What the data for this category looks like
This category typically covers panel furniture, solid wood furniture, custom home goods, and wooden door manufacturers. Daily generated documents include cutting diagrams, cutting lists, edge banding and drilling programs, coating process documents, hardware material lists, and assembly and packaging records. Data mainly comes from order design systems, sheet and wood batch ledgers, CNC cutting and drilling equipment, ERP/MES systems, and moisture content testing equipment. Update frequency follows order cycles, daily cycles, or material batch cycles. Individual documents often include fields such as dimensional parameters, batch numbers, and process parameters.
Constraints on knowledge base retrieval and recall
Cutting diagrams, cutting lists, and similar documents contain precise dimensions, batch numbers, and process parameters. Retrieval results must match specific numerical values and batch dimensions to avoid panel size deviations or batch mismatches. Real-time updated data from multiple sources requires the retrieval pipeline to support synchronization triggered by order or batch, to prevent calling expired sheet moisture content or cutting programs. Custom orders are associated with a large number of panels, so the scale of single-recall results must be controlled to avoid context overflow that disrupts audit logic. Additionally, associated retrieval of hardware lists and installation records requires ensuring field matching accuracy to guarantee verification accuracy during on-site installation.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Documents in this category include precise parameters such as dimensions and batch numbers. Filter low-match results to avoid incorrectly recalling unrelated panels or process data. |
limit | Top 8–12 entries | Custom orders associate with large numbers of panels and process documents. Controlling recalled entry count avoids context overflow while covering core audit information. |
searchMode | embedding | Use semantic/vector retrieval. Exact-field, metadata, batch, and document-type constraints require separate filters and are not enforced by this mode. |
usingReRank | Enabled | Multiple process documents from the same batch may have similar semantics but inconsistent parameter details. Reranking improves parameter matching accuracy. |
datasetSearchUsingExtensionQuery | Expand queries by order number | Custom orders require associating multiple documents such as cutting diagrams, cutting lists, and hardware lists. Expanded queries cover all data required for end-to-end audits. |
collectionIds | Bind specified datasets by production workshop/material batch | Dividing datasets by production link or material batch narrows retrieval scope, improving recall efficiency and accuracy. |
Three Common Configuration Mistakes
- Phenomenon: The
similarityscores of retrieved results are generally higher than 0.9, but the actual matched panel sizes do not meet audit requirements. Cause: The similarity threshold was not adjusted to match the precise parameter requirements of this category. Relying solely on semantic matching causes irrelevant documents to be recalled with high weight. - Phenomenon: After filling in the knowledge base ID or referencing variables in the workflow, fixed results that do not meet audit requirements are returned, or a call error is triggered. Cause: Workflow variables were not correctly associated with datasets for the corresponding production batch or workshop per FastGPT v4.17.0 variable binding rules. This results in a retrieval scope that does not match actual needs.
- Phenomenon: After importing an Excel-format sheet batch ledger, retrieval fails to match the corresponding batch's moisture content or cutting parameters. Cause: Fields such as batch numbers and dimensional parameters in the Excel file were not mapped to retrieval query parameters. This invalidates the field correspondence of the dataset.
How to Verify Proper Configuration
- Retrieve a single custom order's cutting diagram and cutting list, perform a retrieval test, and confirm the recalled results include all core documents associated with that order.
- Adjust the
similarityparameter value, observe changes in recalled result matching accuracy, and confirm the threshold meets the parameter matching requirements of this category. - Import a single batch of sheet ledger data, run a batch number retrieval, and confirm recalled results only include documents from the corresponding batch with no cross-batch interference.
- Trigger a retrieval task bound to workflow variables, and confirm returned results match the production link or order number specified by the variables.
The values provided are common starting points and should be measured against the reader's own samples.