What this category of data looks like
Air pollution control-related financial report data comes from three main sources: corporate annual environmental special reports, regional monitoring ledgers publicly released by ecological environment departments, and compliance reports from third-party environmental testing institutions. Data update cycles include full annual financial reports, quarterly special monitoring data, and real-time spot monitoring records. Document structures include standard fields: monitoring spot codes, pollutant concentration values, governance facility operating hours, and environmental protection investment amounts. Concentration units are typically μg/m³. Duration units are hours. Monetary amount units are ten thousand yuan. Some documents are structured tables. Others are narrative reports. Individual document lengths vary widely, from a few pages of special reports to dozens of pages of annual financial reports.
What constraints do these characteristics impose on citation sources and traceability
Data sources are scattered, covering internal corporate ledgers, officially published data, and third-party reports. The traceability link requires associating unique identifiers across multiple data sources to avoid confusing records from different monitoring spots. Update cycles vary significantly. Time stamps for real-time data, quarterly data, and annual financial reports differ widely. Precise matching of the query time range is required during recall to prevent referencing invalid data across cycles. There are many specialized fields, with strict unit consistency requirements. Original field names and units must be retained when citing. Unauthorized conversions or omissions are not allowed. Document lengths span a wide range, from a few pages of monitoring records to dozens of pages of annual reports. Adaptation to different document lengths for segmentation and recall is required to avoid missing key governance cost data.
How to configure the settings
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
recall_top_k | Top 8-12 results | Air pollution control financial reports contain multi-dimensional specialized fields, requiring sufficient recall coverage for different monitoring spots and financial report entries to avoid missing key data |
rag_chunk_max_length | 1000-1500 characters | Content in fields such as governance facility operation records and investment details in financial reports is lengthy. This value range prevents truncation of key unit and numerical information |
source_visibility | Configured by data source type | Data publicly released by ecological environment departments allows display of original links, while internal corporate environmental protection investment data requires restricted visibility to comply with compliance requirements |
parse_timeout | 300-600 seconds | Parsing large annual financial report documents takes a long time. This value range prevents early termination of parsing tasks that would cause data loss |
similarity_threshold | 0.75-0.85 | There are many specialized terms in the air pollution control field. A higher similarity threshold can filter irrelevant general environmental protection data and accurately match query requirements |
rerank_top_n | Top 3-5 results | Recall results should focus on highly matched monitoring data and financial report entries, avoiding excessive redundant content that interferes with final output |
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 mistakes
- The symptom is that the workspace knowledge base application only returns 1 citation result. The cause is that
recall_top_kis set to 1, and the multi-document parallel recall switch is not enabled. - The symptom is that some parsed files directly write the original text without generating question-answer pairs. The cause is that
rag_chunk_max_lengthis set too small, causing specialized fields to be truncated and unable to recognize and generate structured question-answer pairs. - The symptom is that a configuration model test prompts an error, but the citation function works normally after forcibly ignoring it. The cause is that
test_prompt_templateis not adapted to specialized terms, triggering context length limit during testing, and runtime context parameters were automatically adjusted.
How to confirm correct configuration
- Upload a single air pollution control special financial report document, and check whether the recall results include exclusive fields such as corresponding monitoring spots and pollutant concentrations.
- View the citation source panel to confirm that the visibility of different data sources meets preset configurations: original links are displayed for public data, and complete original text is hidden for internal data.
- Test multiple documents with different update cycles, and check that the recall results cover monitoring data and financial report entries within the corresponding time range.
- Trigger a parsing task, check that there are no field truncation or timeout errors in the parsing log, and confirm that all documents are parsed normally.
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.