What the data for this category looks like
Market and yield data for investment platforms primarily comes from exchange public market APIs and held underlying asset data APIs. Data is updated via tick-by-tick real-time pushes after trading hours open on trading days. Full daily report documents are generated after market close. Each daily report contains multiple asset records. Fixed fields for each record include asset code, product name, daily yield, cumulative yield, trading volume, trading amount, and unit net value. Yield is measured in percentage, trading volume in shares, and trading amount in CNY. Field names may differ slightly across investment platforms, but the core identifier field is always asset code.
What constraints these characteristics impose on model access and configuration
Market and yield data have strict real-time requirements. High-frequency scheduled pull intervals must be used to prevent data lag from reducing announcement accuracy. Bulk multi-asset data structures require configuring field mapping rules to ensure the model can accurately extract target fields. Differences in field names across investment platforms require configuring data source adaptation parameters to unify data formats. Daily reports are generated at a fixed time after trading day close. This requires limiting scheduled task triggers to trading days to avoid invalid announcements during non-trading hours. Market APIs have rate limits. Bulk pull scale must match API limits to avoid triggering rate limit errors.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
workflow_trigger_cron | 0 18 * * 1-5 | Investment platform daily reports are typically generated at 18:00 on trading days. This setting matches the report generation time to trigger pull and announcement tasks |
datasource_update_interval | 300 seconds | Market data updates every 5 minutes on trading days. This interval ensures real-time performance of announcement data |
batch_fetch_size | 20 items per request | Individual asset data has a small size. This batch size avoids triggering rate limits on market APIs |
field_mapping_rule | Match uniquely by asset code | Asset code is the unique identifier for investment platform data. This prevents mixing data across different assets |
response_timeout | 10 seconds | Mainstream market APIs typically have response delays under 5 seconds. This setting reserves reasonable redundancy to avoid timeouts |
data_cleanup_rule | Filter non-business fields | Redundant fields returned by APIs, such as debugging and metadata, can interfere with model output. Cleanup rules must be configured to remove these |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by data format, data volume, and business rules. Specific scenarios require individual analysis. It is recommended to test with your own samples before finalizing settings.
Three common configuration errors
- Symptom: Generated announcement content includes unrelated zero-value fields or debugging information. Cause: The
data_cleanup_ruleparameter is not configured, and redundant non-business fields returned by the API are not filtered. - Symptom: Scheduled announcement tasks trigger during non-trading hours and generate invalid content. Cause: The
workflow_trigger_cronparameter does not limit triggers to trading days, resulting in daily task runs. - Symptom: Pulled market data does not match actual portfolio yields. Cause: The
field_mapping_ruledoes not bind the correspondence between asset codes and portfolio IDs, leading to misaligned data.
How to confirm correct configuration
- Manually trigger a data pull task. Check that returned fields match the structure of the investment platform's daily report documents.
- Review workflow run logs. Confirm that scheduled tasks trigger normally at the specified time on trading days, with no timeout or rate limit errors.
- Call the model to generate test announcement content. Check that only valid yield and market data fields are included, with no redundant unrelated information.
- Temporarily adjust the
datasource_update_intervalparameter to60 seconds. Verify that the pull frequency updates as expected.
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.