What data for this category looks like
Data for this category comes from daily outputs of enterprise metrology rooms and laboratories operating under the CNAS system. It includes measuring tool ledgers, calibration plans, calibration certificates, periodic verification records, original test records, reference material ledgers, and deviation handling documents. Data sources cover internal metrology management systems, external calibration institution certificates, standard instrument data, LIMS systems, and environmental monitoring records. Updates align with calibration cycles, single test completion milestones, or when abnormal events occur. Each single document contains fixed fields: instrument number, calibration date, measured value, uncertainty, and qualification judgment. Some documents require association with instrument traceability chain information.
Constraints imposed on knowledge base retrieval and recall
Data sources are scattered and linked to multiple systems. Retrieval must support cross-dataset recall of associated information. Document structures are fixed but fields are highly specialized. Retrieval must match semantic recall for metrology domain-specific terminology. Update frequency has no fixed cycle. Retrieval must support incremental updates and real-time synchronization. Single document lengths vary widely. Entries range from tens of characters for ledger items to thousands of characters for calibration certificates. Retrieval must adapt to different chunking rules. Traceability chain association requires retrieval results to retain field-to-field relationships, while providing association basis for independent fragments. Some documents require association with external calibration institution certificate data. Retrieval must support cross-data source permission verification and recall.
How to set configurations
These configurations are formulated based on FastGPT v4.17.0 rules.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Professional documents such as calibration certificates and test records require high semantic matching to filter low-correlation results |
limit | Top 8–12 entries | Covers key fragments of a single calibration certificate, while controlling total context length within model limits |
searchMode | embedding | Adapts to semantic recall for metrology domain-specific terminology, improving retrieval accuracy |
usingReRank | Enabled | Performs secondary ranking on recall results to correct semantic bias in initial retrieval |
collectionIds | Set based on actual measurement | Only recall datasets related to measuring instruments and calibration records, avoiding mixing irrelevant business data |
datasetSearchUsingExtensionQuery | Enabled | Performs specialized terminology expansion on user traceability queries to improve retrieval matching rate |
Three common misconfigurations
- Scenario: After uploading an Excel-format measuring tool ledger, retrieval result matching rate is very low. Cause: Structured parsing configuration for table documents is not enabled, causing plain text chunking to lose association between table fields.
- Scenario: After calling the FastGPT chat interface, returned results do not include associated knowledge base ID. Cause: Metadata return configuration for retrieval results is not enabled, or associated information return is not specified in request parameters.
- Scenario: After uploading a scanned PDF-format calibration certificate, valid text content cannot be retrieved. Cause: OCR recognition switch is not enabled, or resolution adaptation parameters for scanned documents are not configured.
How to confirm configurations are set correctly
- Upload a single calibration certificate document, run a retrieval test, and check whether recall results include core fields such as instrument number and calibration date.
- Send a cross-dataset retrieval request, and check that only document data within the specified
collectionIdsrange is recalled. - Send a traceability query with specialized terminology, and check whether the order of retrieval results aligns with semantic matching requirements.
- View the knowledge base upload log, and confirm that scanned PDF documents have completed OCR recognition and generated valid text chunks.
The values provided are common starting points and should be measured against the reader's own samples.