What the data for this category looks like
Data sources for coke-related financial reports and industry data include Dalian Commodity Exchange public delivery warehouse receipt data, China Coke Industry Association monthly industry analysis reports, and quarterly and annual financial reports of listed coal and coke enterprises. Update frequencies follow three schedules: daily spot quotes, monthly industry supply and demand and inventory data, and quarterly and annual enterprise operating data. Documents are split into two categories: industry reports and enterprise financial reports. Industry reports contain supply and demand balance sheets, regional inventory details, import and export data, and price indices. Enterprise financial report documents include fields such as coke business segment revenue, capacity utilization rate, procurement costs, and average sales price. Common units are tons, yuan per ton, and ten thousand RMB.
What constraints these characteristics impose on deployment and upgrade
Differences in update rhythms across multiple data sources increase scheduled task scheduling complexity. Pull cycles for different data sources must be split separately. The diversity of document structures and field units requires custom document parsing rules and unit conversion mappings to be configured during deployment. The specialized nature of coke industry data requires the knowledge base to preset dedicated classification tags. Industry term dictionaries must be updated synchronously during upgrades. For offline deployment scenarios, locally stored historical data volumes are large. Disk quotas and upload size limits must be adjusted accordingly.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Coke financial reports include multi-page industry supply and demand tables, with longer parsing times than generic documents. The default timeout is insufficient. |
UPLOAD_FILE_MAX_SIZE | 1500–2000 MB | Single annual industry reports or bundled batches of enterprise financial reports have large file sizes. Upload limits must be relaxed. |
maxContext | 8000–12000 characters | Long text fields such as coke supply and demand balance sheets and inventory details require sufficient context to support accurate analysis. |
SIMILARITY_THRESHOLD | 0.72–0.78 | Coke industry terminology is highly specialized. The matching threshold must be lowered to retrieve relevant segmented data. |
RERANK_TOP_K | Top 8–10 entries | Coke data has multiple dimensions. More candidate results must be retained for reranking to filter valid information. |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, 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 misconfigurations
- Symptom: An uncaught exception prompt appears when the workflow calls the AI chat component. Logs show configuration file loading failed. Cause: During private deployment, the knowledge base path parameter in
config.prod.yamlwas not modified correctly. Parsing rule files for coke financial reports cannot be read. - Symptom: Specified numbers of industry report documents fail to load after offline deployment. The system prompts insufficient storage resources. Cause:
UPLOAD_FILE_MAX_SIZEand local storage quota configurations were not adjusted. The setup does not adapt to the large volume of coke documents. - Symptom: Parsed coke financial report field units are inconsistent, with mixed use of "tons" and "kilograms". Cause: Custom unit conversion mapping rules were not configured. The generic parsing module cannot recognize specialized unit expressions used in the coke industry.
How to confirm configurations are properly set
- Upload a single monthly coke industry report. Verify that the parsed text fully retains core fields such as supply and demand tables and price indices, with no truncation or garbled characters.
- Manually trigger a scheduled pull task. Confirm that the pulled data source matches the configured pull address, and that updated data covers current cycle coke industry data.
- Test the workflow calling the AI chat component. Input a question related to coke inventory. Check that returned results include uniformly formatted industry data and specialized terminology.
- Review system operation logs. Confirm there are no error messages such as parsing timeouts or failed file uploads, and that service operation status is stable.
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.