feat: Datenmodell Sprachnotizen (T7 L2 Task 4)

VoiceCapture und VoiceCaptureAudio nach Spec §3 mit CHECKs (audioBytes 1..2 MiB, durationSec 1..240, CONFIRMED genau mit inspectionId, audioMime nur bei AUDIO), Recht voice-captures:use für ADMIN, BEEKEEPER und STAFF, Sperrreihenfolge mit VoiceCapture am Ende, Ausschluss aus dem Backup mit Begründung, Wächter gegen Suche und Export. Refs #27 Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
2026-10-16 09:20:04 +02:00
1 parent 257286fbef
commit af9df0afe4
@@ -0,0 +1,96 @@
+-- CreateEnum
+CREATE TYPE "VoiceCaptureSource" AS ENUM ('AUDIO', 'TEXT');
+
+-- CreateEnum
+CREATE TYPE "VoiceCaptureStatus" AS ENUM ('QUEUED', 'TRANSCRIBING', 'TRANSCRIBED', 'EXTRACTING', 'READY', 'CONFIRMED', 'DISCARDED', 'FAILED', 'EXPIRED');
+
+-- CreateTable
+CREATE TABLE "VoiceCapture" (
+    "id" TEXT NOT NULL,
+    "ownerUserId" TEXT NOT NULL,
+    "colonyId" TEXT NOT NULL,
+    "inspectionDate" DATE NOT NULL,
+    "source" "VoiceCaptureSource" NOT NULL,
+    "status" "VoiceCaptureStatus" NOT NULL DEFAULT 'QUEUED',
+    "audioMime" TEXT,
+    "audioBytes" INTEGER,
+    "durationSec" INTEGER,
+    "transcriptEnc" TEXT,
+    "suggestionEnc" TEXT,
+    "attempts" INTEGER NOT NULL DEFAULT 0,
+    "availableAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "claimedAt" TIMESTAMP(3),
+    "claimToken" TEXT,
+    "lastErrorCode" TEXT,
+    "model" TEXT,
+    "inputTokens" INTEGER,
+    "outputTokens" INTEGER,
+    "inspectionId" TEXT,
+    "createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "readyAt" TIMESTAMP(3),
+    "contentExpiresAt" TIMESTAMP(3) NOT NULL,
+    "contentDeletedAt" TIMESTAMP(3),
+    "updatedAt" TIMESTAMP(3) NOT NULL,
+
+    CONSTRAINT "VoiceCapture_pkey" PRIMARY KEY ("id")
+);
+
+-- CreateTable
+CREATE TABLE "VoiceCaptureAudio" (
+    "captureId" TEXT NOT NULL,
+    "dataEnc" BYTEA NOT NULL,
+    "createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+
+    CONSTRAINT "VoiceCaptureAudio_pkey" PRIMARY KEY ("captureId")
+);
+
+-- CreateIndex
+CREATE INDEX "VoiceCapture_status_availableAt_idx" ON "VoiceCapture"("status", "availableAt");
+
+-- CreateIndex
+CREATE INDEX "VoiceCapture_ownerUserId_status_idx" ON "VoiceCapture"("ownerUserId", "status");
+
+-- CreateIndex
+CREATE INDEX "VoiceCapture_contentExpiresAt_idx" ON "VoiceCapture"("contentExpiresAt");
+
+-- AddForeignKey
+ALTER TABLE "VoiceCapture" ADD CONSTRAINT "VoiceCapture_ownerUserId_fkey" FOREIGN KEY ("ownerUserId") REFERENCES "User"("id") ON DELETE RESTRICT ON UPDATE CASCADE;
+
+-- AddForeignKey
+ALTER TABLE "VoiceCapture" ADD CONSTRAINT "VoiceCapture_colonyId_fkey" FOREIGN KEY ("colonyId") REFERENCES "Colony"("id") ON DELETE RESTRICT ON UPDATE CASCADE;
+
+-- AddForeignKey
+ALTER TABLE "VoiceCaptureAudio" ADD CONSTRAINT "VoiceCaptureAudio_captureId_fkey" FOREIGN KEY ("captureId") REFERENCES "VoiceCapture"("id") ON DELETE CASCADE ON UPDATE CASCADE;
+
+
+-- Constraints Prisma cannot express.
+ALTER TABLE "VoiceCapture"
+  ADD CONSTRAINT "VoiceCapture_audioBytes_range"
+  CHECK ("audioBytes" IS NULL OR "audioBytes" BETWEEN 1 AND 2097152);
+
+ALTER TABLE "VoiceCapture"
+  ADD CONSTRAINT "VoiceCapture_durationSec_range"
+  CHECK ("durationSec" IS NULL OR "durationSec" BETWEEN 1 AND 240);
+
+ALTER TABLE "VoiceCapture"
+  ADD CONSTRAINT "VoiceCapture_confirmed_matches_inspection"
+  CHECK (("status" = 'CONFIRMED') = ("inspectionId" IS NOT NULL));
+
+ALTER TABLE "VoiceCapture"
+  ADD CONSTRAINT "VoiceCapture_audioMime_only_for_audio"
+  CHECK ("source" = 'AUDIO' OR "audioMime" IS NULL);
+
+-- Permission: the boot-time catalog sync only grants new permissions to
+-- system roles (ADMIN), so the grants are made here (same as T6). READONLY
+-- and API_CLIENT get nothing: voice notes are for people who inspect.
+INSERT INTO "Permission" (id, key, resource, action, "labelDe", "groupDe", "sortOrder")
+VALUES
+  (gen_random_uuid()::text, 'voice-captures:use', 'voice-captures', 'use', 'Benutzen', 'Sprache', 4000)
+ON CONFLICT (key) DO NOTHING;
+
+INSERT INTO "RolePermission" ("roleId", "permissionId", "grantedAt")
+SELECT r.id, p.id, now()
+FROM "Role" r, "Permission" p
+WHERE p.key = 'voice-captures:use'
+  AND r.key IN ('ADMIN', 'BEEKEEPER', 'STAFF')
+ON CONFLICT DO NOTHING;
@@ -104,6 +104,7 @@ model User {
  refreshTokens RefreshToken[]

  inspections            Inspection[]
+  voiceCaptures          VoiceCapture[]
  hiveChanges            HiveChange[]
  varroaMeasurements     VarroaMeasurement[]
  swarmControlChecks     SwarmControlCheck[]
@@ -333,6 +334,7 @@ model Colony {
  queenEvaluations   QueenEvaluation[]
  suggestions        Suggestion[]
  hiveSetup          HiveSetup?
+  voiceCaptures      VoiceCapture[]

  @@index([apiaryId])
  @@index([status])
@@ -2285,3 +2287,73 @@ enum HiveChangeSource {
  INSPECTION
  AI_SUGGESTION
 }
+
+enum VoiceCaptureSource {
+  AUDIO
+  TEXT
+}
+
+enum VoiceCaptureStatus {
+  QUEUED // audio uploaded, waits for Whisper
+  TRANSCRIBING // claimed by the CronJob
+  TRANSCRIBED // text present, waits for Claude (also source TEXT)
+  EXTRACTING // claimed by the worker
+  READY // suggestions present, waits for the user
+  CONFIRMED // inspection saved, content deleted
+  DISCARDED // discarded by the user, content deleted
+  FAILED // finally failed, code in lastErrorCode
+  EXPIRED // deadline passed, content deleted
+}
+
+// A voice note. Audio and text are personal data: content lives here only
+// encrypted and only until it is taken over, at most 7 days. The row itself
+// holds no content and stays 90 days for the error and cost overview.
+model VoiceCapture {
+  id             String             @id @default(cuid())
+  ownerUserId    String
+  owner          User               @relation(fields: [ownerUserId], references: [id])
+  colonyId       String
+  colony         Colony             @relation(fields: [colonyId], references: [id])
+  inspectionDate DateTime           @db.Date // date of the inspection per the form
+  source         VoiceCaptureSource
+  status         VoiceCaptureStatus @default(QUEUED)
+
+  audioMime   String? // audio/webm, audio/ogg, audio/mp4
+  audioBytes  Int?
+  durationSec Int? // reported by the device; Whisper measures again
+
+  transcriptEnc String? // key-ring ciphertext
+  suggestionEnc String? // key-ring ciphertext of the checked suggestion
+
+  attempts      Int       @default(0)
+  availableAt   DateTime  @default(now())
+  claimedAt     DateTime?
+  claimToken    String? // guards against completion by a process that lost the claim
+  lastErrorCode String? // fixed code, never content
+
+  model        String? // e.g. "claude-sonnet-5-5"
+  inputTokens  Int?
+  outputTokens Int?
+
+  inspectionId     String? // set when CONFIRMED
+  createdAt        DateTime  @default(now())
+  readyAt          DateTime?
+  contentExpiresAt DateTime // createdAt + 7 days
+  contentDeletedAt DateTime? // when audio/text/suggestion were deleted
+  updatedAt        DateTime  @updatedAt
+
+  audio VoiceCaptureAudio?
+
+  @@index([status, availableAt])
+  @@index([ownerUserId, status])
+  @@index([contentExpiresAt])
+}
+
+// The audio lives in its own table: deleting is a DELETE of the row (not
+// nulling a big column), and the row is never loaded into lists.
+model VoiceCaptureAudio {
+  captureId String       @id
+  capture   VoiceCapture @relation(fields: [captureId], references: [id], onDelete: Cascade)
+  dataEnc   Bytes // key-ring ciphertext of the audio, at most 2 MB plaintext
+  createdAt DateTime     @default(now())
+}
@@ -415,6 +415,8 @@ export const PERMISSION_CATALOG = [
  // Beute in 3D: hive systems and their part-type catalog. Reading the
  // catalog and writing a colony's setup use the colonies permissions.
  ...group('Beutensysteme', 'hive-systems', [['manage', 'Pflegen']], 3900),
+  // T7 Freihand: voice notes. Every endpoint also needs inspections:create.
+  ...group('Sprache', 'voice-captures', [['use', 'Benutzen']], 4000),
 ] as const satisfies readonly PermissionCatalogEntry[];

 export type PermissionKey = (typeof PERMISSION_CATALOG)[number]['key'];
@@ -1536,6 +1536,20 @@ describe('BackupService', () => {
      );
    });

+    it('excludes VoiceCapture and VoiceCaptureAudio with a reason', async () => {
+      const bundle = await service.exportAll();
+
+      for (const [model, key] of [
+        ['VoiceCapture', 'voiceCapture'],
+        ['VoiceCaptureAudio', 'voiceCaptureAudio'],
+      ] as const) {
+        expect(bundle.data).not.toHaveProperty(key);
+        expect(MODELS_EXCLUDED_FROM_BACKUP[model]).toMatch(
+          /ephemeral, personal data, not deliverable after a restore/,
+        );
+      }
+    });
+
    it('restores phenology reports after the apiaries, remapping a user that does not exist', async () => {
      await importData({
        apiary: [{ id: 'apiary-1' }],
@@ -158,6 +158,10 @@ export const MODELS_EXCLUDED_FROM_BACKUP: Readonly<
  NotificationDelivery:
    'Delivery queue - restoring it would resend old messages.',
  TelegramPollState: 'Update offset and poll lease of this deployment.',
+  VoiceCapture:
+    'Voice notes are ephemeral, personal data, not deliverable after a restore (content is deleted after at most 7 days).',
+  VoiceCaptureAudio:
+    'Encrypted audio of voice notes: ephemeral, personal data, not deliverable after a restore.',
 };

 const DOCUMENT_ENTITY_TYPES = new Set<string>(
@@ -259,6 +259,7 @@ describe('lock keys per table', () => {
      'AuthorityNotice',
      'ApiaryRestriction',
      'ComplianceAcknowledgement',
+      'VoiceCapture',
    ]);
  });

@@ -59,6 +59,7 @@ const LOCKABLE_TABLES = {
  AuthorityNotice: lockable('AuthorityNotice'),
  ApiaryRestriction: lockable('ApiaryRestriction'),
  ComplianceAcknowledgement: lockable('ComplianceAcknowledgement'),
+  VoiceCapture: lockable('VoiceCapture'),
 } as const;

 export type LockableTable = keyof typeof LOCKABLE_TABLES;
@@ -1,3 +1,5 @@
+import * as fs from 'node:fs';
+import * as path from 'node:path';
 import { INestApplication } from '@nestjs/common';
 import * as bcrypt from 'bcryptjs';
 import request from 'supertest';
@@ -95,3 +97,33 @@ describe('Search and filters (e2e)', () => {
    expect(colonies.map((entry) => entry.id)).toEqual([dissolvedColonyId]);
  });
 });
+
+describe('Voice capture content guard', () => {
+  // VoiceCapture holds personal data that is deleted after at most 7 days. It
+  // must never be reachable through search or any export/report builder.
+  const sourceRoots = ['search', 'export'].map((dir) =>
+    path.resolve(__dirname, '../src', dir),
+  );
+
+  function sourceFiles(dir: string): string[] {
+    return fs.readdirSync(dir, { withFileTypes: true }).flatMap((entry) => {
+      const full = path.join(dir, entry.name);
+      if (entry.isDirectory()) return sourceFiles(full);
+      return entry.name.endsWith('.ts') ? [full] : [];
+    });
+  }
+
+  it('VoiceCapture appears in no search provider and no export', () => {
+    const files = sourceRoots.flatMap(sourceFiles);
+
+    expect(files.length).toBeGreaterThan(0);
+    for (const file of files) {
+      expect({
+        file: path.basename(file),
+        mentions: /voiceCapture|VoiceCapture/.test(
+          fs.readFileSync(file, 'utf8'),
+        ),
+      }).toEqual({ file: path.basename(file), mentions: false });
+    }
+  });
+});
@@ -0,0 +1,134 @@
+import { PrismaService } from '../src/prisma/prisma.service';
+
+/**
+ * Schema-level guarantees of the T7 migration that the Prisma schema alone
+ * cannot express (CHECK constraints, cascade) against a real PostgreSQL. The
+ * test database is shared and not empty: every row here is created by this
+ * file and removed again.
+ */
+describe('T7 schema (e2e)', () => {
+  let prisma: PrismaService;
+  const suffix = `${Date.now()}-${Math.round(Math.random() * 1e6)}`;
+  const CHECK_VIOLATION = /check constraint|violates/i;
+
+  let userId: string;
+  let apiaryId: string;
+  let colonyId: string;
+  let inspectionId: string;
+
+  function capture(extra: Record<string, unknown> = {}) {
+    return prisma.voiceCapture.create({
+      data: {
+        ownerUserId: userId,
+        colonyId,
+        inspectionDate: new Date('2026-10-16'),
+        source: 'AUDIO',
+        contentExpiresAt: new Date('2026-10-23T00:00:00Z'),
+        ...extra,
+      },
+    });
+  }
+
+  beforeAll(async () => {
+    prisma = new PrismaService();
+    const role = await prisma.role.findUniqueOrThrow({
+      where: { key: 'READONLY' },
+    });
+    const user = await prisma.user.create({
+      data: {
+        email: `t7-schema-${suffix}@example.com`,
+        passwordHash: 'x',
+        name: 'T7 schema',
+        roleId: role.id,
+      },
+    });
+    userId = user.id;
+    const apiary = await prisma.apiary.create({
+      data: { name: `T7 Stand ${suffix}` },
+    });
+    apiaryId = apiary.id;
+    const colony = await prisma.colony.create({
+      data: { code: `T7-${suffix}`, apiaryId, colonyType: 'PRODUCTION_COLONY' },
+    });
+    colonyId = colony.id;
+    const inspection = await prisma.inspection.create({
+      data: { colonyId, userId, date: new Date('2026-10-16') },
+    });
+    inspectionId = inspection.id;
+  });
+
+  afterEach(async () => {
+    await prisma.voiceCapture.deleteMany({ where: { ownerUserId: userId } });
+  });
+
+  afterAll(async () => {
+    await prisma.voiceCapture.deleteMany({ where: { ownerUserId: userId } });
+    await prisma.inspection.deleteMany({ where: { id: inspectionId } });
+    await prisma.colony.deleteMany({ where: { id: colonyId } });
+    await prisma.apiary.deleteMany({ where: { id: apiaryId } });
+    await prisma.user.deleteMany({ where: { id: userId } });
+    await prisma.$disconnect();
+  });
+
+  it('accepts a plain queued audio capture with defaults', async () => {
+    const row = await capture({
+      audioMime: 'audio/webm',
+      audioBytes: 2097152,
+      durationSec: 240,
+    });
+
+    expect(row.status).toBe('QUEUED');
+    expect(row.attempts).toBe(0);
+    expect(row.claimToken).toBeNull();
+  });
+
+  it('rejects audioBytes 0 and 2097153', async () => {
+    await expect(capture({ audioBytes: 0 })).rejects.toThrow(CHECK_VIOLATION);
+    await expect(capture({ audioBytes: 2097153 })).rejects.toThrow(
+      CHECK_VIOLATION,
+    );
+  });
+
+  it('rejects durationSec 241', async () => {
+    await expect(capture({ durationSec: 241 })).rejects.toThrow(
+      CHECK_VIOLATION,
+    );
+    await expect(capture({ durationSec: 0 })).rejects.toThrow(CHECK_VIOLATION);
+  });
+
+  it('rejects CONFIRMED without inspectionId and inspectionId on a non-confirmed row', async () => {
+    await expect(capture({ status: 'CONFIRMED' })).rejects.toThrow(
+      CHECK_VIOLATION,
+    );
+    await expect(capture({ status: 'READY', inspectionId })).rejects.toThrow(
+      CHECK_VIOLATION,
+    );
+    const confirmed = await capture({ status: 'CONFIRMED', inspectionId });
+    expect(confirmed.inspectionId).toBe(inspectionId);
+  });
+
+  it('rejects audioMime on a TEXT capture', async () => {
+    await expect(
+      capture({ source: 'TEXT', audioMime: 'audio/webm' }),
+    ).rejects.toThrow(CHECK_VIOLATION);
+    const text = await capture({ source: 'TEXT' });
+    expect(text.audioMime).toBeNull();
+  });
+
+  it('deleting a capture cascades to its audio row', async () => {
+    const row = await capture({
+      audioMime: 'audio/webm',
+      audioBytes: 3,
+      audio: { create: { dataEnc: Buffer.from([1, 2, 3]) } },
+    });
+    expect(
+      await prisma.voiceCaptureAudio.count({ where: { captureId: row.id } }),
+    ).toBe(1);
+
+    await prisma.voiceCapture.delete({ where: { id: row.id } });
+
+    expect(
+      await prisma.voiceCaptureAudio.count({ where: { captureId: row.id } }),
+    ).toBe(0);
+  });
+});
@@ -0,0 +1,56 @@
+import { PERMISSION_CATALOG } from '../src/auth/permissions/permission-catalog';
+import { PrismaService } from '../src/prisma/prisma.service';
+
+const KEY = 'voice-captures:use';
+
+/**
+ * The migration grants the T7 permission to the system roles (the boot-time
+ * catalog sync only grants new permissions to ADMIN). Checked against the
+ * migrated test database.
+ */
+describe('T7 voice permissions (e2e)', () => {
+  let prisma: PrismaService;
+
+  beforeAll(() => {
+    prisma = new PrismaService();
+  });
+
+  afterAll(async () => {
+    await prisma.$disconnect();
+  });
+
+  async function holds(roleKey: string): Promise<boolean> {
+    const grant = await prisma.rolePermission.findFirst({
+      where: { role: { key: roleKey }, permission: { key: KEY } },
+    });
+    return grant !== null;
+  }
+
+  it('ADMIN, BEEKEEPER and STAFF hold voice-captures:use', async () => {
+    expect(await holds('ADMIN')).toBe(true);
+    expect(await holds('BEEKEEPER')).toBe(true);
+    expect(await holds('STAFF')).toBe(true);
+  });
+
+  it('READONLY and API_CLIENT do not', async () => {
+    expect(await holds('READONLY')).toBe(false);
+    expect(await holds('API_CLIENT')).toBe(false);
+  });
+
+  it('matches the code catalog in the Sprache group at sortOrder 4000', async () => {
+    const entry = PERMISSION_CATALOG.find((e) => e.key === KEY);
+    const row = await prisma.permission.findUniqueOrThrow({
+      where: { key: KEY },
+    });
+
+    expect(entry).toBeDefined();
+    expect(row).toMatchObject({
+      resource: 'voice-captures',
+      action: 'use',
+      groupDe: 'Sprache',
+      labelDe: 'Benutzen',
+      sortOrder: 4000,
+    });
+    expect(entry).toMatchObject({ groupDe: 'Sprache', sortOrder: 4000 });
+  });
+});