What the Data for This Category Looks Like
Typical entities are factories producing bread, pastries, biscuits, instant noodles, and frozen dough products. Daily documents include formulas and ingredient lists, production records, baking or steaming curves, metal detection records, cleaning and disinfection records, and factory inspection reports. Data sources cover raw material qualification certificates, ERP/MES systems, weighing systems, ovens or frying lines, laboratory test data, and food traceability ledgers. Update frequency follows production batches, shifts, or daily updates. Document fields include allergen labels, ingredient dosage, baking temperature and duration, packaging date, shelf life, and more. Some documents have numeric parameters and formatting requirements.
Constraints Imposed on Citation Sourcing and Traceability
The high-frequency update nature of this category requires citation traceability to associate with the latest production data in real time, and avoid using outdated static documents. The need for multiple document types and specific fields requires the traceability function to accurately match core verification fields such as allergens and ingredient dosage, to ensure the accuracy of cross-review. In scenarios with frequent line changes, it is necessary to associate cleaning and disinfection records with the current batch’s production documents to avoid the risk of allergen cross-contamination. The consequences of errors include product recall risks, so traceability must directly locate the original data node, and cannot rely on vague associated information.
The following configurations are based on the functional logic of FastGPT v4.17.0.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
quoteMaxToken | 300–500 token per quote | Controls the length of a single citation, fully covers core verification fields such as ingredient dosage and baking temperature, and avoids truncating critical information |
responseData | Includes document path, update time, batch number | Matches the traceability requirements of food traceability ledgers, and quickly associates original production nodes with data sources |
detail | Enable original format retention | Restores the layout details of baking curves and ingredient lists, and avoids unformatted plain text citations |
| Maximum context token count | 800–1200 token | Adapts to the total length of multiple documents associated with a single batch, and ensures all associated documents can be fully included in the context |
| Number of recalled citations per single request | Top 6 | Covers multiple document types such as raw material qualification certificates, production records, and metal detection records involved in a single batch |
Three Common Mistakes
- Unformatted plain text appears in context citations, losing tables or line breaks from original documents. Cause: The original format retention setting for the
detailconfiguration was not enabled correctly, causing the split document format to be stripped. - Some queries fail to return corresponding knowledge base citation documents. Cause: The number of recalled citations per single request is set too low, failing to cover all document types associated with the current batch, or the maximum context token count is insufficient, causing critical documents to be truncated.
- Non-associated citation display content cannot be removed. Cause: The return range of
responseDatawas not configured correctly, and documents from unrelated batches were mistakenly included in the citation list.
How to Verify Successful Configuration
- Upload a single batch’s formula sheet, production records, and metal detection report, submit a cross-review request, and check if the context citations retain the original Markdown formatting such as tables and paragraphs.
- Search for a specific allergen field, verify that the returned citation documents include cleaning and disinfection records and ingredient list information for the corresponding batch, and ensure accurate field matching.
- Check the additional information for each citation, confirm that it includes the document source path and update time, and can directly locate the original data node.
- Adjust the configuration for the number of recalled citations per single request, verify that the number of returned citations matches the preset value range.
The values provided here are common starting points, and should be evaluated against the reader's own samples.