What the data for this category looks like
Coke-related due diligence data comes from multiple sources: Dalian Commodity Exchange public settlement data, major coastal port spot quotes, steel industry association procurement statistics, and public APIs from third-party bulk commodity information platforms. Update frequencies follow multiple tiers: spot prices are updated daily, port inventory is updated every two days, monthly industry reports are released each month, and futures settlement prices are updated after each trading day closes. Data documents include structured CSV/Excel reports and semi-structured industry research PDF files. Core fields include origin listed price (unit: yuan per ton), total port inventory (unit: 10,000 tons), railway shipment volume (unit: 10,000 tons), and downstream steel mill procurement volume (unit: 10,000 tons). No custom unit fields are used.
Constraints imposed on model access and configuration by these characteristics
The multi-source, multi-update frequency, mixed-format, and fixed-field characteristics of coke due diligence data create multiple constraints for model access and configuration. Multi-platform public API data sources require configuring model call isolation rules across multiple suppliers to prevent parameter conflicts. Data sources with different update frequencies need matching scheduled scheduling parameters to align call timing with data update rhythms. Fixed fields and units require configuring data validation rules to filter abnormal input. Mixed structured and semi-structured document formats need parsing nodes adapted to both formats to improve data extraction efficiency. Long-document research reports require adjusting the model's context window parameters to accommodate longer input content.
How to Set Configurations
| Config Item | Recommended Approach | Rationale |
|---|---|---|
modelProviderConfig | Assign a unique UUID to the same model as the model ID; use different UUIDs for the same model from different suppliers | Complies with the platform's requirements for model ID uniqueness, avoiding configuration conflicts between the same model from different suppliers |
maxContextWindow | 8000–16000 characters | Covers most input length requirements for coke due diligence reports, from daily spot summaries to monthly research reports |
errorCatchEnable | Enable | Captures errors such as model request timeouts and empty returns, triggering retry or exception alert workflows |
thinkingModeSwitch | Enabled by default, can be temporarily adjusted via node parameters | Due diligence report generation requires multi-step reasoning; enabling thinking by default improves output accuracy, and temporary adjustments adapt to quick query scenarios |
apiRequestTimeout | 30 seconds | Adapts to the stable response duration of coke data interfaces, avoiding unnecessary waiting |
parseStructuredData | Enable, specify field mapping rules | Automatically extract core fields from structured reports, improving the accuracy and efficiency of model input |
The parameter values provided on this page are common recommended starting points for determining configuration baselines. Actual values are affected by material form, data volume, and business rules. Specific issues require case-by-case analysis, and it is recommended to test on your own samples before finalizing settings.
Three Common Configuration Mistakes
- Phenomenon: The model node returns status code 500 with the prompt "model id duplicate". Cause: Unique identifiers are not assigned to the same model from different suppliers, causing the platform to fail to distinguish configurations from different suppliers.
- Phenomenon: The generated due diligence report does not show the thinking process, but the background is still executing reasoning steps. Cause: Only the interface switch for "display thinking" is turned off, and the configuration item that controls the reasoning logic is not adjusted.
- Phenomenon: Model call exceptions occur but no corresponding log records are generated. Cause: The
errorCatchEnableparameter is not enabled, causing exceptions to not be captured when a request error occurs, and logs cannot be generated.
How to Confirm Successful Configuration
- Enter the model supplier configuration page, verify that the IDs of the same model under different suppliers are unique, and confirm that the configuration has been saved successfully.
- Trigger a test call, check whether the log contains relevant records of error capture, and verify that the error handling logic is effective.
- Upload a coke monthly research report, check whether the content length of the model input meets expectations, and verify that the
maxContextWindowconfiguration is effective. - Adjust the
thinkingModeSwitchparameter, generate a test report, and confirm that the display status of the thinking process matches the configuration.
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.