What This Category of Data Looks Like
This category of data centers on post-incident root cause analysis and incident management. Daily documents include RCA reports, incident tickets, review minutes, and rectification tracking sheets. Data sources are incident timelines, operation logs, monitoring metrics, and code commit records. Updates are triggered by each major incident, with a stable monthly update frequency. Data fields include incident occurrence time, root cause description, rectification measures, tracking status, and more. There are no unified fixed units; values adjust based on document type. Data volume grows with incident complexity, with the core focus on review and rectification information across the full incident lifecycle.
How These Characteristics Impact Context and Tokens
The data sources for this category are scattered, requiring integration of multi-source information to form a complete evidence chain. This places clear requirements on total context token control. Data update frequency is once per incident or monthly, with large fluctuations in single data volume. This requires flexible adjustment of context recall limits and token allocation. Daily document structures vary significantly: long documents such as RCA reports have dense paragraphs, while short documents such as incident tickets have clear entries. This demands context splicing logic tailored to different scenarios. Root cause analysis relies on complete historical evidence; missing or truncated context will introduce bias into root cause hypotheses. It is therefore necessary to clarify context recall scope and priority to avoid loss of key evidence. Rectification tracking sheets have strong field correlation; context must retain cross-document related information, and token allocation should prioritize preserving the integrity of core review fields.
Configuration Settings
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to context integration needs for long documents such as RCA reports and review minutes, preventing truncation of key root cause evidence |
quoteMaxToken | 2000–3000 token | Preserves the integrity of core fields for short documents such as incident tickets and rectification tracking sheets, preventing overloading of quoted content |
| Context recall count | Top 3–5 entries | Incident data updates monthly, single recall volume does not need to be large, avoiding token overlimit |
| Context splicing order | Sorted by incident timeline | Root cause analysis relies on chronological evidence chains; time-sorted context aligns better with analysis logic |
| Context filtering rules | Calibrated via actual testing | Core evidence varies widely across different incident scenarios; filtering thresholds need adjustment based on actual data |
Three Common Configuration Mistakes
- Symptom: Context truncation still occurs after configuring
maxContext, with key incident evidence missing from root cause analysis conclusions. Cause: ThemaxContextvalue was not adjusted based on the actual length of the RCA report, leading to early truncation of core incident timeline data. - Symptom: When classifying questions across multiple knowledge bases, classification results consistently match the fallback category. Cause: No context recall rules were configured; only the current question text was used for matching, without introducing historical questions and incident context, leading to insufficient classification basis.
- Symptom: Context association fails after manually adjusting document segments. Cause: Segment length was not matched to the
quoteMaxTokenlimit, leading to segmented context that cannot fully retain the root cause logic of the original document.
How to Verify Proper Configuration
- Upload a typical RCA report, check whether the recalled context includes key information such as incident timelines and core log fragments, and adjust the
maxContextvalue to cover core content. This configuration applies to FastGPT v4.17.0. - Test the multi-knowledge base question classification scenario, input a query with historical incident context, confirm that the classification result matches the question category of the corresponding knowledge base, and adjust the context recall count.
- Upload segmented document fragments, verify that the associated context fully retains the root cause description of the original document, and adjust the
quoteMaxTokenvalue. - Check system logs to confirm no context token overlimit errors occur, and adjust configuration parameters based on error prompts.
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.