What the data for this category looks like
Typical entities in this category include carbonated drink, juice, tea, functional beverage, and packaged drinking water factories. Daily production documents include formula sheets, water treatment records, mixing records, sterilization or aseptic filling records, packaging material inspection reports, and finished product inspection reports. Data sources cover water source and raw material testing systems, formula management systems, online sugar and acidity monitoring equipment, filling equipment, visual inspection equipment, and LIMS laboratory management systems. Data is updated per batch, per shift, or via continuous collection. Each individual document is bound to a unique batch number, and includes core fields such as feeding parameters, test values, and timestamps. Some equipment collects data with precision down to the second.
What constraints do these characteristics impose on the "citation sources and traceability" link
Production data for this category is mostly real-time or near-real-time records bound to a single batch. Field differences across different documents are significant: formula sheets focus on feeding volume and ratio, while water treatment records focus on turbidity and colony count values. During the citation sources and traceability link, multi-source data for a single batch must be accurately matched to avoid mixing across batches. The business characteristics of high-speed filling and multi-category switching require that traceability results can quickly link to original records of corresponding production nodes, and cover full-link documents from raw materials to finished products to ensure the accuracy of cross-review. The high data update frequency also requires that the traceability system can link to the latest production records in real time, avoiding citing outdated data.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Matches the average length of individual beverage production documents, ensuring cited content fully covers core parameters |
responseData | Return original fields of associated documents | Must display beverage production-specific core data such as feeding volume on formula sheets and turbidity values on water treatment records |
detail | Enabled | Must display the storage path and collection timestamp of cited documents to meet the traceability requirements of cross-review |
| Single search limit | Top 6 recall results | Filters the most matching batch records from multi-source production data to avoid interference from redundant information |
| Maximum context | Top 3 rounds of conversation context | Focuses on issues related to the current production batch, avoiding interference from excessive historical conversations on accurate matching |
Three common mistakes
- Phenomenon: When accessing the application via a non-login share link, the citation source module shows that the original text cannot be viewed or associated fields are empty. Cause: The citation permission for non-login scenarios is not enabled in the application configuration, causing the share link to not synchronize traceability configurations.
- Phenomenon: During continuous questioning, subsequent questions cannot associate previous batch information, and responses are off-topic. Cause: The multi-turn conversation context association configuration is not enabled, or the maximum context round setting is too low to retain the batch identifier from the previous question.
- Phenomenon: No citation marker appears at the end of the knowledge base answer paragraph. Cause: The citation display switch for
responseDatais not enabled, or thedetailconfiguration is not enabled, causing the system to not generate citation identifiers.
How to confirm the configuration is complete
- Upload a single batch of beverage production formula sheet and water treatment record, initiate the question "What is the concentrate feeding volume for this batch", and verify that the answer in FastGPT v4.17.0 includes a clickable citation marker at its end.
- Initiate continuous questioning: first ask "Abnormal filling liquid level records for production line 1 today", then follow up with "What is the corresponding batch number", confirm that the second response associates the production line and time information from the first question.
- Generate a non-login share link, access the link, click the citation marker in the response, and confirm that it jumps to the original content of the corresponding document.
- Upload a sterilization record document exceeding 1200 tokens, initiate the question "What is the sterilization temperature for this batch", and check that the cited content fully retains core parameters without truncation.
The configuration values listed above are common starting points and should be evaluated against the relevant samples.