What Data for This Category Looks Like
Data sources related to coke financial reports include Dalian Commodity Exchange coke futures market data, monthly supply and demand reports released by the China Coking Industry Association, quarterly/annual financial reports of A-share listed coking enterprises, and customs statistics on coke import and export data. Update frequencies include daily (futures market and spot prices), monthly (industry supply and demand data), and quarterly/annual (enterprise financial reports). Document structures include structured quantitative data tables and unstructured analytical text. Fields include coke production capacity (unit: 10,000 tons), average daily output (unit: tons/day), port inventory (unit: 10,000 tons), ex-factory tax-included price (unit: yuan/ton), import and export volume (unit: 10,000 tons), and others. Most document formats are official PDF reports or structured Excel/CSV files.
Constraints Imposed by These Characteristics on Model Access and Configuration
There is a need for mixed access to multi-source data, with both high-frequency daily spot/futures data and low-frequency quarterly financial report data. This requires configuring different synchronization trigger rules. Fields use multiple industrial measurement units. Unit verification rules for field mapping must be configured during model access to prevent the model from misusing data. Financial report documents are mostly long texts or structured reports. This requires configuring long-text-adapted segmentation and recall parameters. Some data sources are official static documents. Regular pull timeout thresholds must be configured to ensure complete data retrieval.
How to Set Configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Coke financial reports contain industry terminology and long-form quantitative data. This range adapts to the model's context window while retaining core analytical logic |
chunkSize | 800–1200 characters | Coke industry reports are mostly coherent supply and demand analysis texts. This length preserves complete logic of individual sections and avoids segmentation breaks |
similarityThreshold | 0.72–0.78 | Coke data is mostly quantitative indicators. High matching precision is required to distinguish data from other categories under the broad coal category |
recallTopK | Top 6–8 results | Core relevant data of coke financial reports is scattered across different sections. This recall volume covers key indicators while avoiding redundant information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Official coke financial report documents have large file sizes. Sufficient time is needed for parsing and pulling to avoid mid-process interruptions |
apiRequestTimeout | 300 seconds | Adapts to the demand for high-frequency daily spot data pulling, balancing call stability and real-time performance |
The parameter values provided on this page are all conventional recommendations used as a starting point 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 the configuration.
Three Common Misconfigurations
- Phenomenon: Model API calls return
404 status code (no body). Cause: API address and path for model access are not configured correctly, or access permissions for the corresponding model are not enabled. - Phenomenon: Model responses do not reference uploaded coke financial report documents. Cause: Recall-related parameters are not set correctly, or the RAG recall switch is not enabled, causing the model to generate results only using its own training data.
- Phenomenon: Local deployed model testing prompts
Message field is requirederror. Cause: Required fields for conversation context are not filled correctly in the model access configuration, or the API request format requirements for the local model are not adapted.
How to Confirm Configuration Is Complete
- Initiate a single coke data query, verify that the model's returned results include exclusive fields and units from the uploaded document, and adjust field mapping rules based on the returned content.
- Simulate high-frequency calls to the daily coke price data interface, check for timeout or current-limiting errors, and adjust timeout configurations and current-limiting rules to meet requirements.
- Upload a complete quarterly coke financial report document, verify that the parsed segments retain complete supply and demand analysis logic without critical content breaks, and adjust segmentation parameters to optimize results.
- Test queries for different coal category data, confirm that the model does not confuse coke indicators with other coal category indicators, and verify that the similarity threshold and recall configuration are effective.
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.