What the data for this category looks like
The main entities in this category are skincare, haircare, makeup, and special cosmetics manufacturers. Daily generated documents include formula work sheets, weighing and feeding records, emulsification or filling records, cleaning and sanitization records, semi-finished and finished product inspection reports, and label review records. Data sources include registered or filed formulas, raw material safety documents, weighing and emulsification equipment logs, ERP/MES systems, laboratory test data, and packaging material version documents. The system triggers data updates with each production batch and each line change. Individual documents contain fields such as raw material names, restricted substance content, process parameters like emulsification temperature and homogenization duration, batch numbers, and operating staff. Common units include degrees Celsius, minutes, grams, and other production-standard units.
What constraints do these characteristics impose on source citation and traceability?
First, data is updated dynamically with each production batch, and batch numbers are the core associated fields. Citation and traceability must be strictly bound to the current production batch to avoid cross-batch citation errors. Second, data sources are scattered across multiple systems and documents. A unified traceability standard is required to ensure that cited raw material safety documents, filed formulas, and on-site production records can be accurately matched. Third, individual documents have detailed fields that are tightly bound to compliance requirements. Precise retrieval of content for specific fields is needed to avoid irrelevant information interfering with cross-review. Fourth, error outcomes are directly linked to compliance risks. Traceability results must be traceable to specific document versions and generation times to ensure review results can be rechecked.
How to configure the system
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Core single documents for cosmetics manufacturing, such as formula work sheets, are typically 500–1000 tokens in length. This setting reserves context space to fully display cited content |
responseData | Return full citation metadata | Key compliance fields for cosmetics manufacturing, such as source file names, batch numbers, and document sources, must be included in traceability |
detail | Enable detailed traceability | Specific cited paragraphs and corresponding document versions must be displayed to meet compliance review requirements |
| Single retrieval limit | Top 6 results | The number of core documents for cosmetics manufacturing is concentrated. Excessive retrieval will interfere with the large language model's judgment of compliance content |
| Batch association filter | Match by current production batch number | Documents for each production batch only correspond to that batch. This avoids compliance risks caused by cross-batch citations |
Three common configuration errors
- Phenomenon: Returned citation documents include formula work sheets from non-current production batches. Cause: Batch association filter is not configured, and retrieval does not limit the document scope to the current production batch.
- Phenomenon: The cited file name field returned by the API call is empty. Cause: The
responseDataconfiguration is not enabled, and document metadata fields are not properly bound, resulting in traceability information not being correctly returned. - Phenomenon: Cross-review results do not cite the top-ranked filed formula document in the knowledge base. Cause: The
quoteMaxTokensetting is too small, causing the filed document to be truncated and unrecognizable, or retrieval does not prioritize matching official registered and filed data.
How to confirm the configuration is correct
- Upload a single current batch formula work sheet and its corresponding filed formula document, initiate a cross-review request, and verify that the returned citation list only includes documents related to the current batch and filed content.
- Call the API interface, check whether the returned results include complete citation metadata, and whether the fields contain key information such as file names and batch numbers.
- Adjust the
quoteMaxTokenparameter to different ranges, and observe whether the retrieval scope of cited documents meets the single document length requirement. - Simulate submitting a request that includes non-current batch documents, and confirm that the system does not retrieve production documents from non-target batches.
The values provided here are common starting points and should be measured against the samples used in the actual deployment.