schema.prisma

// This is your Prisma schema file,
// learn more about it in the docs: https://pris.ly/d/prisma-schema

// Looking for ways to speed up your queries, or scale easily with your serverless or edge functions?
// Try Prisma Accelerate: https://pris.ly/cli/accelerate-init

// TODO: Standardize created_by/creator and updated_by/updator columns.
// TODO: Standardize case formatting. Decide between snake_case (bcgov recommended) or camelCase (prisma recommended).

generator client {
  provider = "prisma-client-js"
  // Output path will be specified per-app via prisma generate command
}

datasource db {
  provider = "postgresql"
}

enum WorkflowKind {
  primary
  benchmark_candidate
}

model Document {
  id                    String                 @id @default(cuid())
  title                 String
  original_filename     String
  file_path             String
  /// Normalized PDF blob key for OCR, in-app view, and workflows (always application/pdf).
  normalized_file_path  String?
  file_type             String
  file_size             Int
  metadata              Json?
  source                String
  status                DocumentStatus         @default(pre_ocr)
  created_at            DateTime               @default(now())
  updated_at            DateTime               @updatedAt
  /// Set by the ephemeral-document cleanup janitor once the document's blobs
  /// and Temporal execution record have been purged. Null = not yet purged.
  purged_at             DateTime?
  apim_request_id       String?
  workflow_id           String? // @deprecated: Use workflow_config_id and workflow_execution_id instead
  workflow_config_id    String? // Reference to WorkflowVersion.id (pinned graph config)
  workflow_execution_id String?                @unique // Temporal workflow execution ID
  model_id              String                 @default("prebuilt-layout")
  ocr_result            OcrResult?
  review_sessions       ReviewSession[]
  lock                  DocumentLock?
  groundTruthJob        DatasetGroundTruthJob?
  group_id              String
  group                 Group                  @relation(fields: [group_id], references: [id])
  workflowVersion       WorkflowVersion?       @relation(fields: [workflow_config_id], references: [id], onDelete: SetNull)

  @@index([group_id])
  @@index([workflow_config_id])
  // Supports the documents list endpoint's common access pattern:
  // filter by group_id, order by created_at DESC, limit/offset. Lets Postgres
  // serve the default sort from the index instead of sorting the group's rows.
  @@index([group_id, created_at(sort: Desc)])
  // Partial index `documents_purge_scan_idx` on (workflow_config_id, status)
  // WHERE purged_at IS NULL supports the ephemeral-cleanup janitor. Prisma
  // cannot express partial indexes, so it is managed via the raw SQL migration
  // 20260624000000_add_documents_purge_index. Do not let `migrate dev` drop it.
  //
  // Trigram GIN indexes `documents_title_trgm_idx` /
  // `documents_original_filename_trgm_idx` back the list endpoint's ILIKE
  // search. Prisma cannot express GIN/trigram indexes, so they are managed via
  // the raw SQL migration 20260626000000_add_documents_list_indexes. Do not let
  // `migrate dev` drop them.
  @@map("documents")
}

model LabelingDocument {
  id                String            @id @default(cuid())
  title             String
  original_filename String
  file_path         String
  /// Normalized PDF blob key for OCR and in-app view (always application/pdf).
  normalized_file_path String?
  file_type         String
  file_size         Int
  metadata          Json?
  source            String
  status            DocumentStatus    @default(pre_ocr)
  created_at        DateTime          @default(now())
  updated_at        DateTime          @updatedAt
  apim_request_id   String?
  model_id          String            @default("prebuilt-layout")
  ocr_result        Json?
  labeled_documents LabeledDocument[]
  group_id          String
  group             Group             @relation(fields: [group_id], references: [id])

  @@index([group_id])
  @@map("labeling_documents")
}

model OcrResult {
  id                 String   @id @default(cuid())
  document_id        String   @unique
  document           Document @relation(fields: [document_id], references: [id], onDelete: Cascade)
  keyValuePairs      Json?
  /// Structured OCR output: { format: "text"|"markdown", text: string,
  /// markdown?: string, pages: [{ pageNumber, content, lines }] }. Populated
  /// for prebuilt read/layout/document models where there are no fields to extract.
  content            Json?
  enrichment_summary Json?
  processed_at       DateTime @default(now())

  @@map("ocr_results")
}

model ApiKey {
  id                 String    @id @default(cuid())
  key_hash           String    @unique
  key_prefix         String
  /// The user who most recently generated or regenerated this key. Recorded for
  /// audit purposes only; not used for authentication identity resolution.
  generating_user_id String
  generating_user    User      @relation(fields: [generating_user_id], references: [id])
  group_id           String    @unique
  group              Group     @relation(fields: [group_id], references: [id])
  created_at         DateTime  @default(now())
  last_used          DateTime?
  actor_id           String    @unique
  actor              Actor     @relation(fields: [actor_id], references: [id])

  @@index([group_id])
  @@map("api_keys")
}

model User {
  id              String    @id
  email           String    @unique
  last_login_at   DateTime?
  created_at      DateTime  @default(now())
  updated_at      DateTime  @updatedAt
  is_system_admin Boolean   @default(false)
  actor_id        String    @unique
  actor           Actor     @relation(fields: [actor_id], references: [id])

  apiKeys            ApiKey[]
  userGroups         UserGroup[]
  membershipRequests GroupMembershipRequest[] @relation(name: "MembershipRequestUser")
  @@map("user")
}

model Actor {
  id     String  @id @default(cuid())
  apiKey ApiKey?
  user   User?

  templateModels            TemplateModel[]
  classifierModelsCreated   ClassifierModel[]        @relation(name: "ClassifierCreatedByActor")
  classifierModelsUpdated   ClassifierModel[]        @relation(name: "ClassifierUpdatedByActor")
  createdGroups             Group[]                  @relation(name: "GroupCreatedBy")
  updatedGroups             Group[]                  @relation(name: "GroupUpdatedBy")
  deletedGroups             Group[]                  @relation(name: "GroupDeletedBy")
  reviewSessions            ReviewSession[]
  datasets                  Dataset[]
  benchmarkProjects         BenchmarkProject[]
  benchmarkAuditLogs        BenchmarkAuditLog[]
  createdMembershipRequests GroupMembershipRequest[] @relation(name: "MembershipRequestCreatedBy")
  updatedMembershipRequests GroupMembershipRequest[] @relation(name: "MembershipRequestUpdatedBy")
  workflowLineages          WorkflowLineage[]

  @@map("actor")
}

/// Stable workflow identity (list in UI). Config lives on WorkflowVersion rows.
model WorkflowLineage {
  id                   String            @id @default(cuid())
  name                 String
  /// URL/CLI-friendly stable handle for the workflow, unique within the group.
  /// Auto-derived from `name` on create (kebab-case, deduped). Immutable across renames.
  slug                 String
  description          String?
  actor_id             String
  actor                Actor             @relation(fields: [actor_id], references: [id])
  group_id             String
  group                Group             @relation(fields: [group_id], references: [id])
  workflow_kind        WorkflowKind      @default(primary)
  source_workflow_id   String? // Source/base lineage id (only for benchmark candidates)
  sourceWorkflow       WorkflowLineage?  @relation("LineageSource", fields: [source_workflow_id], references: [id], onDelete: SetNull)
  head_version_id      String?           @unique
  headVersion          WorkflowVersion?  @relation("LineageHead", fields: [head_version_id], references: [id], onDelete: SetNull)
  created_at           DateTime          @default(now())
  updated_at           DateTime          @updatedAt
  versions             WorkflowVersion[] @relation("LineageVersions")
  benchmarkCandidateLineages WorkflowLineage[] @relation("LineageSource")

  @@unique([group_id, slug])
  @@index([group_id])
  @@map("workflow_lineages")
}

/// Immutable graph config revision; revert = change pin to an older version id.
model WorkflowVersion {
  id                   String                  @id @default(cuid())
  lineage_id           String
  lineage              WorkflowLineage         @relation("LineageVersions", fields: [lineage_id], references: [id], onDelete: Cascade)
  version_number       Int
  config               Json // GraphWorkflowConfig
  created_at           DateTime                @default(now())
  headOfLineage        WorkflowLineage?        @relation("LineageHead")
  benchmarkDefinitions BenchmarkDefinition[]
  documents            Document[]
  groundTruthJobs      DatasetGroundTruthJob[]

  @@unique([lineage_id, version_number])
  @@index([lineage_id])
  @@map("workflow_versions")
}

// ========== TEMPLATE MODELS ==========

model TemplateModel {
  id             String              @id @default(cuid())
  name           String
  model_id       String              @unique
  description    String?
  created_by     String
  creator        Actor               @relation(fields: [created_by], references: [id])
  created_at     DateTime            @default(now())
  updated_at     DateTime            @updatedAt
  status         TemplateModelStatus @default(draft)
  field_schema   FieldDefinition[]
  documents      LabeledDocument[]
  training_jobs  TrainingJob[]
  trained_models TrainedModel[]
  group_id       String
  group          Group               @relation(fields: [group_id], references: [id])

  @@index([group_id])
  @@map("template_models")
}

model FieldDefinition {
  id                String        @id @default(cuid())
  template_model_id String
  template_model    TemplateModel @relation(fields: [template_model_id], references: [id], onDelete: Cascade)
  field_key         String
  field_type        FieldType     @default(string)
  field_format      String?
  format_spec       String?       // JSON format spec for normalization/validation
  display_order     Int           @default(0)

  @@unique([template_model_id, field_key])
  @@map("field_definitions")
}

model LabeledDocument {
  id                   String           @id @default(cuid())
  template_model_id    String
  template_model       TemplateModel    @relation(fields: [template_model_id], references: [id], onDelete: Cascade)
  labeling_document_id String
  labeling_document    LabelingDocument @relation(fields: [labeling_document_id], references: [id], onDelete: Cascade)
  status               LabelingStatus   @default(unlabeled)
  labels               DocumentLabel[]
  created_at           DateTime         @default(now())
  updated_at           DateTime         @updatedAt

  @@unique([template_model_id, labeling_document_id])
  @@map("labeled_documents")
}

model DocumentLabel {
  id             String          @id @default(cuid())
  labeled_doc_id String
  labeled_doc    LabeledDocument @relation(fields: [labeled_doc_id], references: [id], onDelete: Cascade)
  field_key      String
  label_name     String
  value          String?
  page_number    Int
  bounding_box   Json
  created_at     DateTime        @default(now())

  @@index([labeled_doc_id, field_key])
  @@map("document_labels")
}

// ========== TRAINING MODELS ==========

model TrainingJob {
  id                String        @id @default(cuid())
  template_model_id String
  template_model    TemplateModel @relation(fields: [template_model_id], references: [id], onDelete: Cascade)

  status         TrainingStatus @default(PENDING)
  container_name String
  sas_url        String?
  blob_count     Int            @default(0)
  /// Azure build mode used for this training run.
  build_mode     BuildMode      @default(template)
  /// Optional max training hours budget. Only meaningful for neural builds;
  /// passed straight through to the Azure `maxTrainingHours` parameter.
  max_training_hours Float?

  /// The versioned Azure model ID this job targets (e.g. "km-invoice-v3").
  /// Captured at startTraining time so the poller and downstream consumers
  /// don't have to re-derive it from naming conventions.
  target_model_id String?
  /// The version number this job will produce. Persisted alongside
  /// target_model_id for the same reason.
  target_version  Int?

  operation_id  String?
  error_message String?

  started_at   DateTime  @default(now())
  completed_at DateTime?

  trained_model TrainedModel?

  @@index([template_model_id])
  @@index([status])
  @@map("training_jobs")
}

model TrainedModel {
  id                String        @id @default(cuid())
  template_model_id String
  template_model    TemplateModel @relation(fields: [template_model_id], references: [id], onDelete: Cascade)
  training_job_id   String        @unique
  training_job      TrainingJob   @relation(fields: [training_job_id], references: [id])

  /// Versioned Azure model identifier. v1 keeps the bare TemplateModel.model_id
  /// for backwards compatibility; v2+ append "-v<n>".
  model_id    String  @unique
  /// Sequential version per TemplateModel, starting at 1.
  version     Int     @default(1)
  /// Exactly one row per TemplateModel should be active. Documents and benchmarks
  /// referencing TemplateModel.model_id resolve to whichever version is active.
  is_active   Boolean @default(false)
  /// Tombstone for deleted versions. The Azure artifact is removed but the row
  /// is kept so audit trails ("v2 — deleted on …") still resolve.
  deleted_at  DateTime?
  /// Snapshot of the labeled documents + labels used to train this version.
  /// Shape: { documents: Array<{ labelingDocumentId, originalFilename, labels: Array<{ fieldKey, labelName, value, pageNumber, boundingBox }> }> }
  dataset_snapshot Json?
  description String?
  doc_types   Json?
  field_count Int     @default(0)
  /// Mirrors TrainingJob.build_mode at the time of success.
  build_mode        BuildMode @default(template)
  /// Mirrors TrainingJob.max_training_hours at the time of success.
  max_training_hours Float?
  /// Read-only `trainingHours` from Azure GET /documentModels/{id} response,
  /// captured by the poller on successful build. Always null for template.
  actual_training_hours Float?

  created_at DateTime @default(now())

  @@unique([template_model_id, version])
  @@index([template_model_id, is_active])
  @@map("trained_models")
}

model ClassifierModel {
  name               String
  description        String
  created_by         String
  updated_by         String
  creator            Actor            @relation(name: "ClassifierCreatedByActor", fields: [created_by], references: [id])
  updator            Actor            @relation(name: "ClassifierUpdatedByActor", fields: [updated_by], references: [id])
  group_id           String
  group              Group            @relation(fields: [group_id], references: [id])
  created_at         DateTime         @default(now())
  updated_at         DateTime         @updatedAt
  last_used_at       DateTime?
  version            Int              @default(1)
  source             ClassifierSource
  status             ClassifierStatus @default(PRETRAINING)
  operation_location String?

  @@id([name, group_id])
  @@index([group_id])
  @@map("classifier_model")
}

model Group {
  id                  String                   @id @default(cuid())
  name                String                   @unique
  description         String?
  created_by          String
  created_at          DateTime                 @default(now())
  updated_by          String?
  updated_at          DateTime                 @updatedAt
  deleted_at          DateTime?
  deleted_by          String?
  creator             Actor                    @relation(name: "GroupCreatedBy", fields: [created_by], references: [id])
  updator             Actor?                   @relation(name: "GroupUpdatedBy", fields: [updated_by], references: [id])
  deleter             Actor?                   @relation(name: "GroupDeletedBy", fields: [deleted_by], references: [id])
  classifier_models   ClassifierModel[]
  user_groups         UserGroup[]
  membership_requests GroupMembershipRequest[]
  documents           Document[]
  workflowLineages    WorkflowLineage[]
  template_models     TemplateModel[]
  labeling_documents  LabelingDocument[]
  api_keys            ApiKey[]
  datasets            Dataset[]
  benchmark_projects  BenchmarkProject[]
  confusionProfiles   ConfusionProfile[]
  reference_tables    ReferenceTable[]

  @@map("group")
}

model ConfusionProfile {
  id          String   @id @default(cuid())
  name        String
  description String?
  matrix      Json
  metadata    Json?
  group_id    String
  group       Group    @relation(fields: [group_id], references: [id])
  created_at  DateTime @default(now())
  updated_at  DateTime @updatedAt

  @@index([group_id])
  @@map("confusion_profiles")
}

model UserGroup {
  user_id    String
  user       User?     @relation(fields: [user_id], references: [id])
  group_id   String
  group      Group?    @relation(fields: [group_id], references: [id])
  role       GroupRole @default(MEMBER)
  created_at DateTime  @default(now())

  @@id([user_id, group_id])
  @@index([group_id])
  @@map("user_group")
}

model GroupMembershipRequest {
  id          String                       @id @default(cuid())
  user_id     String
  user        User                         @relation(name: "MembershipRequestUser", fields: [user_id], references: [id])
  group_id    String
  group       Group                        @relation(fields: [group_id], references: [id])
  status      GroupMembershipRequestStatus @default(PENDING)
  reason      String?
  resolved_at DateTime?
  created_at  DateTime                     @default(now())
  created_by  String
  creator     Actor                        @relation(name: "MembershipRequestCreatedBy", fields: [created_by], references: [id])
  updated_at  DateTime                     @updatedAt
  updated_by  String
  updator     Actor                        @relation(name: "MembershipRequestUpdatedBy", fields: [updated_by], references: [id])

  @@unique([group_id, user_id, status])
  @@map("group_membership_request")
}

// ========== AUDIT ==========

// TODO: Should these have references to the other tables?
model AuditEvent {
  id                    String   @id @default(cuid())
  occurred_at           DateTime @default(now())
  event_type            String
  actor_id              String?
  resource_type         String
  resource_id           String
  document_id           String?
  workflow_execution_id String?
  group_id              String?
  request_id            String?
  payload               Json?

  @@index([occurred_at])
  @@index([event_type])
  @@index([resource_type])
  @@index([document_id])
  @@index([workflow_execution_id])
  @@index([group_id])
  @@map("audit_events")
}

// ========== HITL MODELS ==========

model ReviewSession {
  id           String            @id @default(cuid())
  document_id  String
  document     Document          @relation(fields: [document_id], references: [id], onDelete: Cascade)
  actor_id     String
  actor        Actor             @relation(fields: [actor_id], references: [id])
  status       ReviewStatus      @default(in_progress)
  started_at   DateTime          @default(now())
  completed_at DateTime?
  corrections  FieldCorrection[]
  lock         DocumentLock?

  @@map("review_sessions")
}

model DocumentLock {
  id             String        @id @default(cuid())
  document_id    String        @unique
  document       Document      @relation(fields: [document_id], references: [id], onDelete: Cascade)
  reviewer_id    String
  session_id     String        @unique
  session        ReviewSession @relation(fields: [session_id], references: [id], onDelete: Cascade)
  acquired_at    DateTime      @default(now())
  last_heartbeat DateTime      @default(now())
  expires_at     DateTime

  @@index([expires_at])
  @@map("document_locks")
}

model FieldCorrection {
  id              String           @id @default(cuid())
  session_id      String
  session         ReviewSession    @relation(fields: [session_id], references: [id], onDelete: Cascade)
  field_key       String
  original_value  String?
  corrected_value String?
  original_conf   Float?
  action          CorrectionAction @default(confirmed)
  created_at      DateTime         @default(now())

  @@map("field_corrections")
}

// ========== ENUMS ==========
// TODO: Standarize these enums as constant casing
enum TemplateModelStatus {
  draft
  training
  trained
  failed
}

enum BuildMode {
  template
  neural
}

enum FieldType {
  string
  number
  date
  selectionMark
  signature
}

enum LabelingStatus {
  unlabeled
  in_progress
  labeled
}

enum ReviewStatus {
  in_progress
  approved
  escalated
  skipped
}

enum CorrectionAction {
  confirmed
  corrected
  flagged
  deleted
}

enum TrainingStatus {
  PENDING
  UPLOADING
  UPLOADED
  TRAINING
  SUCCEEDED
  FAILED
}

enum ClassifierSource {
  AZURE
}

enum ClassifierStatus {
  PRETRAINING
  FAILED
  TRAINING
  READY
}

enum DocumentStatus {
  pre_ocr
  ongoing_ocr
  /// OCR extraction complete; workflow may continue (e.g. HITL review).
  extracted
  /// OCR complete; document paused awaiting human review.
  awaiting_review
  /// Entire process complete (OCR + optional HITL review).
  complete
  failed
  /// Original stored but image/PDF could not be normalized to a display/OCR PDF.
  conversion_failed
}

// ========== BENCHMARKING MODELS ==========

model Dataset {
  id          String           @id @default(uuid())
  name        String
  description String?
  metadata    Json             @default("{}")
  storagePath String           @default("")
  createdBy   String
  creator     Actor            @relation(fields: [createdBy], references: [id])
  createdAt   DateTime         @default(now())
  updatedAt   DateTime         @updatedAt
  group_id    String
  group       Group            @relation(fields: [group_id], references: [id])
  versions    DatasetVersion[]

  @@unique([name, group_id])
  @@index([group_id])
  @@map("datasets")
}

// TODO: Does this need created/updated info?
model DatasetVersion {
  id                   String                  @id @default(uuid())
  datasetId            String
  dataset              Dataset                 @relation(fields: [datasetId], references: [id], onDelete: Cascade)
  version              String
  name                 String?
  storagePrefix        String?
  manifestPath         String
  documentCount        Int
  groundTruthSchema    Json?
  frozen               Boolean                 @default(false)
  createdAt            DateTime                @default(now())
  splits               Split[]
  benchmarkDefinitions BenchmarkDefinition[]
  groundTruthJobs      DatasetGroundTruthJob[]

  @@index([datasetId])
  @@map("dataset_versions")
}

// TODO: Does this need created/updated info?
model Split {
  id                   String                @id @default(uuid())
  datasetVersionId     String
  datasetVersion       DatasetVersion        @relation(fields: [datasetVersionId], references: [id], onDelete: Cascade)
  name                 String
  type                 SplitType
  sampleIds            Json
  stratificationRules  Json?
  frozen               Boolean               @default(false)
  createdAt            DateTime              @default(now())
  benchmarkDefinitions BenchmarkDefinition[]

  @@index([datasetVersionId])
  @@map("splits")
}

// TODO: No updated by?
model BenchmarkProject {
  id                   String                @id @default(uuid())
  name                 String
  description          String?
  createdBy            String
  creator              Actor                 @relation(fields: [createdBy], references: [id])
  createdAt            DateTime              @default(now())
  updatedAt            DateTime              @updatedAt
  group_id             String
  group                Group                 @relation(fields: [group_id], references: [id])
  benchmarkDefinitions BenchmarkDefinition[]
  benchmarkRuns        BenchmarkRun[]

  @@unique([name, group_id])
  @@index([group_id])
  @@map("benchmark_projects")
}

model BenchmarkDefinition {
  id                      String             @id @default(uuid())
  projectId               String
  project                 BenchmarkProject   @relation(fields: [projectId], references: [id], onDelete: Cascade)
  name                    String
  datasetVersionId        String
  datasetVersion          DatasetVersion     @relation(fields: [datasetVersionId], references: [id])
  splitId                 String?
  split                   Split?             @relation(fields: [splitId], references: [id])
  workflowVersionId       String
  workflowVersion         WorkflowVersion    @relation(fields: [workflowVersionId], references: [id])
  workflowConfigHash      String
  workflowConfigOverrides Json?              @default("{}") @map("workflow_config_overrides")
  evaluatorType           String
  evaluatorConfig         Json
  runtimeSettings         Json
  immutable               Boolean            @default(false)
  revision                Int                @default(1)
  /// Structured debug log from the last OCR improvement pipeline run (overwritten each run)
  pipelineDebugLog        Json?              @map("pipeline_debug_log")
  scheduleEnabled         Boolean            @default(false)
  scheduleCron            String?
  scheduleId              String?            @unique
  createdAt               DateTime           @default(now())
  updatedAt               DateTime           @updatedAt
  benchmarkRuns           BenchmarkRun[]

  @@index([projectId])
  @@index([datasetVersionId])
  @@index([splitId])
  @@index([workflowVersionId])
  @@map("benchmark_definitions")
}

model BenchmarkRun {
  id                   String                @id @default(uuid())
  definitionId         String
  definition           BenchmarkDefinition   @relation(fields: [definitionId], references: [id], onDelete: Cascade)
  projectId            String
  project              BenchmarkProject      @relation(fields: [projectId], references: [id], onDelete: Cascade)
  status               BenchmarkRunStatus    @default(pending)
  temporalWorkflowId   String
  workerImageDigest    String?
  workerGitSha         String
  startedAt            DateTime?
  completedAt          DateTime?
  metrics              Json                  @default("{}")
  params               Json                  @default("{}")
  tags                 Json                  @default("{}")
  error                String?
  isBaseline           Boolean               @default(false)
  baselineThresholds   Json?
  baselineComparison   Json?
  createdAt            DateTime              @default(now())
  ocrCacheRows         BenchmarkOcrCache[]

  @@index([definitionId])
  @@index([projectId])
  @@map("benchmark_runs")
}

/// Cached Azure OCR poll response per benchmark run sample (for replay without re-calling Azure).
model BenchmarkOcrCache {
  id           String       @id @default(uuid())
  sourceRunId  String
  sourceRun    BenchmarkRun @relation(fields: [sourceRunId], references: [id], onDelete: Cascade)
  sampleId     String
  ocrResponse  Json
  createdAt    DateTime     @default(now())

  @@unique([sourceRunId, sampleId])
  @@index([sourceRunId])
  @@map("benchmark_ocr_cache")
}

model BenchmarkAuditLog {
  id         String      @id @default(uuid())
  timestamp  DateTime    @default(now())
  actor_id   String
  actor      Actor       @relation(fields: [actor_id], references: [id])
  action     AuditAction
  entityType String
  entityId   String
  metadata   Json?

  @@index([timestamp])
  @@index([actor_id])
  @@index([entityType, entityId])
  @@map("benchmark_audit_logs")
}

// ========== BENCHMARKING ENUMS ==========

enum SplitType {
  train
  val
  test
  golden
}

enum BenchmarkRunStatus {
  pending
  running
  completed
  failed
  cancelled
}

enum AuditAction {
  dataset_created
  version_published
  run_started
  run_completed
  baseline_promoted
}

// ========== GROUND TRUTH GENERATION ==========

enum GroundTruthJobStatus {
  pending
  processing
  awaiting_review
  completed
  failed
}

model DatasetGroundTruthJob {
  id                 String               @id @default(uuid())
  datasetVersionId   String
  datasetVersion     DatasetVersion       @relation(fields: [datasetVersionId], references: [id], onDelete: Cascade)
  sampleId           String
  documentId         String?              @unique
  document           Document?            @relation(fields: [documentId], references: [id])
  workflowVersionId  String
  workflowVersion    WorkflowVersion      @relation(fields: [workflowVersionId], references: [id], onDelete: Restrict)
  temporalWorkflowId String?
  status             GroundTruthJobStatus @default(pending)
  workflowConfigOverrides Json?
  groundTruthPath    String?
  error              String?
  createdAt          DateTime             @default(now())
  updatedAt          DateTime             @updatedAt
  // TODO: created_by and updated_by?

  @@index([datasetVersionId])
  @@index([documentId])
  @@index([status])
  @@index([workflowVersionId])
  @@map("dataset_ground_truth_jobs")
}

enum GroupMembershipRequestStatus {
  PENDING
  APPROVED
  DENIED
  CANCELLED
}

enum GroupRole {
  ADMIN
  MEMBER
}

// ========== REFERENCE DATA TABLES ==========

model ReferenceTable {
  id          String   @id @default(cuid())
  group_id    String
  group       Group    @relation(fields: [group_id], references: [id], onDelete: Cascade)
  table_id    String   // unique within group
  label       String
  description String?
  columns     Json     // ColumnDef[]
  lookups     Json     // LookupDef[]
  created_at  DateTime @default(now())
  updated_at  DateTime @updatedAt

  rows ReferenceTableRow[]

  @@unique([group_id, table_id])
  @@index([group_id])
  @@map("reference_tables")
}

model ReferenceTableRow {
  id         String         @id @default(cuid())
  group_id   String
  table_id   String
  table      ReferenceTable @relation(fields: [group_id, table_id], references: [group_id, table_id], onDelete: Cascade)
  data       Json
  created_at DateTime       @default(now())
  updated_at DateTime       @updatedAt

  @@index([group_id, table_id])
  @@map("reference_table_rows")
}