What the data for this category looks like
TCM-related marketing content data is primarily sourced from public standards released by the National Pharmacopoeia Committee, TCM filing databases of provincial drug regulatory authorities, proprietary processing documentation from pharmaceutical companies, and clinical reference materials. In terms of update cadence, the official pharmacopoeia completes formal revisions every 5 years. Filing data updates in real time alongside corporate compliance filings. Internal process documents are adjusted irregularly alongside production optimizations. Most individual documents are semi-structured, containing fields such as medicinal material origin, nature and tropism, functions and indications, dosage and administration, and compliant script templates. Units are typically grams, milliliters, treatment course days, and similar units. Some marketing documents also include decoction piece specifications and extract concentration parameters.
What constraints these characteristics impose on workflow orchestration
TCM marketing content for the finance industry must strictly follow official pharmacopoeia and regulatory compliance scope. Workflows must therefore include compliance check nodes to compare generated marketing content against standard fields from the pharmacopoeia. Semi-structured documents have clear requirements for field units, so workflows must add field validation steps to prevent content errors caused by mismatched units such as gram counts or treatment course days. Update cadences vary significantly across different data sources: official pharmacopoeia content must be synchronized periodically per its revision cycle, while internal corporate filing data can be refreshed as needed. Workflows must therefore configure independent update trigger rules for each data source. Marketing scenarios require matching user inquiries about symptoms to corresponding medicinal material information, so workflows must link structured fields from the knowledge base. Relying solely on unstructured text retrieval cannot meet demand for precise matching.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size | 800–1200 characters | Fields such as functions and indications, dosage and administration in TCM documents are mostly short paragraphs. This segment length preserves field integrity, avoids disrupting semantics through splitting, and aligns with the information density requirements of finance industry marketing content |
Recall count | Top 6 results | TCM marketing content must cover core efficacy and compliant scripts. Excessive recall leads to content redundancy, while insufficient recall fails to cover necessary information. This aligns with user inquiry matching needs for finance customer acquisition scenarios |
Similarity threshold | 0.75–0.85 | Precise matching between user inquiries about symptoms and medicinal material information is required. A threshold that is too low introduces irrelevant content, while a threshold that is too high fails to retrieve valid information. This ensures the targeting of marketing content |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Some TCM processing document files are lengthy. This timeout ensures complete parsing of all content, preventing compliance information from being lost due to parsing interruptions |
workflow_global_var_lifespan | 24 hours | TCM marketing consultations in the finance industry mostly involve continuous interactions within a single day’s session. An overly long expiration period causes variable conflicts, while an overly short period fails to retain session context |
mcp_multi_param_auto_fill | Enabled | TCM MCP calls often require multiple parameters such as medicinal material name, symptoms, and applicable population. Automatic supplementation reduces user input costs and improves the efficiency of marketing content generation |
The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis. It is recommended to test on one’s own samples before finalizing settings.
Three common mistakes
- Issue: Global variables set in the workflow cannot be read after the front-end session restarts. Cause: A reasonable duration for
workflow_global_var_lifespanwas not set. The default expiration cycle does not cover the single-day interaction scenario of marketing consultations, causing variables to be cleared prematurely. - Issue: Only some of the user-input parameters are retrieved when calling the MCP node, preventing the call from completing. Cause: The
mcp_multi_param_auto_fillconfiguration was not enabled, and no parameter collection node was added to the workflow, resulting in missing additional parameters required by the MCP. - Issue: Local workflow debug results do not match front-end actual test results. Cause: Local debugging did not load the latest knowledge base synchronization content, or did not use configuration parameters consistent with the online environment, leading to differences in recall results.
How to confirm proper configuration
- Review knowledge base synchronization records to confirm that update trigger rules for different data sources match configuration requirements.
- Input simulated user inquiry scenarios, trigger the workflow, and verify that generated marketing content matches field units and compliance scope.
- Test the setup and reading process for global variables, confirming that variables can be read normally after a session restart.
- Call the MCP node, verify that the call completes normally and returns valid results after inputting multiple parameters.
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.