What the Data for This Category Looks Like
Data for this category comes from code repositories, test framework reports, requirement specifications, and interface definitions. Update frequency is daily or per code commit. Commonly used documents include structured test cases, coverage statistics reports, Mock interface descriptions, and defect tracking tickets. Test cases contain fields for case ID, preconditions, execution steps, and expected results. Coverage reports contain file paths, coverage status, and uncovered branch information. Mock descriptions contain fields for interface addresses, request parameters, and return formats. Data is stored in plain text or structured JSON format.
Constraints Imposed on Reranking and Hybrid Retrieval
Data characteristics for this category impose three constraints on the reranking and hybrid retrieval link. First, the daily or per-commit update frequency requires the retrieval pipeline to support incremental synchronization, to avoid retrieving expired test cases or coverage reports. Second, the data includes both structured fields and unstructured text. Hybrid retrieval must match both keywords and field attributes, such as matching the execution steps of test cases and the reproduction step descriptions of defect tickets. Third, there is a strong association between test cases and defect tickets. The reranking link must prioritize highly relevant test data and ticket content, while adapting to differences in document lengths to avoid losing critical information due to truncation of long documents.
Configuration Settings
These configurations apply to FastGPT v4.17.0 and above.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Similarity Threshold | 0.72–0.85 | Matches the semantic similarity between test cases and defect tickets. Avoids introducing irrelevant results if set too low, and avoids missing relevant test documents if set too high |
| Maximum Returned Results per Request | Top 10 | The operations duty scenario requires quick problem location. Limiting the number of returned results reduces reading burden while covering most relevant data |
| Retrieval Mode | Hybrid Retrieval | This category of data includes both structured fields and unstructured text. Hybrid retrieval mode matches both keywords and semantic features, adapting to multiple types of test documents |
| Enable Reranking | Yes | The association between test data and tickets is complex. The reranking link performs secondary sorting on retrieval results based on semantic relevance, improving the display priority of critical information |
| Reranking Recall Count | Top 5 | The reranking link focuses on highly relevant results. Limiting the number of documents processed for reranking reduces pipeline latency and meets the real-time requirements of operations duty scenarios |
| Retrieval Timeout Threshold | 600 seconds | Prevents retrieval timeouts caused by overly long test documents or large data volumes, ensuring response speed for operations duty scenarios |
Three Common Misconfigurations
- Symptom: After enabling the reranking function switch, retrieval results are not secondary sorted, and returned results match the initial retrieval results exactly. Cause: AI Proxy environment variables are not configured correctly, or the AI Proxy service is not running normally, preventing the reranking model from being called.
- Symptom: Hybrid retrieval only matches keywords and does not cover semantic features, and retrieval results include a large number of irrelevant test cases. Cause: Retrieval mode is not set to hybrid retrieval, and only keyword retrieval mode is used.
- Symptom: After deployment in an arm architecture environment, the reranking function fails to run normally, and logs prompt architecture incompatibility. Cause: A reranking model image that only supports amd architecture was used. An arm architecture-compatible model image was not selected, or the reranking model was not recompiled.
How to Confirm Successful Configuration
- Check the FastGPT service startup logs to confirm no AI Proxy environment variable missing or verification failure prompts are displayed.
- Send a retrieval request for test cases or defect tickets, review the sorting logic of returned results, and confirm the reranking link has taken effect.
- Check the number of retrieved results to confirm they match the configured maximum returned results per request.
- Review the matching features of retrieved results to confirm hybrid retrieval mode matches both keywords and semantic information.
The values provided are common starting points and should be measured 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.