What the data for this category looks like
Core data for this category comes from water source and raw material testing, formulation systems, online sugar and acidity monitoring equipment, filling equipment, visual inspection systems, and LIMS systems. Daily generated documents include formulation sheets, water treatment records, blending records, sterilization or aseptic filling records, packaging material inspection reports, and finished product inspection reports. Data updates follow production batches or shifts, with some online monitoring data collected continuously. Data fields include raw material batch numbers, sugar and acidity values, filling liquid levels, seal tightness, label batch numbers, and other production parameters and test results corresponding to different production links.
What constraints these characteristics impose on the "source citation and traceability" workflow
Production data for this category is scattered across multiple devices and systems. Field formats vary across different documents, requiring precise alignment with traceability needs for corresponding production links. Continuously collected online monitoring data has a large volume and will occupy context windows, so the length and total number of single citations must be limited. Documents updated per batch or shift need to be bound to a unique production batch identifier, to ensure cited traceability information fully corresponds to the relevant production link. Complex production processes involve multi-link associated data, so citation sources must cover the full chain from raw materials to finished products to avoid traceability gaps. Data anomalies directly relate to product safety, so cited traceability information must have clear traceability.
How to set up the configuration
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum context length | 8000–12000 characters | Covers multi-link production documents and online monitoring data, avoids truncating critical traceability information |
| Single search citation limit | Top 8 entries | Matches the volume of associated data across multiple links for this category, while controlling context load |
| Reranking | Enabled | Improves sorting priority of highly relevant traceability data, reduces invalid citations |
quoteMaxToken | 1500–2000 token | Limits the length of a single citation, adapts to the volume of online monitoring data, avoids context overflow |
| Citation source return format | Include document source, production batch number, collection time | Matches the traceability requirements of this category, facilitates binding production batch and link information |
detail | Enabled | Displays complete inspection parameters and collection times, meets the traceability detail requirements for quality audits |
Three common misconfigurations
- Enabled reranking results in empty citations: Search returns empty reference fragments, or no qualifying traceability data. This occurs when the reranking threshold is set too high, filtering out low-relevance but necessary traceability data that matches production batch criteria.
- Cited content is truncated or times out: Generated responses include incomplete document citation fragments, or timeout errors occur. This happens when
quoteMaxTokenis not restricted, leading to excessively long single citations that occupy too much context space. - Citations without bound production batch identifiers: Traceability information cannot be mapped to specific production batches, making quality audit traceability impossible. This occurs when the citation source return format is not configured to include batch number fields, so no unique production link identifiers can be matched.
How to confirm the configuration is correct
- Import typical production documents and online monitoring data for this category, run quality audit-related queries, check if returned citations include water treatment, filling, and other link data for the corresponding batch.
- View the displayed fields of citation sources, confirm that key information such as document source, production batch number, and collection time is included.
- After enabling reranking, run multiple sets of queries, check if citation results prioritize highly relevant production link data, with no significant empty citations.
- Adjust the value of
quoteMaxToken, verify that the length of single citations meets expectations, and no context overflow or truncation occurs.
The values provided are common starting points and should be measured against the reader's own samples.