What this category of data looks like
This category of data originates from PLM, ERP, MES, supplier materials, verification test results, and customer-specific requirements. Updates are triggered irregularly with each process or engineering change. Daily produced documents include ECN/ECR forms, change assessment reports, drawings and BOM version records, process verification reports, production switching plans, inventory disposal documents, and customer approval documents. Data fields include change number, version number, effective node, material code, verification conclusion, and other items. There is no unified fixed batch update rhythm.
What constraints these characteristics impose on the "source citation and traceability" link
Data sources are scattered and updated independently for each change. Precise matching of change numbers and version numbers is required to avoid cross-version citation errors. Document types are diverse and fields are non-standardized. Retrieval and matching logic must be adjusted for different documents such as ECN forms and process verification reports. Irregular updates require the traceability function to associate the latest version data in real time. Fixed-cycle index refresh cannot be relied upon. Changes are associated with documents produced through cross-departmental collaboration. Verification records from multiple sources including R&D, process, and quality must be traced simultaneously. This expands the retrieval scope and matching complexity, and increases the difficulty of result filtering.
Configuration Settings
This configuration applies to FastGPT v4.17.0, with details as follows:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Similarity Threshold | 0.75–0.85 | Matches precise fields such as change numbers and version numbers, requiring high similarity to ensure citation accuracy |
| Single Search Limit | Top 8 entries | Covers change-related documents from multiple sources, avoiding omission of key verification records |
| Reranking | Enabled | Prioritizes highly relevant fragments with change numbers and version numbers, improving traceability efficiency |
quoteMaxToken | 1200–1500 token | Accommodates complete citation fragments of long documents such as ECN and process verification reports |
| Maximum Context | 8000–10000 characters | Carries change-related data from multiple sources, ensuring complete traceability logic |
| Query Optimization | Enabled | Automatically supplements key retrieval conditions such as change numbers and version numbers, improving retrieval accuracy |
Three Common Configuration Errors
- Symptom: Retrieval results are empty or citation fragments are missing after enabling reranking. Cause: The reranking threshold is set too high, filtering out low-similarity but critical document fragments with change numbers.
- Symptom: Retrieval timeout causes citation failure. Cause: The single search limit is set too high, and retrieving a large number of documents from multiple sources exceeds the
quoteMaxTokenlimit. - Symptom: The cited document version does not match the current change. Cause: Version number filtering is not added to the retrieval conditions, resulting in incorrect citation of old cross-version documents.
How to Verify Proper Configuration
- Submit a test question that includes a change number, and check whether the returned results include the change number and version number of the corresponding document.
- After enabling the reranking function, verify whether the retrieval results prioritize fragments with change identifiers.
- After adjusting the
quoteMaxTokenparameter, confirm that complete citation fragments of long documents are not truncated. - Check system logs to confirm there are no abnormal records of retrieval timeouts or empty results.
The values provided are common starting points and should be measured against the reader's own samples.