reports, incident tickets, review minutes, and rectification tracking sheets. Fields include unique fault identifiers, root cause analysis conclusions, associated log snippets, code commit hashes, rectification tracking status, and more. There are no unified fixed units; values adjust based on document type.
Constraints imposed on the "source citation and traceability" link
Data sources for this category are scattered and structurally diverse. The citation and traceability link must support cross-source data association and matching, and avoid citation bias across data sources. Update frequency is triggered by major incidents, so the system must adapt to ad-hoc batch pull scenarios, and eliminate redundant configurations for continuous syncing. Daily deliverables include multiple types of structured and unstructured content, so the system must support automatic adaptation of field extraction rules based on document type. Evidence collection is the most time-consuming step. The traceability link must reduce manual verification costs, cut down on invalid citation generation, and avoid root cause analysis bias caused by incorrect traceability. Unlike incident response categories, this category focuses on post-incident root cause tracing, so traceability must accurately associate fault-specific data.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | The core citation content for this category (log snippets, RCA snippets) falls within this range, and can cover key evidence |
responseData | format with traceability metadata | This category needs to associate metadata such as fault IDs and code commit hashes. This format can be directly used for automatic association in review minutes |
detail | enabled | Root cause analysis requires verification of associated evidence. When enabled, original citation snippets are displayed, which helps SREs verify root cause conclusions |
| Maximum number of citations per request | top 3–5 | Core fault evidence for this category is usually concentrated in 3-5 key logs or commit records. Too many citations will distract analysis attention |
| Citation filtering threshold | calibrated via actual testing | Log correlation varies widely across fault scenarios. Adjustments must be made based on actual fault data |
Three common configuration mistakes
- Symptom: Citation content not associated with the local fault knowledge base appears in conversations. Cause: The data source scope is not limited in the root cause analysis task, and the global knowledge base is called by default, resulting in display of non-fault-related content.
- Symptom: Too many citation entries are returned even after citation count limits are configured. Cause: The value of
quoteMaxTokenis not set correctly, or the fault data source filtering rules exclusive to this category are not bound, resulting in irrelevant logs and code commit records being included in the citation scope. - Symptom: Citation snippets have no direct connection to the current root cause analysis conclusion. Cause: The citation filtering threshold is not adjusted, or sorting is not done based on category-specific data sources such as fault timelines and monitoring metrics, resulting in low-correlation content being prioritized for recall.
How to verify proper configuration
Initiate a simulated fault root cause analysis task under FastGPT v4.17.0, and check whether returned results include citation snippets from specified data sources. View the citation area in the conversation interface, and confirm that the number of displayed citations matches the configured maximum per request. Verify that metadata of citation snippets (such as fault ID, log timestamp) matches the currently analyzed fault. Adjust the value of quoteMaxToken, and check whether the length of returned citation snippets changes with the configuration.
The values provided are common starting points and should be evaluated against local sample data.
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.