What the data for this category looks like
Core data sources for coking coal include mine production ledgers, third-party port quality inspection documents, futures delivery warehouse receipts, and industry monitoring weekly reports. Data updates trigger along goods circulation nodes. Port quality inspection documents update in real time when goods arrive at ports. Industry weekly reports sync on a weekly basis. Single due diligence documents have a fixed structure, including fields such as mine point code, dry basis ash content, dry basis volatile matter, total sulfur content, caking index, coke residue characteristics, delivery grade, port arrival date, and cargo right subject. There are no nested multi-level attachments. Core quality inspection indicators must comply with national standard general formats.
What constraints these characteristics impose on workflow orchestration
Coking coal data sources are scattered, and update nodes are not fixed. This requires workflows to support parallel pulling from multiple data sources, trigger data synchronization based on cargo right transfer nodes, and avoid invalid repeated requests. Single documents have high precision requirements for quality inspection indicators. Field validation nodes must be configured to filter abnormal values. Delivery grade is a core decision field. Workflows must support dynamically matching downstream knowledge bases and AI question-answering rules based on this field to avoid confusion across grade data. Some port quality inspection documents are encrypted PDF files. Decryption nodes must be configured to adapt to different encryption standards and ensure normal data reading.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
recallPriority | Sorted as "delivery grade > dry basis ash content > caking index" | Delivery grade is the core screening dimension in coking coal due diligence, followed by matching core quality inspection indicators |
globalVarInitRule | Bound to coalGrade, portArrivalDate, ownerEntity fields | These three fields are dynamic core parameters for each due diligence, and must be automatically extracted when the workflow starts |
similarityThreshold | Set to 0.85 | Coking coal quality inspection indicators are highly professional. A higher threshold can filter irrelevant content and ensure AI question-answering accuracy |
parseTimeout | Set to 600 seconds | Some port quality inspection documents are multi-page PDFs containing a large amount of test data, resulting in long parsing time |
dynamicKnowledgeBaseRule | Match corresponding knowledge base ID by coalGrade | Different delivery grades of coking coal correspond to exclusive quality standards and industry data, requiring calls to exclusive knowledge bases |
fieldValidateConfig | Configure validation conditions: caking index ≥5, total sulfur content ≤1.5% | Complies with national standards for basic coking coal quality requirements, filters invalid data |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require case-by-case analysis. It is recommended to test on your own samples before finalizing settings.
Three common mistakes
- Phenomenon: The AI conversation module in the workflow cannot read specified knowledge base content, or returns industry data unrelated to the current due diligence. Cause: The
dynamicKnowledgeBaseRulefor dynamically matching knowledge bases bycoalGradeis not configured. A fixed knowledge base ID is incorrectly bound, and knowledge base switching driven by global variables is not implemented. - Phenomenon: The workflow execution throws the
PARSE_FILE_TIMEOUTerror code, and the node status shows timeout. Cause: TheparseTimeoutparameter is not adjusted, and the default short timeout period is used, which cannot complete parsing of multi-page port quality inspection documents containing a large amount of test data. - Phenomenon: The quality inspection indicator values returned by AI conversation have obvious abnormalities, such as a caking index of 0 or total sulfur content exceeding 3%. Cause: The
fieldValidateConfigvalidation rule is not configured, and abnormal data that does not meet national standards or industry conventional ranges is not filtered.
How to confirm the configuration is complete
- Trigger a simulated workflow run, check whether the
coalGrade,portArrivalDate, andownerEntityfields extracted by the upstream data node are correct, and confirm that global variable initialization is normal. - Test coking coal data inputs with different delivery grades, check whether the system automatically switches to the knowledge base corresponding to the grade, and verify that the dynamic knowledge base matching rule takes effect.
- Upload a test document containing abnormal quality inspection indicators, check whether the field validation node intercepts abnormal data, and confirm that the validation rule takes effect.
- Run a test case containing a multi-page PDF quality inspection document, check whether the node execution duration meets expectations, and confirm that the
parseTimeoutparameter configuration is reasonable.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.