What the data for this category looks like
Data for joint-stock bank intelligent due diligence reports comes primarily from internal credit approval archives, quarterly/annual financial report disclosure documents, People's Bank of China credit reporting system interfaces, and interbank transaction records. Update schedules: Internal credit archives update on T+1. Financial report disclosure documents update on a fixed quarterly or annual basis. People's Bank of China credit reporting data has a T+N update cycle. Document structures include structured credit approval forms, semi-structured financial report notes, and unstructured due diligence meeting minutes. Fields include credit exposure (unit: ten thousand yuan), days past due (unit: days), related transaction amount (unit: ten thousand yuan), and others. Some fields use nested formatting.
What constraints these characteristics impose on workflow orchestration
Multi-source heterogeneous data requires configuring multi-node data access. Separate processing for structured, semi-structured, and unstructured data avoids issues where a single node cannot adapt to all data formats. Data sources with different update schedules require the workflow to support both manual and scheduled triggers. Set priority rules for data sources to ensure real-time data is called first. Nested fields and unit differences require configuring a variable standardization processing node. This unifies field formats and units to prevent errors in subsequent calculations or outputs. Long document content requires configuring a long text segmentation parsing node. Split ultra-long documents to fit the knowledge base's processing limits while retaining semantic integrity.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Recall count | Top 15 entries | Joint-stock bank due diligence reports cover multi-dimensional data sources. 15 entries balances recall completeness and processing efficiency |
Similarity threshold | 0.72–0.78 | Due diligence data uses mostly professional terminology. A threshold that is too low introduces irrelevant content. A threshold that is too high misses valid related data |
PARSE_FILE_TIMEOUT_SECONDS | 900 seconds | Single financial report note documents can be dozens of pages long. Sufficient time is needed to complete long text parsing |
Chunk size | 1000–1200 characters | Adapts to the structured extraction accuracy of bank financial report fields. Avoids semantic breaks caused by overly short segments |
Global Variable Storage | Enable persistent storage | Global default variables such as regulatory compliance thresholds must be retained. Prevents variable resets after each conversation |
HTTP Request Retry Count | 2 retries | Addresses temporary network fluctuations when connecting to People's Bank of China credit reporting interfaces. Reduces the probability of call failures |
The parameter values provided on this page are common recommendations used as a starting point for determining configurations. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing settings.
Three common mistakes
- Symptom: The knowledge base search node in the workflow returns an empty array, and the preset fallback logic is not triggered. Cause: The
搜索为空分支configuration is not set, and no fallback data source switching rule is configured. This cannot handle scenarios where credit reporting data returns no results temporarily. - Symptom: In version v4.8.10, global default variables such as regulatory compliance thresholds are cleared after restarting the workflow. Cause: The
Global Variable持久化存储configuration is not enabled. Variables are stored only at the session level by default, and are automatically deleted after the session ends. - Symptom: Among the multiple variables output by the HTTP request node, a specified variable cannot be bound to the knowledge base search node. Cause: No variable mapping rule is set in the workflow. The data source ID variable output by the HTTP request is not bound to the data source parameter of the knowledge base node, resulting in parameter mismatch.
How to confirm the configuration is complete
- Initiate a test call, observe the number of return results from the knowledge base search node, confirm it matches the
Recall countconfiguration, and adjust the threshold to match the recall requirements of due diligence data. - Edit the global default variables, exit the configuration interface, and re-enter to verify whether the variables are retained. Confirm that the
Global Variable持久化存储configuration is effective. - Simulate a scenario where the knowledge base search returns no results, trigger the workflow, and check whether the preset fallback processing logic is executed. Verify that the empty result branch configuration is correct.
- Fill in application variables for the knowledge base node, submit a test, and check the system return status code. Confirm that variable parsing and parameter binding have no abnormalities.
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.