What the data for this category looks like
Daily yield report data for financial leases comes from project revenue snapshots generated daily by leasing company core business systems. The update cadence is full stock project data for the previous day, generated in batches each early morning. Each single data entry includes fields such as project unique identifier, lease asset category, initial principal, ongoing revenue level, lease start date, and remaining repayment periods. Field units include yuan, calendar days, and number of periods. No additional aggregated statistical content is included.
What constraints these characteristics impose on conversation logs and auditing
Internal business system data sources require conversation logs to associate project unique identifiers. This prevents cross-project data confusion. Daily batch updates require audit logs to use date-sharded storage. This stops single log files from growing too large for retrieval. Fields contain core business revenue and project information. Logs must fully retain original request parameters and all AI-returned fields. No arbitrary truncation is allowed. Different business dimension field units require audit steps to verify unit consistency. This avoids unit errors in broadcast content. Daily dataset updates require logs to record the corresponding dataset version identifier. This enables tracing back conversation content for specific dates.
How to Configure Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
logLevel | info | Required to record request parameters, returned fields, and dataset versions to meet full-link tracing needs for auditing |
maxContext | 10000 characters | Single entries in financial lease daily report data have limited length. Retaining sufficient context allows complete display of project-related information and prevents key fields from being truncated |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Batch-generated daily report data files can be large, so sufficient time is needed to complete log parsing and storage |
vectorRecallTopK | Top 10 entries | Financial lease projects are numerous. Too many recalls increase log storage pressure, while too few fail to cover all relevant projects |
auditLogRetentionDays | 365 days | Meets general requirements for business log retention in the financial industry |
logFieldWhitelist | ["projectId", "leaseAssetType", "initialPrincipal", "incomeLevel", "startDate"] | Only retain core fields required for auditing to reduce invalid log storage usage |
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 individual analysis, and it is recommended to test on your own samples before finalizing settings.
Three Common Configuration Mistakes
- Symptom: Truncated vector dimension error message appears in logs:
The current vector dimension is 2560, and the vector. Cause: ThevectorDimensionparameter was not configured to match the vector dimension of the imported financial lease daily report data, resulting in incompatibility between the vector storage and recall process. - Symptom: Conversation logs do not record branch jump information from the question classification node in the workflow, making it impossible to trace decision paths during audits. Cause: The
workflowLogEnabledparameter was not enabled, and log collection rules for workflow nodes were not configured. - Symptom: FastGPT runtime logs cannot be found via the default path after Docker deployment. Cause: The
logDirparameter was not modified to specify a custom log storage directory, and the default log path was not mapped to a location accessible to the host machine.
How to Confirm Successful Configuration
- Initiate a yield query conversation for a specific lease project, and check whether the logs include the project unique identifier, query parameters, and all fields returned by the AI.
- View the log storage directory to confirm that logs are generated as date-sharded files, with no single files too large for retrieval.
- Check the vector database dimension configuration to confirm it matches the value of the
vectorDimensionparameter, with no dimension mismatch errors. - Verify the log collection switch for workflow nodes, and confirm that conversation logs after classification jumps can be associated with the classification identifier of the corresponding branch.
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.