generated: '2026-07-21'
method: derived
source: openapi/unitary-ai-content-classification-openapi-original.yaml
notes: >-
  Derived from the schemas of Unitary's content classification OpenAPI. The API is
  job-centric: every asynchronous submission creates a classification job identified
  by job_id, whose result payload varies by endpoint family.
entities:
  - name: ClassificationJob
    id_field: job_id
    description: Asynchronous unit of work created by every queued classification request (JobIDResponse); status exposed via AsyncResponseStatus.
  - name: PolicyClassificationResult
    description: Policy categories with risk levels and scores (PolicyClassificationResponse / VideoPolicyClassificationResponse; PolicyCategoryFull, PolicyCategoryRiskLevels, PolicyCategoryScores).
  - name: ItemsCharacteristicsResult
    description: Granular item/characteristic signals (ItemsCharacteristicsResponse / VideoItemsCharacteristicsResponse; category families FullNSFWCategory, FullViolenceCategory, FullSubstancesCategory, FullHateSymbols, FullOtherCategory, FullVisualContentCategory).
  - name: DetoxifyResult
    description: Text toxicity scores (DetoxifyResponse, LanguageToxicityCategory) produced by a selected DetoxifyModel.
  - name: VideoMetadata
    description: Media facts attached to video results (width, height, fps, duration, seconds_processed).
  - name: VirtualModeratorCase
    id_field: moderation_job_id
    description: Blended AI + human moderation case (VirtualModeratorRequest / VirtualModeratorQueuedResponse) with main content (TextMainContent / MediaMainContent) and attachments (TextAttachment, LinkAttachment, MetadataAttachment, UserReportAttachment); result includes decision_type, escalate, escalate_reasons, policy_categories.
  - name: ApiToken
    description: 24-hour Bearer credential returned by /authenticate (LoginResponse api_token + expiry).
relationships:
  - from: ClassificationJob
    to: PolicyClassificationResult
    kind: has_one
    via: job_id (AsyncResponse_PolicyClassificationResponse_ / AsyncResponse_VideoPolicyClassificationResponse_)
  - from: ClassificationJob
    to: ItemsCharacteristicsResult
    kind: has_one
    via: job_id (AsyncResponse_ItemsCharacteristicsResponse_ / AsyncResponse_VideoItemsCharacteristicsResponse_)
  - from: ClassificationJob
    to: DetoxifyResult
    kind: has_one
    via: job_id (AsyncResponse_DetoxifyResponse_)
  - from: PolicyClassificationResult
    to: VideoMetadata
    kind: has_one
    via: metadata (video results)
  - from: VirtualModeratorCase
    to: ClassificationJob
    kind: has_one
    via: classification_job_id (Virtual Moderator responses reference the underlying classification job)