What the data for this category looks like
Thermal coal research report data mainly comes from industry association public data, futures exchange listed information, securities firm industry analysis reports, and professional coal information platforms. Update frequency is not fixed. Securities firm reports update when industry events occur. Industry supply and demand data updates weekly or monthly. Most document structures include core indicator tables, supply and demand balance analysis, and price trend charts. Fields include thermal coal closing price (unit: yuan/ton), calorific value (unit: large calories per kilogram), port inventory (unit: 10,000 tons), railway transportation volume, and more. Some reports include regional price difference comparison details.
What constraints do these characteristics impose on workflow orchestration
The scattered data sources and non-fixed update rhythm of thermal coal research reports require workflows to support two update modes: parallel pulling from multiple data sources, and event-triggered or scheduled polling updates. The high proportion of structured tables and charts in documents requires workflows to use dedicated structured data parsing nodes. This avoids disrupting the logical association between indicators via plain text parsing. Differences in units across sources — such as different pricing benchmarks for closing prices, or differences in unit expressions for calorific value — require built-in standardized conversion links in workflows. Some in-depth reports exceed 10,000 words. Segmented recall and context splicing rules must be configured to ensure large models can fully access core analysis data.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
ragRecallTopN | Top 10-15 entries | Core indicators of thermal coal research reports are concentrated in the top 10 recall results. This adapts to the context loading logic of FastGPT 4.10.0, and avoids interference from redundant information. |
maxContext | 20-30 entries | Matches the long-text analysis needs of thermal coal research reports. Balances context length and large model token consumption, and conforms to conventional window requirements for industry analysis. |
CODE_RUNNER_MAX_COUNT | 3 | Supports parallel verification and format conversion for multiple data sources. Adapts to processing needs of multi-source research report data, and avoids component conflicts. |
fileParseMode | raw | Skips default text parsing, directly passes uploaded research report PDF/Excel files to the large model, and retains the cell association relationship of original tables. |
MODEL_ENV_VAR_ENABLED | Enabled | Allows the question optimization node to call preset thermal coal industry parameter environment variables, and unifies indicator calibers. |
PARSE_FILE_MAX_SIZE | 50 MB | Adapts to the file size limit of single in-depth thermal coal research reports, and avoids parsing timeout. |
The parameter values provided on this page are common recommendations used as starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require on-site analysis, and it is recommended to test on your own samples before finalizing.
Three Common Mistakes
- Phenomenon: After setting
maxContextto 30 entries, the conversation details only display 2 context entries, and the large model reply does not associate with core research report data. Cause: The value ofragRecallTopNis not adjusted synchronously. The default recall count is too low, resulting in valid associated research report fragments not being loaded. This issue is relatively common in FastGPT 4.10.0. - Phenomenon: After uploading an Excel file of a thermal coal research report, the large model cannot identify structured fields such as closing price and calorific value in the table. Cause:
fileParseModeis not configured asraw. The default text parsing process splits the association relationship of table cells, losing indicator correspondence. - Phenomenon: After adding the 4th code running component to the workflow, a configuration limit error pops up on the interface, with error code 400. Cause: The default value of
CODE_RUNNER_MAX_COUNTis not modified. The system defaults to limiting the workflow to only support 3 parallel code components.
How to Confirm Configuration is Complete
- Enter the RAG configuration module of the workflow, check the context retention count and recall count configuration, and adjust them to adapt to the content volume of thermal coal research reports.
- Upload a single thermal coal research report file, trigger the parsing process, and check the parsing result preview to confirm that the association relationship of structured tables is not damaged.
- Initiate a test call to the workflow, check the component running logs to confirm that all configured code running components can execute normally.
- Trigger the test run of the question optimization node to confirm that the node can normally call the preset industry parameter 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.