Developers can now integrate more deeply with Model Armor's feedback loop. The GoogleCloudDiscoveryengineV1Feedback schema gains an optional modelArmorDisputeInfo field, referencing the new GoogleCloudDiscoveryengineV1FeedbackModelArmorDisputeInfo schema. This allows recording user-provided expectedLabel, raw sanitizationResultJson, and the templateName when a user disputes a BLOCK or FLAG verdict, crucial for improving AI moderation accuracy.
Feedback granularity is also enhanced with new positive reasons in GoogleCloudDiscoveryengineV1Feedback: RELEVANT, COMPREHENSIVE, GOOD_CITATION, and ACCURATE. These provide richer signals for evaluating search and recommendation quality.
For project provisioning, GoogleCloudDiscoveryengineV1ProvisionProjectRequest now accepts an optional idpConfig to set up identity provider configurations and ACLs during creation.
Finally, new schemas ApiservingMcpMcpToolVisibility and ApiservingMcpMcpToolVisibilityFieldVisibility have been added, seemingly related to internal tool visibility restrictions. Note that the visibilityEnforcementStrategy within these schemas is explicitly marked as 'DO NOT USE' and 'not yet implemented', indicating they are likely for internal or very early preview use.
