This document applies to FastGPT v4.17.0.
What data in this category looks like
Data in this category primarily comes from raw audio files, annotation job platforms, proprietary business vocabularies, and spot-check samples. Daily circulating documents include transcription specifications, audio batch lists, quality inspection reports, and speaker annotation forms.
Data updates follow a daily or per-batch rhythm. Each data entry includes fields such as audio clips, corresponding transcription drafts, annotator identifiers, and quality inspection scores. Data must be associated with raw audio sources and annotation rules per batch, which differs from text annotation workflows that directly process text content.
What constraints do these characteristics impose on permissions and data isolation
Data in this category includes sensitive raw audio clips, annotator operation traces, and proprietary business vocabularies, with frequent updates per batch. Permission boundaries for different job roles (annotators, quality inspectors, algorithm teams) must align with their job scopes. For example, annotators can only access assigned audio batches and corresponding transcription specifications. Quality inspectors can view full batch quality inspection data but cannot modify original annotations.
Cross-team data transfers require strict association verification to prevent unauthorized disclosure of audio clips and annotation results. Temporary permission configuration for single-batch data must also be supported.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Team permission groups | Divided by business roles: annotation group, quality inspection group, algorithm group | Matches job scopes of different roles, prevents cross-role access to unauthorized data |
teamId | Bind to the team ID of the assigned single audio batch | Ensures only team members can access audio and annotation data for that batch |
teamId | Associate with the team that owns exclusive transcription specifications | Restricts only team members to view or modify the team’s own transcription rules |
tmbId | Associate with the unique identifier of the member’s affiliated team | Validates team permissions for access requests, blocks cross-team unauthorized access |
| Data access validity period | Set according to single-batch job cycle, 3–7 days | Adapts to daily or per-batch data update rhythm, avoids retaining unnecessary access permissions long-term |
| Permission verification logs | Enable full recording, retain for 30 days | Meets compliance review requirements, facilitates tracing abnormal access behavior |
Three common misconfigurations
- Phenomenon: A "no permission to operate this conversation record" prompt pops up when accessing a specified audio batch. Cause: The member’s
tmbIdis not associated with theteamIdbound to the batch, or the member is not added to the corresponding team permission group. - Phenomenon: A permission error occurs during testing after configuring a locally deployed model. Cause: The access interface of the deployment program is not bound to an exclusive
teamId, leading to cross-team access requests being blocked. - Phenomenon: Cross-team members can view quality inspection reports of other batches. Cause: Permission groups are not configured by business role, or the access scope of the permission group is not limited to the corresponding data batches.
How to confirm the configuration is correct
- Log in to the system with an annotator account, attempt to access an unassigned audio batch, and confirm that an insufficient permissions prompt appears.
- View the team permission group configuration, confirm that the annotation group can only see assigned audio lists and transcription specifications, and cannot access quality inspection reports.
- Initiate a data access request, check the system logs to confirm that verification information for the corresponding
teamIdandtmbIdis recorded. - Adjust the access validity period for a single batch of data, confirm that access to the batch’s audio and annotation data is no longer possible after expiration.
The values provided are common starting points and should be measured 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.