What Data for This Category Looks Like
Data for this category primarily comes from order and inventory modules, BOM lists, process route files, equipment capacity ledgers, staff shift schedules, work-in-progress ledgers, and supplier arrival records in manufacturing enterprise ERP systems. Daily generated documents include sales orders, master production schedules, material requirements plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Data is updated when order changes or abnormal events occur, or on a daily or weekly cycle. Document fields include order number, material code, process name, capacity value, delivery time, and more. Units include pieces, hours, days, and others.
Constraints Imposed on Source Citation and Traceability
Data sources for this category are scattered and have hierarchical relationships. The logical chain from order to work order to scheduling must be fully matched during traceability, otherwise the cited basis will lack coherence. Multi-source data have large differences in update frequencies. Real-time changing capacity and work-in-progress data must be traced to the latest version first, to avoid using expired information that affects scheduling audit results. Most document fields are standardized codes and values. During traceability, the format and unit of corresponding fields must be strictly matched to prevent citation errors caused by inconsistent units. Additionally, the volume of this category’s data grows with order volume. The scope of associated data must be controlled during traceability to avoid excessive context that reduces large model call efficiency.
Configuration Settings
This configuration is based on FastGPT v4.17.0
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 8000–12000 token | Production planning and shop scheduling data has multiple associated levels, requiring coverage of multi-dimensional document content including orders, schedules, work orders, etc., to avoid losing key associated information due to context truncation |
responseData | Structured field mapping | This category of documents contains a large number of coded fields. Structured mapping ensures accurate matching of key identifiers such as order numbers and material codes during traceability |
detail | Enabled | Auxiliary associated documents such as BOMs and process routes must be recalled to fully present the basis chain for scheduling and inspection |
| Single search citation limit | Top 6–10 entries | The amount of data associated with a single work order in this category is moderate. Excessive citations will lead to excessive context and affect large model reasoning efficiency |
| Traceability matching accuracy | Calibrated based on actual measurements | Most data fields in this category are standardized codes. Matching rules must be adjusted based on actual data matching effects |
Three Common Misconfigurations
- Phenomenon: Uploaded scheduling table CSV files show garbled characters in FastGPT. Cause: The encoding format of the CSV file was not specified, resulting in character conversion errors during parsing.
- Phenomenon: When
quoteMaxTokenis set above 3000, the large model does not receive context content. Cause: The system’s default total context length threshold is exceeded, resulting in context being truncated and discarded, and not correctly passed to the large model. - Phenomenon: Configuring HTTP response data as a citation has no effect. Cause: The response data was not converted to the structured document format supported by FastGPT, or the citation rules of the knowledge base were not correctly associated.
How to Verify Proper Configuration
- Upload a test scheduling table CSV file, check that the parsed fields are complete, with no garbled characters or missing content.
- Adjust
quoteMaxTokento a test value, initiate a scheduling-related query, confirm that the large model's reply contains cited documents such as work orders and orders. - After enabling the
detailswitch, query a scheduling question that includes a process route, confirm that the reply associates specific content of the process route. - View the knowledge base's citation logs, confirm that each query correctly recalls the corresponding number of associated documents.
The values provided are common starting points and should be evaluated against the reader’s own samples.