What the data for this category looks like
Securities financial report data mainly comes from public disclosure platforms of domestic and overseas stock exchanges. Updates follow fixed cycles: quarterly (quarterly reports), semi-annual (half-year reports), and annual (annual reports). Temporary announcements are triggered alongside major events. Most documents are in PDF format, with a structure including cover page, table of contents, audit report, core financial statements, and detailed notes. Fields include financial indicators such as attributable net profit, earnings per share, return on net assets, etc. Units are uniformly yuan, ten thousand yuan, or hundred million yuan, and some fields have year-over-year and quarter-over-year markers.
What constraints do these characteristics impose on workflow orchestration
The fixed-cycle disclosure requirement for securities financial reports means workflows must be configured with scheduled trigger nodes to align with disclosure periods. Unstructured raw PDF documents require structured parsing strategies to avoid field loss from only extracting plain text. Long document structures require workflows to use segment processing nodes to adapt to large model context limits. Fixed field names and units require workflows to configure standardized mapping rules to ensure accuracy of extracted data. Random triggers for temporary announcements require workflows to support a combination of manual and scheduled trigger modes.
How to configure
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_PDF_STRATEGY | Structured field extraction + full-text parsing | Securities financial reports contain fixed structured financial fields and detailed note text. Covering both ensures complete extraction of core data |
MAX_CONTEXT_LENGTH | 8000–12000 characters | The notes section of a single annual report is lengthy. This range prevents key content from being truncated |
RECALL_TOP_K | Top 3–5 entries | Core data of securities financial reports is concentrated in fixed sections. Too many recall results will introduce non-core announcement content |
WORKFLOW_TRIGGER_CRON | 0 0 10 1–5 | Aligns with fixed working hour periods for domestic financial report disclosures to automatically trigger data processing workflows |
HTTP_REQUEST_TIMEOUT | 600 seconds | Large models require longer response times to process long financial report text and multi-round field validation |
VARIABLE_MAPPING_MODE | Precise matching by field name | Securities financial report field names are fixed. Precise matching prevents variables from being confused with non-target fields |
The parameter values provided on this page are common recommended starting points for configuring workflows. 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 Misconfigurations
- Issue: When passing the
company_codevariable to call the knowledge base in the workflow, an error message stating "knowledge base does not exist" is returned. Cause: No mapping rule between variables and knowledge base ID is configured, and a hard-coded knowledge base ID is used directly, causing the variable to not take effect. - Issue: The units of financial report data output by the workflow are inconsistent, with values in both yuan and ten thousand yuan appearing at the same time. Cause: No unit standardization conversion node is configured, and field values from the original document are extracted directly without unifying unit formats.
- Issue: After setting
RECALL_TOP_Kto "Top 8 entries", the returned results include announcement content unrelated to financial reports. Cause: No classification label for recalled documents is specified, causing recall of temporary announcements unrelated to financial reports.
How to Confirm Proper Configuration
- Upload a single listed company annual report PDF, trigger the workflow, and check whether the parsed core fields match the original document.
- Pass a preset listed company code variable, and check whether the workflow automatically matches the knowledge base documents of the corresponding entity.
- Adjust the value of
MAX_CONTEXT_LENGTH, and check whether truncation or timeout errors occur during long text parsing. - Set a scheduled trigger task, and check whether the workflow automatically starts and generates the corresponding report during the specified time period.
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.