What this category of data looks like
This category of data includes raw business datasets, ETL operation logs, annotation results exported from annotation platforms, sampling tables and abnormal data lists generated by sampling quality inspection, and final delivered quality reports. Daily maintained documents include cleaning rule configuration files and abnormal data processing ledgers. Data updates daily or per batch processing task. Fields include data unique identifier, pre- and post-cleaning comparison values, anomaly type tags, quality inspection sampling ratio, delivery compliance identifier, and more. No unified fixed units exist; values vary with the original dataset and business scenario. This document applies to FastGPT v4.17.0.
Constraints imposed by these characteristics on permissions and data isolation
Raw business datasets may contain sensitive business fields. Isolation must be implemented based on data ownership to prevent leaks. ETL logs and abnormal data lists record complete processing processes. Access must be restricted to only core project members. The daily or per-batch update feature requires permission configurations to take effect dynamically with task batches, to avoid old permissions being bound to new data. Delivered reports must be bound to the corresponding project team to prevent cross-project misuse or leakage. Operation scopes differ across roles. Permissions must be allocated according to responsibilities, to prevent unauthorized modification of core cleaning rules or quality inspection results.
How to configure settings
| Configuration Item | Recommended Approach | Basis for This Approach |
|---|---|---|
| Team Binding Configuration | Bind the teamId of the corresponding project | Raw datasets and cleaning tasks belong to the corresponding team, preventing cross-team access to sensitive data |
| Member Permission Grouping | Divide permissions into three levels: data engineer, quality inspector, project manager by role | Operation scopes differ across roles, matching daily work responsibilities |
| Guest Link Quota | Configure 10-100 according to the package | Adapt to the team's need to batch-share cleaning task links, avoiding overexposure of a single link |
| Conversation Record Verification | Enable permission verification based on tmbId | Prevent unauthorized members from viewing others' cleaning task conversations and processing records |
| Permission Refresh Configuration | Set to 1 day | Match the daily/per-batch data update rhythm, ensuring permission configurations take effect synchronously with data updates |
| Data Storage Isolation | Bind the exclusive storage directory of the corresponding project | Daily/per-batch updated data must be isolated by batch to avoid data mixing |
Three common misconfigurations
- Symptom: Interface prompts "No permission to operate this conversation record". Cause: Conversation record verification based on
tmbIdis not enabled, or the current user is not assigned the permission group for the corresponding cleaning task. - Symptom: Guest window cannot create more than 10 links. Cause: The guest link quota configuration is not adjusted according to the team package; the default quota is only 10.
- Symptom: Cross-team cleaning rules are accidentally modified. Cause: The
teamIdof the corresponding project is not bound, causing configurations from different teams to overwrite each other.
How to confirm the configuration is complete
- Use a non-project member account to access the cleaning task page, confirm that the corresponding data cannot be viewed.
- Enter the team settings page, adjust the guest link quota to 20, verify that 20 valid links can be created.
- Switch accounts with different roles, confirm that data engineers can edit cleaning rules, and quality inspectors can only view sampling results.
- View the data storage directory, confirm that cleaning task data for different projects is stored in exclusive paths, with no cross-project data mixing.
Note: The configuration values provided are common starting points, and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.