What the Data for This Category Looks Like
Coking coal research report data comes from the China Coal Industry Association, regional coal producing area industry associations, Dalian Commodity Exchange, and third-party coal consulting institutions. Update frequencies include daily spot prices, weekly port inventory and transportation data, monthly supply and demand balance reports, and quarterly and annual in-depth industry analyses. Document structure includes publishing organization, publish time, coking coal grade (main coking coal, 1/3 coking coal, etc.), core indicators (ash content, sulfur content, bonding index G value, colloidal layer thickness Y value), price range, inventory and trading volume data. Price units are yuan/ton. Inventory and trading volume units are 10,000 tons.
Constraints on Multi-turn Conversation and Prompting
Coking coal research reports contain a large number of professional indicators and specific units. Multi-turn conversations must retain indicator definitions and unit conventions from historical interactions to avoid confusion. High-frequency updated data requires the conversation system to limit the time range of retrieved documents, ensuring the timeliness of returned content. Single in-depth research reports are lengthy, so the length of retrieved segments must be properly configured. Prompts must explicitly specify extraction of coking coal-specific indicators to avoid mixing irrelevant content from general coal categories. Data calibers vary across different publishing organizations, so prompts must require labeling of data sources and statistical calibers to ensure consistency of conversation results.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | First 3000 characters | Coking coal single in-depth research reports often exceed 2000 characters; retaining sufficient context avoids disruption of professional indicators |
Recall count | First 4 entries | Coking coal research reports have multiple data dimensions; excessive retrieved entries will dilute core supply and demand and price information |
Similarity threshold | 0.75–0.85 | High matching precision is required for coking coal professional indicators, to avoid mixing low-relevance general industry research reports |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Coking coal research reports often contain large amounts of tabular data, so parsing takes longer than general documents |
Rerank result count | First 2 entries | Focus on core price and supply and demand data, reducing interference from unnecessary industry background information |
single_turn_prompt | Concatenate according to conversation history | Retain previously mentioned coking coal indicator definitions, avoiding repeated correction of units and statistical rules |
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 individual analysis. It is recommended to test against your own samples before finalizing settings.
Three Common Misconfigurations
- Only setting single-turn trigger rules when configuring
single_turn_prompt, which causes subsequent conversations to fail to retain previously agreed coking coal indicator definitions, resulting in unit confusion error messages. - Failing to carry the
conversationIdparameter when calling the conversation record interface, leading to empty conversation record fields and inability to view historical interaction content. - Setting the number of retrieved entries to the first 8 and the similarity threshold to 0.6, which introduces non-coking coal research reports from general coal categories, resulting in insufficient result matching accuracy.
How to Verify Proper Configuration
- Initiate an initial conversation containing coking coal professional indicators, such as bonding index G value and colloidal layer thickness Y value. Verify that the system automatically associates historical indicator definitions when following up with additional questions.
- Call the conversation record interface, confirm that complete historical interaction content is returned after carrying the
conversationIdparameter. - View the knowledge base retrieval log, confirm that all returned research reports are labeled with coking coal-related keywords, and the quantity matches the configured number of retrieved entries.
- Upload a single coking coal in-depth research report, verify that parsing time does not exceed the configured
PARSE_FILE_TIMEOUT_SECONDSthreshold.
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.