chore(frontend): active strict, ajoute .form-select et documente le spec ciblé

- tsconfig.json : "strict": true, vérifié sans erreur sur app et specs
- _forms.scss : .form-select, select natif habillé comme .form-input avec un
  chevron, documenté dans le design système
- TESTING.md : commande pour jouer un seul fichier ou dossier de specs

.env.example

@@ -17,3 +17,39 @@ APP_LOG_LEVEL=INFO
 APP_SECRET_KEY=change_me
 APP_CORS_ORIGINS=http://localhost:4200
 BACKEND_PORT=8000
+FRONTEND_PORT=3000
+
+# Mailpit capture les courriels du backend, rien ne sort vers l'extérieur.
+MAILPIT_SMTP_PORT=1025
+MAILPIT_UI_PORT=8025
+
+# API Mock EnerVision
+APP_MOCK_API_BASE_URL=https://api-mock.charlieandre.fr
+APP_MOCK_API_USERNAME=change_me
+APP_MOCK_API_PASSWORD=change_me
+APP_MOCK_API_TIMEOUT_SECONDS=10
+
+# Airflow (webserver + scheduler, LocalExecutor). Base de métadonnées dédiée `airflow` dans le
+# même conteneur `db` (cf. db/init/120-airflow-database.sql), pas un conteneur de plus.
+AIRFLOW_PORT=8080
+# Chiffre les connexions/variables stockées par Airflow. Générer la vôtre :
+# python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
+AIRFLOW_FERNET_KEY=change_me
+# Clé Flask du webserver Airflow (signature de session), distincte de la précédente. Générer la
+# vôtre : python -c "import secrets; print(secrets.token_urlsafe(48))"
+AIRFLOW_WEBSERVER_SECRET_KEY=change_me
+AIRFLOW_ADMIN_USERNAME=admin
+# Compte Airflow créé au premier démarrage (service `airflow-init`), sans rapport avec les
+# comptes `app_user` d'EnerVision.
+AIRFLOW_ADMIN_PASSWORD=change_me
+AIRFLOW_ADMIN_EMAIL=admin@enervision.fr
+# `APP_SECRET_KEY` du backend, que le DAG `alertes` lance en sous-processus. Distincte de
+# celle de l'API : la détection ne signe aucun jeton, et Airflow exécute du code depuis son
+# interface (cf. ADR 0008). Générer la vôtre :
+# python -c "import secrets; print(secrets.token_urlsafe(48))"
+AIRFLOW_APP_SECRET_KEY=change_me
+
+# Stack complète derrière le reverse proxy (docker-compose.prod.yml).
+# PUBLIC_HOST alimente l'origine CORS, le lien de réinitialisation et le certificat.
+PUBLIC_HOST=enervision.local
+ACME_EMAIL=

.github/dependabot.yml

@@ -38,3 +38,9 @@ updates:
     directory: "/apps/frontend"
     schedule:
       interval: "weekly"
+
+  # Images du reverse proxy et du compagnon ACME, épinglées dans les fichiers Compose
+  - package-ecosystem: "docker-compose"
+    directory: "/"
+    schedule:
+      interval: "weekly"

.github/workflows/airflow.yml

@@ -0,0 +1,101 @@
+name: Airflow
+
+# Piège : la version de Python vient de etl/airflow/.python-version. C'est 3.12 et non 3.14
+# (contrairement à backend.yml et ml.yml) : apache-airflow 2.10 ne supporte pas 3.14. Le 3.14 de
+# ml/ ne vit que dans l'image Docker, dans son propre environnement (cf. etl/airflow/Dockerfile).
+#
+# Piège : l'image COPY les fichiers de dépendances et le code de ml/ et de apps/backend/. Une
+# modification de l'un ou de l'autre peut donc casser sa construction, d'où ces chemins dans
+# les déclencheurs, alors même que ce workflow ne teste ni le modèle ni l'API.
+
+on:
+  push:
+    paths:
+      - "etl/airflow/**"
+      - "ml/pyproject.toml"
+      - "ml/uv.lock"
+      - "ml/enervision_ml/**"
+      - "apps/backend/pyproject.toml"
+      - "apps/backend/uv.lock"
+      - "apps/backend/app/**"
+      - ".github/workflows/airflow.yml"
+  pull_request:
+    paths:
+      - "etl/airflow/**"
+      - "ml/pyproject.toml"
+      - "ml/uv.lock"
+      - "ml/enervision_ml/**"
+      - "apps/backend/pyproject.toml"
+      - "apps/backend/uv.lock"
+      - "apps/backend/app/**"
+      - ".github/workflows/airflow.yml"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: airflow-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  verification:
+    name: Lint et intégrité des DAGs
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: etl/airflow
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: etl/airflow/uv.lock
+
+      - name: Installe l'interpréteur déclaré par .python-version
+        run: uv python install
+
+      - name: Synchronise les dépendances sans dévier du verrou
+        run: uv sync --all-groups --frozen
+
+      - name: Vérifie le formatage
+        run: uv run ruff format --check .
+
+      - name: Analyse statique
+        run: uv run ruff check --output-format=github .
+
+      # Aucun test ne lance de tâche ni de scheduler : DagBag charge les fichiers de dags/ et
+      # vérifie import, planification, plafonds d'exécution et commande de chaque tâche.
+      - name: Tests d'intégrité des DAGs
+        run: uv run pytest
+
+  image:
+    name: Construction de l'image
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      - name: Construit l'image (contexte à la racine, elle COPY ml/ et apps/backend/)
+        run: docker build -f etl/airflow/Dockerfile -t enervision-airflow:ci .
+
+      # Vérifie ce qui ne casse qu'à l'exécution, pas à la construction : libgomp1 absent
+      # (`OSError: libgomp.so.1` au premier import) ou environnement ml/ non figé.
+      - name: Vérifie que le pipeline ML s'importe sans réseau
+        run: >
+          docker run --rm --network none enervision-airflow:ci
+          bash -c "cd /opt/ml && env -u VIRTUAL_ENV uv run --no-sync python -m enervision_ml.train --help"
+
+      # `--help` sort par argparse avant `get_settings()` : ni base ni secret requis, et
+      # l'import du module prouve que l'environnement /opt/backend est complet. Les deux
+      # commandes du DAG `alertes` sont couvertes, `app.cli` tirant tout FastAPI derrière lui.
+      - name: Vérifie que les deux commandes du DAG alertes s'importent sans réseau
+        run: >
+          docker run --rm --network none enervision-airflow:ci
+          bash -c "cd /opt/backend
+          && env -u VIRTUAL_ENV uv run --no-sync python -m app.detection.internal_alerts --help
+          && env -u VIRTUAL_ENV uv run --no-sync python -m app.cli generate-recommendations --help"

.github/workflows/backend.yml

@@ -56,3 +56,113 @@ jobs:
       # Le marqueur `integration` est exclu par défaut, donc aucune base n'est nécessaire ici.
       - name: Tests et couverture
         run: uv run pytest --cov-fail-under=85
+
+  # Piège : l'image est celle de docker-compose.yml, pas une image `postgres` nue. La première
+  # migration (`5353c0e4f094`) échoue volontairement si l'extension TimescaleDB manque, et un
+  # écart d'image entre la CI et le poste rendrait ce job vert sur une base qui n'est pas la nôtre.
+  integration:
+    name: Tests exigeant une base
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: apps/backend
+
+    services:
+      db:
+        image: timescale/timescaledb-ha:pg17
+        env:
+          POSTGRES_USER: enervision
+          POSTGRES_PASSWORD: change_me
+          POSTGRES_DB: enervision_test
+        ports:
+          - "5433:5432"
+        options: >-
+          --health-cmd "pg_isready -U enervision -d enervision_test"
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 12
+          --health-start-period 40s
+
+    env:
+      DATABASE_URL: postgresql+asyncpg://enervision:change_me@localhost:5433/enervision_test
+      APP_SECRET_KEY: secret-de-test-assez-long-pour-le-validateur
+      PGPASSWORD: change_me
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: apps/backend/uv.lock
+
+      - name: Installe l'interpréteur déclaré par .python-version
+        run: uv python install
+
+      - name: Synchronise les dépendances sans dévier du verrou
+        run: uv sync --all-groups --frozen
+
+      # Sur le poste, c'est db/init/110-test-database.sql qui pose l'extension. Ce fichier n'est
+      # pas monté ici, et sans lui `alembic upgrade head` s'arrête sur la garde de la révision 1.
+      - name: Active TimescaleDB sur la base de test
+        run: psql -h localhost -p 5433 -U enervision -d enervision_test -c "CREATE EXTENSION IF NOT EXISTS timescaledb"
+
+      - name: Applique les migrations
+        run: uv run alembic upgrade head
+
+      # `-m` en ligne de commande écrase celui d'`addopts`. La couverture est désactivée : ce job
+      # ne joue qu'une partie de la suite, son taux n'aurait aucun sens face au seuil de 85 %.
+      - name: Tests d'intégration
+        run: uv run pytest -m integration --no-cov
+
+  security-audit:
+    name: Audit des dépendances
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: apps/backend
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: apps/backend/uv.lock
+
+      # L'audit porte sur le verrou, pas sur l'environnement : sinon pip-audit auditerait
+      # aussi les paquets que son propre `--with` injecte, hors dépendances du projet.
+      - name: Audite les dépendances livrées
+        # Piège : sans `shell: bash`, un échec de `uv export` serait masqué par le pipe.
+        shell: bash
+        run: uv export --frozen --no-dev --no-emit-project --no-hashes | uvx pip-audit --requirement /dev/stdin --no-deps
+
+  sast:
+    name: Analyse statique de sécurité
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: apps/backend
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      # Pourquoi : pas de cache ici. uvx n'installe pas le projet, le verrou n'alimente donc
+      # aucune clé de cache ; la seule roue téléchargée est celle de Bandit.
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+
+      # Pourquoi : le périmètre est `app`, le code livré. Les tests emploient légitimement des
+      # secrets factices et des `assert` que Bandit signalerait sans qu'aucun n'atteigne la prod.
+      - name: Analyse le code livré (bloquant à partir de MEDIUM)
+        run: uvx bandit==1.9.4 --recursive app --severity-level medium --confidence-level medium
+
+      # Piège : sans cette seconde passe, un constat LOW disparaîtrait du journal sans trace.
+      - name: Rapport complet, tous niveaux
+        continue-on-error: true
+        run: uvx bandit==1.9.4 --recursive app

.github/workflows/frontend.yml

@@ -1,20 +1,6 @@
 name: Frontend
-# Pipeline à choix multiple
 
 on:
-  # workflow_dispatch -> lancement manuel des jobs
-  workflow_dispatch:
-    inputs:
-      job_choice:
-        required: true
-        description: "Choix du job"
-        type: choice
-        default: all
-        options:
-            - build
-            - sonarqube
-            - test
-            - all # lancer tous les jobs
   push:
     paths:
       - "apps/frontend/**"
@@ -23,8 +9,9 @@ on:
     paths:
       - "apps/frontend/**"
       - ".github/workflows/frontend.yml"
-# Ordre de lancement des jobs
-# build -> test -> sonarqube -> deploy
+
+permissions:
+  contents: read
 
 jobs:
   build:
@@ -36,13 +23,25 @@ jobs:
           node-version: 24
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
-      
-      - run: npm ci 
+
+      - run: npm ci
         working-directory: apps/frontend
       - run: npm run build
         working-directory: apps/frontend
 
-  test:  
+  security-audit:
+    name: Audit des dépendances
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: 24
+      # Seuil high : une vulnérabilité moderate de devDependency ne doit pas bloquer une livraison.
+      - run: npm audit --audit-level=high --package-lock-only
+        working-directory: apps/frontend
+
+  test:
     needs: build
     runs-on: ubuntu-latest
     steps:
@@ -52,26 +51,14 @@ jobs:
           node-version: 24
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
-      - run: npm ci 
+      - name : Installation des dépendances (Front)
+        run: npm ci 
         working-directory: apps/frontend
-      - run: npm test -- --watch=false
+      - name : Lancement des tests et génénration du rapport de couverture (Front)
+        run: npm test --watch=false --code-coverage --coverageReporters=lcov
         working-directory: apps/frontend
-
-  sonarqube:
-    needs: [build, test]
-    name: SonarQube
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+      - name: Upload coverage
+        uses: actions/upload-artifact@v4
         with:
-          fetch-depth: 0  # Shallow clones should be disabled for a better relevancy of analysis
-      - name: SonarQube Scan
-        uses: SonarSource/sonarqube-scan-action@7006c4492b2e0ee0f816d36501671557c97f5995 # v8.1.0
-        env:
-          SONAR_TOKEN: ${{ secrets.SONAR_TOKEN }}
-
-    
-  # deploy:
-  #   runs-on: ubuntu-latest
-  #   steps:
-  #   - run: echo "DEPLOY job is running"
+          name: frontend-coverage
+          path: apps/frontend/coverage/frontend/lcov.info

.github/workflows/ml.yml

@@ -0,0 +1,82 @@
+name: ML
+
+# Piège : la version de Python vient de ml/.python-version, et doit rester en 3.14 (cf.
+# .github/workflows/backend.yml, même contrainte).
+
+on:
+  push:
+    paths:
+      - "ml/**"
+      - ".github/workflows/ml.yml"
+  pull_request:
+    paths:
+      - "ml/**"
+      - ".github/workflows/ml.yml"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ml-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  verification:
+    name: Lint, typage et tests
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ml
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: ml/uv.lock
+
+      - name: Installe l'interpréteur déclaré par .python-version
+        run: uv python install
+
+      - name: Synchronise les dépendances sans dévier du verrou
+        run: uv sync --all-groups --frozen
+
+      - name: Vérifie le formatage
+        run: uv run ruff format --check .
+
+      - name: Analyse statique
+        run: uv run ruff check --output-format=github .
+
+      - name: Typage
+        run: uv run mypy enervision_ml tests
+
+      # Aucun test ne touche PostgreSQL ni MLflow distant : tout tourne sur donnees
+      # synthetiques ou un magasin SQLite local jetable (cf. ml/tests/test_train.py).
+      - name: Tests
+        run: uv run pytest
+
+  sast:
+    name: Analyse statique de sécurité
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ml
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v4
+
+      # Pourquoi : pas de cache ici. uvx n'installe pas le projet, le verrou n'alimente donc
+      # aucune clé de cache ; la seule roue téléchargée est celle de Bandit.
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+
+      - name: Analyse le code livré (bloquant à partir de MEDIUM)
+        run: uvx bandit==1.9.4 --recursive enervision_ml --severity-level medium --confidence-level medium
+
+      - name: Rapport complet, tous niveaux
+        continue-on-error: true
+        run: uvx bandit==1.9.4 --recursive enervision_ml

.github/workflows/sonarqube.yml

@@ -0,0 +1,169 @@
+name: SonarQube
+
+on:
+  push:
+    paths:
+      - "apps/frontend/**"
+      - "apps/backend/**"
+      - "ml/**"
+      - "etl/airflow/**"
+      - ".github/workflows/sonarqube.yml"
+  pull_request:
+    paths:
+      - "apps/frontend/**"
+      - "apps/backend/**"
+      - "ml/**"
+      - "etl/airflow/**"
+      - ".github/workflows/sonarqube.yml"
+
+
+# Build l'ensemble du projet, puis lance les tests
+# Génère les rapports de couverture, puis lance l'analyse SonarQube
+
+jobs:
+  build-front:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: 24
+          cache: npm
+          cache-dependency-path: apps/frontend/package-lock.json
+      
+      - run: npm ci 
+        working-directory: apps/frontend
+      - run: npm run build
+        working-directory: apps/frontend
+
+  test-front:  
+    needs: build-front
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - uses: actions/setup-node@v6
+        with:
+          node-version: 24
+          cache: npm
+          cache-dependency-path: apps/frontend/package-lock.json
+
+      - name : Installation des dépendances (Front)
+        run: npm ci 
+        working-directory: apps/frontend
+
+      - name : Lancement des tests et génénration du rapport de couverture (Front)
+        run: npm test --watch=false --code-coverage --coverageReporters=lcov
+        working-directory: apps/frontend
+
+      - name: Upload coverage
+        uses: actions/upload-artifact@v4
+        with:
+          name: frontend-coverage
+          path: apps/frontend/coverage/frontend/lcov.info
+
+  build-back:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: apps/backend/uv.lock
+      - name: Installe l'interpréteur déclaré par .python-version
+        run: uv python install
+        working-directory: apps/backend
+
+      - name: Synchronise les dépendances sans dévier du verrou
+        run: uv sync --all-groups --frozen
+        working-directory: apps/backend
+
+      - name: Vérifie le formatage
+        run: uv run ruff format --check .
+        working-directory: apps/backend
+
+      - name: Analyse statique
+        run: uv run ruff check --output-format=github .
+        working-directory: apps/backend
+
+      - name: Typage
+        run: uv run mypy app
+        working-directory: apps/backend
+
+
+  test-back:
+    needs: build-back
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: apps/backend/uv.lock
+
+      - name : Lancement des tests et génénration du rapport de couverture (Back)
+        run: uv run pytest --cov-fail-under=85 --cov-report=xml
+        working-directory: apps/backend
+
+      - name: Upload coverage
+        uses: actions/upload-artifact@v4
+        with:
+          name: backend-coverage
+          path: apps/backend/coverage.xml
+
+  test-ml:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: ml/uv.lock
+
+      - name: Installe l'interpréteur déclaré par .python-version
+        run: uv python install
+        working-directory: ml
+
+      - name: Synchronise les dépendances sans dévier du verrou
+        run: uv sync --all-groups --frozen
+        working-directory: ml
+
+      - name: Lancement des tests et génération du rapport de couverture (ML)
+        run: uv run pytest --cov-report=xml
+        working-directory: ml
+
+      - name: Upload coverage
+        uses: actions/upload-artifact@v4
+        with:
+          name: ml-coverage
+          path: ml/coverage.xml
+
+  sonarqube:
+    needs: [build-front, build-back, test-front, test-back, test-ml]
+    name: SonarQube
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - name: Téléchargement du rapport de couverture (Front)
+        uses: actions/download-artifact@v4
+        with:
+          name: frontend-coverage
+          path: apps/frontend/coverage/frontend
+      - name: Téléchargement du rapport de couverture (Back)
+        uses: actions/download-artifact@v4
+        with:
+          name: backend-coverage
+          path: apps/backend
+      - name: Téléchargement du rapport de couverture (ML)
+        uses: actions/download-artifact@v4
+        with:
+          name: ml-coverage
+          path: ml
+      - name: SonarQube Scan
+        uses: SonarSource/sonarqube-scan-action@v8
+        env:
+          SONAR_TOKEN: ${{ secrets.SONAR_TOKEN }}

.gitignore

@@ -52,11 +52,26 @@ standalone_admin_password.txt
 secrets/
 
 # Donnees locales
-data/
+data/raw/*
+!data/raw/.gitkeep
 *.sqlite3
 monitoring/grafana/data/
 monitoring/prometheus/data/
 
+# ML : jeu de donnees, modeles entraines et suivi MLflow local, tous generes/volumineux
+ml/data/
+ml/models/*
+!ml/models/.gitkeep
+ml/mlruns/
+ml/mlartifacts/
+ml/mlflow.db
+
+# Airflow : base sqlite locale generee par les tests d'integrite des DAGs (etl/airflow/tests)
+etl/airflow/tests/.airflow_home/
+
+# TLS : certificats du reverse proxy, générés par script ou par certbot
+infra/proxy/tls/*.pem
+
 # IDE et OS
 .idea/
 .vscode/

Makefile

@@ -1,15 +1,31 @@
 BACKEND := apps/backend
 FRONTEND := apps/frontend
+ML := ml
+AIRFLOW := etl/airflow
+COMPOSE_PROD := docker compose -f docker-compose.yml -f docker-compose.prod.yml
+
+# Piège : sans `export`, une valeur passée en ligne de commande n'atteindrait pas docker compose.
+# PUBLIC_HOST retombe sur le `.env`, que make ne lit pas, puis sur la valeur de `.env.example`.
+PUBLIC_HOST ?= $(shell sed -n 's/^PUBLIC_HOST=//p' .env 2>/dev/null | tail -1)
+PUBLIC_HOST := $(or $(strip $(PUBLIC_HOST)),enervision.local)
+export PUBLIC_HOST
+ifdef ACME_EMAIL
+export ACME_EMAIL
+endif
 
 .DEFAULT_GOAL := help
-.PHONY: help install install-backend install-frontend dev dev-backend dev-frontend \
+.PHONY: help install install-backend install-frontend install-ml install-airflow \
+        dev dev-backend dev-frontend \
         lint format typecheck test test-cov test-integration check \
-        openapi docker-build db-up db-down db-reset db-logs db-psql migrate bootstrap-admin
+        openapi docker-build db-up db-down db-reset db-logs db-psql migrate bootstrap-admin \
+        ml-lint ml-typecheck ml-test ml-check ml-train ml-score detect-alerts recommendations \
+        airflow-lint airflow-test airflow-check airflow-up airflow-down airflow-logs \
+        tls-selfsigned tls-acme tls-renew stack-up stack-down stack-logs
 
 help: ## Liste les cibles disponibles
 	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | awk 'BEGIN {FS = ":.*?## "}; {printf "  \033[36m%-16s\033[0m %s\n", $$1, $$2}'
 
-install: install-backend install-frontend ## Installe les dépendances backend et frontend
+install: install-backend install-frontend install-ml install-airflow ## Installe les dépendances backend, frontend, ML et Airflow
 
 install-backend: ## Installe les dépendances du backend
 	cd $(BACKEND) && uv sync --all-groups
@@ -17,6 +33,12 @@ install-backend: ## Installe les dépendances du backend
 install-frontend: ## Installe les dépendances du frontend
 	cd $(FRONTEND) && npm ci
 
+install-ml: ## Installe les dépendances du pipeline ML
+	cd $(ML) && uv sync --all-groups
+
+install-airflow: ## Installe les dépendances de lint/test des DAGs Airflow
+	cd $(AIRFLOW) && uv sync --all-groups
+
 dev: ## Lance toute la stack (backend + frontend) en rechargement à chaud
 	@trap 'kill 0' EXIT INT TERM; \
 	$(MAKE) --no-print-directory dev-backend & \
@@ -55,9 +77,79 @@ check: lint typecheck test ## Chaîne de vérification complète
 openapi: ## Régénère apps/backend/openapi.json depuis les routes déclarées
 	cd $(BACKEND) && uv run python -m app.cli export-openapi
 
+ml-lint: ## Analyse statique du pipeline ML
+	cd $(ML) && uv run ruff check .
+
+ml-typecheck: ## Vérifie le typage du pipeline ML
+	cd $(ML) && uv run mypy enervision_ml tests
+
+ml-test: ## Exécute les tests du pipeline ML (donnees synthetiques, sans base ni serveur MLflow)
+	cd $(ML) && uv run pytest
+
+ml-check: ml-lint ml-typecheck ml-test ## Chaîne de vérification complète du pipeline ML
+
+ml-train: ## Entraine le modele LightGBM. CSV=chemin optionnel, sinon lit ML_DATABASE_URL
+	cd $(ML) && uv run python -m enervision_ml.train $(if $(CSV),--csv $(CSV),)
+
+ml-score: ## Score le prochain pas horaire et l'ecrit dans `prediction`. CSV=chemin optionnel
+	cd $(ML) && uv run python -m enervision_ml.score $(if $(CSV),--csv $(CSV),)
+
+detect-alerts: ## Détecte les alertes internes depuis les lectures en base. SITE= et NOW= optionnels
+	cd $(BACKEND) && uv run python -m app.detection.internal_alerts $(if $(SITE),--site-id $(SITE),) $(if $(NOW),--now $(NOW),)
+
+recommendations: ## Genere les recommandations depuis les alertes en base. SITE=identifiant optionnel
+	cd $(BACKEND) && uv run python -m app.cli generate-recommendations $(if $(SITE),--site-id $(SITE),)
+
+airflow-lint: ## Analyse statique des DAGs Airflow
+	cd $(AIRFLOW) && uv run ruff check .
+
+airflow-test: ## Verifie que les DAGs s'importent sans erreur et ont la structure attendue
+	cd $(AIRFLOW) && uv run pytest
+
+airflow-check: airflow-lint airflow-test ## Chaîne de vérification complète des DAGs Airflow
+
+airflow-up: ## Démarre Airflow (webserver + scheduler, LocalExecutor). db-up requis avant.
+	docker compose up -d airflow-init airflow-webserver airflow-scheduler
+	@echo "airflow  -> http://localhost:$${AIRFLOW_PORT:-8080}"
+
+airflow-down: ## Arrête le webserver et le scheduler Airflow
+	docker compose stop airflow-webserver airflow-scheduler
+
+airflow-logs: ## Suit les journaux du scheduler Airflow (où tournent les tâches, LocalExecutor)
+	docker compose logs -f airflow-scheduler
+
 docker-build: ## Construit l'image du backend
 	docker build -t enervision-backend:local $(BACKEND)
 
+tls-selfsigned: ## Génère le certificat de démonstration. PUBLIC_HOST=..., FORCE=1 pour écraser
+	./scripts/tls-selfsigned.sh $(if $(FORCE),--force,)
+
+stack-up: ## Démarre la stack complète derrière le reverse proxy (80/443). PUBLIC_HOST=... au besoin
+	@test -f infra/proxy/tls/fullchain.pem \
+		|| { echo "Aucun certificat dans infra/proxy/tls. Lancer d'abord make tls-selfsigned"; exit 1; }
+	@openssl x509 -in infra/proxy/tls/fullchain.pem -noout -checkhost "$(PUBLIC_HOST)" >/dev/null \
+		|| { echo "Le certificat ne couvre pas $(PUBLIC_HOST). Relancer make tls-selfsigned PUBLIC_HOST=$(PUBLIC_HOST) FORCE=1"; exit 1; }
+	$(COMPOSE_PROD) up -d --build
+
+stack-down: ## Arrête la stack complète en conservant les données
+	$(COMPOSE_PROD) stop
+
+stack-logs: ## Suit les journaux du reverse proxy
+	$(COMPOSE_PROD) logs -f proxy
+
+tls-acme: ## Demande un certificat Let's Encrypt. PUBLIC_HOST public et ACME_EMAIL requis
+	@test "$(PUBLIC_HOST)" != enervision.local \
+		|| { echo "PUBLIC_HOST doit être un domaine public résolvable, pas le nom de démonstration"; exit 1; }
+	$(COMPOSE_PROD) --profile acme run --rm certbot certonly --webroot -w /var/www/certbot \
+		-d $(PUBLIC_HOST) \
+		--email $${ACME_EMAIL:?ACME_EMAIL=... requis} \
+		--agree-tos --no-eff-email --deploy-hook /deploy-hook.sh
+	$(COMPOSE_PROD) exec proxy nginx -s reload
+
+tls-renew: ## Renouvelle les certificats Let's Encrypt et recharge le proxy
+	$(COMPOSE_PROD) --profile acme run --rm certbot renew --deploy-hook /deploy-hook.sh
+	$(COMPOSE_PROD) exec proxy nginx -s reload
+
 db-up: ## Démarre la base PostgreSQL TimescaleDB
 	docker compose up -d db
 

README.md

@@ -21,10 +21,12 @@ Ce que la documentation apporte à chacun : [docs/architecture/00-vue-ensemble.m
 | Backend    | FastAPI, Python 3.14                | `apps/backend`      | Initialise    |
 | Frontend   | Angular 22, Node 24 LTS             | `apps/frontend`     | Tableau de bord |
 | Base       | PostgreSQL 17 + TimescaleDB         | `db`                | Initialise    |
-| ETL        | Apache Airflow                      | `etl/airflow`       | A initialiser |
+| ETL        | Apache Airflow                      | `etl/airflow`       | Trois DAGs    |
 | Infra      | Terraform (k3s single-node)         | `infra/terraform`   | Initialise    |
+| Reverse proxy | Nginx, TLS                       | `infra/proxy`       | En place      |
 | CI/CD      | GitHub Actions                      | `.github/workflows` | Backend en place |
 | Monitoring | Prometheus, Grafana, Alertmanager   | `monitoring`        | A initialiser |
+| ML         | LightGBM, MLflow                    | `ml`                | Entrainement initialise |
 
 Le backend, la base et l'infrastructure (Terraform/k3s) sont initialises a ce stade. Le frontend
 sert un tableau de bord sur `/dashboard`, dont les données proviennent de fixtures : les endpoints
@@ -46,13 +48,16 @@ L'etat detaille de chaque brique et les vues d'architecture sont dans
 │   ├── migrations/     Migrations SQL versionnees
 │   └── seeds/          Jeux de donnees de reference
 ├── etl/airflow/
-│   ├── dags/           DAGs d'ingestion et d'agregation
+│   ├── dags/           DAGs d'orchestration (pipeline ML, alertes)
 │   ├── plugins/        Operateurs et hooks maison
 │   ├── include/        Requetes SQL et ressources des DAGs
 │   └── tests/          Tests d'integrite des DAGs
-├── infra/terraform/
-│   ├── modules/        Modules reutilisables
-│   └── environments/   Racines Terraform, une par environnement
+├── infra/
+│   ├── proxy/          Reverse proxy Nginx : terminaison TLS et routage
+│   └── terraform/
+│       ├── modules/        Modules reutilisables
+│       └── environments/   Racines Terraform, une par environnement
+├── ml/                 Pipeline d'entrainement LightGBM, suivi MLflow
 ├── monitoring/
 │   ├── prometheus/     Collecte et regles d'alerte
 │   ├── grafana/        Provisioning et dashboards
@@ -96,6 +101,21 @@ Verifier que la base repond et que l'extension est chargee :
 curl -s localhost:8000/api/v1/health/ready
 ```
 
+## Stack complète derrière le reverse proxy
+
+Pour servir l'application comme sur la machine cible, en HTTPS et sous une seule origine.
+L'overlay emploie `!override` et `!reset`, donc **Docker Compose 2.24.4 ou plus récent** :
+
+```bash
+make tls-selfsigned PUBLIC_HOST=enervision.local   # certificat de démonstration
+make stack-up PUBLIC_HOST=enervision.local         # nginx en 80/443, rien d'autre n'est publié
+```
+
+Le navigateur avertit d'un émetteur inconnu : Let's Encrypt reste hors d'atteinte tant qu'aucun
+nom de domaine public ne résout vers la machine. Routage, mode ACME et renouvellement dans
+[`infra/proxy/README.md`](infra/proxy/README.md) ; la décision et ses motifs dans
+[l'ADR 0007](docs/adr/0007-terminaison-tls-et-reverse-proxy-nginx.md).
+
 ## Conventions
 
 - Branches : `feat/`, `fix/`, `chore/`, `docs/`, `test/` suivi d'un libelle court.

apps/backend/.env.example

@@ -8,3 +8,17 @@ APP_SECRET_KEY=change_me
 
 APP_CORS_ORIGINS=http://localhost:4200
 DATABASE_URL=postgresql+asyncpg://enervision:change_me@localhost:5433/enervision
+
+# Mot de passe oublié : lien à usage unique valable 15 minutes par défaut.
+APP_FRONTEND_RESET_PASSWORD_URL=http://localhost:4200/reset-password
+
+# SMTP local de dev (Mailpit, cf. docker-compose.yml) : aucune authentification, aucun TLS.
+# À remplacer par un vrai relais en staging/prod.
+APP_SMTP_HOST=localhost
+APP_SMTP_PORT=1025
+APP_SMTP_USE_TLS=false
+APP_SMTP_FROM_ADDRESS=no-reply@enervision.fr
+APP_MOCK_API_BASE_URL=https://api-mock.charlieandre.fr
+APP_MOCK_API_USERNAME=change_me
+APP_MOCK_API_PASSWORD=change_me
+APP_MOCK_API_TIMEOUT_SECONDS=10

apps/backend/Dockerfile

@@ -11,13 +11,14 @@ WORKDIR /app
 RUN --mount=type=cache,target=/root/.cache/uv \
     --mount=type=bind,source=uv.lock,target=uv.lock \
     --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
-    uv sync --locked --no-install-project --no-dev
+    uv sync --locked --no-install-project --no-dev --no-build
 
+# Le projet lui-meme n'est pas installe (pas de second `uv sync`) : il tourne depuis /app, le
+# repertoire de travail, et rien ne lit ses metadonnees. L'installer imposerait de le construire
+# (backend hatchling), donc de retirer `--no-build` de l'etape ci-dessus, qui garantit que
+# l'installation des dependances n'execute aucun script de build (regle Sonar docker:S8541).
 COPY . /app
 
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv sync --locked --no-dev
-
 
 FROM python:3.14-slim AS runtime
 

apps/backend/README.md

@@ -103,6 +103,8 @@ Le sens de dependance est unique : `endpoints` vers `services` vers `repositorie
 | `/api/v1/auth/logout` | Ferme la session courante | cookie, idempotente |
 | `/api/v1/auth/logout-all` | Ferme toutes les sessions du compte | jeton |
 | `/api/v1/auth/password` | Change son propre mot de passe | jeton |
+| `/api/v1/auth/forgot-password` | Demande un lien de réinitialisation par email | public |
+| `/api/v1/auth/reset-password` | Choisit un nouveau mot de passe depuis ce lien | public |
 | `/api/v1/auth/me` | Décrit le compte connecté | jeton |
 | `/api/v1/users` | Liste et crée des comptes | `admin` |
 | `/api/v1/users/{id}` | Change le rôle ou l'activation | `admin` |
@@ -111,6 +113,7 @@ Le sens de dependance est unique : `endpoints` vers `services` vers `repositorie
 | `/api/v1/sites/{site_id}` | Décrit un site | `lecteur` |
 | `/api/v1/recommendations` | Liste les recommandations | `lecteur` |
 | `/api/v1/recommendations/{recommendation_id}` | Décrit une recommandation | `lecteur` |
+| `/api/v1/recommendations/generate` | Génère les recommandations depuis les alertes (POST) | `admin` |
 | `/metrics` | Métriques au format Prometheus | jeton si `APP_METRICS_TOKEN` |
 | `/docs`, `/openapi.json` | Documentation, fermée en `staging` et `prod` | public sinon |
 

apps/backend/TESTING.md

@@ -123,6 +123,11 @@ async def test_repository_reads_back_what_it_wrote(session: AsyncSession) -> Non
 defaut, ce qui garde `make check` jouable sans Docker. Tout autre marqueur doit etre
 declare dans `pyproject.toml` : `--strict-markers` refuse les marqueurs inconnus.
 
+Ces tests ne sont pas pour autant facultatifs : le job `integration` de
+`.github/workflows/backend.yml` monte un service TimescaleDB, applique les migrations et
+les joue a chaque poussee. Un test `integration` casse donc la CI comme un autre. En local,
+`make db-up` puis `make test-integration`.
+
 ## Couverture
 
 Les branches sont mesurees, pas seulement les lignes. Le seuil de 85 % ne s'applique
@@ -142,14 +147,27 @@ uv run pytest tests/api/test_health.py           # un seul fichier
 uv run pytest -k readiness                       # par motif de nom
 ```
 
-## Trois fichiers à connaître avant de toucher à l'authentification
+## Quatre fichiers à connaître avant de toucher à l'authentification
+
+`tests/api/acces.py` porte la classification des routes du contrat, en quatre ensembles :
+`ROUTES_PUBLIQUES`, `ROUTE_COOKIE`, `ROUTES_SANS_ROLE` et la table `ROLE_MINIMUM`. Ce n'est pas
+un fichier de test, c'est la référence que les trois autres confrontent au comportement observé.
+**Toute route ajoutée doit y être classée** : `test_every_declared_route_is_classified` échoue
+sinon, et échoue aussi sur une entrée qui ne correspond plus à aucune route.
 
 `tests/api/test_route_protection.py` interroge réellement chaque route sans identifiant et
 échoue si l'une d'elles répond autre chose qu'un 401 ou un 403. Il n'inspecte pas l'arbre de
 dépendances : celui-ci n'est accessible que par l'API privée de FastAPI, et surtout une route
 peut porter la bonne dépendance tout en répondant quand même. **Rendre une route publique impose
-donc de modifier la liste `ROUTES_PUBLIQUES` de ce fichier**, ce qui apparaît en clair dans la
-diff d'une pull request.
+donc de modifier `ROUTES_PUBLIQUES` dans `acces.py`**, ce qui apparaît en clair dans la diff
+d'une pull request.
+
+`tests/api/test_matrice_acces.py` croise chaque route gardée avec chacun des trois rôles, dans
+les deux sens : un rôle insuffisant reçoit un 403 `Droits insuffisants`, un rôle suffisant ne le
+reçoit jamais. Le second sens est ce qui rend visible une garde posée trop haut, par exemple
+`AdminDep` sur une route de lecture. La même matrice est rejouée sous `integration` avec de vrais
+jetons, donc en traversant le décodage du JWT et la relecture du compte en base, que
+`dependency_overrides` court-circuite.
 
 `tests/services/test_auth.py` donne au faux hacheur un **compteur d'appels**. C'est ce qui rend
 possibles les deux assertions qui prouvent la conception, et qu'aucune autre forme de test

apps/backend/alembic/versions/c0adab96238c_jetons_de_reinitialisation.py

@@ -0,0 +1,96 @@
+"""jetons et tentatives de reinitialisation de mot de passe
+
+Revision ID: c0adab96238c
+Revises: e6d2026091501
+Create Date: 2026-09-17 10:37:12.571314
+
+Meme schema que `refresh_token` pour `password_reset_token` : seule l'empreinte SHA-256 du
+jeton est stockee, jamais le jeton lui-meme, pour la meme raison (revocation en cascade,
+aucune session utilisable dans un pg_dump qui fuiterait).
+
+`password_reset_attempt` vit hors de `audit_log`, comme `login_attempt`, car son volume est
+pilote par l'attaquant : une campagne de demandes y ecrirait des lignes que l'audit, en ajout
+seul, ne devrait jamais purger.
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+revision: str = "c0adab96238c"
+down_revision: str | Sequence[str] | None = "e6d2026091501"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+JETONS_VIVANTS = "consumed_at is null"
+
+
+def upgrade() -> None:
+    op.create_table(
+        "password_reset_attempt",
+        sa.Column("id", sa.BigInteger(), sa.Identity(always=True), nullable=False),
+        sa.Column(
+            "occurred_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column("email_tried", sa.String(length=320), nullable=False),
+        sa.Column("client_ip", postgresql.INET(), nullable=True),
+        sa.PrimaryKeyConstraint("id", name="pk_password_reset_attempt"),
+    )
+    op.create_index(
+        "ix_password_reset_attempt_email_date",
+        "password_reset_attempt",
+        ["email_tried", "occurred_at"],
+    )
+    op.create_index(
+        "ix_password_reset_attempt_ip_date", "password_reset_attempt", ["client_ip", "occurred_at"]
+    )
+
+    op.create_table(
+        "password_reset_token",
+        sa.Column("id", sa.UUID(), server_default=sa.text("gen_random_uuid()"), nullable=False),
+        sa.Column("user_id", sa.UUID(), nullable=False),
+        sa.Column("token_hash", sa.LargeBinary(), nullable=False),
+        sa.Column(
+            "issued_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column("expires_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("consumed_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("client_ip", postgresql.INET(), nullable=True),
+        sa.Column("user_agent", sa.Text(), nullable=True),
+        sa.ForeignKeyConstraint(
+            ["user_id"],
+            ["app_user.id"],
+            name="fk_password_reset_token_user",
+            ondelete="CASCADE",
+        ),
+        sa.PrimaryKeyConstraint("id", name="pk_password_reset_token"),
+        sa.UniqueConstraint("token_hash", name="uq_password_reset_token_hash"),
+    )
+    op.create_index("ix_password_reset_token_user", "password_reset_token", ["user_id"])
+    op.create_index(
+        "ix_password_reset_token_vivants",
+        "password_reset_token",
+        ["user_id"],
+        postgresql_where=JETONS_VIVANTS,
+    )
+
+
+def downgrade() -> None:
+    op.drop_index(
+        "ix_password_reset_token_vivants",
+        table_name="password_reset_token",
+        postgresql_where=JETONS_VIVANTS,
+    )
+    op.drop_index("ix_password_reset_token_user", table_name="password_reset_token")
+    op.drop_table("password_reset_token")
+    op.drop_index("ix_password_reset_attempt_ip_date", table_name="password_reset_attempt")
+    op.drop_index("ix_password_reset_attempt_email_date", table_name="password_reset_attempt")
+    op.drop_table("password_reset_attempt")

apps/backend/app/api/deps.py

@@ -16,6 +16,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.core.config import Settings, get_settings
 from app.core.hashing import Argon2Hasher, build_hasher
+from app.core.mailer import Mailer, SmtpConfig
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role, has_at_least
 from app.core.security import TokenExpiredError, TokenInvalidError, TokenPolicy
@@ -24,13 +25,18 @@ from app.db.session import get_session
 from app.repositories.alert import AlertRepository
 from app.repositories.audit_log import AuditLogRepository
 from app.repositories.login_attempt import LoginAttemptRepository
+from app.repositories.password_reset_attempt import PasswordResetAttemptRepository
+from app.repositories.password_reset_token import PasswordResetTokenRepository
+from app.repositories.prediction import PredictionRepository
 from app.repositories.reading import ReadingRepository
 from app.repositories.recommendation import RecommendationRepository
 from app.repositories.refresh_token import RefreshTokenRepository
 from app.repositories.site import SiteRepository
 from app.repositories.user import UserRepository
 from app.services.alert import AlertService
-from app.services.auth import AuthService, LoginPolicy
+from app.services.auth import AuthService, LoginPolicy, PasswordResetPolicy
+from app.services.prediction import PredictionService
+from app.services.reading import ReadingService
 from app.services.recommendation import RecommendationService
 from app.services.sensor import SensorService
 from app.services.site import SiteService
@@ -97,11 +103,27 @@ def get_client_ip(request: Request, settings: SettingsDep) -> str | None:
     return request.client.host if request.client else None
 
 
+def get_mailer(settings: SettingsDep) -> Mailer:
+    return Mailer(
+        SmtpConfig(
+            host=settings.smtp_host,
+            port=settings.smtp_port,
+            username=settings.smtp_username,
+            password=(
+                settings.smtp_password.get_secret_value() if settings.smtp_password else None
+            ),
+            use_tls=settings.smtp_use_tls,
+            from_address=settings.smtp_from_address,
+        )
+    )
+
+
 def get_auth_service(
     session: SessionDep,
     settings: SettingsDep,
     hasher: Annotated[Argon2Hasher, Depends(get_hasher)],
     token_policy: Annotated[TokenPolicy, Depends(get_token_policy)],
+    mailer: Annotated[Mailer, Depends(get_mailer)],
 ) -> AuthService:
     return AuthService(
         users=UserRepository(session),
@@ -118,6 +140,16 @@ def get_auth_service(
             max_failures_per_identifier=settings.login_max_failures_per_identifier,
         ),
         refresh_ttl=timedelta(seconds=settings.refresh_token_ttl_seconds),
+        reset_tokens=PasswordResetTokenRepository(session),
+        reset_attempts=PasswordResetAttemptRepository(session),
+        reset_policy=PasswordResetPolicy(
+            window_seconds=settings.password_reset_window_seconds,
+            max_requests_per_identifier=settings.password_reset_max_requests_per_identifier,
+            max_requests_per_ip=settings.password_reset_max_requests_per_ip,
+            token_ttl=timedelta(seconds=settings.password_reset_ttl_seconds),
+            frontend_reset_url=settings.frontend_reset_password_url,
+        ),
+        mailer=mailer,
     )
 
 
@@ -141,21 +173,30 @@ UserServiceDep = Annotated[UserService, Depends(get_user_service)]
 
 
 def get_site_service(session: SessionDep) -> SiteService:
-    return SiteService(sites=SiteRepository(session))
+    return SiteService(sites=SiteRepository(session), readings=ReadingRepository(session))
 
 
 SiteServiceDep = Annotated[SiteService, Depends(get_site_service)]
 
 
 def get_alert_service(session: SessionDep) -> AlertService:
-    return AlertService(alerts=AlertRepository(session))
+    return AlertService(
+        alerts=AlertRepository(session),
+        readings=ReadingRepository(session),
+        predictions=PredictionRepository(session),
+        sites=SiteRepository(session),
+    )
 
 
 AlertServiceDep = Annotated[AlertService, Depends(get_alert_service)]
 
 
 def get_recommendation_service(session: SessionDep) -> RecommendationService:
-    return RecommendationService(recommendations=RecommendationRepository(session))
+    return RecommendationService(
+        recommendations=RecommendationRepository(session),
+        alerts=AlertRepository(session),
+        transaction=session,
+    )
 
 
 RecommendationServiceDep = Annotated[RecommendationService, Depends(get_recommendation_service)]
@@ -168,6 +209,13 @@ def get_stats_service(session: SessionDep) -> StatsService:
 StatsServiceDep = Annotated[StatsService, Depends(get_stats_service)]
 
 
+def get_reading_service(session: SessionDep) -> ReadingService:
+    return ReadingService(readings=ReadingRepository(session))
+
+
+ReadingServiceDep = Annotated[ReadingService, Depends(get_reading_service)]
+
+
 def get_sensor_service(session: SessionDep) -> SensorService:
     return SensorService(sites=SiteRepository(session), readings=ReadingRepository(session))
 
@@ -175,6 +223,15 @@ def get_sensor_service(session: SessionDep) -> SensorService:
 SensorServiceDep = Annotated[SensorService, Depends(get_sensor_service)]
 
 
+def get_prediction_service(session: SessionDep) -> PredictionService:
+    return PredictionService(
+        sites=SiteRepository(session), predictions=PredictionRepository(session)
+    )
+
+
+PredictionServiceDep = Annotated[PredictionService, Depends(get_prediction_service)]
+
+
 async def get_current_principal(
     credentials: CredentialsDep,
     session: SessionDep,

apps/backend/app/api/openapi.py

@@ -63,7 +63,7 @@ TAGS: Final[list[dict[str, Any]]] = [
         "name": "recommendations",
         "description": (
             "Consultation des recommandations issues des alertes. Accessible à partir du rôle "
-            "`lecteur`."
+            "`lecteur`. Leur génération par le moteur de règles est réservée au rôle `admin`."
         ),
     },
     {
@@ -71,10 +71,25 @@ TAGS: Final[list[dict[str, Any]]] = [
         "description": "Statistiques agrégées de consommation. Accessible à partir du rôle "
         "`lecteur`.",
     },
+    {
+        "name": "readings",
+        "description": (
+            "Historique des lectures de consommation. Fenêtre temporelle plafonnée à 90 jours, "
+            "24 dernières heures par défaut si `start`/`end` sont omis. Accessible à partir du "
+            "rôle `lecteur`."
+        ),
+    },
     {
         "name": "sensors",
         "description": "État de santé des capteurs par site. Réservé au rôle `admin`.",
     },
+    {
+        "name": "predictions",
+        "description": (
+            "Dernière prévision de consommation par site, calculée hors ligne par le pipeline "
+            "de scoring (`ml/`) et simplement lue ici. Accessible à partir du rôle `lecteur`."
+        ),
+    },
 ]
 
 cookie_de_rafraichissement = APIKeyCookie(
@@ -156,3 +171,16 @@ REPONSE_ORIGINE_REFUSEE: Final[Reponses] = {
         "description": "Origine non autorisée (protection CSRF de `require_trusted_origin`).",
     },
 }
+
+REPONSE_LIMITE: Final[Reponses] = {
+    429: {
+        "model": ErrorResponse,
+        "description": "Trop de demandes sur cette fenêtre glissante.",
+        "headers": {
+            "Retry-After": {
+                "description": "Secondes à attendre avant une nouvelle tentative.",
+                "schema": {"type": "integer"},
+            }
+        },
+    },
+}

apps/backend/app/api/v1/endpoints/auth.py

@@ -2,7 +2,7 @@
 # d'accès ne va jamais dans un cookie. C'est ce qui réduit la surface CSRF aux trois routes de
 # ce module : partout ailleurs, le navigateur n'attache rien de lui-même.
 
-from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
+from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Request, Response, status
 
 from app.api.deps import (
     AuthServiceDep,
@@ -12,6 +12,7 @@ from app.api.deps import (
     require_trusted_origin,
 )
 from app.api.openapi import (
+    REPONSE_LIMITE,
     REPONSE_ORIGINE_REFUSEE,
     REPONSE_VALIDATION,
     REPONSES_AUTHENTIFIEES,
@@ -21,15 +22,19 @@ from app.api.openapi import (
 from app.core.cookies import RefreshCookie, cookie_name
 from app.core.logging import get_logger
 from app.schemas.auth import (
+    ForgotPasswordRequest,
     LoginRequest,
     PasswordChangeRequest,
     PrincipalResponse,
+    ResetPasswordRequest,
+    ResetTokenValidationResponse,
     TokenResponse,
 )
 from app.schemas.errors import ErrorResponse
 from app.services.auth import (
     AuthenticatedSession,
     InvalidCredentialsError,
+    InvalidOrExpiredResetTokenError,
     RateLimitedError,
     SessionRejectedError,
 )
@@ -39,6 +44,7 @@ logger = get_logger(__name__)
 
 DETAIL_IDENTIFIANTS = "Identifiants invalides"
 DETAIL_SESSION = "Session invalide"
+DETAIL_LIEN_RESET = "Lien invalide ou expiré"
 
 REPONSES_LOGIN: Reponses = {
     **REPONSE_VALIDATION,
@@ -85,6 +91,20 @@ REPONSES_MOT_DE_PASSE: Reponses = {
     },
 }
 
+REPONSES_FORGOT_PASSWORD: Reponses = {
+    **REPONSE_VALIDATION,
+    **REPONSE_LIMITE,
+}
+
+REPONSES_RESET_PASSWORD: Reponses = {
+    **REPONSE_VALIDATION,
+    **REPONSE_ORIGINE_REFUSEE,
+    400: {
+        "model": ErrorResponse,
+        "description": "Lien invalide, déjà utilisé, ou expiré (durée de vie : 15 minutes).",
+    },
+}
+
 
 def repond(
     response: Response, settings: SettingsDep, session: AuthenticatedSession
@@ -267,3 +287,79 @@ async def change_password(
 
     logger.info("auth.password_changed user_id=%s", principal.id)
     return repond(response, settings, session)
+
+
+@router.post(
+    "/forgot-password",
+    status_code=status.HTTP_202_ACCEPTED,
+    summary="Demande un lien de réinitialisation par email",
+    responses=REPONSES_FORGOT_PASSWORD,
+)
+async def forgot_password(
+    payload: ForgotPasswordRequest,
+    request: Request,
+    response: Response,
+    service: AuthServiceDep,
+    background_tasks: BackgroundTasks,
+    client_ip: str | None = Depends(get_client_ip),
+) -> None:
+    response.headers["Cache-Control"] = "no-store"
+
+    try:
+        await service.request_password_reset(
+            email=payload.email,
+            client_ip=client_ip,
+            user_agent=request.headers.get("user-agent"),
+            background_tasks=background_tasks,
+        )
+    except RateLimitedError as erreur:
+        logger.warning("auth.password_reset.rate_limited ip=%s", client_ip)
+        raise HTTPException(
+            status_code=status.HTTP_429_TOO_MANY_REQUESTS,
+            detail="Trop de demandes, réessayez plus tard",
+            headers={"Retry-After": str(erreur.retry_after)},
+        ) from erreur
+
+
+@router.get(
+    "/reset-password/validate",
+    response_model=ResetTokenValidationResponse,
+    summary="Vérifie sans le consommer si un lien de réinitialisation est encore valide",
+    responses=REPONSE_VALIDATION,
+)
+async def validate_reset_token(token: str, service: AuthServiceDep) -> ResetTokenValidationResponse:
+    return ResetTokenValidationResponse(valid=await service.is_reset_token_valid(token=token))
+
+
+@router.post(
+    "/reset-password",
+    response_model=TokenResponse,
+    summary="Choisit un nouveau mot de passe depuis un lien reçu par email",
+    dependencies=[Depends(require_trusted_origin)],
+    responses=REPONSES_RESET_PASSWORD,
+)
+async def reset_password(
+    payload: ResetPasswordRequest,
+    request: Request,
+    response: Response,
+    settings: SettingsDep,
+    service: AuthServiceDep,
+    client_ip: str | None = Depends(get_client_ip),
+) -> TokenResponse:
+    response.headers["Cache-Control"] = "no-store"
+
+    try:
+        session = await service.confirm_password_reset(
+            token=payload.token,
+            new_password=payload.new_password,
+            client_ip=client_ip,
+            user_agent=request.headers.get("user-agent"),
+        )
+    except InvalidOrExpiredResetTokenError as erreur:
+        logger.warning("auth.password_reset.invalid_token ip=%s", client_ip)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST, detail=DETAIL_LIEN_RESET
+        ) from erreur
+
+    logger.info("auth.password_reset.success user_id=%s", session.principal.id)
+    return repond(response, settings, session)

apps/backend/app/api/v1/endpoints/predictions.py

@@ -0,0 +1,18 @@
+from fastapi import APIRouter
+
+from app.api.deps import LecteurDep, PredictionServiceDep
+from app.schemas.prediction import PredictionSummaryResponse
+
+router = APIRouter()
+
+
+@router.get(
+    "",
+    response_model=PredictionSummaryResponse,
+    summary="Dernière prédiction de consommation par site",
+)
+async def get_predictions(
+    _: LecteurDep, service: PredictionServiceDep
+) -> PredictionSummaryResponse:
+    resume = await service.summary()
+    return PredictionSummaryResponse.model_validate(resume)

apps/backend/app/api/v1/endpoints/readings.py

@@ -0,0 +1,54 @@
+from datetime import datetime
+
+from fastapi import APIRouter, HTTPException, Query, status
+
+from app.api.deps import LecteurDep, ReadingServiceDep
+from app.api.openapi import REPONSE_VALIDATION, Reponses
+from app.schemas.errors import ErrorResponse
+from app.schemas.reading import ReadingResponse
+from app.services.reading import FenetreInverseeError, FenetreTropLargeError
+
+router = APIRouter()
+
+REPONSES_FENETRE: Reponses = {
+    **REPONSE_VALIDATION,
+    400: {
+        "model": ErrorResponse,
+        "description": (
+            "Fenêtre temporelle invalide : `start` postérieur ou égal à `end`, ou écart entre "
+            "les deux supérieur à 90 jours."
+        ),
+    },
+}
+
+
+@router.get(
+    "",
+    response_model=list[ReadingResponse],
+    summary="Liste l'historique des lectures",
+    responses=REPONSES_FENETRE,
+)
+async def list_readings(
+    _: LecteurDep,
+    service: ReadingServiceDep,
+    site_id: str | None = None,
+    start: datetime | None = None,
+    end: datetime | None = None,
+    limit: int = Query(500, ge=1, le=2000),
+    offset: int = Query(0, ge=0),
+) -> list[ReadingResponse]:
+    try:
+        lectures = await service.list_history(
+            site_id=site_id, start=start, end=end, limit=limit, offset=offset
+        )
+    except FenetreInverseeError as erreur:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="`start` doit être strictement antérieur à `end`",
+        ) from erreur
+    except FenetreTropLargeError as erreur:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="L'écart entre `start` et `end` ne peut pas dépasser 90 jours",
+        ) from erreur
+    return [ReadingResponse.model_validate(lecture) for lecture in lectures]

apps/backend/app/api/v1/endpoints/recommendations.py

@@ -1,13 +1,18 @@
 from fastapi import APIRouter, HTTPException, status
 
-from app.api.deps import LecteurDep, RecommendationServiceDep
-from app.api.openapi import REPONSE_VALIDATION, Reponses
+from app.api.deps import AdminDep, LecteurDep, RecommendationServiceDep
+from app.api.openapi import REPONSE_VALIDATION, REPONSES_ADMIN, Reponses
 from app.schemas.errors import ErrorResponse
-from app.schemas.recommendation import RecommendationResponse
+from app.schemas.recommendation import (
+    RecommendationGenerationResponse,
+    RecommendationResponse,
+)
 from app.services.recommendation import RecommendationNotFoundError
 
 router = APIRouter()
 
+REPONSES_GENERATION: Reponses = {**REPONSES_ADMIN, **REPONSE_VALIDATION}
+
 REPONSES_INTROUVABLE: Reponses = {
     **REPONSE_VALIDATION,
     404: {"model": ErrorResponse, "description": "Aucune recommandation ne porte cet identifiant."},
@@ -38,3 +43,22 @@ async def get_recommendation(
             status_code=status.HTTP_404_NOT_FOUND, detail="Recommandation introuvable"
         ) from erreur
     return RecommendationResponse.model_validate(recommendation)
+
+
+@router.post(
+    "/generate",
+    response_model=RecommendationGenerationResponse,
+    summary="Génère les recommandations à partir des alertes",
+    responses=REPONSES_GENERATION,
+)
+async def generate_recommendations(
+    _: AdminDep,
+    service: RecommendationServiceDep,
+    site_id: str | None = None,
+) -> RecommendationGenerationResponse:
+    rapport = await service.generate(site_id=site_id)
+    return RecommendationGenerationResponse(
+        alerts_examined=rapport.alertes_examinees,
+        recommendations_created=rapport.recommandations_creees,
+        already_present=rapport.deja_presentes,
+    )

apps/backend/app/api/v1/endpoints/sites.py

@@ -3,7 +3,7 @@ from fastapi import APIRouter, HTTPException, status
 from app.api.deps import LecteurDep, SiteServiceDep
 from app.api.openapi import REPONSE_VALIDATION, Reponses
 from app.schemas.errors import ErrorResponse
-from app.schemas.site import SiteResponse
+from app.schemas.site import SiteCurrentResponse, SiteResponse
 from app.services.site import SiteNotFoundError
 
 router = APIRouter()
@@ -34,3 +34,19 @@ async def get_site(site_id: str, _: LecteurDep, service: SiteServiceDep) -> Site
             status_code=status.HTTP_404_NOT_FOUND, detail="Site introuvable"
         ) from erreur
     return SiteResponse.model_validate(site)
+
+
+@router.get(
+    "/{site_id}/current",
+    response_model=SiteCurrentResponse,
+    summary="Dernière mesure d'un site",
+    responses=REPONSES_INTROUVABLE,
+)
+async def get_current(site_id: str, _: LecteurDep, service: SiteServiceDep) -> SiteCurrentResponse:
+    try:
+        actuel = await service.current(site_id)
+    except SiteNotFoundError as erreur:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND, detail="Site introuvable"
+        ) from erreur
+    return SiteCurrentResponse.model_validate(actuel)

apps/backend/app/api/v1/router.py

@@ -1,7 +1,18 @@
 from fastapi import APIRouter
 
 from app.api.openapi import REPONSE_SERVEUR, REPONSES_ADMIN, REPONSES_LECTEUR
-from app.api.v1.endpoints import alerts, auth, health, recommendations, sensors, sites, stats, users
+from app.api.v1.endpoints import (
+    alerts,
+    auth,
+    health,
+    predictions,
+    readings,
+    recommendations,
+    sensors,
+    sites,
+    stats,
+    users,
+)
 
 api_router = APIRouter(responses=REPONSE_SERVEUR)
 api_router.include_router(health.router, prefix="/health", tags=["health"])
@@ -18,6 +29,12 @@ api_router.include_router(
     responses=REPONSES_LECTEUR,
 )
 api_router.include_router(stats.router, prefix="/stats", tags=["stats"], responses=REPONSES_LECTEUR)
+api_router.include_router(
+    readings.router, prefix="/readings", tags=["readings"], responses=REPONSES_LECTEUR
+)
 api_router.include_router(
     sensors.router, prefix="/sensors", tags=["sensors"], responses=REPONSES_ADMIN
 )
+api_router.include_router(
+    predictions.router, prefix="/predictions", tags=["predictions"], responses=REPONSES_LECTEUR
+)

apps/backend/app/cli.py

@@ -9,6 +9,7 @@ import argparse
 import asyncio
 import json
 import secrets
+import string
 import sys
 from getpass import getpass
 from pathlib import Path
@@ -21,10 +22,13 @@ from app.core.hashing import build_hasher
 from app.core.roles import Role
 from app.db.session import get_session_factory
 from app.main import create_app
+from app.repositories.alert import AlertRepository
+from app.repositories.recommendation import RecommendationRepository
 from app.repositories.user import UserRepository
+from app.schemas.auth import PASSWORD_MIN_LENGTH, SPECIAL_CHARACTERS, valide_complexite
+from app.services.recommendation import RecommendationService
 
 LONGUEUR_MOT_DE_PASSE_GENERE = 24
-LONGUEUR_MINIMALE = 12
 CHEMIN_CONTRAT = Path(__file__).resolve().parent.parent / "openapi.json"
 
 
@@ -62,6 +66,22 @@ async def create_admin(
     )
 
 
+async def generate_recommendations(*, site_id: str | None) -> str:
+    async with get_session_factory()() as session:
+        service = RecommendationService(
+            recommendations=RecommendationRepository(session),
+            alerts=AlertRepository(session),
+            transaction=session,
+        )
+        rapport = await service.generate(site_id=site_id)
+
+    return (
+        f"{rapport.alertes_examinees} alerte(s) examinée(s), "
+        f"{rapport.recommandations_creees} recommandation(s) créée(s), "
+        f"{rapport.deja_presentes} déjà présente(s)"
+    )
+
+
 # Piège : le schéma ne doit dépendre ni du `.env` du poste ni des variables `APP_*`, sinon le
 # fichier versionné changerait de machine en machine et le test de dérive deviendrait un oracle
 # de configuration locale. Tout ce qui atteint le schéma est donc posé ici, `_env_file` compris.
@@ -108,18 +128,45 @@ def build_parser() -> argparse.ArgumentParser:
         "export-openapi", help="Écrit le contrat OpenAPI sur disque"
     )
     contrat.add_argument("--output", default=str(CHEMIN_CONTRAT))
+
+    recommandations = sous_commandes.add_parser(
+        "generate-recommendations",
+        help="Applique le moteur de règles aux alertes en base",
+    )
+    recommandations.add_argument(
+        "--site-id", default=None, help="Limite le traitement aux alertes d'un site"
+    )
     return parser
 
 
+def genere_mot_de_passe() -> str:
+    tirage = secrets.SystemRandom()
+    classes = [
+        string.ascii_uppercase,
+        string.ascii_lowercase,
+        string.digits,
+        SPECIAL_CHARACTERS,
+    ]
+    reste = LONGUEUR_MOT_DE_PASSE_GENERE - len(classes)
+    caracteres = [tirage.choice(classe) for classe in classes]
+    caracteres += [tirage.choice("".join(classes)) for _ in range(reste)]
+    tirage.shuffle(caracteres)
+    return "".join(caracteres)
+
+
 def read_password(*, generate: bool) -> str:
     if generate:
-        mot_de_passe = secrets.token_urlsafe(LONGUEUR_MOT_DE_PASSE_GENERE)
+        mot_de_passe = genere_mot_de_passe()
         print(f"Mot de passe généré, il ne sera plus affiché : {mot_de_passe}")
         return mot_de_passe
 
     mot_de_passe = getpass("Mot de passe : ")
-    if len(mot_de_passe) < LONGUEUR_MINIMALE:
-        raise SystemExit(f"Le mot de passe doit faire au moins {LONGUEUR_MINIMALE} caractères")
+    if len(mot_de_passe) < PASSWORD_MIN_LENGTH:
+        raise SystemExit(f"Le mot de passe doit faire au moins {PASSWORD_MIN_LENGTH} caractères")
+    try:
+        valide_complexite(mot_de_passe)
+    except ValueError as erreur:
+        raise SystemExit(str(erreur)) from erreur
     if mot_de_passe != getpass("Confirmation : "):
         raise SystemExit("Les deux saisies diffèrent")
     return mot_de_passe
@@ -132,6 +179,10 @@ def main(argv: list[str] | None = None) -> int:
         print(export_openapi(Path(arguments.output)))
         return 0
 
+    if arguments.commande == "generate-recommendations":
+        print(asyncio.run(generate_recommendations(site_id=arguments.site_id)))
+        return 0
+
     mot_de_passe = read_password(generate=arguments.generate)
 
     succes, message = asyncio.run(

apps/backend/app/core/config.py

@@ -34,6 +34,11 @@ class Settings(BaseSettings):
     database_pool_size: int = 5
     database_max_overflow: int = 10
 
+    mock_api_base_url: str = "https://api-mock.charlieandre.fr"
+    mock_api_username: str | None = None
+    mock_api_password: SecretStr | None = None
+    mock_api_timeout_seconds: float = Field(default=10.0, gt=0)
+
     jwt_issuer: str = "enervision-api"
     jwt_audience: str = "enervision-web"
     access_token_ttl_seconds: int = Field(default=900, ge=60, le=3600)
@@ -54,6 +59,19 @@ class Settings(BaseSettings):
     login_max_failures_per_ip: int = Field(default=20, ge=1)
     login_max_failures_per_identifier: int = Field(default=50, ge=1)
 
+    password_reset_ttl_seconds: int = Field(default=900, ge=60, le=3600)
+    password_reset_window_seconds: int = Field(default=900, ge=60)
+    password_reset_max_requests_per_identifier: int = Field(default=3, ge=1)
+    password_reset_max_requests_per_ip: int = Field(default=10, ge=1)
+
+    smtp_host: str = "localhost"
+    smtp_port: int = Field(default=587, ge=1, le=65535)
+    smtp_username: str | None = None
+    smtp_password: SecretStr | None = None
+    smtp_use_tls: bool = False
+    smtp_from_address: str = "no-reply@enervision.fr"
+    frontend_reset_password_url: str = "http://localhost:4200/reset-password"  # noqa: S105
+
     trust_proxy_headers: bool = False
     expose_api_docs: bool | None = None
     metrics_token: SecretStr | None = None

apps/backend/app/core/mailer.py

@@ -0,0 +1,48 @@
+# Piège : l'URL de réinitialisation porte le jeton en clair. Ne jamais la journaliser :
+# `send_password_reset_email()` ne logue que le destinataire, jamais `reset_url`.
+
+from dataclasses import dataclass
+from email.message import EmailMessage
+
+import aiosmtplib
+
+from app.core.logging import get_logger
+
+logger = get_logger(__name__)
+
+
+@dataclass(frozen=True, slots=True)
+class SmtpConfig:
+    host: str
+    port: int
+    username: str | None
+    password: str | None
+    use_tls: bool
+    from_address: str
+
+
+class Mailer:
+    def __init__(self, config: SmtpConfig) -> None:
+        self._config = config
+
+    async def send_password_reset_email(self, *, to: str, reset_url: str) -> None:
+        message = EmailMessage()
+        message["From"] = self._config.from_address
+        message["To"] = to
+        message["Subject"] = "Réinitialisation de votre mot de passe EnerVision"
+        message.set_content(
+            "Une réinitialisation de mot de passe a été demandée pour ce compte.\n\n"
+            f"Ouvrez ce lien dans les 15 minutes pour choisir un nouveau mot de passe : "
+            f"{reset_url}\n\n"
+            "Si vous n'êtes pas à l'origine de cette demande, ignorez cet email."
+        )
+
+        _, message_recu = await aiosmtplib.send(
+            message,
+            hostname=self._config.host,
+            port=self._config.port,
+            username=self._config.username,
+            password=self._config.password,
+            use_tls=self._config.use_tls,
+        )
+        logger.info("mailer.password_reset_sent to=%s smtp_response=%s", to, message_recu)

apps/backend/app/detection/internal_alerts.py

@@ -0,0 +1,68 @@
+# Détection d'alertes internes EnerVision (issue #104) : script lancé à la main pour l'instant,
+# comme `enervision_ml.score` côté ML, sans automatisation Airflow pour l'ordonnancer.
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import sys
+from datetime import UTC, datetime
+
+from app.core.config import get_settings
+from app.db.session import get_session_factory
+from app.repositories.alert import AlertRepository
+from app.repositories.prediction import PredictionRepository
+from app.repositories.reading import ReadingRepository
+from app.repositories.site import SiteRepository
+from app.services.alert import AlertService
+
+
+async def run_detection(*, now: datetime | None = None, site_id: str | None = None) -> int:
+    """Exécute les cinq règles de détection et enregistre les nouvelles alertes. Rend le nombre de
+    lignes effectivement insérées (les doublons de `source_alert_id` sont silencieusement
+    ignorés)."""
+    async with get_session_factory()() as session:
+        service = AlertService(
+            alerts=AlertRepository(session),
+            readings=ReadingRepository(session),
+            predictions=PredictionRepository(session),
+            sites=SiteRepository(session),
+        )
+        nouvelles = await service.detect(now=now, site_id=site_id)
+        await session.commit()
+    return len(nouvelles)
+
+
+def _parse_instant(valeur: str) -> datetime:
+    instant = datetime.fromisoformat(valeur)
+    return instant if instant.tzinfo is not None else instant.replace(tzinfo=UTC)
+
+
+def parse_args(argv: list[str] | None = None) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        prog="python -m app.detection.internal_alerts",
+        description="Détection d'alertes internes EnerVision",
+    )
+    parser.add_argument("--site-id", default=None, help="Limite la détection à un seul site.")
+    parser.add_argument(
+        "--now",
+        type=_parse_instant,
+        default=None,
+        help=(
+            "Instant de référence (ISO 8601, UTC si le fuseau est omis). Défaut : l'heure courante."
+        ),
+    )
+    return parser.parse_args(argv)
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = parse_args(argv)
+    # Échoue tôt si `APP_SECRET_KEY`/`DATABASE_URL` manquent, avant toute requête à la base.
+    get_settings()
+    nombre = asyncio.run(run_detection(now=args.now, site_id=args.site_id))
+    print(f"{nombre} nouvelle(s) alerte(s) enregistrée(s).")
+    return 0
+
+
+if __name__ == "__main__":  # pragma: no cover
+    sys.exit(main())

apps/backend/app/etl/historical_import.py

@@ -0,0 +1,621 @@
+from __future__ import annotations
+
+import argparse
+import asyncio
+import hashlib
+import json
+from pathlib import Path
+from typing import Any, cast
+
+import pandas as pd
+from sqlalchemy import text
+from sqlalchemy.ext.asyncio import AsyncConnection, create_async_engine
+
+from app.core.config import get_settings
+
+REQUIRED_COLUMNS = {
+    "timestamp",
+    "site_id",
+    "site_type",
+    "site_name",
+    "consumption_kwh",
+    "consumption_euros",
+    "temperature_celsius",
+    "humidity_percent",
+    "solar_irradiance_wm2",
+    "hour",
+    "day_of_week",
+    "day_name",
+    "month",
+    "is_weekend",
+    "is_working_hours",
+}
+
+MEASURE_COLUMNS = [
+    "consumption_kwh",
+    "consumption_euros",
+    "temperature_celsius",
+    "humidity_percent",
+    "solar_irradiance_wm2",
+]
+
+SOURCE_NAME = "csv"
+
+
+def compute_sha256(path: Path) -> str:
+    """Calcule l'empreinte SHA-256 du fichier source."""
+    sha256 = hashlib.sha256()
+
+    with path.open("rb") as source:
+        for block in iter(lambda: source.read(1024 * 1024), b""):
+            sha256.update(block)
+
+    return sha256.hexdigest()
+
+
+def load_metadata(path: Path) -> dict[str, Any]:
+    """Charge les métadonnées fournies avec le dataset."""
+    with path.open("r", encoding="utf-8") as source:
+        metadata = json.load(source)
+
+    if not isinstance(metadata, dict):
+        raise ValueError("Le fichier de métadonnées doit contenir un objet JSON.")
+
+    return cast(dict[str, Any], metadata)
+
+
+def classify_quality(
+    row: dict[str, Any],
+) -> tuple[str, list[str]]:
+    """
+    Déduit une qualité technique à partir des champs manquants.
+
+    Les valeurs NULL sont conservées. On ne cherche pas ici à
+    déterminer la cause physique exacte de leur absence.
+    """
+    missing = [column for column in MEASURE_COLUMNS if pd.isna(row.get(column))]
+
+    if not missing:
+        quality = "good"
+    elif len(missing) == len(MEASURE_COLUMNS):
+        quality = "critical"
+    elif "consumption_kwh" in missing:
+        quality = "degraded"
+    else:
+        quality = "partial"
+
+    reasons = [f"missing:{column}" for column in missing]
+
+    return quality, reasons
+
+
+def validate_source(
+    frame: pd.DataFrame,
+    metadata: dict[str, Any],
+) -> None:
+    """Valide le dataset avant tout chargement en base."""
+    missing_columns = REQUIRED_COLUMNS.difference(frame.columns)
+
+    if missing_columns:
+        raise ValueError(f"Colonnes obligatoires absentes : {sorted(missing_columns)}")
+
+    expected_records = int(metadata["total_records"])
+
+    if len(frame) != expected_records:
+        raise ValueError(f"Nombre de lignes inattendu : {len(frame)} au lieu de {expected_records}")
+
+    expected_sites = set(metadata["sites"].keys())
+    actual_sites = set(frame["site_id"].unique())
+
+    if actual_sites != expected_sites:
+        raise ValueError(
+            f"Sites incohérents. Attendus={sorted(expected_sites)}, trouvés={sorted(actual_sites)}"
+        )
+
+    duplicated = frame.duplicated(subset=["site_id", "timestamp"]).sum()
+
+    if duplicated:
+        raise ValueError(f"{duplicated} doublons (site_id, timestamp) détectés")
+
+    static_variants = frame.groupby("site_id")[["site_type", "site_name"]].nunique()
+
+    if (static_variants > 1).any().any():
+        raise ValueError("Un site possède plusieurs valeurs de site_type ou site_name.")
+
+    # Vérifie également que tous les timestamps
+    # peuvent être interprétés correctement.
+    pd.to_datetime(
+        frame["timestamp"],
+        errors="raise",
+    )
+
+
+def normalize_timestamps(
+    frame: pd.DataFrame,
+    source_timezone: str,
+) -> pd.DataFrame:
+    """
+    Normalise les timestamps et leur associe une timezone.
+
+    Les timestamps originaux sont conservés dans une colonne
+    temporaire afin de pouvoir les stocker dans raw_data.
+    """
+    normalized = frame.copy()
+
+    normalized["_source_timestamp"] = normalized["timestamp"]
+
+    timestamps = pd.to_datetime(
+        normalized["timestamp"],
+        errors="raise",
+    )
+
+    if timestamps.dt.tz is None:
+        timestamps = timestamps.dt.tz_localize(source_timezone)
+    else:
+        timestamps = timestamps.dt.tz_convert(source_timezone)
+
+    normalized["timestamp"] = timestamps
+
+    return normalized
+
+
+def to_json_value(value: Any) -> Any:
+    """
+    Convertit une valeur Pandas/Numpy en valeur
+    compatible JSON.
+    """
+    if value is None:
+        return None
+
+    try:
+        if pd.isna(value):
+            return None
+    except TypeError, ValueError:
+        pass
+
+    if isinstance(value, pd.Timestamp):
+        return value.isoformat()
+
+    if hasattr(value, "item"):
+        return value.item()
+
+    return value
+
+
+async def ensure_dataset(
+    connection: AsyncConnection,
+    metadata: dict[str, Any],
+    sha256: str,
+    source_timezone: str,
+    storage_uri: str,
+) -> int:
+    """
+    Crée l'entrée dataset si elle n'existe pas.
+
+    Le SHA-256 permet de reconnaître un fichier déjà importé
+    et participe à l'idempotence et à la traçabilité.
+    """
+    result = await connection.execute(
+        text(
+            """
+            SELECT dataset_id
+            FROM dataset
+            WHERE archive_sha256 = :sha256
+            LIMIT 1
+            """
+        ),
+        {
+            "sha256": sha256,
+        },
+    )
+
+    existing = result.scalar_one_or_none()
+
+    if existing is not None:
+        return int(existing)
+
+    metadata_summary = {
+        "generator_version": metadata.get("generator_version"),
+        "total_sites": metadata.get("total_sites"),
+        "total_records": metadata.get("total_records"),
+        "date_range": metadata.get("date_range"),
+        "frequency": metadata.get("frequency"),
+        "null_injection_enabled": metadata.get("null_injection_enabled"),
+        "null_strategies": metadata.get("null_strategies"),
+        "importer": "historical_import_v1",
+    }
+
+    result = await connection.execute(
+        text(
+            """
+            INSERT INTO dataset (
+                dataset_name,
+                archive_sha256,
+                storage_uri,
+                source_timezone,
+                "metadata"
+            )
+            VALUES (
+                :dataset_name,
+                :archive_sha256,
+                :storage_uri,
+                :source_timezone,
+                CAST(:metadata AS jsonb)
+            )
+            RETURNING dataset_id
+            """
+        ),
+        {
+            "dataset_name": ("EnerVision historical dataset 2023-2024"),
+            "archive_sha256": sha256,
+            "storage_uri": storage_uri,
+            "source_timezone": source_timezone,
+            "metadata": json.dumps(
+                metadata_summary,
+                ensure_ascii=False,
+            ),
+        },
+    )
+
+    return int(result.scalar_one())
+
+
+async def upsert_sites(
+    connection: AsyncConnection,
+    frame: pd.DataFrame,
+) -> None:
+    """Insère ou met à jour les sites du dataset."""
+    sites = cast(
+        list[dict[str, Any]],
+        frame[
+            [
+                "site_id",
+                "site_type",
+                "site_name",
+            ]
+        ]
+        .drop_duplicates(subset=["site_id"])
+        .to_dict(orient="records"),
+    )
+
+    await connection.execute(
+        text(
+            """
+            INSERT INTO site (
+                site_id,
+                site_type,
+                site_name
+            )
+            VALUES (
+                :site_id,
+                :site_type,
+                :site_name
+            )
+            ON CONFLICT (site_id)
+            DO UPDATE SET
+                site_type = EXCLUDED.site_type,
+                site_name = EXCLUDED.site_name
+            """
+        ),
+        sites,
+    )
+
+
+def build_reading_batch(
+    chunk: pd.DataFrame,
+    dataset_id: int,
+) -> list[dict[str, Any]]:
+    """
+    Transforme un chunk Pandas en lignes prêtes
+    à être chargées dans la table reading.
+    """
+    rows: list[dict[str, Any]] = []
+
+    records = cast(
+        list[dict[str, Any]],
+        chunk.to_dict(orient="records"),
+    )
+
+    for record in records:
+        quality, reasons = classify_quality(record)
+
+        raw_data = {
+            column: to_json_value(value)
+            for column, value in record.items()
+            if column != "_source_timestamp"
+        }
+
+        # Dans raw_data, on conserve le timestamp
+        # exactement tel qu'il était dans le CSV.
+        raw_data["timestamp"] = to_json_value(record["_source_timestamp"])
+
+        rows.append(
+            {
+                "site_id": record["site_id"],
+                "timestamp": record["timestamp"],
+                "source": SOURCE_NAME,
+                "dataset_id": dataset_id,
+                # Non fourni par le dataset historique.
+                "consumption_kw": None,
+                "consumption_kwh": to_json_value(record["consumption_kwh"]),
+                "consumption_euros": to_json_value(record["consumption_euros"]),
+                # Non fournis par le CSV historique.
+                "voltage_v": None,
+                "current_a": None,
+                "power_factor": None,
+                "temperature_celsius": (to_json_value(record["temperature_celsius"])),
+                "humidity_percent": (to_json_value(record["humidity_percent"])),
+                "solar_irradiance_wm2": (to_json_value(record["solar_irradiance_wm2"])),
+                "is_working_hours": bool(record["is_working_hours"]),
+                "data_quality": quality,
+                "null_reasons": reasons,
+                # Aucune imputation pendant l'ingestion RAW.
+                # Les valeurs manquantes sont conservées telles quelles
+                # afin de préserver la donnée source.
+                "imputed_values": None,
+                "imputation_method": None,
+                # Conservation de la donnée source
+                # pour la traçabilité.
+                "raw_data": json.dumps(
+                    raw_data,
+                    ensure_ascii=False,
+                ),
+            }
+        )
+
+    return rows
+
+
+READING_INSERT = text(
+    """
+    INSERT INTO reading (
+        site_id,
+        timestamp,
+        source,
+        dataset_id,
+        consumption_kw,
+        consumption_kwh,
+        consumption_euros,
+        voltage_v,
+        current_a,
+        power_factor,
+        temperature_celsius,
+        humidity_percent,
+        solar_irradiance_wm2,
+        is_working_hours,
+        data_quality,
+        null_reasons,
+        imputed_values,
+        imputation_method,
+        raw_data
+    )
+    VALUES (
+        :site_id,
+        :timestamp,
+        :source,
+        :dataset_id,
+        :consumption_kw,
+        :consumption_kwh,
+        :consumption_euros,
+        :voltage_v,
+        :current_a,
+        :power_factor,
+        :temperature_celsius,
+        :humidity_percent,
+        :solar_irradiance_wm2,
+        :is_working_hours,
+        :data_quality,
+        :null_reasons,
+        CAST(:imputed_values AS jsonb),
+        :imputation_method,
+        CAST(:raw_data AS jsonb)
+    )
+    ON CONFLICT DO NOTHING
+    """
+)
+
+
+async def import_historical(
+    csv_path: Path,
+    metadata_path: Path,
+    source_timezone: str,
+    batch_size: int,
+    dry_run: bool,
+    storage_uri: str,
+) -> None:
+    """
+    Exécute le pipeline ETL historique EnerVision.
+
+    Étapes :
+    1. Extract
+    2. Validate
+    3. Transform
+    4. Load
+    """
+    metadata = load_metadata(metadata_path)
+
+    frame = pd.read_csv(csv_path)
+
+    validate_source(
+        frame,
+        metadata,
+    )
+
+    print(f"Lignes             : {len(frame)}")
+    print(f"Sites              : {frame['site_id'].nunique()}")
+    print(f"Période            : {frame['timestamp'].min()} -> {frame['timestamp'].max()}")
+    print(f"Doublons           : {frame.duplicated(['site_id', 'timestamp']).sum()}")
+
+    print("\nValeurs NULL :")
+    print(frame[MEASURE_COLUMNS].isna().sum())
+
+    sha256 = compute_sha256(csv_path)
+
+    print(f"\nSHA-256            : {sha256}")
+
+    if dry_run:
+        print("\nDry-run terminé : aucune donnée écrite.")
+        return
+
+    normalized = normalize_timestamps(
+        frame,
+        source_timezone,
+    )
+
+    settings = get_settings()
+
+    engine = create_async_engine(
+        str(settings.database_url),
+        pool_pre_ping=True,
+    )
+
+    try:
+        async with engine.begin() as connection:
+            dataset_id = await ensure_dataset(
+                connection=connection,
+                metadata=metadata,
+                sha256=sha256,
+                source_timezone=source_timezone,
+                storage_uri=storage_uri,
+            )
+
+            await upsert_sites(
+                connection,
+                normalized,
+            )
+
+            result = await connection.execute(
+                text(
+                    """
+                    SELECT COUNT(*)
+                    FROM reading
+                    WHERE dataset_id = :dataset_id
+                      AND source = :source
+                    """
+                ),
+                {
+                    "dataset_id": dataset_id,
+                    "source": SOURCE_NAME,
+                },
+            )
+
+            before = int(result.scalar_one())
+
+            for start in range(
+                0,
+                len(normalized),
+                batch_size,
+            ):
+                chunk = normalized.iloc[start : start + batch_size]
+
+                rows = build_reading_batch(
+                    chunk,
+                    dataset_id,
+                )
+
+                await connection.execute(
+                    READING_INSERT,
+                    rows,
+                )
+
+                loaded = min(
+                    start + batch_size,
+                    len(normalized),
+                )
+
+                print(f"Chargement : {loaded}/{len(normalized)}")
+
+            result = await connection.execute(
+                text(
+                    """
+                    SELECT COUNT(*)
+                    FROM reading
+                    WHERE dataset_id = :dataset_id
+                      AND source = :source
+                    """
+                ),
+                {
+                    "dataset_id": dataset_id,
+                    "source": SOURCE_NAME,
+                },
+            )
+
+            after = int(result.scalar_one())
+
+            print("\nImport terminé.")
+            print(f"dataset_id         : {dataset_id}")
+            print(f"lectures avant     : {before}")
+            print(f"lectures après     : {after}")
+            print(f"nouvelles lectures : {after - before}")
+
+    finally:
+        await engine.dispose()
+
+
+def parse_args() -> argparse.Namespace:
+    """Définit les arguments CLI de l'import."""
+    parser = argparse.ArgumentParser(description=("Import historique EnerVision"))
+
+    parser.add_argument(
+        "--csv",
+        type=Path,
+        required=True,
+        help="Chemin vers le CSV historique.",
+    )
+
+    parser.add_argument(
+        "--metadata",
+        type=Path,
+        required=True,
+        help=("Chemin vers le fichier dataset_metadata.json."),
+    )
+
+    parser.add_argument(
+        "--source-timezone",
+        default="UTC",
+        help=("Timezone associée aux timestamps du dataset. Défaut : UTC."),
+    )
+
+    parser.add_argument(
+        "--batch-size",
+        type=int,
+        default=1000,
+        help=("Nombre de lignes insérées par batch. Défaut : 1000."),
+    )
+
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help=("Valide les données sans écrire en base."),
+    )
+
+    return parser.parse_args()
+
+
+def main() -> None:
+    """Point d'entrée CLI du pipeline."""
+    args = parse_args()
+
+    if args.batch_size <= 0:
+        raise ValueError("--batch-size doit être strictement supérieur à 0.")
+
+    # resolve() est volontairement exécuté ici,
+    # dans la partie synchrone du programme.
+    # Cela évite une opération filesystem bloquante
+    # à l'intérieur d'une fonction async.
+    storage_uri = args.csv.resolve().as_uri()
+
+    asyncio.run(
+        import_historical(
+            csv_path=args.csv,
+            metadata_path=args.metadata,
+            source_timezone=(args.source_timezone),
+            batch_size=args.batch_size,
+            dry_run=args.dry_run,
+            storage_uri=storage_uri,
+        )
+    )
+
+
+if __name__ == "__main__":
+    main()

apps/backend/app/etl/mock_api_import.py

@@ -0,0 +1,416 @@
+# Contrainte : la réponse de l'API Mock est une entrée hostile, pas une source de confiance.
+# Voir OWASP API10 dans docs/architecture/owasp-traceabilite.md. Rien de ce qu'elle renvoie
+# n'atteint la base sans passer par build_site_row() ou build_reading_row() : seuls les champs
+# attendus sont recopiés, les grandeurs physiques sont bornées par PHYSICAL_BOUNDS et la taille
+# des tableaux est plafonnée par MAX_SITES et par --limit. Une valeur hors bornes devient NULL
+# et laisse sa trace dans null_reasons plutôt que de lever : le mock émet des anomalies par
+# construction, et raw_data conserve de toute façon la réponse d'origine intacte.
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import json
+from datetime import datetime
+from typing import Any
+
+import httpx
+from sqlalchemy import text
+from sqlalchemy.ext.asyncio import AsyncConnection, create_async_engine
+
+from app.core.config import get_settings
+
+SOURCE_HISTORY = "api_history"
+
+MAX_SITES = 100
+
+MAX_LIMIT = 1000
+
+# Les quatre seules valeurs que la contrainte ck_reading_quality accepte.
+ACCEPTED_QUALITIES = frozenset({"good", "partial", "degraded", "critical"})
+
+PHYSICAL_BOUNDS: dict[str, tuple[float, float]] = {
+    "consumption_kw": (0.0, 100_000.0),
+    "consumption_kwh": (0.0, 100_000.0),
+    "voltage_v": (0.0, 1_000.0),
+    "current_a": (0.0, 10_000.0),
+    "power_factor": (0.0, 1.0),
+    "temperature_celsius": (-90.0, 60.0),
+    "humidity_percent": (0.0, 100.0),
+}
+
+CAPACITY_BOUNDS = (0.0, 100_000.0)
+
+
+def create_mock_api_client() -> httpx.AsyncClient:
+    settings = get_settings()
+
+    if settings.mock_api_username is None or settings.mock_api_password is None:
+        raise ValueError("Les identifiants de l'API Mock ne sont pas configurés.")
+
+    return httpx.AsyncClient(
+        base_url=settings.mock_api_base_url.rstrip("/"),
+        auth=(
+            settings.mock_api_username,
+            settings.mock_api_password.get_secret_value(),
+        ),
+        timeout=settings.mock_api_timeout_seconds,
+    )
+
+
+def read_text(payload: dict[str, Any], key: str) -> str:
+    value = payload.get(key)
+
+    if not isinstance(value, str) or not value:
+        raise ValueError(f"Champ {key} absent ou invalide dans la réponse de l'API Mock.")
+
+    return value
+
+
+def optional_text(value: Any) -> str | None:
+    return value if isinstance(value, str) else None
+
+
+def coerce_measure(
+    value: Any,
+    bounds: tuple[float, float],
+) -> float | None:
+    if isinstance(value, bool) or not isinstance(value, int | float):
+        return None
+
+    lower, upper = bounds
+
+    # Écarte aussi NaN et les infinis, qu'aucune comparaison de bornes ne retient.
+    return float(value) if lower <= value <= upper else None
+
+
+def resolve_quality(
+    value: Any,
+    rejected: list[str],
+) -> str | None:
+    quality = value if isinstance(value, str) and value in ACCEPTED_QUALITIES else None
+
+    if rejected:
+        return "critical" if quality == "critical" else "degraded"
+
+    return quality
+
+
+def resolve_null_reasons(
+    value: Any,
+    rejected: list[str],
+) -> list[str]:
+    reported = [str(reason) for reason in value] if isinstance(value, list) else []
+
+    return reported + rejected
+
+
+async def fetch_sites(
+    client: httpx.AsyncClient,
+) -> list[dict[str, Any]]:
+    response = await client.get("/api/v1/sites")
+
+    response.raise_for_status()
+
+    payload = response.json()
+
+    if not isinstance(payload, list):
+        raise ValueError("La réponse /api/v1/sites doit être une liste.")
+
+    if len(payload) > MAX_SITES:
+        raise ValueError(f"La réponse /api/v1/sites dépasse le plafond de {MAX_SITES} sites.")
+
+    return payload
+
+
+def build_site_row(
+    site: dict[str, Any],
+) -> dict[str, Any]:
+    return {
+        "site_id": read_text(site, "site_id"),
+        "site_type": read_text(site, "site_type"),
+        "site_name": read_text(site, "site_name"),
+        "location": optional_text(site.get("location")),
+        "capacity_kw": coerce_measure(site.get("capacity_kw"), CAPACITY_BOUNDS),
+        "status": optional_text(site.get("status")),
+    }
+
+
+async def upsert_sites(
+    connection: AsyncConnection,
+    sites: list[dict[str, Any]],
+) -> None:
+    rows = [build_site_row(site) for site in sites]
+
+    if not rows:
+        return
+
+    await connection.execute(
+        text(
+            """
+            INSERT INTO site (
+                site_id,
+                site_type,
+                site_name,
+                location,
+                capacity_kw,
+                status
+            )
+            VALUES (
+                :site_id,
+                :site_type,
+                :site_name,
+                :location,
+                :capacity_kw,
+                :status
+            )
+            ON CONFLICT (site_id)
+            DO UPDATE SET
+                site_type = EXCLUDED.site_type,
+                site_name = EXCLUDED.site_name,
+                location = EXCLUDED.location,
+                capacity_kw = EXCLUDED.capacity_kw,
+                status = EXCLUDED.status
+            """
+        ),
+        rows,
+    )
+
+
+async def fetch_readings(
+    client: httpx.AsyncClient,
+    site_id: str,
+    start_time: datetime,
+    end_time: datetime,
+    limit: int = MAX_LIMIT,
+) -> list[dict[str, Any]]:
+    response = await client.get(
+        "/api/v1/readings",
+        params={
+            "site_id": site_id,
+            "start_time": start_time.isoformat(),
+            "end_time": end_time.isoformat(),
+            "limit": limit,
+        },
+    )
+
+    response.raise_for_status()
+
+    payload = response.json()
+
+    if not isinstance(payload, list):
+        raise ValueError("La réponse /api/v1/readings doit être une liste.")
+
+    if len(payload) > limit:
+        raise ValueError(f"La réponse /api/v1/readings dépasse la limite demandée de {limit}.")
+
+    return payload
+
+
+def build_reading_row(
+    reading: dict[str, Any],
+) -> dict[str, Any]:
+    measures: dict[str, float | None] = {}
+    rejected: list[str] = []
+
+    for name, bounds in PHYSICAL_BOUNDS.items():
+        received = reading.get(name)
+        measures[name] = coerce_measure(received, bounds)
+
+        if received is not None and measures[name] is None:
+            rejected.append(f"out_of_physical_bounds:{name}")
+
+    return {
+        "site_id": read_text(reading, "site_id"),
+        "timestamp": parse_datetime(read_text(reading, "timestamp")),
+        "source": SOURCE_HISTORY,
+        "dataset_id": None,
+        **measures,
+        "consumption_euros": None,
+        "solar_irradiance_wm2": None,
+        "is_working_hours": None,
+        "data_quality": resolve_quality(reading.get("data_quality"), rejected),
+        "null_reasons": resolve_null_reasons(reading.get("null_reasons"), rejected),
+        "imputed_values": None,
+        "imputation_method": None,
+        "raw_data": json.dumps(
+            reading,
+            ensure_ascii=False,
+        ),
+    }
+
+
+# Le conflit vise l'index unique uq_reading_source plutôt que la table entière : sans cible
+# nommée, DO NOTHING avalerait aussi une violation de clé primaire.
+READING_INSERT = text(
+    """
+    INSERT INTO reading (
+        site_id,
+        timestamp,
+        source,
+        dataset_id,
+        consumption_kw,
+        consumption_kwh,
+        consumption_euros,
+        voltage_v,
+        current_a,
+        power_factor,
+        temperature_celsius,
+        humidity_percent,
+        solar_irradiance_wm2,
+        is_working_hours,
+        data_quality,
+        null_reasons,
+        imputed_values,
+        imputation_method,
+        raw_data
+    )
+    VALUES (
+        :site_id,
+        :timestamp,
+        :source,
+        :dataset_id,
+        :consumption_kw,
+        :consumption_kwh,
+        :consumption_euros,
+        :voltage_v,
+        :current_a,
+        :power_factor,
+        :temperature_celsius,
+        :humidity_percent,
+        :solar_irradiance_wm2,
+        :is_working_hours,
+        :data_quality,
+        :null_reasons,
+        CAST(:imputed_values AS jsonb),
+        :imputation_method,
+        CAST(:raw_data AS jsonb)
+    )
+    ON CONFLICT (site_id, timestamp, source, (coalesce(dataset_id, 0)))
+    DO NOTHING
+    """
+)
+
+
+def build_reading_batch(
+    readings: list[dict[str, Any]],
+) -> list[dict[str, Any]]:
+    return [build_reading_row(reading) for reading in readings]
+
+
+async def import_mock_api_history(
+    start_time: datetime,
+    end_time: datetime,
+    limit: int,
+    dry_run: bool,
+) -> None:
+    settings = get_settings()
+
+    async with create_mock_api_client() as client:
+        sites = await fetch_sites(client)
+
+        print(f"Sites récupérés : {len(sites)}")
+
+        all_readings: list[dict[str, Any]] = []
+
+        for site in sites:
+            site_id = read_text(site, "site_id")
+
+            readings = await fetch_readings(
+                client=client,
+                site_id=site_id,
+                start_time=start_time,
+                end_time=end_time,
+                limit=limit,
+            )
+
+            print(f"{site_id}: {len(readings)} lectures")
+
+            all_readings.extend(readings)
+
+    print(f"Lectures récupérées : {len(all_readings)}")
+
+    if dry_run:
+        print("Dry-run terminé : aucune donnée écrite.")
+        return
+
+    engine = create_async_engine(
+        str(settings.database_url),
+        pool_pre_ping=True,
+    )
+
+    try:
+        async with engine.begin() as connection:
+            await upsert_sites(
+                connection,
+                sites,
+            )
+
+            rows = build_reading_batch(all_readings)
+
+            if rows:
+                await connection.execute(
+                    READING_INSERT,
+                    rows,
+                )
+
+    finally:
+        await engine.dispose()
+
+    print("Import API Mock terminé.")
+
+
+def parse_datetime(value: str) -> datetime:
+    return datetime.fromisoformat(value.replace("Z", "+00:00"))
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=("Import historique depuis l'API Mock EnerVision"))
+
+    parser.add_argument(
+        "--start-time",
+        required=True,
+        type=parse_datetime,
+    )
+
+    parser.add_argument(
+        "--end-time",
+        required=True,
+        type=parse_datetime,
+    )
+
+    parser.add_argument(
+        "--limit",
+        type=int,
+        default=MAX_LIMIT,
+    )
+
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+    )
+
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    if args.limit < 1 or args.limit > MAX_LIMIT:
+        raise ValueError(f"--limit doit être compris entre 1 et {MAX_LIMIT}.")
+
+    if args.start_time >= args.end_time:
+        raise ValueError("--start-time doit être antérieur à --end-time.")
+
+    asyncio.run(
+        import_mock_api_history(
+            start_time=args.start_time,
+            end_time=args.end_time,
+            limit=args.limit,
+            dry_run=args.dry_run,
+        )
+    )
+
+
+if __name__ == "__main__":
+    main()

apps/backend/app/main.py

@@ -1,9 +1,14 @@
 from collections.abc import AsyncIterator
 from contextlib import asynccontextmanager
+from pathlib import Path
 
 from fastapi import Depends, FastAPI
 from fastapi.middleware.cors import CORSMiddleware
+from fastapi.openapi.docs import get_redoc_html, get_swagger_ui_html
+from fastapi.staticfiles import StaticFiles
 from prometheus_fastapi_instrumentator import Instrumentator
+from starlette.requests import Request
+from starlette.responses import HTMLResponse
 
 from app.api.errors import register_error_handlers
 from app.api.middleware import SecurityHeadersMiddleware
@@ -18,6 +23,8 @@ logger = get_logger(__name__)
 
 METHODES_AUTORISEES = ["GET", "POST", "PATCH", "PUT", "DELETE", "OPTIONS"]
 EN_TETES_AUTORISES = ["Authorization", "Content-Type"]
+STATIC_DIR = Path(__file__).parent / "static"
+LOGO_URL = "/static/logo-icon.png"
 
 
 @asynccontextmanager
@@ -43,11 +50,41 @@ def create_app(settings: Settings | None = None) -> FastAPI:
         openapi_tags=TAGS,
         debug=resolved.debug,
         lifespan=lifespan,
-        docs_url="/docs" if documentee else None,
-        redoc_url="/redoc" if documentee else None,
+        docs_url=None,
+        redoc_url=None,
         openapi_url="/openapi.json" if documentee else None,
     )
 
+    if documentee:
+        application.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
+
+        # ReDoc supporte nativement `info.x-logo` (extension Redocly) pour afficher un logo
+        # en en-tête ; Swagger UI n'a pas d'equivalent, il ne reprend que le favicon.
+        openapi_original = application.openapi
+
+        def openapi_avec_logo() -> dict[str, object]:
+            schema = openapi_original()
+            schema["info"]["x-logo"] = {"url": LOGO_URL, "altText": "EnerVision"}
+            return schema
+
+        application.openapi = openapi_avec_logo  # type: ignore[method-assign]
+
+        @application.get("/docs", include_in_schema=False)
+        async def docs_swagger(_: Request) -> HTMLResponse:
+            return get_swagger_ui_html(
+                openapi_url="/openapi.json",
+                title=f"{application.title} · Swagger UI",
+                swagger_favicon_url=LOGO_URL,
+            )
+
+        @application.get("/redoc", include_in_schema=False)
+        async def docs_redoc(_: Request) -> HTMLResponse:
+            return get_redoc_html(
+                openapi_url="/openapi.json",
+                title=f"{application.title} · ReDoc",
+                redoc_favicon_url=LOGO_URL,
+            )
+
     application.add_middleware(SecurityHeadersMiddleware)
 
     if resolved.allowed_origins:

apps/backend/app/models/__init__.py

@@ -4,6 +4,8 @@
 from app.models.audit_log import AuditLog
 from app.models.energy import Alert, Dataset, Prediction, Reading, Recommendation, Site
 from app.models.login_attempt import LoginAttempt
+from app.models.password_reset_attempt import PasswordResetAttempt
+from app.models.password_reset_token import PasswordResetToken
 from app.models.refresh_token import RefreshToken
 from app.models.user import AppUser
 
@@ -13,6 +15,8 @@ __all__ = [
     "AuditLog",
     "Dataset",
     "LoginAttempt",
+    "PasswordResetAttempt",
+    "PasswordResetToken",
     "Prediction",
     "Reading",
     "Recommendation",

apps/backend/app/models/audit_log.py

@@ -29,6 +29,8 @@ class AuditAction(StrEnum):
     COMPTE_ACTIVE = "user.enabled"
     COMPTE_MOT_DE_PASSE_REINITIALISE = "user.password_reset_by_admin"
     COMPTE_MOT_DE_PASSE_CHANGE = "user.password_changed"
+    MOT_DE_PASSE_OUBLIE_DEMANDE = "auth.password_reset_requested"
+    MOT_DE_PASSE_REINITIALISE_PAR_SOI = "auth.password_reset_self_service"
     REFRESH_REUTILISE = "auth.refresh_reuse_detected"
     SESSIONS_REVOQUEES = "auth.all_sessions_revoked"
     LIMITE_PAR_IDENTIFIANT = "auth.identifier_throttled"

apps/backend/app/models/password_reset_attempt.py

@@ -0,0 +1,27 @@
+# Pourquoi : même séparation que `login_attempt` par rapport à `audit_log` : ce compteur est
+# piloté par l'attaquant (une campagne de demandes) et se purge, l'audit log est en ajout seul.
+# Piège : la tentative est enregistrée même quand l'email est inconnu, sinon le 429 apprendrait
+# qu'un compte existe.
+
+from datetime import datetime
+
+from sqlalchemy import BigInteger, DateTime, Identity, Index, String, func
+from sqlalchemy.dialects.postgresql import INET
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.db.base import Base
+
+
+class PasswordResetAttempt(Base):
+    __tablename__ = "password_reset_attempt"
+    __table_args__ = (
+        Index("ix_password_reset_attempt_email_date", "email_tried", "occurred_at"),
+        Index("ix_password_reset_attempt_ip_date", "client_ip", "occurred_at"),
+    )
+
+    id: Mapped[int] = mapped_column(BigInteger, Identity(always=True), primary_key=True)
+    occurred_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )
+    email_tried: Mapped[str] = mapped_column(String(320), nullable=False)
+    client_ip: Mapped[str | None] = mapped_column(INET, nullable=True)

apps/backend/app/models/password_reset_token.py

@@ -0,0 +1,40 @@
+# Pourquoi : même schéma que `refresh_token` (chaîne opaque, jamais un JWT) pour la même
+# raison : un jeton de réinitialisation doit être révocable d'un coup, et un JWT ne figure
+# dans aucune ligne à invalider.
+
+import uuid
+from datetime import datetime
+
+from sqlalchemy import DateTime, ForeignKey, Index, LargeBinary, Text, func
+from sqlalchemy.dialects.postgresql import INET
+from sqlalchemy.dialects.postgresql import UUID as PG_UUID
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.db.base import Base
+
+
+class PasswordResetToken(Base):
+    __tablename__ = "password_reset_token"
+    __table_args__ = (
+        Index("ix_password_reset_token_user", "user_id"),
+        Index(
+            "ix_password_reset_token_vivants",
+            "user_id",
+            postgresql_where="consumed_at is null",
+        ),
+    )
+
+    id: Mapped[uuid.UUID] = mapped_column(
+        PG_UUID(as_uuid=True), primary_key=True, server_default=func.gen_random_uuid()
+    )
+    user_id: Mapped[uuid.UUID] = mapped_column(
+        PG_UUID(as_uuid=True), ForeignKey("app_user.id", ondelete="CASCADE"), nullable=False
+    )
+    token_hash: Mapped[bytes] = mapped_column(LargeBinary, nullable=False, unique=True)
+    issued_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )
+    expires_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False)
+    consumed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
+    client_ip: Mapped[str | None] = mapped_column(INET, nullable=True)
+    user_agent: Mapped[str | None] = mapped_column(Text, nullable=True)

apps/backend/app/repositories/alert.py

@@ -1,6 +1,7 @@
 from collections.abc import Sequence
 
 from sqlalchemy import select
+from sqlalchemy.dialects.postgresql import insert
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Alert
@@ -19,3 +20,36 @@ class AlertRepository:
         if severity is not None:
             requete = requete.where(Alert.severity == severity)
         return (await self._session.scalars(requete)).all()
+
+    async def create_many(self, alerts: Sequence[Alert]) -> Sequence[Alert]:
+        # `ON CONFLICT DO NOTHING` sur `uq_alert_source_reference` : rejouer la détection sur une
+        # fenêtre qui recouvre une exécution précédente ne doit pas dupliquer une alerte déjà
+        # enregistrée. `RETURNING` ne renvoie donc que les lignes effectivement insérées.
+        if not alerts:
+            return []
+        valeurs = [
+            {
+                "source_alert_id": alerte.source_alert_id,
+                "site_id": alerte.site_id,
+                "source": alerte.source,
+                "timestamp": alerte.timestamp,
+                "type": alerte.type,
+                "severity": alerte.severity,
+                "message": alerte.message,
+                "value": alerte.value,
+                "threshold": alerte.threshold,
+                "metric": alerte.metric,
+                "prediction_id": alerte.prediction_id,
+                "raw_data": alerte.raw_data,
+            }
+            for alerte in alerts
+        ]
+        requete = (
+            insert(Alert)
+            .values(valeurs)
+            .on_conflict_do_nothing(constraint="uq_alert_source_reference")
+            .returning(Alert)
+        )
+        resultat = await self._session.execute(requete)
+        await self._session.flush()
+        return resultat.scalars().all()

apps/backend/app/repositories/password_reset_attempt.py

@@ -0,0 +1,42 @@
+from dataclasses import dataclass
+from datetime import UTC, datetime, timedelta
+
+from sqlalchemy import func, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models.password_reset_attempt import PasswordResetAttempt
+
+
+@dataclass(frozen=True, slots=True)
+class ResetRequestCounts:
+    per_identifier: int
+    per_ip: int
+
+
+class PasswordResetAttemptRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def record(self, *, email: str, client_ip: str | None) -> None:
+        self._session.add(
+            PasswordResetAttempt(email_tried=email.strip().lower(), client_ip=client_ip)
+        )
+
+    async def count_recent(
+        self, *, email: str, client_ip: str | None, window_seconds: int
+    ) -> ResetRequestCounts:
+        identifiant = email.strip().lower()
+        meme_email = PasswordResetAttempt.email_tried == identifiant
+        meme_ip = PasswordResetAttempt.client_ip == client_ip
+
+        requete = select(
+            func.count().filter(meme_email),
+            func.count().filter(meme_ip),
+        ).where(
+            PasswordResetAttempt.occurred_at
+            > datetime.now(UTC) - timedelta(seconds=window_seconds),
+            meme_email | meme_ip,
+        )
+
+        par_identifiant, par_ip = (await self._session.execute(requete)).one()
+        return ResetRequestCounts(per_identifier=par_identifiant, per_ip=par_ip)

apps/backend/app/repositories/password_reset_token.py

@@ -0,0 +1,78 @@
+# Piège : `consume()` est une seule instruction, sur le modèle de `claim_for_rotation()` du
+# jeton de rafraîchissement. Un SELECT puis un UPDATE laisseraient une fenêtre où deux
+# soumissions concurrentes du même lien réussiraient toutes les deux.
+
+from dataclasses import dataclass
+from datetime import datetime
+from uuid import UUID
+
+from sqlalchemy import func, select, update
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models.password_reset_token import PasswordResetToken
+
+
+@dataclass(frozen=True, slots=True)
+class ConsumedResetToken:
+    id: UUID
+    user_id: UUID
+
+
+class PasswordResetTokenRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def create(
+        self,
+        *,
+        user_id: UUID,
+        token_hash: bytes,
+        expires_at: datetime,
+        client_ip: str | None,
+        user_agent: str | None,
+    ) -> PasswordResetToken:
+        jeton = PasswordResetToken(
+            user_id=user_id,
+            token_hash=token_hash,
+            expires_at=expires_at,
+            client_ip=client_ip,
+            user_agent=user_agent,
+        )
+        self._session.add(jeton)
+        await self._session.flush()
+        return jeton
+
+    async def consume(self, token_hash: bytes) -> ConsumedResetToken | None:
+        requete = (
+            update(PasswordResetToken)
+            .where(
+                PasswordResetToken.token_hash == token_hash,
+                PasswordResetToken.consumed_at.is_(None),
+                PasswordResetToken.expires_at > func.clock_timestamp(),
+            )
+            .values(consumed_at=func.clock_timestamp())
+            .returning(PasswordResetToken.id, PasswordResetToken.user_id)
+        )
+        ligne = (await self._session.execute(requete)).one_or_none()
+        if ligne is None:
+            return None
+        return ConsumedResetToken(id=ligne.id, user_id=ligne.user_id)
+
+    # Piège : simple SELECT, volontairement pas atomique avec la consommation. Sert seulement
+    # au feedback UX (jeton encore valide ?) ; `consume()` reste la seule source de vérité.
+    async def exists_valid(self, token_hash: bytes) -> bool:
+        requete = select(PasswordResetToken.id).where(
+            PasswordResetToken.token_hash == token_hash,
+            PasswordResetToken.consumed_at.is_(None),
+            PasswordResetToken.expires_at > func.clock_timestamp(),
+        )
+        return (await self._session.execute(requete)).first() is not None
+
+    async def invalidate_all_for_user(self, user_id: UUID) -> int:
+        resultat = await self._session.execute(
+            update(PasswordResetToken)
+            .where(PasswordResetToken.user_id == user_id, PasswordResetToken.consumed_at.is_(None))
+            .values(consumed_at=func.clock_timestamp())
+            .returning(PasswordResetToken.id)
+        )
+        return len(resultat.all())

apps/backend/app/repositories/prediction.py

@@ -0,0 +1,49 @@
+from collections.abc import Sequence
+from datetime import datetime
+
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models.energy import Prediction
+
+
+class PredictionRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def list_since(
+        self, *, since: datetime, site_id: str | None = None
+    ) -> Sequence[Prediction]:
+        # Restreint à `available` : une prévision `insufficient_data`/`error` n'a pas de
+        # `predicted_value` à comparer à une lecture réelle (détection d'anomalie).
+        # Piège : `prediction` n'a pas d'unicité sur `(site_id, target_at)` (cf.
+        # `enervision_ml.score`, qui insère toujours une nouvelle ligne plutôt que d'écraser la
+        # précédente). `prediction_id` en dernier départage donc les égalités de `target_at` par
+        # ordre croissant : `_detect_anomaly` construit un dict qui garde le dernier rencontré,
+        # c'est-à-dire le run le plus récent plutôt qu'une ligne choisie au hasard par le plan
+        # d'exécution.
+        requete = (
+            select(Prediction)
+            .where(Prediction.target_at >= since, Prediction.status == "available")
+            .order_by(Prediction.site_id, Prediction.target_at, Prediction.prediction_id)
+        )
+        if site_id is not None:
+            requete = requete.where(Prediction.site_id == site_id)
+        return (await self._session.scalars(requete)).all()
+
+    async def latest_by_site(self) -> Sequence[Prediction]:
+        # `.distinct(site_id)` compile en `DISTINCT ON (site_id)` sous PostgreSQL : une seule
+        # ligne par site, la plus récente grâce à l'ordre composite qui suit. Même mécanisme que
+        # `ReadingRepository.latest_by_site`. Trié sur `target_at` (couvert par
+        # `ix_prediction_site_target`) plutôt que `created_at` : c'est la prévision la plus
+        # récente qui compte pour un tableau de bord, pas forcément le dernier run de scoring.
+        requete = (
+            select(Prediction)
+            .distinct(Prediction.site_id)
+            .order_by(
+                Prediction.site_id,
+                Prediction.target_at.desc(),
+                Prediction.prediction_id.desc(),
+            )
+        )
+        return (await self._session.scalars(requete)).all()

apps/backend/app/repositories/reading.py

@@ -1,4 +1,5 @@
 from collections.abc import Sequence
+from datetime import datetime
 
 from sqlalchemy import select
 from sqlalchemy.ext.asyncio import AsyncSession
@@ -12,10 +13,58 @@ class ReadingRepository:
 
     async def latest_by_site(self) -> Sequence[Reading]:
         # `.distinct(site_id)` compile en `DISTINCT ON (site_id)` sous PostgreSQL : une seule
-        # ligne par site, la plus récente grâce à l'ordre composite qui suit.
+        # ligne par site, la plus récente grâce à l'ordre composite qui suit. `reading_id` départage
+        # les égalités de timestamp, que `uq_reading_source` autorise à `source` différente.
         requete = (
             select(Reading)
             .distinct(Reading.site_id)
-            .order_by(Reading.site_id, Reading.timestamp.desc())
+            .order_by(Reading.site_id, Reading.timestamp.desc(), Reading.reading_id.desc())
         )
         return (await self._session.execute(requete)).scalars().all()
+
+    async def latest_for_site(self, site_id: str) -> Reading | None:
+        # Piège : `uq_reading_source` autorise deux lignes au même `site_id`+`timestamp` quand la
+        # `source` diffère. Sans `reading_id` en départage, le `LIMIT 1` renverrait au hasard.
+        requete = (
+            select(Reading)
+            .where(Reading.site_id == site_id)
+            .order_by(Reading.timestamp.desc(), Reading.reading_id.desc())
+            .limit(1)
+        )
+        lecture: Reading | None = await self._session.scalar(requete)
+        return lecture
+
+    async def list_since(self, *, since: datetime, site_id: str | None = None) -> Sequence[Reading]:
+        # Trié par site puis par heure croissante : la détection d'alertes (spike) a besoin de
+        # comparer chaque lecture à celle qui la précède immédiatement pour le même site.
+        # `reading_id` en dernier départage : `uq_reading_source` autorise deux lignes au même
+        # `site_id`+`timestamp` quand la `source` diffère (même piège que `latest_for_site`), sans
+        # quoi l'ordre entre elles ne serait pas garanti d'un appel à l'autre.
+        requete = (
+            select(Reading)
+            .where(Reading.timestamp >= since)
+            .order_by(Reading.site_id, Reading.timestamp, Reading.reading_id)
+        )
+        if site_id is not None:
+            requete = requete.where(Reading.site_id == site_id)
+        return (await self._session.scalars(requete)).all()
+
+    async def list_history(
+        self,
+        *,
+        start: datetime,
+        end: datetime,
+        site_id: str | None = None,
+        limit: int,
+        offset: int,
+    ) -> Sequence[Reading]:
+        requete = (
+            select(Reading)
+            .where(Reading.timestamp >= start, Reading.timestamp < end)
+            .order_by(Reading.timestamp.desc(), Reading.reading_id.desc())
+            .limit(limit)
+            .offset(offset)
+        )
+        if site_id is not None:
+            requete = requete.where(Reading.site_id == site_id)
+        return (await self._session.scalars(requete)).all()

apps/backend/app/repositories/recommendation.py

@@ -1,11 +1,24 @@
 from collections.abc import Sequence
+from dataclasses import asdict, dataclass
 
 from sqlalchemy import select
+from sqlalchemy.dialects.postgresql import insert
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Recommendation
 
 
+@dataclass(frozen=True, slots=True)
+class NouvelleRecommandation:
+    alert_id: int
+    action: str
+    explanation: str
+    rule_reference: str
+
+
+TAILLE_DE_LOT = 1000
+
+
 class RecommendationRepository:
     def __init__(self, session: AsyncSession) -> None:
         self._session = session
@@ -20,3 +33,19 @@ class RecommendationRepository:
         )
         recommendation: Recommendation | None = await self._session.scalar(requete)
         return recommendation
+
+    # Pourquoi : l'idempotence est déléguée à `uq_recommendation_alert_rule` plutôt qu'à une
+    # lecture préalable, qui laisserait une fenêtre entre le contrôle et l'insertion.
+    async def create_missing(self, nouvelles: Sequence[NouvelleRecommandation]) -> int:
+        creees = 0
+        # Piège : asyncpg plafonne une requête à 32 767 paramètres, soit 8 191 lignes de quatre
+        # colonnes. Au-delà de ce seuil un `INSERT` d'un seul tenant échouerait.
+        for debut in range(0, len(nouvelles), TAILLE_DE_LOT):
+            requete = (
+                insert(Recommendation)
+                .values([asdict(nouvelle) for nouvelle in nouvelles[debut : debut + TAILLE_DE_LOT]])
+                .on_conflict_do_nothing(constraint="uq_recommendation_alert_rule")
+                .returning(Recommendation.recommendation_id)
+            )
+            creees += len((await self._session.scalars(requete)).all())
+        return creees

apps/backend/app/schemas/auth.py

@@ -1,17 +1,45 @@
 # Contrainte : le mot de passe est borné à 128 caractères. Sans plafond, une chaîne de dix
 # mégaoctets ferait travailler Argon2 gratuitement, à la charge du serveur.
+# Contrainte : `SPECIAL_CHARACTERS` doit rester identique à `password.validator.ts` côté
+# frontend. `\w`/`\d` divergent entre Python (Unicode) et JavaScript (ASCII) : une classe
+# explicite, plutôt qu'une négation, évite qu'un mot de passe soit accepté d'un côté et
+# rejeté de l'autre (ex. "Sécurité1", où "é" comptait comme "spécial" pour Python seul).
 
+import re
 from typing import Literal, Self
 from uuid import UUID
 
-from pydantic import BaseModel, ConfigDict, EmailStr, Field
+from pydantic import BaseModel, ConfigDict, EmailStr, Field, field_validator
 
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role
 
-PASSWORD_MIN_LENGTH = 12
+PASSWORD_MIN_LENGTH = 8
 PASSWORD_MAX_LENGTH = 128
 
+SPECIAL_CHARACTERS = "!@#$%^&*()-_=+[]{};:,.?"
+
+_MAJUSCULE = re.compile(r"[A-ZÀ-ÖØ-Þ]")
+_MINUSCULE = re.compile(r"[a-zà-öø-þ]")
+_CHIFFRE = re.compile(r"[0-9]")
+_SPECIAL = re.compile(r"[" + re.escape(SPECIAL_CHARACTERS) + r"]")
+
+
+def valide_complexite(mot_de_passe: str) -> str:
+    manquants = [
+        nom
+        for nom, motif in (
+            ("une majuscule", _MAJUSCULE),
+            ("une minuscule", _MINUSCULE),
+            ("un chiffre", _CHIFFRE),
+            ("un caractère spécial", _SPECIAL),
+        )
+        if not motif.search(mot_de_passe)
+    ]
+    if manquants:
+        raise ValueError(f"Le mot de passe doit contenir au moins {', '.join(manquants)}")
+    return mot_de_passe
+
 
 class LoginRequest(BaseModel):
     email: EmailStr
@@ -22,6 +50,25 @@ class PasswordChangeRequest(BaseModel):
     current_password: str = Field(min_length=1, max_length=PASSWORD_MAX_LENGTH)
     new_password: str = Field(min_length=PASSWORD_MIN_LENGTH, max_length=PASSWORD_MAX_LENGTH)
 
+    @field_validator("new_password")
+    @classmethod
+    def _new_password_est_complexe(cls, valeur: str) -> str:
+        return valide_complexite(valeur)
+
+
+class ForgotPasswordRequest(BaseModel):
+    email: EmailStr
+
+
+class ResetPasswordRequest(BaseModel):
+    token: str = Field(min_length=1)
+    new_password: str = Field(min_length=PASSWORD_MIN_LENGTH, max_length=PASSWORD_MAX_LENGTH)
+
+    @field_validator("new_password")
+    @classmethod
+    def _new_password_est_complexe(cls, valeur: str) -> str:
+        return valide_complexite(valeur)
+
 
 class PrincipalResponse(BaseModel):
     model_config = ConfigDict(from_attributes=True)
@@ -37,6 +84,10 @@ class PrincipalResponse(BaseModel):
         return cls.model_validate(principal)
 
 
+class ResetTokenValidationResponse(BaseModel):
+    valid: bool
+
+
 class TokenResponse(BaseModel):
     access_token: str
     token_type: Literal["bearer"] = "bearer"  # noqa: S105

apps/backend/app/schemas/prediction.py

@@ -0,0 +1,43 @@
+from datetime import datetime
+from enum import StrEnum
+
+from pydantic import BaseModel, ConfigDict
+
+
+class PredictionTargetMetric(StrEnum):
+    CONSUMPTION_KWH = "consumption_kwh"
+    CONSUMPTION_KW = "consumption_kw"
+
+
+class PredictionStatus(StrEnum):
+    AVAILABLE = "available"
+    INSUFFICIENT_DATA = "insufficient_data"
+    ERROR = "error"
+
+
+class SitePredictionResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    target_at: datetime
+    target_metric: PredictionTargetMetric
+    period_minutes: int | None
+    predicted_value: float | None
+    status: PredictionStatus
+    failure_reason: str | None
+    model_reference: str
+    created_at: datetime
+
+
+class SitePredictionSummaryResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    site_id: str
+    site_name: str
+    prediction: SitePredictionResponse | None
+
+
+class PredictionSummaryResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    timestamp: datetime
+    sites: list[SitePredictionSummaryResponse]

apps/backend/app/schemas/reading.py

@@ -0,0 +1,45 @@
+from datetime import datetime
+from decimal import Decimal
+from enum import StrEnum
+from typing import Any
+
+from pydantic import BaseModel, ConfigDict
+
+
+class ReadingSource(StrEnum):
+    CSV = "csv"
+    API_CURRENT = "api_current"
+    API_HISTORY = "api_history"
+
+
+class ReadingDataQuality(StrEnum):
+    GOOD = "good"
+    PARTIAL = "partial"
+    DEGRADED = "degraded"
+    CRITICAL = "critical"
+
+
+class ReadingResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    reading_id: int
+    site_id: str
+    timestamp: datetime
+    source: ReadingSource
+    consumption_kw: float | None
+    consumption_kwh: float | None
+    # Piège : `Decimal` (miroir de `Numeric(14, 2)` en base, pour ne pas arrondir un montant)
+    # sérialise en chaîne dans le JSON, pas en nombre — un consommateur qui ferait un `parseFloat`
+    # naïf perdrait la précision que ce choix visait à garder.
+    consumption_euros: Decimal | None
+    voltage_v: float | None
+    current_a: float | None
+    power_factor: float | None
+    temperature_celsius: float | None
+    humidity_percent: float | None
+    solar_irradiance_wm2: float | None
+    is_working_hours: bool | None
+    data_quality: ReadingDataQuality | None
+    null_reasons: list[str] | None
+    imputed_values: dict[str, Any] | None
+    imputation_method: str | None

apps/backend/app/schemas/recommendation.py

@@ -12,3 +12,9 @@ class RecommendationResponse(BaseModel):
     explanation: str
     rule_reference: str
     created_at: datetime
+
+
+class RecommendationGenerationResponse(BaseModel):
+    alerts_examined: int
+    recommendations_created: int
+    already_present: int

apps/backend/app/schemas/site.py

@@ -1,3 +1,6 @@
+from datetime import datetime
+from typing import Literal
+
 from pydantic import BaseModel, ConfigDict
 
 
@@ -10,3 +13,20 @@ class SiteResponse(BaseModel):
     location: str | None
     capacity_kw: float | None
     status: str | None
+
+
+class SiteCurrentResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    timestamp: datetime | None
+    site_id: str
+    site_type: str
+    consumption_kw: float | None
+    consumption_kwh: float | None
+    voltage_v: float | None
+    current_a: float | None
+    power_factor: float | None
+    temperature_celsius: float | None
+    humidity_percent: float | None
+    null_reasons: list[str]
+    data_quality: Literal["good", "partial", "degraded", "critical"]

apps/backend/app/services/alert.py

@@ -1,14 +1,323 @@
 from collections.abc import Sequence
+from datetime import UTC, datetime, timedelta
 
-from app.models.energy import Alert
+from app.models.energy import Alert, Prediction, Reading, Site
 from app.repositories.alert import AlertRepository
+from app.repositories.prediction import PredictionRepository
+from app.repositories.reading import ReadingRepository
+from app.repositories.site import SiteRepository
+
+# Fenêtre de lectures/prédictions analysée à chaque exécution : assez large pour couvrir une paire
+# de lectures consécutives (spike) et une coupure prolongée (outage), sans réanalyser tout
+# l'historique à chaque lancement manuel du script de détection.
+LOOKBACK = timedelta(hours=48)
+
+# Cadence nominale d'une lecture : le CSV historique comme l'API Mock livrent un pas horaire.
+EXPECTED_INTERVAL = timedelta(hours=1)
+# Au-delà de trois pas manqués, on parle de coupure plutôt que d'un simple retard d'ingestion.
+OUTAGE_THRESHOLD = EXPECTED_INTERVAL * 3
+
+# +/-50% entre deux lectures consécutives du même site.
+SPIKE_RELATIVE_THRESHOLD = 0.5
+# 30% d'écart entre la consommation réelle et la prévision du même site/instant.
+ANOMALY_RELATIVE_THRESHOLD = 0.3
+# Une prévision quasi nulle rend l'écart relatif ininterprétable ; on l'ignore plutôt.
+ANOMALY_MINIMUM_PREDICTED_VALUE = 1e-6
+
+THRESHOLD_METRIC = "consumption_kw"
+ANOMALY_METRIC = "consumption_kwh"
+# `data_quality` -> sévérité du capteur défaillant. `good` est volontairement absent : il ne
+# déclenche jamais d'alerte.
+QUALITE_VERS_SEVERITE: dict[str, str] = {
+    "partial": "low",
+    "degraded": "medium",
+    "critical": "critical",
+}
 
 
 class AlertService:
-    def __init__(self, *, alerts: AlertRepository) -> None:
+    def __init__(
+        self,
+        *,
+        alerts: AlertRepository,
+        readings: ReadingRepository,
+        predictions: PredictionRepository,
+        sites: SiteRepository,
+    ) -> None:
         self._alerts = alerts
+        self._readings = readings
+        self._predictions = predictions
+        self._sites = sites
 
     async def list_all(
         self, *, site_id: str | None = None, severity: str | None = None
     ) -> Sequence[Alert]:
         return await self._alerts.list_all(site_id=site_id, severity=severity)
+
+    async def detect(
+        self, *, now: datetime | None = None, site_id: str | None = None
+    ) -> Sequence[Alert]:
+        """Compare les lectures/prévisions récentes aux cinq règles internes et enregistre les
+        alertes déclenchées (`source='enervision'`). Idempotent grâce à `source_alert_id` :
+        rejouer sur une fenêtre déjà analysée ne recrée pas les mêmes lignes."""
+        instant = now or datetime.now(UTC)
+        depuis = instant - LOOKBACK
+
+        sites = await self._sites.list_all()
+        if site_id is not None:
+            sites = [site for site in sites if site.site_id == site_id]
+        sites_par_id = {site.site_id: site for site in sites}
+        if not sites_par_id:
+            return []
+
+        lectures = [
+            lecture
+            for lecture in await self._readings.list_since(since=depuis, site_id=site_id)
+            if lecture.site_id in sites_par_id
+        ]
+        predictions = [
+            prediction
+            for prediction in await self._predictions.list_since(since=depuis, site_id=site_id)
+            if prediction.site_id in sites_par_id
+        ]
+        dernieres_lectures = {
+            lecture.site_id: lecture
+            for lecture in await self._readings.latest_by_site()
+            if lecture.site_id in sites_par_id
+        }
+
+        candidates = [
+            *_detect_threshold(lectures, sites_par_id),
+            *_detect_spike(lectures),
+            *_detect_anomaly(lectures, predictions),
+            *_detect_outage(sites, dernieres_lectures, instant),
+            *_detect_sensor(lectures),
+        ]
+        if not candidates:
+            return []
+        return await self._alerts.create_many(candidates)
+
+
+def _severity_from_ratio(ratio: float) -> str:
+    if ratio >= 2.0:
+        return "critical"
+    if ratio >= 1.5:
+        return "high"
+    if ratio >= 1.2:
+        return "medium"
+    return "low"
+
+
+def _detect_threshold(lectures: Sequence[Reading], sites_par_id: dict[str, Site]) -> list[Alert]:
+    # Seuil fixe = la capacité déclarée du site : dépasser `capacity_kw` est un dépassement
+    # matériel, pas une simple variation, et évite un seuil arbitraire non fourni par le domaine.
+    alertes = []
+    for lecture in lectures:
+        site = sites_par_id[lecture.site_id]
+        valeur = lecture.consumption_kw
+        if site.capacity_kw is None or site.capacity_kw <= 0 or valeur is None:
+            continue
+        if valeur <= site.capacity_kw:
+            continue
+        alertes.append(
+            Alert(
+                source_alert_id=f"threshold:{THRESHOLD_METRIC}:{lecture.timestamp.isoformat()}",
+                site_id=lecture.site_id,
+                source="enervision",
+                timestamp=lecture.timestamp,
+                type="threshold",
+                severity=_severity_from_ratio(valeur / site.capacity_kw),
+                message=(
+                    f"Puissance appelée {valeur:.1f} kW au-dessus de la capacité du site "
+                    f"({site.capacity_kw:.1f} kW)"
+                ),
+                value=valeur,
+                threshold=site.capacity_kw,
+                metric=THRESHOLD_METRIC,
+                prediction_id=None,
+                raw_data={},
+            )
+        )
+    return alertes
+
+
+def _detect_spike(lectures: Sequence[Reading]) -> list[Alert]:
+    # `lectures` est triée par site, heure puis `reading_id` (cf. `ReadingRepository.list_since`) :
+    # deux lignes consécutives du même site sont donc deux mesures consécutives dans le temps,
+    # sauf lorsqu'elles partagent le même horodatage (deux `source` différentes pour le même
+    # instant, permises par `uq_reading_source`) : ce n'est alors pas une variation réelle, on
+    # l'ignore plutôt que de générer une fausse alerte figée par son `source_alert_id`.
+    alertes = []
+    precedente: Reading | None = None
+    for lecture in lectures:
+        if (
+            precedente is None
+            or precedente.site_id != lecture.site_id
+            or precedente.timestamp == lecture.timestamp
+        ):
+            precedente = lecture
+            continue
+        avant, apres = precedente.consumption_kw, lecture.consumption_kw
+        precedente = lecture
+        if avant is None or apres is None:
+            continue
+        if avant == 0:
+            # Une variation relative n'a pas de sens depuis zéro, mais un redémarrage direct à
+            # une consommation positive reste le signal le plus alarmant du lot : `critical`
+            # plutôt qu'un ratio indéfini.
+            if apres > 0:
+                alertes.append(_spike_alert(lecture, avant, apres, severity="critical"))
+            continue
+        variation = abs(apres - avant) / abs(avant)
+        if variation < SPIKE_RELATIVE_THRESHOLD:
+            continue
+        alertes.append(
+            _spike_alert(
+                lecture,
+                avant,
+                apres,
+                severity=_severity_from_ratio(variation / SPIKE_RELATIVE_THRESHOLD),
+            )
+        )
+    return alertes
+
+
+def _spike_alert(lecture: Reading, avant: float, apres: float, *, severity: str) -> Alert:
+    return Alert(
+        source_alert_id=f"spike:{THRESHOLD_METRIC}:{lecture.timestamp.isoformat()}",
+        site_id=lecture.site_id,
+        source="enervision",
+        timestamp=lecture.timestamp,
+        type="spike",
+        severity=severity,
+        message=(
+            f"Variation brutale entre deux lectures consécutives ({avant:.1f} kW -> {apres:.1f} kW)"
+        ),
+        value=apres,
+        threshold=avant,
+        metric=THRESHOLD_METRIC,
+        prediction_id=None,
+        raw_data={},
+    )
+
+
+def _detect_anomaly(lectures: Sequence[Reading], predictions: Sequence[Prediction]) -> list[Alert]:
+    # Alignement strict (site_id, target_at == timestamp) : `enervision_ml.score` produit une
+    # cible à l'heure pile suivant la dernière lecture, sur la même grille horaire que `reading`.
+    predictions_par_cle = {
+        (prediction.site_id, prediction.target_at): prediction
+        for prediction in predictions
+        if prediction.target_metric == ANOMALY_METRIC
+    }
+    alertes = []
+    for lecture in lectures:
+        prediction = predictions_par_cle.get((lecture.site_id, lecture.timestamp))
+        reel = lecture.consumption_kwh
+        if prediction is None or reel is None or prediction.predicted_value is None:
+            continue
+        predite = prediction.predicted_value
+        if abs(predite) < ANOMALY_MINIMUM_PREDICTED_VALUE:
+            continue
+        ecart = abs(reel - predite) / abs(predite)
+        if ecart < ANOMALY_RELATIVE_THRESHOLD:
+            continue
+        alertes.append(
+            Alert(
+                source_alert_id=f"anomaly:{ANOMALY_METRIC}:{lecture.timestamp.isoformat()}",
+                site_id=lecture.site_id,
+                source="enervision",
+                timestamp=lecture.timestamp,
+                type="anomaly",
+                severity=_severity_from_ratio(ecart / ANOMALY_RELATIVE_THRESHOLD),
+                message=(
+                    f"Écart de {ecart * 100:.0f}% entre la consommation mesurée ({reel:.1f} kWh) "
+                    f"et la prévision ({predite:.1f} kWh)"
+                ),
+                value=reel,
+                threshold=predite,
+                metric=ANOMALY_METRIC,
+                prediction_id=prediction.prediction_id,
+                raw_data={},
+            )
+        )
+    return alertes
+
+
+def _detect_outage(
+    sites: Sequence[Site], dernieres_lectures: dict[str, Reading], now: datetime
+) -> list[Alert]:
+    alertes = []
+    for site in sites:
+        derniere = dernieres_lectures.get(site.site_id)
+        if derniere is None:
+            alertes.append(
+                _outage_alert(
+                    site.site_id,
+                    now,
+                    reference=None,
+                    message="Aucune lecture n'a jamais été reçue pour ce site",
+                    severity="critical",
+                )
+            )
+            continue
+        absence = now - derniere.timestamp
+        if absence < OUTAGE_THRESHOLD:
+            continue
+        alertes.append(
+            _outage_alert(
+                site.site_id,
+                now,
+                reference=derniere.timestamp,
+                message=(
+                    f"Aucune lecture depuis {absence} (dernière lecture : "
+                    f"{derniere.timestamp.isoformat()})"
+                ),
+                severity=_severity_from_ratio(absence / OUTAGE_THRESHOLD),
+            )
+        )
+    return alertes
+
+
+def _outage_alert(
+    site_id: str, now: datetime, *, reference: datetime | None, message: str, severity: str
+) -> Alert:
+    return Alert(
+        source_alert_id=f"outage:{reference.isoformat() if reference is not None else 'jamais'}",
+        site_id=site_id,
+        source="enervision",
+        timestamp=now,
+        type="outage",
+        severity=severity,
+        message=message,
+        value=None,
+        threshold=None,
+        metric=None,
+        prediction_id=None,
+        raw_data={},
+    )
+
+
+def _detect_sensor(lectures: Sequence[Reading]) -> list[Alert]:
+    alertes = []
+    for lecture in lectures:
+        severite = QUALITE_VERS_SEVERITE.get(lecture.data_quality or "")
+        if severite is None:
+            continue
+        raisons = ", ".join(lecture.null_reasons or []) or "raison non précisée"
+        alertes.append(
+            Alert(
+                source_alert_id=f"sensor:{lecture.timestamp.isoformat()}",
+                site_id=lecture.site_id,
+                source="enervision",
+                timestamp=lecture.timestamp,
+                type="sensor",
+                severity=severite,
+                message=f"Qualité de mesure {lecture.data_quality} ({raisons})",
+                value=None,
+                threshold=None,
+                metric=None,
+                prediction_id=None,
+                raw_data={},
+            )
+        )
+    return alertes

apps/backend/app/services/auth.py

@@ -14,7 +14,11 @@ from datetime import UTC, datetime, timedelta
 from typing import NoReturn, Protocol
 from uuid import UUID, uuid4
 
+from fastapi import BackgroundTasks
+
 from app.core.hashing import Argon2Hasher
+from app.core.logging import get_logger
+from app.core.mailer import Mailer
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role
 from app.core.security import (
@@ -28,9 +32,13 @@ from app.models.login_attempt import LoginOutcome
 from app.models.refresh_token import RevocationReason
 from app.repositories.audit_log import AuditLogRepository
 from app.repositories.login_attempt import LoginAttemptRepository
+from app.repositories.password_reset_attempt import PasswordResetAttemptRepository
+from app.repositories.password_reset_token import PasswordResetTokenRepository
 from app.repositories.refresh_token import RefreshTokenRepository
 from app.repositories.user import UserRepository
 
+logger = get_logger(__name__)
+
 
 class Transaction(Protocol):
     async def commit(self) -> None: ...
@@ -54,6 +62,10 @@ class RateLimitedError(AuthError):
         self.retry_after = retry_after
 
 
+class InvalidOrExpiredResetTokenError(AuthError):
+    pass
+
+
 @dataclass(frozen=True, slots=True)
 class LoginPolicy:
     window_seconds: int
@@ -62,6 +74,15 @@ class LoginPolicy:
     max_failures_per_identifier: int
 
 
+@dataclass(frozen=True, slots=True)
+class PasswordResetPolicy:
+    window_seconds: int
+    max_requests_per_identifier: int
+    max_requests_per_ip: int
+    token_ttl: timedelta
+    frontend_reset_url: str
+
+
 @dataclass(frozen=True, slots=True)
 class AuthenticatedSession:
     principal: Principal
@@ -83,6 +104,10 @@ class AuthService:
         token_policy: TokenPolicy,
         login_policy: LoginPolicy,
         refresh_ttl: timedelta,
+        reset_tokens: PasswordResetTokenRepository,
+        reset_attempts: PasswordResetAttemptRepository,
+        reset_policy: PasswordResetPolicy,
+        mailer: Mailer,
     ) -> None:
         self._users = users
         self._attempts = attempts
@@ -93,6 +118,10 @@ class AuthService:
         self._token_policy = token_policy
         self._login_policy = login_policy
         self._refresh_ttl = refresh_ttl
+        self._reset_tokens = reset_tokens
+        self._reset_attempts = reset_attempts
+        self._reset_policy = reset_policy
+        self._mailer = mailer
 
     async def authenticate(
         self, *, email: str, password: str, client_ip: str | None, user_agent: str | None
@@ -200,6 +229,102 @@ class AuthService:
         rafraichi = await self._users.get_by_id(principal.id)
         return self._session(self._en_principal(rafraichi or compte), secret)
 
+    async def request_password_reset(
+        self,
+        *,
+        email: str,
+        client_ip: str | None,
+        user_agent: str | None,
+        background_tasks: BackgroundTasks,
+    ) -> None:
+        await self._refuse_si_limite_reset(email=email, client_ip=client_ip)
+
+        compte = await self._users.get_by_email(email)
+        # Piège : le hachage factice équilibre le temps de réponse sur un compte inconnu, comme
+        # `authenticate()`. La réponse et sa forme restent identiques dans tous les cas : compte
+        # inconnu, compte inactif, ou email envoyé avec succès. L'envoi SMTP lui-même est différé
+        # en tâche de fond : le laisser dans le chemin de réponse rouvrirait le même oracle par le
+        # temps (aller-retour réseau) et par la forme (500 si le relais SMTP échoue, contre 202).
+        if compte is None or not compte.is_active or compte.kind != AccountKind.HUMAIN.value:
+            await self._hasher.verify_dummy()
+            await self._reset_attempts.record(email=email, client_ip=client_ip)
+            await self._transaction.commit()
+            return
+
+        await self._reset_tokens.invalidate_all_for_user(compte.id)
+        secret = generate_refresh_secret()
+        await self._reset_tokens.create(
+            user_id=compte.id,
+            token_hash=fingerprint_refresh(secret),
+            expires_at=datetime.now(UTC) + self._reset_policy.token_ttl,
+            client_ip=client_ip,
+            user_agent=user_agent,
+        )
+        await self._reset_attempts.record(email=email, client_ip=client_ip)
+        await self._audit.record(
+            action=AuditAction.MOT_DE_PASSE_OUBLIE_DEMANDE,
+            actor_label=compte.email,
+            target_type="app_user",
+            target_id=str(compte.id),
+            client_ip=client_ip,
+            user_agent=user_agent,
+        )
+        await self._transaction.commit()
+
+        lien = f"{self._reset_policy.frontend_reset_url}?token={secret}"
+        background_tasks.add_task(self._envoie_email_reset, compte.email, lien)
+
+    async def _envoie_email_reset(self, email: str, reset_url: str) -> None:
+        try:
+            await self._mailer.send_password_reset_email(to=email, reset_url=reset_url)
+        except Exception:
+            logger.exception("auth.password_reset.mail_failed")
+
+    # Piège : lecture seule, pas d'appel à `consume()`. Aucune limitation de débit n'est
+    # nécessaire ici : le jeton est un secret de 256 bits (`generate_refresh_secret`), donc
+    # non brute-forçable, et cette route n'apprend rien sur l'existence d'un compte ou d'un
+    # email, seulement si le lien déjà en main du visiteur est encore valide.
+    async def is_reset_token_valid(self, token: str) -> bool:
+        return await self._reset_tokens.exists_valid(fingerprint_refresh(token))
+
+    async def confirm_password_reset(
+        self, *, token: str, new_password: str, client_ip: str | None, user_agent: str | None
+    ) -> AuthenticatedSession:
+        revendique = await self._reset_tokens.consume(fingerprint_refresh(token))
+        if revendique is None:
+            raise InvalidOrExpiredResetTokenError("Lien invalide ou expiré")
+
+        # Piège : le jeton peut avoir été émis avant une désactivation du compte. Sans cette
+        # relecture, un lien encore valide (15 min) changerait quand même le mot de passe d'un
+        # compte désactivé, réutilisable dès sa réactivation.
+        compte = await self._users.get_by_id(revendique.user_id)
+        if compte is None or not compte.is_active or compte.kind != AccountKind.HUMAIN.value:
+            raise InvalidOrExpiredResetTokenError("Lien invalide ou expiré")
+
+        await self._users.update_password(
+            revendique.user_id, await self._hasher.hash(new_password), must_change_password=False
+        )
+        revoquees = await self._refresh.revoke_all_for_user(
+            revendique.user_id, RevocationReason.CHANGEMENT_MOT_DE_PASSE
+        )
+        secret = await self._ouvre_une_famille(
+            user_id=revendique.user_id, client_ip=client_ip, user_agent=user_agent
+        )
+        await self._audit.record(
+            action=AuditAction.MOT_DE_PASSE_REINITIALISE_PAR_SOI,
+            target_type="app_user",
+            target_id=str(revendique.user_id),
+            client_ip=client_ip,
+            user_agent=user_agent,
+            detail={"sessions_revoquees": revoquees},
+        )
+        await self._transaction.commit()
+
+        compte = await self._users.get_by_id(revendique.user_id)
+        if compte is None:
+            raise SessionRejectedError("Compte introuvable")
+        return self._session(self._en_principal(compte), secret)
+
     async def logout_all(self, principal: Principal) -> int:
         revoquees = await self._refresh.revoke_all_for_user(
             principal.id, RevocationReason.DECONNEXION
@@ -307,6 +432,23 @@ class AuthService:
         await self._transaction.commit()
         raise RateLimitedError(politique.window_seconds)
 
+    async def _refuse_si_limite_reset(self, *, email: str, client_ip: str | None) -> None:
+        politique = self._reset_policy
+        compteurs = await self._reset_attempts.count_recent(
+            email=email, client_ip=client_ip, window_seconds=politique.window_seconds
+        )
+
+        depasse = (
+            compteurs.per_identifier >= politique.max_requests_per_identifier
+            or compteurs.per_ip >= politique.max_requests_per_ip
+        )
+        if not depasse:
+            return
+
+        await self._reset_attempts.record(email=email, client_ip=client_ip)
+        await self._transaction.commit()
+        raise RateLimitedError(politique.window_seconds)
+
     async def _echoue(
         self,
         email: str,

apps/backend/app/services/data_quality.py

@@ -0,0 +1,18 @@
+# Contrainte : `ck_reading_quality` accepte NULL et quatre valeurs seulement, alors que le contrat
+# frontend n'a aucune valeur pour l'absence de qualité. `qualite_ou_critique()` replie donc sur
+# `critical`, la seule des quatre qui n'induise pas une confiance qu'on n'a pas. `QUALITES_CONNUES`
+# reste exposé pour les appelants qui doivent distinguer un `critical` stocké d'un repli.
+
+from typing import Literal, get_args
+
+DataQuality = Literal["good", "partial", "degraded", "critical"]
+
+QUALITES_CONNUES: frozenset[str] = frozenset(get_args(DataQuality))
+
+_PAR_VALEUR: dict[str, DataQuality] = {valeur: valeur for valeur in get_args(DataQuality)}
+
+
+def qualite_ou_critique(valeur: str | None) -> DataQuality:
+    if valeur is None:
+        return "critical"
+    return _PAR_VALEUR.get(valeur, "critical")

apps/backend/app/services/prediction.py

@@ -0,0 +1,68 @@
+from dataclasses import dataclass
+from datetime import UTC, datetime
+
+from app.models.energy import Prediction, Site
+from app.repositories.prediction import PredictionRepository
+from app.repositories.site import SiteRepository
+
+
+@dataclass(frozen=True, slots=True)
+class SitePrediction:
+    target_at: datetime
+    target_metric: str
+    period_minutes: int | None
+    predicted_value: float | None
+    status: str
+    failure_reason: str | None
+    model_reference: str
+    created_at: datetime
+
+
+@dataclass(frozen=True, slots=True)
+class SitePredictionSummary:
+    site_id: str
+    site_name: str
+    prediction: SitePrediction | None
+
+
+@dataclass(frozen=True, slots=True)
+class PredictionSummary:
+    timestamp: datetime
+    sites: list[SitePredictionSummary]
+
+
+class PredictionService:
+    def __init__(self, sites: SiteRepository, predictions: PredictionRepository) -> None:
+        self._sites = sites
+        self._predictions = predictions
+
+    async def summary(self) -> PredictionSummary:
+        sites = await self._sites.list_all()
+        dernieres = {p.site_id: p for p in await self._predictions.latest_by_site()}
+
+        return PredictionSummary(
+            timestamp=datetime.now(UTC),
+            sites=[_resume_site(site, dernieres.get(site.site_id)) for site in sites],
+        )
+
+
+def _resume_site(site: Site, derniere: Prediction | None) -> SitePredictionSummary:
+    # Piège : l'absence de ligne signifie « jamais scoré », pas une valeur pseudo-statut, qui
+    # n'existe pas dans la contrainte de la table. `prediction` reste `None` plutôt que de
+    # fabriquer un statut absent du domaine `available`/`insufficient_data`/`error`.
+    prediction = None
+    if derniere is not None:
+        prediction = SitePrediction(
+            target_at=derniere.target_at,
+            target_metric=derniere.target_metric,
+            period_minutes=derniere.period_minutes,
+            predicted_value=derniere.predicted_value,
+            status=derniere.status,
+            failure_reason=derniere.failure_reason,
+            model_reference=derniere.model_reference,
+            created_at=derniere.created_at,
+        )
+
+    return SitePredictionSummary(
+        site_id=site.site_id, site_name=site.site_name, prediction=prediction
+    )

apps/backend/app/services/reading.py

@@ -0,0 +1,59 @@
+from collections.abc import Sequence
+from datetime import UTC, datetime, timedelta
+
+from app.models.energy import Reading
+from app.repositories.reading import ReadingRepository
+
+FENETRE_PAR_DEFAUT = timedelta(hours=24)
+FENETRE_MAXIMALE = timedelta(days=90)
+
+
+class FenetreInverseeError(Exception):
+    """`start` est postérieur ou égal à `end`."""
+
+
+class FenetreTropLargeError(Exception):
+    """L'écart entre `start` et `end` dépasse `FENETRE_MAXIMALE`."""
+
+
+class ReadingService:
+    def __init__(self, *, readings: ReadingRepository) -> None:
+        self._readings = readings
+
+    async def list_history(
+        self,
+        *,
+        site_id: str | None = None,
+        start: datetime | None = None,
+        end: datetime | None = None,
+        limit: int,
+        offset: int,
+    ) -> Sequence[Reading]:
+        debut, fin = self._resoudre_fenetre(start, end)
+        return await self._readings.list_history(
+            site_id=site_id, start=debut, end=fin, limit=limit, offset=offset
+        )
+
+    @staticmethod
+    def _resoudre_fenetre(
+        start: datetime | None, end: datetime | None
+    ) -> tuple[datetime, datetime]:
+        # Piège : un datetime naïf (sans fuseau dans la chaîne ISO reçue) fait échouer la
+        # comparaison à `reading.timestamp` (`timestamptz`) au niveau du pilote, en 500 plutôt
+        # qu'un refus propre. On le traite comme de l'UTC plutôt que de le rejeter.
+        debut = _vers_utc(start)
+        fin = _vers_utc(end) or datetime.now(UTC)
+        if debut is None:
+            debut = fin - FENETRE_PAR_DEFAUT
+
+        if debut >= fin:
+            raise FenetreInverseeError
+        if fin - debut > FENETRE_MAXIMALE:
+            raise FenetreTropLargeError
+        return debut, fin
+
+
+def _vers_utc(instant: datetime | None) -> datetime | None:
+    if instant is None:
+        return None
+    return instant if instant.tzinfo is not None else instant.replace(tzinfo=UTC)

apps/backend/app/services/recommendation.py

@@ -1,7 +1,15 @@
 from collections.abc import Sequence
+from dataclasses import dataclass
+from typing import Protocol
 
 from app.models.energy import Recommendation
+from app.repositories.alert import AlertRepository
 from app.repositories.recommendation import RecommendationRepository
+from app.services.recommendation_rules import applique_les_regles
+
+
+class Transaction(Protocol):
+    async def commit(self) -> None: ...
 
 
 class RecommendationError(Exception):
@@ -12,9 +20,24 @@ class RecommendationNotFoundError(RecommendationError):
     pass
 
 
+@dataclass(frozen=True, slots=True)
+class RapportGeneration:
+    alertes_examinees: int
+    recommandations_creees: int
+    deja_presentes: int
+
+
 class RecommendationService:
-    def __init__(self, *, recommendations: RecommendationRepository) -> None:
+    def __init__(
+        self,
+        *,
+        recommendations: RecommendationRepository,
+        alerts: AlertRepository,
+        transaction: Transaction,
+    ) -> None:
         self._recommendations = recommendations
+        self._alerts = alerts
+        self._transaction = transaction
 
     async def list_all(self) -> Sequence[Recommendation]:
         return await self._recommendations.list_all()
@@ -24,3 +47,16 @@ class RecommendationService:
         if recommendation is None:
             raise RecommendationNotFoundError(recommendation_id)
         return recommendation
+
+    async def generate(self, *, site_id: str | None = None) -> RapportGeneration:
+        alertes = await self._alerts.list_all(site_id=site_id)
+        nouvelles = [nouvelle for alerte in alertes for nouvelle in applique_les_regles(alerte)]
+
+        creees = await self._recommendations.create_missing(nouvelles)
+        await self._transaction.commit()
+
+        return RapportGeneration(
+            alertes_examinees=len(alertes),
+            recommandations_creees=creees,
+            deja_presentes=len(nouvelles) - creees,
+        )

apps/backend/app/services/recommendation_rules.py

@@ -0,0 +1,117 @@
+# Piège : `rule_reference` est la clé d'idempotence en base, portée par la contrainte
+# `uq_recommendation_alert_rule`. Renommer une référence déjà livrée ne remplace pas les
+# recommandations existantes, il en crée de nouvelles à côté. Une règle qui change de sens
+# prend donc une référence suffixée `-v2` - REGLES.
+
+from collections.abc import Callable
+from dataclasses import dataclass
+from typing import Final
+
+from app.models.energy import Alert
+from app.repositories.recommendation import NouvelleRecommandation
+from app.schemas.alert import AlertSeverity, AlertType
+
+FACTEUR_DEPASSEMENT_MAJEUR: Final = 1.2
+POURCENTAGE_DEPASSEMENT_MAJEUR: Final = round((FACTEUR_DEPASSEMENT_MAJEUR - 1) * 100)
+
+
+@dataclass(frozen=True, slots=True)
+class Regle:
+    reference: str
+    action: str
+    declencheur: Callable[[Alert], bool]
+    motif: Callable[[Alert], str]
+
+
+def _du_type(attendu: AlertType) -> Callable[[Alert], bool]:
+    return lambda alerte: alerte.type == attendu
+
+
+def _de_severite(attendue: AlertSeverity) -> Callable[[Alert], bool]:
+    return lambda alerte: alerte.severity == attendue
+
+
+# Un seuil nul ou négatif rendrait le rapport `value / threshold` arbitraire : l'alerte ne
+# renseigne alors aucun dépassement exploitable, et la règle ne se déclenche pas.
+def _depasse_largement_le_seuil(alerte: Alert) -> bool:
+    if alerte.value is None or alerte.threshold is None or alerte.threshold <= 0:
+        return False
+    return alerte.value >= alerte.threshold * FACTEUR_DEPASSEMENT_MAJEUR
+
+
+REGLES: Final[tuple[Regle, ...]] = (
+    Regle(
+        reference="spike-delestage-v1",
+        action="Délester les équipements non prioritaires sur le créneau du pic",
+        declencheur=_du_type(AlertType.SPIKE),
+        motif=lambda alerte: f"Pic de consommation signalé sur le site {alerte.site_id}",
+    ),
+    Regle(
+        reference="threshold-reduction-v1",
+        action="Ramener la puissance appelée sous le seuil contractuel",
+        declencheur=_du_type(AlertType.THRESHOLD),
+        motif=lambda alerte: f"Seuil de consommation dépassé sur le site {alerte.site_id}",
+    ),
+    Regle(
+        reference="outage-secours-v1",
+        action="Basculer sur l'alimentation de secours et prévenir l'exploitant",
+        declencheur=_du_type(AlertType.OUTAGE),
+        motif=lambda alerte: (
+            f"Risque de surcharge ou de coupure imminente sur le site {alerte.site_id}"
+        ),
+    ),
+    Regle(
+        reference="sensor-maintenance-v1",
+        action="Planifier une intervention de maintenance sur le capteur",
+        declencheur=_du_type(AlertType.SENSOR),
+        motif=lambda alerte: (
+            f"Capteur défaillant sur le site {alerte.site_id}, les mesures ne sont plus fiables"
+        ),
+    ),
+    Regle(
+        reference="anomaly-verification-v1",
+        action="Confronter la mesure à la prévision et vérifier le paramétrage du site",
+        declencheur=_du_type(AlertType.ANOMALY),
+        motif=lambda alerte: (
+            f"Écart anormal entre la mesure et le comportement attendu du site {alerte.site_id}"
+        ),
+    ),
+    Regle(
+        reference="escalade-astreinte-v1",
+        action="Escalader à l'astreinte sous une heure",
+        declencheur=_de_severite(AlertSeverity.CRITICAL),
+        motif=lambda alerte: f"Alerte de sévérité critique sur le site {alerte.site_id}",
+    ),
+    Regle(
+        reference="contrat-puissance-v1",
+        action="Réévaluer la puissance souscrite au contrat",
+        declencheur=_depasse_largement_le_seuil,
+        motif=lambda alerte: (
+            f"Dépassement d'au moins {POURCENTAGE_DEPASSEMENT_MAJEUR} % du seuil "
+            f"sur le site {alerte.site_id}"
+        ),
+    ),
+)
+
+
+def applique_les_regles(alerte: Alert) -> list[NouvelleRecommandation]:
+    contexte = _contexte_de_mesure(alerte)
+    return [
+        NouvelleRecommandation(
+            alert_id=alerte.alert_id,
+            action=regle.action,
+            explanation=f"{regle.motif(alerte)}{contexte}.",
+            rule_reference=regle.reference,
+        )
+        for regle in REGLES
+        if regle.declencheur(alerte)
+    ]
+
+
+def _contexte_de_mesure(alerte: Alert) -> str:
+    if alerte.value is None:
+        return ""
+    grandeur = alerte.metric or "valeur"
+    if alerte.threshold is None:
+        return f" ({grandeur} mesurée à {alerte.value})"
+    return f" ({grandeur} mesurée à {alerte.value}, seuil {alerte.threshold})"

apps/backend/app/services/sensor.py

@@ -5,12 +5,11 @@ from typing import Literal
 from app.models.energy import Reading, Site
 from app.repositories.reading import ReadingRepository
 from app.repositories.site import SiteRepository
+from app.services.data_quality import qualite_ou_critique
 
 CapteurStatus = Literal["ok", "failing"]
 OverallStatus = Literal["ok", "degraded", "critical"]
 
-QUALITES_CONNUES: frozenset[str] = frozenset({"good", "partial", "degraded", "critical"})
-
 RAISON_VERS_CAPTEUR: dict[str, str] = {
     "consumption_sensor_failure": "consumption",
     "electrical_sensor_failure": "electrical",
@@ -80,7 +79,7 @@ def _sante_site(site: Site, derniere: Reading | None) -> SanteSite:
             overall="critical",
         )
 
-    qualite = derniere.data_quality if derniere.data_quality in QUALITES_CONNUES else "critical"
+    qualite = qualite_ou_critique(derniere.data_quality)
     overall = _overall_depuis_qualite(qualite)
 
     if overall == "critical":

apps/backend/app/services/site.py

@@ -1,7 +1,11 @@
 from collections.abc import Sequence
+from dataclasses import dataclass
+from datetime import datetime
 
 from app.models.energy import Site
+from app.repositories.reading import ReadingRepository
 from app.repositories.site import SiteRepository
+from app.services.data_quality import DataQuality, qualite_ou_critique
 
 
 class SiteError(Exception):
@@ -12,9 +16,26 @@ class SiteNotFoundError(SiteError):
     pass
 
 
+@dataclass(frozen=True, slots=True)
+class SiteCurrentReading:
+    timestamp: datetime | None
+    site_id: str
+    site_type: str
+    consumption_kw: float | None
+    consumption_kwh: float | None
+    voltage_v: float | None
+    current_a: float | None
+    power_factor: float | None
+    temperature_celsius: float | None
+    humidity_percent: float | None
+    null_reasons: list[str]
+    data_quality: DataQuality
+
+
 class SiteService:
-    def __init__(self, *, sites: SiteRepository) -> None:
+    def __init__(self, *, sites: SiteRepository, readings: ReadingRepository) -> None:
         self._sites = sites
+        self._readings = readings
 
     async def list_all(self) -> Sequence[Site]:
         return await self._sites.list_all()
@@ -24,3 +45,38 @@ class SiteService:
         if site is None:
             raise SiteNotFoundError(site_id)
         return site
+
+    async def current(self, site_id: str) -> SiteCurrentReading:
+        site = await self.get_by_id(site_id)
+        derniere = await self._readings.latest_for_site(site_id)
+
+        if derniere is None:
+            return SiteCurrentReading(
+                timestamp=None,
+                site_id=site.site_id,
+                site_type=site.site_type,
+                consumption_kw=None,
+                consumption_kwh=None,
+                voltage_v=None,
+                current_a=None,
+                power_factor=None,
+                temperature_celsius=None,
+                humidity_percent=None,
+                null_reasons=[],
+                data_quality="critical",
+            )
+
+        return SiteCurrentReading(
+            timestamp=derniere.timestamp,
+            site_id=site.site_id,
+            site_type=site.site_type,
+            consumption_kw=derniere.consumption_kw,
+            consumption_kwh=derniere.consumption_kwh,
+            voltage_v=derniere.voltage_v,
+            current_a=derniere.current_a,
+            power_factor=derniere.power_factor,
+            temperature_celsius=derniere.temperature_celsius,
+            humidity_percent=derniere.humidity_percent,
+            null_reasons=derniere.null_reasons or [],
+            data_quality=qualite_ou_critique(derniere.data_quality),
+        )

apps/backend/app/services/stats.py

@@ -1,14 +1,10 @@
 from dataclasses import dataclass
 from datetime import UTC, datetime
-from typing import Literal
 
 from app.models.energy import Reading, Site
 from app.repositories.reading import ReadingRepository
 from app.repositories.site import SiteRepository
-
-DataQuality = Literal["good", "partial", "degraded", "critical"]
-
-QUALITES_CONNUES: frozenset[str] = frozenset({"good", "partial", "degraded", "critical"})
+from app.services.data_quality import QUALITES_CONNUES, DataQuality, qualite_ou_critique
 
 
 @dataclass(frozen=True, slots=True)
@@ -58,13 +54,10 @@ class StatsService:
     @staticmethod
     def _resume_site(site: Site, derniere: Reading | None) -> SiteConsumption:
         capacite = site.capacity_kw or 0
-        # Piège : `data_quality` est nul dès qu'un site n'a jamais reçu de lecture, ou que le
-        # producteur n'a pas su la qualifier. Le contrat frontend n'a pas de valeur pour ce cas,
-        # `critical` est la seule des quatre qui n'induit pas une confiance qu'on n'a pas.
         qualite: DataQuality = "critical"
         consommation = None
         if derniere is not None and derniere.data_quality in QUALITES_CONNUES:
-            qualite = derniere.data_quality  # type: ignore[assignment]
+            qualite = qualite_ou_critique(derniere.data_quality)
             consommation = derniere.consumption_kw
 
         charge = (

apps/backend/pyproject.toml

@@ -16,6 +16,9 @@ dependencies = [
     "pyjwt>=2.10",
     "argon2-cffi>=23.1",
     "anyio>=4.0",
+    "aiosmtplib>=5.1.3",
+    "httpx>=0.28.1",
+    "pandas>=3.0.5",
 ]
 
 [dependency-groups]
@@ -25,7 +28,7 @@ dev = [
     "pytest>=9.1.1",
     "pytest-asyncio>=1.4.0",
     "pytest-cov>=7.1.0",
-    "httpx>=0.28.1",
+    "pandas-stubs>=3.0.5.260914",
 ]
 
 [build-system]

apps/backend/tests/api/acces.py

@@ -0,0 +1,88 @@
+# Pourquoi : classification unique des routes du contrat, lue par test_route_protection.py,
+# test_openapi.py et test_matrice_acces.py. Trois listes séparées dérivaient auparavant chacune
+# de leur côté, et deux entrées de ROUTES_A_ROLE ne correspondaient plus à aucune route sans que
+# rien ne le signale.
+# Piège : les trois ensembles doivent rester disjoints et couvrir tout le schéma. C'est
+# `test_every_declared_route_is_classified` qui le vérifie, pas la relecture.
+
+from typing import Final
+
+from app.core.roles import Role
+
+Route = tuple[str, str]
+
+ROUTES_PUBLIQUES: Final[frozenset[Route]] = frozenset(
+    {
+        ("GET", "/api/v1/health/live"),
+        ("GET", "/api/v1/health/ready"),
+        ("POST", "/api/v1/auth/login"),
+        # Sans cookie, la déconnexion ne fait rien et répond 204 : elle est idempotente.
+        ("POST", "/api/v1/auth/logout"),
+        ("POST", "/api/v1/auth/forgot-password"),
+        # Protégée par le jeton dans le corps de la requête, pas par un `Principal` : aucune
+        # authentification préalable ne s'applique, c'est la validité du jeton qui tranche.
+        ("POST", "/api/v1/auth/reset-password"),
+        # Même raison : lecture seule, protégée par le jeton passé en paramètre, pas par un
+        # `Principal`. Le jeton est un secret de 256 bits, non brute-forçable.
+        ("GET", "/api/v1/auth/reset-password/validate"),
+        ("GET", "/metrics"),
+    }
+)
+
+# Le cookie opaque porte seul l'autorisation : sans lui la route rend 401, mais aucun `Principal`
+# n'est construit et `require_role` n'entre jamais en jeu.
+ROUTE_COOKIE: Final[frozenset[Route]] = frozenset({("POST", "/api/v1/auth/refresh")})
+
+# Authentifiées par `CurrentPrincipalDep` nu, donc hors de `require_role` et, avec lui, hors du
+# refus `password_change_required`. Volontaire pour `/auth/password`, qui est la sortie de l'état
+# provisoire ; subi pour `/auth/logout-all`, cf. test_matrice_acces.py.
+ROUTES_SANS_ROLE: Final[frozenset[Route]] = frozenset(
+    {
+        ("GET", "/api/v1/auth/me"),
+        ("POST", "/api/v1/auth/password"),
+        ("POST", "/api/v1/auth/logout-all"),
+    }
+)
+
+ROLE_MINIMUM: Final[dict[Route, Role]] = {
+    ("GET", "/api/v1/sites"): Role.LECTEUR,
+    ("GET", "/api/v1/sites/{site_id}"): Role.LECTEUR,
+    ("GET", "/api/v1/sites/{site_id}/current"): Role.LECTEUR,
+    ("GET", "/api/v1/alerts"): Role.LECTEUR,
+    ("GET", "/api/v1/recommendations"): Role.LECTEUR,
+    ("GET", "/api/v1/recommendations/{recommendation_id}"): Role.LECTEUR,
+    ("POST", "/api/v1/recommendations/generate"): Role.ADMIN,
+    ("GET", "/api/v1/stats/summary"): Role.LECTEUR,
+    ("GET", "/api/v1/readings"): Role.LECTEUR,
+    ("GET", "/api/v1/predictions"): Role.LECTEUR,
+    ("GET", "/api/v1/sensors/status"): Role.ADMIN,
+    ("GET", "/api/v1/users"): Role.ADMIN,
+    ("POST", "/api/v1/users"): Role.ADMIN,
+    ("PATCH", "/api/v1/users/{user_id}"): Role.ADMIN,
+    ("POST", "/api/v1/users/{user_id}/password-reset"): Role.ADMIN,
+}
+
+# Piège : `{recommendation_id}` est typé `int` et `{user_id}` est un UUID. Une substitution
+# uniforme par une chaîne quelconque rendrait 422 avant d'atteindre la garde de rôle, et le test
+# passerait en prouvant autre chose que ce qu'il annonce.
+SUBSTITUTIONS: Final[dict[str, str]] = {
+    "{user_id}": "00000000-0000-0000-0000-000000000000",
+    "{site_id}": "site-absent-du-jeu-de-donnees",
+    "{recommendation_id}": "999999999",
+}
+
+
+def chemin_concret(chemin: str) -> str:
+    for gabarit, valeur in SUBSTITUTIONS.items():
+        chemin = chemin.replace(gabarit, valeur)
+    return chemin
+
+
+def routes_du_schema(schema: dict[str, object]) -> list[Route]:
+    chemins: dict[str, dict[str, object]] = schema["paths"]  # type: ignore[assignment]
+    return [
+        (methode.upper(), chemin)
+        for chemin, operations in chemins.items()
+        for methode in operations
+        if methode.upper() in {"GET", "POST", "PATCH", "PUT", "DELETE"}
+    ]

apps/backend/tests/api/test_auth.py

@@ -11,6 +11,7 @@ from app.core.roles import AccountKind, Role
 from app.services.auth import (
     AuthenticatedSession,
     InvalidCredentialsError,
+    InvalidOrExpiredResetTokenError,
     RateLimitedError,
     SessionRejectedError,
 )
@@ -27,15 +28,27 @@ PRINCIPAL = Principal(
 
 
 class FauxService:
-    def __init__(self, erreur: Exception | None = None) -> None:
+    def __init__(self, erreur: Exception | None = None, *, jeton_valide: bool = True) -> None:
         self._erreur = erreur
+        self._jeton_valide = jeton_valide
 
     async def refresh(self, **_: object) -> AuthenticatedSession:
         return await self.authenticate()
 
+    async def is_reset_token_valid(self, **_: object) -> bool:
+        return self._jeton_valide
+
     async def logout(self, **_: object) -> None:
         return None
 
+    async def request_password_reset(self, **_: object) -> None:
+        if self._erreur is not None:
+            raise self._erreur
+        return None
+
+    async def confirm_password_reset(self, **_: object) -> AuthenticatedSession:
+        return await self.authenticate()
+
     async def authenticate(self, **_: object) -> AuthenticatedSession:
         if self._erreur is not None:
             raise self._erreur
@@ -206,3 +219,126 @@ async def test_a_cookie_bearing_route_accepts_a_request_without_origin(
     response = await client.post("/api/v1/auth/logout")
 
     assert response.status_code != 403
+
+
+async def test_forgot_password_answers_202_when_the_account_exists(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post(
+        "/api/v1/auth/forgot-password", json={"email": "operateur@enervision.fr"}
+    )
+
+    assert response.status_code == 202
+    assert response.headers["cache-control"] == "no-store"
+
+
+async def test_forgot_password_answers_202_identically_when_the_account_is_unknown(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post(
+        "/api/v1/auth/forgot-password", json={"email": "inconnu@enervision.fr"}
+    )
+
+    assert response.status_code == 202
+
+
+async def test_forgot_password_returns_429_with_a_retry_after_when_the_rate_limit_is_reached(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    fake_auth_service[0] = RateLimitedError(900)
+
+    response = await client.post(
+        "/api/v1/auth/forgot-password", json={"email": "operateur@enervision.fr"}
+    )
+
+    assert response.status_code == 429
+    assert response.headers["retry-after"] == "900"
+
+
+async def test_forgot_password_rejects_a_malformed_email(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post("/api/v1/auth/forgot-password", json={"email": "pas-un-email"})
+
+    assert response.status_code == 422
+
+
+@pytest.fixture
+def fake_auth_service_reset_validity(app: FastAPI) -> Iterator[list[bool]]:
+    programme = [True]
+    app.dependency_overrides[get_auth_service] = lambda: FauxService(jeton_valide=programme[0])
+    yield programme
+    app.dependency_overrides.pop(get_auth_service, None)
+
+
+async def test_validate_reset_token_reports_a_living_token(
+    fake_auth_service_reset_validity: list[bool], client: AsyncClient
+) -> None:
+    response = await client.get(
+        "/api/v1/auth/reset-password/validate", params={"token": "un-secret-opaque"}
+    )
+
+    assert response.status_code == 200
+    assert response.json() == {"valid": True}
+
+
+async def test_validate_reset_token_reports_an_invalid_or_expired_token(
+    fake_auth_service_reset_validity: list[bool], client: AsyncClient
+) -> None:
+    fake_auth_service_reset_validity[0] = False
+
+    response = await client.get(
+        "/api/v1/auth/reset-password/validate", params={"token": "un-secret-perime"}
+    )
+
+    assert response.status_code == 200
+    assert response.json() == {"valid": False}
+
+
+async def test_reset_password_returns_the_token_and_the_cookie_on_success(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post(
+        "/api/v1/auth/reset-password",
+        json={"token": "un-secret-opaque", "new_password": "Un-nouveau-mot-de-passe1!"},
+    )
+
+    assert response.status_code == 200
+    assert response.cookies.get("ev_refresh") is not None
+    assert "refresh_secret" not in response.text
+
+
+async def test_reset_password_rejects_an_invalid_or_expired_token(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    fake_auth_service[0] = InvalidOrExpiredResetTokenError("Lien invalide ou expiré")
+
+    response = await client.post(
+        "/api/v1/auth/reset-password",
+        json={"token": "un-secret-perime", "new_password": "Un-nouveau-mot-de-passe1!"},
+    )
+
+    assert response.status_code == 400
+
+
+async def test_reset_password_rejects_a_weak_password(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post(
+        "/api/v1/auth/reset-password",
+        json={"token": "un-secret-opaque", "new_password": "trop-simple"},
+    )
+
+    assert response.status_code == 422
+
+
+async def test_reset_password_refuses_a_foreign_origin(
+    fake_auth_service: list[Exception | None], client: AsyncClient
+) -> None:
+    response = await client.post(
+        "/api/v1/auth/reset-password",
+        json={"token": "un-secret-opaque", "new_password": "Un-nouveau-mot-de-passe1!"},
+        headers={"Origin": "https://malveillant.example"},
+    )
+
+    assert response.status_code == 403

apps/backend/tests/api/test_matrice_acces.py

@@ -0,0 +1,279 @@
+# Pourquoi : la matrice rôle x route sur les routes réelles. `test_authorization.py` la joue déjà,
+# mais contre une route jetable montée par une fixture, ce qui ne dit rien du niveau effectivement
+# posé sur `/sites` ou `/users`. `ROLE_MINIMUM` (tests/api/acces.py) est la référence, et ce
+# fichier est ce qui la confronte au comportement observé.
+# Piège : l'assertion porte sur le refus de la garde, pas sur un 200. Un rôle suffisant peut
+# légitimement recevoir 404 ou 422 selon les données ; ce qui compte est qu'il ne reçoive pas le
+# 403 `Droits insuffisants`. Sans cette nuance, le test dépendrait du contenu de la base.
+# Les tests `integration` en fin de fichier rejouent la même matrice avec de vrais jetons, donc en
+# traversant le décodage du JWT et la relecture du compte, ce que l'override court-circuite.
+
+import uuid
+from collections.abc import AsyncIterator, Callable, Iterator
+
+import pytest
+from fastapi import FastAPI
+from httpx import AsyncClient, Response
+from sqlalchemy import text
+
+from app.api.deps import get_current_principal
+from app.core.hashing import build_hasher
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role, has_at_least
+from app.db.session import get_session, get_session_factory
+from app.repositories.user import UserRepository
+from tests.api.acces import ROLE_MINIMUM, chemin_concret
+
+ROLES = [Role.LECTEUR, Role.OPERATEUR, Role.ADMIN]
+IDS_DE_ROLE = ["lecteur", "operateur", "admin"]
+REFUS_DE_DROITS = "Droits insuffisants"
+REFUS_DE_MOT_DE_PASSE = "password_change_required"
+MOT_DE_PASSE = "un-mot-de-passe-de-recette"
+
+
+# `FakeSession` de tests/factories.py rend un unique objet pour les trois formes d'appel, ce qui
+# suffit à un test d'endpoint ciblé mais pas à balayer 13 routes qui interrogent chacune la base
+# à sa façon. Ce double rend un résultat vide quelle que soit la forme demandée, pour que la
+# réponse observée vienne de la garde de rôle et jamais d'un double mal ajusté.
+class ResultatVide:
+    def scalars(self) -> ResultatVide:
+        return self
+
+    def all(self) -> list[object]:
+        return []
+
+    def first(self) -> None:
+        return None
+
+    def one_or_none(self) -> None:
+        return None
+
+    def scalar_one_or_none(self) -> None:
+        return None
+
+    def mappings(self) -> ResultatVide:
+        return self
+
+    def __iter__(self) -> Iterator[object]:
+        return iter(())
+
+
+class SessionMuette:
+    async def scalar(self, *_: object, **__: object) -> None:
+        return None
+
+    async def execute(self, *_: object, **__: object) -> ResultatVide:
+        return ResultatVide()
+
+    async def scalars(self, *_: object, **__: object) -> ResultatVide:
+        return ResultatVide()
+
+    async def get(self, *_: object, **__: object) -> None:
+        return None
+
+    async def flush(self) -> None:
+        return None
+
+    async def commit(self) -> None:
+        return None
+
+    async def rollback(self) -> None:
+        return None
+
+    def add(self, *_: object, **__: object) -> None:
+        return None
+
+
+@pytest.fixture
+def base_muette(app: FastAPI) -> None:
+    async def override() -> AsyncIterator[SessionMuette]:
+        yield SessionMuette()
+
+    app.dependency_overrides[get_session] = override
+
+
+def principal(role: Role, *, must_change_password: bool = False) -> Principal:
+    return Principal(
+        id=uuid.uuid4(),
+        email=f"matrice-{role.value}@enervision.fr",
+        role=role,
+        kind=AccountKind.HUMAIN,
+        must_change_password=must_change_password,
+    )
+
+
+@pytest.fixture
+def connecte(app: FastAPI) -> Iterator[Callable[[Principal], None]]:
+    def installe(acteur: Principal) -> None:
+        app.dependency_overrides[get_current_principal] = lambda: acteur
+
+    yield installe
+    app.dependency_overrides.pop(get_current_principal, None)
+
+
+async def appelle(client: AsyncClient, methode: str, chemin: str, **kwargs: object) -> Response:
+    return await client.request(methode, chemin_concret(chemin), json={}, **kwargs)  # type: ignore[arg-type]
+
+
+def motif_du_refus(response: Response) -> str | None:
+    if response.status_code != 403:
+        return None
+    detail = response.json().get("detail")
+    return detail if isinstance(detail, str) else None
+
+
+@pytest.mark.parametrize("role", ROLES, ids=IDS_DE_ROLE)
+async def test_a_role_below_the_minimum_is_refused_on_every_guarded_route(
+    connecte: Callable[[Principal], None],
+    client: AsyncClient,
+    base_muette: None,
+    role: Role,
+) -> None:
+    connecte(principal(role))
+    laissees_passer: list[tuple[str, str, int]] = []
+
+    for (methode, chemin), minimum in ROLE_MINIMUM.items():
+        if has_at_least(role, minimum):
+            continue
+        response = await appelle(client, methode, chemin)
+        if motif_du_refus(response) != REFUS_DE_DROITS:
+            laissees_passer.append((methode, chemin, response.status_code))
+
+    assert laissees_passer == []
+
+
+# Le pendant du test précédent : sans lui, une garde posée trop haut, par exemple `AdminDep` sur
+# `/sites`, ne ferait échouer aucun test du dépôt.
+@pytest.mark.parametrize("role", ROLES, ids=IDS_DE_ROLE)
+async def test_a_role_at_or_above_the_minimum_is_never_refused_by_the_guard(
+    connecte: Callable[[Principal], None],
+    client: AsyncClient,
+    base_muette: None,
+    role: Role,
+) -> None:
+    connecte(principal(role))
+    refusees: list[tuple[str, str]] = []
+
+    for (methode, chemin), minimum in ROLE_MINIMUM.items():
+        if not has_at_least(role, minimum):
+            continue
+        response = await appelle(client, methode, chemin)
+        if motif_du_refus(response) == REFUS_DE_DROITS:
+            refusees.append((methode, chemin))
+
+    assert refusees == []
+
+
+async def test_a_pending_password_change_is_refused_on_every_guarded_route(
+    connecte: Callable[[Principal], None],
+    client: AsyncClient,
+    base_muette: None,
+) -> None:
+    connecte(principal(Role.ADMIN, must_change_password=True))
+    laissees_passer: list[tuple[str, str, int]] = []
+
+    for methode, chemin in ROLE_MINIMUM:
+        response = await appelle(client, methode, chemin)
+        if motif_du_refus(response) != REFUS_DE_MOT_DE_PASSE:
+            laissees_passer.append((methode, chemin, response.status_code))
+
+    assert laissees_passer == []
+
+
+@pytest.fixture
+async def comptes_par_role() -> AsyncIterator[dict[Role, str]]:
+    marque = uuid.uuid4().hex[:12]
+    hacheur = build_hasher(time_cost=1, memory_cost_kib=8192, parallelism=1, max_concurrency=2)
+    empreinte = await hacheur.hash(MOT_DE_PASSE)
+    adresses = {role: f"matrice-{marque}-{role.value}@enervision.fr" for role in ROLES}
+
+    async with get_session_factory()() as session:
+        depot = UserRepository(session)
+        for role, email in adresses.items():
+            await depot.create(email=email, password_hash=empreinte, role=role)
+        await session.commit()
+
+    yield adresses
+
+    async with get_session_factory()() as session:
+        await session.execute(
+            text("delete from app_user where email like :motif"), {"motif": f"matrice-{marque}-%"}
+        )
+        await session.commit()
+
+
+async def authentifie(client: AsyncClient, email: str) -> dict[str, str]:
+    reponse = await client.post(
+        "/api/v1/auth/login", json={"email": email, "password": MOT_DE_PASSE}
+    )
+    assert reponse.status_code == 200, reponse.text
+    return {"Authorization": f"Bearer {reponse.json()['access_token']}"}
+
+
+@pytest.mark.integration
+@pytest.mark.parametrize("role", ROLES, ids=IDS_DE_ROLE)
+async def test_a_real_token_reaches_exactly_the_routes_of_its_rank(
+    comptes_par_role: dict[Role, str], client: AsyncClient, role: Role
+) -> None:
+    entetes = await authentifie(client, comptes_par_role[role])
+    ecarts: list[tuple[str, str, int, str]] = []
+
+    for (methode, chemin), minimum in ROLE_MINIMUM.items():
+        response = await appelle(client, methode, chemin, headers=entetes)
+        refuse = motif_du_refus(response) == REFUS_DE_DROITS
+        if refuse is has_at_least(role, minimum):
+            ecarts.append((methode, chemin, response.status_code, response.text[:120]))
+
+    assert ecarts == []
+
+
+# Contrainte : `operateur` n'ouvre aujourd'hui aucune route de plus que `lecteur`, faute d'écriture
+# métier dans l'API. Figer l'égalité rend la régression visible le jour où une route d'opérateur
+# arrive sans que `ROLE_MINIMUM` soit mis à jour.
+@pytest.mark.integration
+async def test_the_operator_rank_opens_nothing_more_than_the_reader_rank(
+    comptes_par_role: dict[Role, str], client: AsyncClient
+) -> None:
+    lecteur = await authentifie(client, comptes_par_role[Role.LECTEUR])
+    operateur = await authentifie(client, comptes_par_role[Role.OPERATEUR])
+    divergences: list[tuple[str, str]] = []
+
+    for methode, chemin in ROLE_MINIMUM:
+        cote_lecteur = await appelle(client, methode, chemin, headers=lecteur)
+        cote_operateur = await appelle(client, methode, chemin, headers=operateur)
+        if cote_lecteur.status_code != cote_operateur.status_code:
+            divergences.append((methode, chemin))
+
+    assert divergences == []
+
+
+# Piège : `/auth/logout-all` prend un `CurrentPrincipalDep` nu, donc elle échappe au gate
+# `must_change_password` que seul `require_role` applique. Comportement figé ici, pas corrigé.
+@pytest.mark.integration
+async def test_a_temporary_password_blocks_the_business_routes_but_not_logout_all(
+    client: AsyncClient,
+) -> None:
+    marque = uuid.uuid4().hex[:12]
+    email = f"matrice-{marque}-provisoire@enervision.fr"
+    hacheur = build_hasher(time_cost=1, memory_cost_kib=8192, parallelism=1, max_concurrency=2)
+    empreinte = await hacheur.hash(MOT_DE_PASSE)
+
+    async with get_session_factory()() as session:
+        await UserRepository(session).create(
+            email=email, password_hash=empreinte, role=Role.ADMIN, must_change_password=True
+        )
+        await session.commit()
+
+    try:
+        entetes = await authentifie(client, email)
+        sites = await client.get("/api/v1/sites", headers=entetes)
+        identite = await client.get("/api/v1/auth/me", headers=entetes)
+        fermeture = await client.post("/api/v1/auth/logout-all", headers=entetes)
+
+        assert motif_du_refus(sites) == REFUS_DE_MOT_DE_PASSE
+        assert identite.status_code == 200
+        assert fermeture.status_code == 204
+    finally:
+        async with get_session_factory()() as session:
+            await session.execute(text("delete from app_user where email = :e"), {"e": email})
+            await session.commit()

apps/backend/tests/api/test_openapi.py

@@ -8,6 +8,7 @@ from typing import Any
 import pytest
 
 from app import cli
+from tests.api.acces import ROLE_MINIMUM
 
 METHODES = {"get", "post", "patch", "put", "delete"}
 
@@ -24,19 +25,11 @@ ORIGINE_VERIFIEE = {
 
 # Toute route derrière `require_role` (LecteurDep, OperateurDep, AdminDep) peut rendre 403 pour
 # `password_change_required`, pas seulement les routes `admin`.
-ROUTES_A_ROLE = {
-    ("GET", "/api/v1/users"),
-    ("POST", "/api/v1/users"),
-    ("PATCH", "/api/v1/users/{id}"),
-    ("POST", "/api/v1/users/{id}/password-reset"),
-    ("GET", "/api/v1/sites"),
-    ("GET", "/api/v1/sites/{site_id}"),
-    ("GET", "/api/v1/alerts"),
-    ("GET", "/api/v1/recommendations"),
-    ("GET", "/api/v1/recommendations/{recommendation_id}"),
-    ("GET", "/api/v1/stats/summary"),
-    ("GET", "/api/v1/sensors/status"),
-}
+# Piège : cette liste était recopiée ici, et deux de ses entrées portaient `{id}` là où le contrat
+# expose `{user_id}`. Elles ne correspondaient donc à aucune opération, et le test ci-dessous
+# passait au vert sans rien vérifier sur ces deux routes. Elle est maintenant dérivée, et
+# `test_every_declared_route_is_classified` interdit l'entrée morte.
+ROUTES_A_ROLE = frozenset(ROLE_MINIMUM)
 
 
 @pytest.fixture(scope="module")

apps/backend/tests/api/test_predictions.py

@@ -0,0 +1,85 @@
+from collections.abc import Callable, Iterator
+from datetime import UTC, datetime
+from uuid import uuid4
+
+import pytest
+from fastapi import FastAPI
+from httpx import AsyncClient
+
+from app.api.deps import get_current_principal, get_prediction_service
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role
+from app.services.prediction import PredictionSummary, SitePrediction, SitePredictionSummary
+
+TARGET_AT = datetime(2026, 9, 16, 13, 0, tzinfo=UTC)
+CREATED_AT = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
+
+
+def lecteur() -> Principal:
+    # Le garde-fou de rôle (`lecteur` minimum) est déjà couvert par l'ensemble `ROUTES_A_ROLE`
+    # de `tests/api/test_openapi.py` : pas besoin ici d'un paramètre de rôle jamais appelé avec
+    # autre chose que sa valeur par défaut.
+    return Principal(
+        id=uuid4(),
+        email="lecteur@enervision.fr",
+        role=Role.LECTEUR,
+        kind=AccountKind.HUMAIN,
+        must_change_password=False,
+    )
+
+
+class FauxService:
+    def __init__(self) -> None:
+        self.resume = PredictionSummary(
+            timestamp=datetime.now(UTC),
+            sites=[
+                SitePredictionSummary(
+                    site_id="SITE001",
+                    site_name="Bureau Paris La Défense",
+                    prediction=SitePrediction(
+                        target_at=TARGET_AT,
+                        target_metric="consumption_kwh",
+                        period_minutes=60,
+                        predicted_value=812.5,
+                        status="available",
+                        failure_reason=None,
+                        model_reference="lightgbm-abc123",
+                        created_at=CREATED_AT,
+                    ),
+                ),
+                SitePredictionSummary(site_id="SITE002", site_name="Usine Lyon", prediction=None),
+            ],
+        )
+
+    async def summary(self) -> PredictionSummary:
+        return self.resume
+
+
+@pytest.fixture
+def servi(app: FastAPI) -> Iterator[Callable[[], FauxService]]:
+    def installe() -> FauxService:
+        service = FauxService()
+        app.dependency_overrides[get_prediction_service] = lambda: service
+        app.dependency_overrides[get_current_principal] = lambda: lecteur()
+        return service
+
+    yield installe
+    app.dependency_overrides.pop(get_prediction_service, None)
+    app.dependency_overrides.pop(get_current_principal, None)
+
+
+async def test_get_predictions_returns_the_service_result(
+    servi: Callable[[], FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.get("/api/v1/predictions")
+
+    assert response.status_code == 200
+    corps = response.json()
+    premier, second = corps["sites"]
+    assert premier["site_id"] == "SITE001"
+    assert premier["prediction"]["predicted_value"] == 812.5
+    assert premier["prediction"]["status"] == "available"
+    assert second["site_id"] == "SITE002"
+    assert second["prediction"] is None

apps/backend/tests/api/test_readings.py

@@ -0,0 +1,198 @@
+from collections.abc import Callable, Iterator
+from datetime import UTC, datetime
+from uuid import uuid4
+
+import pytest
+from fastapi import FastAPI
+from httpx import AsyncClient
+
+from app.api.deps import get_current_principal, get_reading_service
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role
+from app.models.energy import Reading
+from app.services.reading import FenetreInverseeError, FenetreTropLargeError
+
+
+def principal(role: Role = Role.LECTEUR) -> Principal:
+    return Principal(
+        id=uuid4(),
+        email=f"{role.value}@enervision.fr",
+        role=role,
+        kind=AccountKind.HUMAIN,
+        must_change_password=False,
+    )
+
+
+def reading(reading_id: int = 1, site_id: str = "site-1") -> Reading:
+    return Reading(
+        reading_id=reading_id,
+        site_id=site_id,
+        timestamp=datetime(2026, 9, 16, tzinfo=UTC),
+        source="api_current",
+        consumption_kw=42.5,
+        consumption_kwh=None,
+        consumption_euros=None,
+        voltage_v=230.0,
+        current_a=None,
+        power_factor=None,
+        temperature_celsius=None,
+        humidity_percent=None,
+        solar_irradiance_wm2=None,
+        is_working_hours=True,
+        data_quality="good",
+        null_reasons=None,
+        imputed_values=None,
+        imputation_method=None,
+        raw_data={},
+    )
+
+
+class FauxService:
+    def __init__(self, leve: Exception | None = None) -> None:
+        self.reading = reading()
+        self.leve = leve
+        self.appels: list[tuple[str | None, str | None, str | None, int, int]] = []
+
+    async def list_history(
+        self,
+        *,
+        site_id: str | None = None,
+        start: datetime | None = None,
+        end: datetime | None = None,
+        limit: int,
+        offset: int,
+    ) -> list[Reading]:
+        self.appels.append((site_id, start, end, limit, offset))
+        if self.leve is not None:
+            raise self.leve
+        return [self.reading]
+
+
+@pytest.fixture
+def lecteur_connecte(app: FastAPI) -> Iterator[None]:
+    app.dependency_overrides[get_current_principal] = lambda: principal()
+    yield
+    app.dependency_overrides.pop(get_current_principal, None)
+
+
+@pytest.fixture
+def servi(app: FastAPI, lecteur_connecte: None) -> Iterator[Callable[..., FauxService]]:
+    def installe(*, leve: Exception | None = None) -> FauxService:
+        service = FauxService(leve=leve)
+        app.dependency_overrides[get_reading_service] = lambda: service
+        return service
+
+    yield installe
+    app.dependency_overrides.pop(get_reading_service, None)
+
+
+async def test_list_readings_returns_the_readings(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.get("/api/v1/readings")
+
+    assert response.status_code == 200
+    corps = response.json()
+    assert corps == [
+        {
+            "reading_id": 1,
+            "site_id": "site-1",
+            "timestamp": "2026-09-16T00:00:00Z",
+            "source": "api_current",
+            "consumption_kw": 42.5,
+            "consumption_kwh": None,
+            "consumption_euros": None,
+            "voltage_v": 230.0,
+            "current_a": None,
+            "power_factor": None,
+            "temperature_celsius": None,
+            "humidity_percent": None,
+            "solar_irradiance_wm2": None,
+            "is_working_hours": True,
+            "data_quality": "good",
+            "null_reasons": None,
+            "imputed_values": None,
+            "imputation_method": None,
+        }
+    ]
+
+
+async def test_list_readings_transmits_the_filters_and_pagination(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    service = servi()
+
+    response = await client.get(
+        "/api/v1/readings",
+        params={
+            "site_id": "site-1",
+            "start": "2026-09-01T00:00:00Z",
+            "end": "2026-09-02T00:00:00Z",
+            "limit": 50,
+            "offset": 10,
+        },
+    )
+
+    assert response.status_code == 200
+    assert service.appels == [
+        (
+            "site-1",
+            datetime(2026, 9, 1, tzinfo=UTC),
+            datetime(2026, 9, 2, tzinfo=UTC),
+            50,
+            10,
+        )
+    ]
+
+
+async def test_list_readings_returns_400_when_the_window_is_inverted(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi(leve=FenetreInverseeError())
+
+    response = await client.get("/api/v1/readings")
+
+    assert response.status_code == 400
+
+
+async def test_list_readings_returns_400_when_the_window_is_too_large(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi(leve=FenetreTropLargeError())
+
+    response = await client.get("/api/v1/readings")
+
+    assert response.status_code == 400
+
+
+async def test_list_readings_returns_422_for_a_limit_above_the_maximum(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.get("/api/v1/readings", params={"limit": 5000})
+
+    assert response.status_code == 422
+
+
+async def test_list_readings_returns_422_for_a_negative_offset(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.get("/api/v1/readings", params={"offset": -1})
+
+    assert response.status_code == 422
+
+
+async def test_list_readings_returns_an_empty_list_when_there_is_nothing(
+    lecteur_connecte: None, fake_session: Callable[..., None], client: AsyncClient
+) -> None:
+    fake_session(result=[])
+
+    response = await client.get("/api/v1/readings")
+
+    assert response.status_code == 200
+    assert response.json() == []

apps/backend/tests/api/test_recommendations.py

@@ -10,7 +10,7 @@ from app.api.deps import get_current_principal, get_recommendation_service
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role
 from app.models.energy import Recommendation
-from app.services.recommendation import RecommendationNotFoundError
+from app.services.recommendation import RapportGeneration, RecommendationNotFoundError
 
 MOMENT = datetime(2024, 1, 1, tzinfo=UTC)
 
@@ -40,6 +40,7 @@ class FauxService:
     def __init__(self, erreur: Exception | None = None) -> None:
         self._erreur = erreur
         self.recommendation = recommendation()
+        self.site_demande: str | None = None
 
     async def list_all(self) -> list[Recommendation]:
         return [self.recommendation]
@@ -49,6 +50,10 @@ class FauxService:
             raise self._erreur
         return self.recommendation
 
+    async def generate(self, *, site_id: str | None = None) -> RapportGeneration:
+        self.site_demande = site_id
+        return RapportGeneration(alertes_examinees=2, recommandations_creees=3, deja_presentes=1)
+
 
 @pytest.fixture
 def lecteur_connecte(app: FastAPI) -> Iterator[None]:
@@ -142,3 +147,56 @@ async def test_get_recommendation_returns_404_when_the_session_finds_nothing(
     response = await client.get("/api/v1/recommendations/404")
 
     assert response.status_code == 404
+
+
+@pytest.fixture
+def admin_connecte(app: FastAPI) -> Iterator[None]:
+    app.dependency_overrides[get_current_principal] = lambda: principal(Role.ADMIN)
+    yield
+    app.dependency_overrides.pop(get_current_principal, None)
+
+
+@pytest.fixture
+def servi_en_admin(app: FastAPI, admin_connecte: None) -> Iterator[Callable[[], FauxService]]:
+    def installe() -> FauxService:
+        service = FauxService()
+        app.dependency_overrides[get_recommendation_service] = lambda: service
+        return service
+
+    yield installe
+    app.dependency_overrides.pop(get_recommendation_service, None)
+
+
+async def test_generate_recommendations_returns_the_generation_report(
+    servi_en_admin: Callable[[], FauxService], client: AsyncClient
+) -> None:
+    servi_en_admin()
+
+    response = await client.post("/api/v1/recommendations/generate")
+
+    assert response.status_code == 200
+    assert response.json() == {
+        "alerts_examined": 2,
+        "recommendations_created": 3,
+        "already_present": 1,
+    }
+
+
+async def test_generate_recommendations_forwards_the_requested_site(
+    servi_en_admin: Callable[[], FauxService], client: AsyncClient
+) -> None:
+    service = servi_en_admin()
+
+    await client.post("/api/v1/recommendations/generate", params={"site_id": "SITE002"})
+
+    assert service.site_demande == "SITE002"
+
+
+async def test_generate_recommendations_refuses_a_reader(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.post("/api/v1/recommendations/generate")
+
+    assert response.status_code == 403

apps/backend/tests/api/test_route_protection.py

@@ -1,6 +1,6 @@
 # Ce test est le garde-fou de l'autorisation : rendre une route publique oblige à modifier
-# `ROUTES_PUBLIQUES` ci-dessous, ce qui apparaît en clair dans la diff d'une pull request et
-# demande une justification au relecteur.
+# `ROUTES_PUBLIQUES` dans `tests/api/acces.py`, ce qui apparaît en clair dans la diff d'une pull
+# request et demande une justification au relecteur.
 # Pourquoi : il interroge réellement chaque route sans jeton au lieu d'inspecter l'arbre de
 # dépendances. L'arbre n'est accessible que par l'API privée de FastAPI, et surtout une route
 # peut porter la bonne dépendance tout en répondant quand même.
@@ -11,51 +11,64 @@ import pytest
 from fastapi import FastAPI
 from httpx import AsyncClient
 
-ROUTES_PUBLIQUES = frozenset(
-    {
-        ("GET", "/api/v1/health/live"),
-        ("GET", "/api/v1/health/ready"),
-        ("POST", "/api/v1/auth/login"),
-        # Sans cookie, la déconnexion ne fait rien et répond 204 : elle est idempotente.
-        ("POST", "/api/v1/auth/logout"),
-        ("GET", "/metrics"),
-    }
+from tests.api.acces import (
+    ROLE_MINIMUM,
+    ROUTE_COOKIE,
+    ROUTES_PUBLIQUES,
+    ROUTES_SANS_ROLE,
+    Route,
+    chemin_concret,
+    routes_du_schema,
 )
 
-VALEURS_DE_SUBSTITUTION = "00000000-0000-0000-0000-000000000000"
 STATUTS_DE_REFUS = {401, 403}
+HORS_SCHEMA = {("GET", "/metrics")}
 
 
-def routes_declarees(app: FastAPI) -> list[tuple[str, str]]:
+def routes_declarees(app: FastAPI) -> list[Route]:
     schema: dict[str, Any] = app.openapi()
-    return [
-        (methode.upper(), chemin)
-        for chemin, operations in schema["paths"].items()
-        for methode in operations
-        if methode.upper() in {"GET", "POST", "PATCH", "PUT", "DELETE"}
-    ]
+    return routes_du_schema(schema)
 
 
-def routes_protegees(app: FastAPI) -> list[tuple[str, str]]:
+def routes_protegees(app: FastAPI) -> list[Route]:
     return [route for route in routes_declarees(app) if route not in ROUTES_PUBLIQUES]
 
 
 def test_the_public_allow_list_has_no_stale_entry(app: FastAPI) -> None:
-    declarees = set(routes_declarees(app)) | {("GET", "/metrics")}
+    declarees = set(routes_declarees(app)) | HORS_SCHEMA
 
     inconnues = ROUTES_PUBLIQUES - declarees
 
     assert inconnues == set()
 
 
+# Sans lui, une route ajoutée sans être classée n'est vue par aucun test de rôle : elle hérite
+# du seul contrôle anonyme, et une garde posée au mauvais niveau passe inaperçue.
+def test_every_declared_route_is_classified(app: FastAPI) -> None:
+    classees = ROUTES_PUBLIQUES | ROUTE_COOKIE | ROUTES_SANS_ROLE | set(ROLE_MINIMUM)
+
+    non_classees = set(routes_declarees(app)) - classees
+    fantomes = classees - set(routes_declarees(app)) - HORS_SCHEMA
+
+    assert non_classees == set(), "classer la route dans tests/api/acces.py"
+    assert fantomes == set(), "entrée morte : la route n'existe plus sous ce chemin"
+
+
+def test_the_four_classes_of_routes_stay_disjoint() -> None:
+    classes = [ROUTES_PUBLIQUES, ROUTE_COOKIE, ROUTES_SANS_ROLE, frozenset(ROLE_MINIMUM)]
+
+    for rang, classe in enumerate(classes):
+        for autre in classes[rang + 1 :]:
+            assert classe & autre == frozenset()
+
+
 async def test_every_route_rejects_an_anonymous_caller_unless_explicitly_public(
     app: FastAPI, client: AsyncClient
 ) -> None:
     ouvertes: list[tuple[str, str, int]] = []
 
     for methode, chemin in routes_protegees(app):
-        concret = chemin.replace("{user_id}", VALEURS_DE_SUBSTITUTION)
-        response = await client.request(methode, concret, json={})
+        response = await client.request(methode, chemin_concret(chemin), json={})
         if response.status_code not in STATUTS_DE_REFUS:
             ouvertes.append((methode, chemin, response.status_code))
 
@@ -74,3 +87,19 @@ async def test_the_declared_routes_are_actually_reachable(app: FastAPI) -> None:
 )
 def test_the_health_probes_stay_public(app: FastAPI, chemin: str) -> None:
     assert ("GET", chemin) in ROUTES_PUBLIQUES
+
+
+# Piège : ni les routes `include_in_schema=False` (/docs, /redoc) ni un `Mount` Starlette
+# (/static) n'apparaissent dans `app.openapi()["paths"]`. `routes_declarees()` ne les voit
+# donc jamais, et elles échapperaient silencieusement au garde-fou ci-dessus.
+@pytest.mark.parametrize(
+    "chemin",
+    ["/docs", "/redoc", "/static/logo-icon.png"],
+    ids=["swagger_ui", "redoc", "logo_statique"],
+)
+async def test_the_documentation_routes_are_public_by_design(
+    app: FastAPI, client: AsyncClient, chemin: str
+) -> None:
+    response = await client.get(chemin)
+
+    assert response.status_code == 200

apps/backend/tests/api/test_sites.py

@@ -1,4 +1,5 @@
 from collections.abc import Callable, Iterator
+from datetime import UTC, datetime
 from uuid import uuid4
 
 import pytest
@@ -9,7 +10,9 @@ from app.api.deps import get_current_principal, get_site_service
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role
 from app.models.energy import Site
-from app.services.site import SiteNotFoundError
+from app.services.site import SiteCurrentReading, SiteNotFoundError
+
+TIMESTAMP = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
 
 
 def principal(role: Role = Role.LECTEUR) -> Principal:
@@ -33,10 +36,28 @@ def site(site_id: str = "site-1") -> Site:
     )
 
 
+def lecture_actuelle(site_id: str = "site-1") -> SiteCurrentReading:
+    return SiteCurrentReading(
+        timestamp=TIMESTAMP,
+        site_id=site_id,
+        site_type="industriel",
+        consumption_kw=87.34,
+        consumption_kwh=87.34,
+        voltage_v=401.2,
+        current_a=132.5,
+        power_factor=0.923,
+        temperature_celsius=22.1,
+        humidity_percent=58.4,
+        null_reasons=[],
+        data_quality="good",
+    )
+
+
 class FauxService:
     def __init__(self, erreur: Exception | None = None) -> None:
         self._erreur = erreur
         self.site = site()
+        self.actuel = lecture_actuelle()
 
     async def list_all(self) -> list[Site]:
         return [self.site]
@@ -46,6 +67,11 @@ class FauxService:
             raise self._erreur
         return self.site
 
+    async def current(self, site_id: str) -> SiteCurrentReading:
+        if self._erreur is not None:
+            raise self._erreur
+        return self.actuel
+
 
 @pytest.fixture
 def lecteur_connecte(app: FastAPI) -> Iterator[None]:
@@ -109,6 +135,30 @@ async def test_get_site_returns_404_for_an_unknown_site(
     assert response.status_code == 404
 
 
+async def test_get_current_returns_the_latest_reading(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi()
+
+    response = await client.get("/api/v1/sites/site-1/current")
+
+    assert response.status_code == 200
+    corps = response.json()
+    assert corps["site_id"] == "site-1"
+    assert corps["data_quality"] == "good"
+    assert corps["consumption_kw"] == 87.34
+
+
+async def test_get_current_returns_404_for_an_unknown_site(
+    servi: Callable[..., FauxService], client: AsyncClient
+) -> None:
+    servi(SiteNotFoundError("site-inconnu"))
+
+    response = await client.get("/api/v1/sites/site-inconnu/current")
+
+    assert response.status_code == 404
+
+
 async def test_list_sites_reaches_the_repository_through_the_session(
     lecteur_connecte: None, fake_session: Callable[..., None], client: AsyncClient
 ) -> None:

apps/backend/tests/db/test_data_schema.py

@@ -112,8 +112,10 @@ async def test_duplicate_reading_is_rejected_when_key_matches(
     )
     await data_connection.execute(statement)
 
+    savepoint = data_connection.begin_nested()
+
     with pytest.raises(IntegrityError):
-        async with data_connection.begin_nested():
+        async with savepoint:
             await data_connection.execute(statement)
 
 
@@ -147,9 +149,12 @@ async def test_invalid_reading_is_rejected_when_constraints_fail(
     }
     values.update(changes)
 
+    statement = insert(Reading).values(**values)
+    savepoint = data_connection.begin_nested()
+
     with pytest.raises(IntegrityError):
-        async with data_connection.begin_nested():
-            await data_connection.execute(insert(Reading).values(**values))
+        async with savepoint:
+            await data_connection.execute(statement)
 
 
 async def test_prediction_requires_period_when_energy_is_predicted(
@@ -164,8 +169,10 @@ async def test_prediction_requires_period_when_energy_is_predicted(
         model_reference="test-model/1",
     )
 
+    savepoint = data_connection.begin_nested()
+
     with pytest.raises(IntegrityError):
-        async with data_connection.begin_nested():
+        async with savepoint:
             await data_connection.execute(statement)
 
 
@@ -212,21 +219,22 @@ async def test_alert_rejects_prediction_when_site_differs(
         )
     ).scalar_one()
 
+    statement = insert(Alert).values(
+        source_alert_id=str(uuid4()),
+        site_id=other_site,
+        source="enervision",
+        timestamp=MOMENT,
+        type="spike",
+        severity="high",
+        message="Test",
+        prediction_id=prediction_id,
+        raw_data={},
+    )
+    savepoint = data_connection.begin_nested()
+
     with pytest.raises(IntegrityError):
-        async with data_connection.begin_nested():
-            await data_connection.execute(
-                insert(Alert).values(
-                    source_alert_id=str(uuid4()),
-                    site_id=other_site,
-                    source="enervision",
-                    timestamp=MOMENT,
-                    type="spike",
-                    severity="high",
-                    message="Test",
-                    prediction_id=prediction_id,
-                    raw_data={},
-                )
-            )
+        async with savepoint:
+            await data_connection.execute(statement)
 
 
 async def test_recommendation_is_unique_when_alert_and_rule_match(
@@ -256,6 +264,8 @@ async def test_recommendation_is_unique_when_alert_and_rule_match(
     )
     await data_connection.execute(statement)
 
+    savepoint = data_connection.begin_nested()
+
     with pytest.raises(IntegrityError):
-        async with data_connection.begin_nested():
+        async with savepoint:
             await data_connection.execute(statement)

apps/backend/tests/etl/test_historical_import.py

@@ -0,0 +1,245 @@
+import hashlib
+import json
+
+import pandas as pd
+import pytest
+
+from app.etl.historical_import import (
+    SOURCE_NAME,
+    build_reading_batch,
+    classify_quality,
+    compute_sha256,
+    load_metadata,
+    normalize_timestamps,
+    validate_source,
+)
+
+
+def make_metadata() -> dict:
+    return {
+        "total_records": 2,
+        "sites": {
+            "SITE001": {},
+        },
+    }
+
+
+def make_dataframe() -> pd.DataFrame:
+    return pd.DataFrame(
+        [
+            {
+                "timestamp": "2023-01-01 00:00:00",
+                "site_id": "SITE001",
+                "site_type": "office",
+                "site_name": "Site 1",
+                "consumption_kwh": 10.5,
+                "consumption_euros": 2.5,
+                "temperature_celsius": 20.0,
+                "humidity_percent": 50.0,
+                "solar_irradiance_wm2": 0.0,
+                "hour": 0,
+                "day_of_week": 6,
+                "day_name": "Sunday",
+                "month": 1,
+                "is_weekend": True,
+                "is_working_hours": False,
+            },
+            {
+                "timestamp": "2023-01-01 01:00:00",
+                "site_id": "SITE001",
+                "site_type": "office",
+                "site_name": "Site 1",
+                "consumption_kwh": 11.0,
+                "consumption_euros": 2.7,
+                "temperature_celsius": 19.5,
+                "humidity_percent": 52.0,
+                "solar_irradiance_wm2": 0.0,
+                "hour": 1,
+                "day_of_week": 6,
+                "day_name": "Sunday",
+                "month": 1,
+                "is_weekend": True,
+                "is_working_hours": False,
+            },
+        ]
+    )
+
+
+def test_compute_sha256(tmp_path):
+    file_path = tmp_path / "dataset.csv"
+    content = b"hello-enervision"
+
+    file_path.write_bytes(content)
+
+    expected = hashlib.sha256(content).hexdigest()
+
+    assert compute_sha256(file_path) == expected
+
+
+def test_load_metadata(tmp_path):
+    metadata_path = tmp_path / "metadata.json"
+
+    metadata = {
+        "total_records": 2,
+        "sites": {
+            "SITE001": {},
+        },
+    }
+
+    metadata_path.write_text(
+        json.dumps(metadata),
+        encoding="utf-8",
+    )
+
+    assert load_metadata(metadata_path) == metadata
+
+
+def test_validate_source_accepts_valid_dataset():
+    frame = make_dataframe()
+
+    validate_source(
+        frame,
+        make_metadata(),
+    )
+
+
+def test_validate_source_rejects_missing_column():
+    frame = make_dataframe().drop(columns=["consumption_kwh"])
+
+    metadata = make_metadata()
+
+    with pytest.raises(
+        ValueError,
+        match="Colonnes obligatoires absentes",
+    ):
+        validate_source(
+            frame,
+            metadata,
+        )
+
+
+def test_validate_source_rejects_duplicates():
+    frame = make_dataframe()
+
+    frame.loc[1, "timestamp"] = frame.loc[
+        0,
+        "timestamp",
+    ]
+
+    metadata = make_metadata()
+
+    with pytest.raises(
+        ValueError,
+        match="doublons",
+    ):
+        validate_source(
+            frame,
+            metadata,
+        )
+
+
+def test_validate_source_rejects_unknown_site():
+    frame = make_dataframe()
+
+    frame.loc[1, "site_id"] = "SITE999"
+
+    metadata = make_metadata()
+
+    with pytest.raises(
+        ValueError,
+        match="Sites incohérents",
+    ):
+        validate_source(
+            frame,
+            metadata,
+        )
+
+
+def test_normalize_timestamps_adds_timezone():
+    frame = make_dataframe()
+
+    normalized = normalize_timestamps(
+        frame,
+        "UTC",
+    )
+
+    assert normalized["timestamp"].dt.tz is not None
+
+    assert "_source_timestamp" in normalized.columns
+
+
+def test_classify_quality_good():
+    row = make_dataframe().iloc[0].to_dict()
+
+    quality, reasons = classify_quality(row)
+
+    assert quality == "good"
+    assert reasons == []
+
+
+def test_classify_quality_degraded_when_consumption_missing():
+    row = make_dataframe().iloc[0].to_dict()
+    row["consumption_kwh"] = None
+
+    quality, reasons = classify_quality(row)
+
+    assert quality == "degraded"
+
+    assert "missing:consumption_kwh" in reasons
+
+
+def test_build_reading_batch_respects_database_contract():
+    frame = normalize_timestamps(
+        make_dataframe(),
+        "UTC",
+    )
+
+    rows = build_reading_batch(
+        frame.iloc[:1],
+        dataset_id=3,
+    )
+
+    assert len(rows) == 1
+
+    row = rows[0]
+
+    assert row["dataset_id"] == 3
+
+    # Important :
+    # contrainte ck_reading_dataset_source.
+    assert row["source"] == "csv"
+    assert SOURCE_NAME == "csv"
+
+    # Important :
+    # contrainte ck_reading_imputation.
+    assert row["imputed_values"] is None
+    assert row["imputation_method"] is None
+
+    assert row["data_quality"] == "good"
+    assert row["null_reasons"] == []
+
+
+def test_build_reading_batch_keeps_missing_values():
+    frame = make_dataframe()
+
+    frame.loc[0, "temperature_celsius"] = None
+
+    frame = normalize_timestamps(
+        frame,
+        "UTC",
+    )
+
+    rows = build_reading_batch(
+        frame.iloc[:1],
+        dataset_id=3,
+    )
+
+    row = rows[0]
+
+    assert row["temperature_celsius"] is None
+
+    assert "missing:temperature_celsius" in row["null_reasons"]
+
+    # RAW ingestion : aucune imputation.
+    assert row["imputed_values"] is None
+    assert row["imputation_method"] is None

apps/backend/tests/etl/test_mock_api_import.py

@@ -0,0 +1,834 @@
+import json
+import sys
+from datetime import datetime
+from types import SimpleNamespace
+from typing import Any
+from unittest.mock import AsyncMock, MagicMock
+
+import httpx
+import pytest
+from httpx import AsyncClient, MockTransport, Request, Response
+from sqlalchemy import text
+from sqlalchemy.ext.asyncio import AsyncSession
+
+import app.etl.mock_api_import as mock_api_import
+from app.etl.mock_api_import import (
+    MAX_SITES,
+    READING_INSERT,
+    SOURCE_HISTORY,
+    build_reading_batch,
+    build_reading_row,
+    build_site_row,
+    fetch_readings,
+    fetch_sites,
+    upsert_sites,
+)
+
+
+def make_site() -> dict[str, Any]:
+    return {
+        "site_id": "SITE001",
+        "site_type": "office",
+        "site_name": "Bureau Paris La Défense",
+        "location": "Paris, France",
+        "capacity_kw": 200,
+        "status": "active",
+    }
+
+
+def make_reading() -> dict[str, Any]:
+    return {
+        "timestamp": "2024-06-15T12:00:00Z",
+        "site_id": "SITE001",
+        "site_type": "office",
+        "consumption_kw": 87.34,
+        "consumption_kwh": 87.34,
+        "voltage_v": 401.2,
+        "current_a": 132.5,
+        "power_factor": 0.923,
+        "temperature_celsius": 22.1,
+        "humidity_percent": 58.4,
+        "null_reasons": [],
+        "data_quality": "good",
+    }
+
+
+async def test_fetch_sites_returns_sites() -> None:
+    def handler(request: Request) -> Response:
+        assert request.url.path == "/api/v1/sites"
+
+        return Response(
+            status_code=200,
+            json=[make_site()],
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        sites = await fetch_sites(client)
+
+    assert len(sites) == 1
+    assert sites[0]["site_id"] == "SITE001"
+    assert sites[0]["site_type"] == "office"
+
+
+async def test_fetch_sites_rejects_non_list_response() -> None:
+    def handler(request: Request) -> Response:
+        return Response(
+            status_code=200,
+            json={"unexpected": "payload"},
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        with pytest.raises(
+            ValueError,
+            match="La réponse /api/v1/sites doit être une liste",
+        ):
+            await fetch_sites(client)
+
+
+async def test_fetch_readings_sends_expected_query_parameters() -> None:
+    captured_params: dict[str, str] = {}
+
+    def handler(request: Request) -> Response:
+        nonlocal captured_params
+
+        captured_params = dict(request.url.params)
+
+        return Response(
+            status_code=200,
+            json=[make_reading()],
+        )
+
+    transport = MockTransport(handler)
+
+    start_time = datetime.fromisoformat("2024-06-15T12:00:00")
+    end_time = datetime.fromisoformat("2024-06-15T13:00:00")
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        readings = await fetch_readings(
+            client=client,
+            site_id="SITE001",
+            start_time=start_time,
+            end_time=end_time,
+            limit=60,
+        )
+
+    assert len(readings) == 1
+    assert captured_params["site_id"] == "SITE001"
+    assert captured_params["start_time"] == "2024-06-15T12:00:00"
+    assert captured_params["end_time"] == "2024-06-15T13:00:00"
+    assert captured_params["limit"] == "60"
+
+
+async def test_fetch_readings_rejects_non_list_response() -> None:
+    def handler(request: Request) -> Response:
+        return Response(
+            status_code=200,
+            json={"unexpected": "payload"},
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        with pytest.raises(
+            ValueError,
+            match="La réponse /api/v1/readings doit être une liste",
+        ):
+            await fetch_readings(
+                client=client,
+                site_id="SITE001",
+                start_time=datetime.fromisoformat("2024-06-15T12:00:00"),
+                end_time=datetime.fromisoformat("2024-06-15T13:00:00"),
+                limit=60,
+            )
+
+
+async def test_fetch_readings_raises_on_http_error() -> None:
+    def handler(request: Request) -> Response:
+        return Response(
+            status_code=404,
+            json={"detail": "Site non trouvé"},
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        with pytest.raises(httpx.HTTPStatusError):
+            await fetch_readings(
+                client=client,
+                site_id="SITE999",
+                start_time=datetime.fromisoformat("2024-06-15T12:00:00"),
+                end_time=datetime.fromisoformat("2024-06-15T13:00:00"),
+                limit=60,
+            )
+
+
+def test_build_reading_row_respects_database_contract() -> None:
+    reading = make_reading()
+
+    row = build_reading_row(reading)
+
+    assert row["site_id"] == "SITE001"
+    assert row["source"] == SOURCE_HISTORY
+    assert row["source"] == "api_history"
+    assert row["dataset_id"] is None
+
+    assert row["timestamp"] == datetime.fromisoformat("2024-06-15T12:00:00+00:00")
+
+    assert row["consumption_kw"] == 87.34
+    assert row["consumption_kwh"] == 87.34
+    assert row["data_quality"] == "good"
+    assert row["null_reasons"] == []
+
+    assert row["imputed_values"] is None
+    assert row["imputation_method"] is None
+
+
+def test_build_reading_row_keeps_null_values_and_quality() -> None:
+    reading = make_reading()
+
+    reading["consumption_kw"] = None
+    reading["consumption_kwh"] = None
+    reading["voltage_v"] = None
+    reading["current_a"] = None
+    reading["power_factor"] = None
+    reading["data_quality"] = "degraded"
+    reading["null_reasons"] = [
+        "consumption_sensor_failure",
+        "electrical_sensor_failure",
+    ]
+
+    row = build_reading_row(reading)
+
+    assert row["consumption_kw"] is None
+    assert row["consumption_kwh"] is None
+    assert row["voltage_v"] is None
+    assert row["current_a"] is None
+    assert row["power_factor"] is None
+
+    assert row["data_quality"] == "degraded"
+    assert row["null_reasons"] == [
+        "consumption_sensor_failure",
+        "electrical_sensor_failure",
+    ]
+
+    assert row["imputed_values"] is None
+    assert row["imputation_method"] is None
+
+
+def test_build_reading_row_keeps_raw_source_data() -> None:
+    reading = make_reading()
+
+    row = build_reading_row(reading)
+
+    raw_data = json.loads(row["raw_data"])
+
+    assert raw_data == reading
+
+
+def test_build_reading_batch_transforms_all_readings() -> None:
+    first = make_reading()
+
+    second = make_reading()
+    second["timestamp"] = "2024-06-15T12:01:00Z"
+    second["consumption_kw"] = 90.5
+
+    rows = build_reading_batch([first, second])
+
+    assert len(rows) == 2
+
+    assert rows[0]["site_id"] == "SITE001"
+    assert rows[0]["consumption_kw"] == 87.34
+
+    assert rows[1]["site_id"] == "SITE001"
+    assert rows[1]["consumption_kw"] == 90.5
+
+
+def test_create_mock_api_client_requires_credentials(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    settings = SimpleNamespace(
+        mock_api_username=None,
+        mock_api_password=None,
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "get_settings",
+        lambda: settings,
+    )
+
+    with pytest.raises(
+        ValueError,
+        match="Les identifiants de l'API Mock ne sont pas configurés",
+    ):
+        mock_api_import.create_mock_api_client()
+
+
+async def test_create_mock_api_client_uses_configuration(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    password = MagicMock()
+    password.get_secret_value.return_value = "test-password"
+
+    settings = SimpleNamespace(
+        mock_api_base_url="https://mock.test/",
+        mock_api_username="test-user",
+        mock_api_password=password,
+        mock_api_timeout_seconds=10.0,
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "get_settings",
+        lambda: settings,
+    )
+
+    client = mock_api_import.create_mock_api_client()
+
+    try:
+        assert str(client.base_url) == "https://mock.test"
+        assert client.timeout.connect == 10.0
+    finally:
+        await client.aclose()
+
+
+async def test_upsert_sites_with_empty_list_does_nothing() -> None:
+    connection = AsyncMock()
+
+    await upsert_sites(
+        connection,
+        [],
+    )
+
+    connection.execute.assert_not_awaited()
+
+
+async def test_import_mock_api_history_dry_run_does_not_write(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    def handler(request: Request) -> Response:
+        if request.url.path == "/api/v1/sites":
+            return Response(
+                status_code=200,
+                json=[make_site()],
+            )
+
+        if request.url.path == "/api/v1/readings":
+            return Response(
+                status_code=200,
+                json=[make_reading()],
+            )
+
+        return Response(status_code=404)
+
+    transport = MockTransport(handler)
+
+    client = AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "create_mock_api_client",
+        lambda: client,
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "get_settings",
+        lambda: SimpleNamespace(
+            database_url="postgresql+asyncpg://unused",
+        ),
+    )
+
+    create_engine_mock = MagicMock()
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "create_async_engine",
+        create_engine_mock,
+    )
+
+    await mock_api_import.import_mock_api_history(
+        start_time=datetime.fromisoformat("2024-06-15T12:00:00"),
+        end_time=datetime.fromisoformat("2024-06-15T13:00:00"),
+        limit=60,
+        dry_run=True,
+    )
+
+    create_engine_mock.assert_not_called()
+
+
+async def test_import_mock_api_history_loads_data(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    def handler(request: Request) -> Response:
+        if request.url.path == "/api/v1/sites":
+            return Response(
+                status_code=200,
+                json=[make_site()],
+            )
+
+        if request.url.path == "/api/v1/readings":
+            return Response(
+                status_code=200,
+                json=[make_reading()],
+            )
+
+        return Response(status_code=404)
+
+    transport = MockTransport(handler)
+
+    client = AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "create_mock_api_client",
+        lambda: client,
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "get_settings",
+        lambda: SimpleNamespace(
+            database_url="postgresql+asyncpg://test:test@localhost/test",
+        ),
+    )
+
+    connection = AsyncMock()
+
+    transaction_context = MagicMock()
+    transaction_context.__aenter__ = AsyncMock(
+        return_value=connection,
+    )
+    transaction_context.__aexit__ = AsyncMock(
+        return_value=None,
+    )
+
+    engine = MagicMock()
+    engine.begin.return_value = transaction_context
+    engine.dispose = AsyncMock()
+
+    create_engine_mock = MagicMock(
+        return_value=engine,
+    )
+
+    upsert_sites_mock = AsyncMock()
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "create_async_engine",
+        create_engine_mock,
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "upsert_sites",
+        upsert_sites_mock,
+    )
+
+    await mock_api_import.import_mock_api_history(
+        start_time=datetime.fromisoformat("2024-06-15T12:00:00"),
+        end_time=datetime.fromisoformat("2024-06-15T13:00:00"),
+        limit=60,
+        dry_run=False,
+    )
+
+    create_engine_mock.assert_called_once_with(
+        "postgresql+asyncpg://test:test@localhost/test",
+        pool_pre_ping=True,
+    )
+
+    upsert_sites_mock.assert_awaited_once_with(
+        connection,
+        [make_site()],
+    )
+
+    connection.execute.assert_awaited_once()
+    engine.dispose.assert_awaited_once()
+
+
+def test_parse_datetime_accepts_z_suffix() -> None:
+    result = mock_api_import.parse_datetime(
+        "2024-06-15T12:00:00Z",
+    )
+
+    assert result == datetime.fromisoformat(
+        "2024-06-15T12:00:00+00:00",
+    )
+
+
+def test_parse_args_reads_cli_parameters(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        sys,
+        "argv",
+        [
+            "mock_api_import",
+            "--start-time",
+            "2024-06-15T12:00:00Z",
+            "--end-time",
+            "2024-06-15T13:00:00Z",
+            "--limit",
+            "60",
+            "--dry-run",
+        ],
+    )
+
+    args = mock_api_import.parse_args()
+
+    assert args.start_time == datetime.fromisoformat(
+        "2024-06-15T12:00:00+00:00",
+    )
+    assert args.end_time == datetime.fromisoformat(
+        "2024-06-15T13:00:00+00:00",
+    )
+    assert args.limit == 60
+    assert args.dry_run is True
+
+
+def test_main_rejects_limit_out_of_bounds(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        sys,
+        "argv",
+        [
+            "mock_api_import",
+            "--start-time",
+            "2024-06-15T12:00:00Z",
+            "--end-time",
+            "2024-06-15T13:00:00Z",
+            "--limit",
+            "0",
+        ],
+    )
+
+    with pytest.raises(
+        ValueError,
+        match="--limit doit être compris entre 1 et 1000",
+    ):
+        mock_api_import.main()
+
+
+def test_main_rejects_invalid_period(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        sys,
+        "argv",
+        [
+            "mock_api_import",
+            "--start-time",
+            "2024-06-15T14:00:00Z",
+            "--end-time",
+            "2024-06-15T13:00:00Z",
+            "--limit",
+            "60",
+        ],
+    )
+
+    with pytest.raises(
+        ValueError,
+        match="--start-time doit être antérieur à --end-time",
+    ):
+        mock_api_import.main()
+
+
+def test_main_runs_import(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    start_time = datetime.fromisoformat(
+        "2024-06-15T12:00:00+00:00",
+    )
+    end_time = datetime.fromisoformat(
+        "2024-06-15T13:00:00+00:00",
+    )
+
+    import_mock = AsyncMock()
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "parse_args",
+        lambda: SimpleNamespace(
+            start_time=start_time,
+            end_time=end_time,
+            limit=60,
+            dry_run=True,
+        ),
+    )
+
+    monkeypatch.setattr(
+        mock_api_import,
+        "import_mock_api_history",
+        import_mock,
+    )
+
+    mock_api_import.main()
+
+    import_mock.assert_awaited_once_with(
+        start_time=start_time,
+        end_time=end_time,
+        limit=60,
+        dry_run=True,
+    )
+
+
+async def test_fetch_sites_rejects_a_response_above_the_cap() -> None:
+    def handler(request: Request) -> Response:
+        return Response(
+            status_code=200,
+            json=[make_site() for _ in range(MAX_SITES + 1)],
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        with pytest.raises(
+            ValueError,
+            match=f"dépasse le plafond de {MAX_SITES} sites",
+        ):
+            await fetch_sites(client)
+
+
+async def test_fetch_readings_rejects_a_response_above_the_requested_limit() -> None:
+    def handler(request: Request) -> Response:
+        return Response(
+            status_code=200,
+            json=[make_reading(), make_reading(), make_reading()],
+        )
+
+    transport = MockTransport(handler)
+
+    async with AsyncClient(
+        transport=transport,
+        base_url="https://mock.test",
+    ) as client:
+        with pytest.raises(
+            ValueError,
+            match="dépasse la limite demandée de 2",
+        ):
+            await fetch_readings(
+                client=client,
+                site_id="SITE001",
+                start_time=datetime.fromisoformat("2024-06-15T12:00:00"),
+                end_time=datetime.fromisoformat("2024-06-15T13:00:00"),
+                limit=2,
+            )
+
+
+def test_build_reading_row_neutralises_values_outside_physical_bounds() -> None:
+    reading = make_reading()
+
+    reading["power_factor"] = 42.0
+    reading["temperature_celsius"] = 1e30
+    reading["humidity_percent"] = -1.0
+
+    row = build_reading_row(reading)
+
+    assert row["power_factor"] is None
+    assert row["temperature_celsius"] is None
+    assert row["humidity_percent"] is None
+
+    assert row["null_reasons"] == [
+        "out_of_physical_bounds:power_factor",
+        "out_of_physical_bounds:temperature_celsius",
+        "out_of_physical_bounds:humidity_percent",
+    ]
+
+    assert row["data_quality"] == "degraded"
+
+    assert json.loads(row["raw_data"])["power_factor"] == 42.0
+
+
+def test_build_reading_row_rejects_a_measure_that_is_not_a_number() -> None:
+    reading = make_reading()
+
+    reading["consumption_kw"] = "87.34"
+
+    row = build_reading_row(reading)
+
+    assert row["consumption_kw"] is None
+    assert "out_of_physical_bounds:consumption_kw" in row["null_reasons"]
+
+
+def test_build_reading_row_drops_a_quality_the_database_refuses() -> None:
+    reading = make_reading()
+
+    reading["data_quality"] = "unknown"
+
+    row = build_reading_row(reading)
+
+    assert row["data_quality"] is None
+
+
+def test_build_reading_row_requires_an_identifier() -> None:
+    reading = make_reading()
+
+    del reading["site_id"]
+
+    with pytest.raises(
+        ValueError,
+        match="Champ site_id absent ou invalide",
+    ):
+        build_reading_row(reading)
+
+
+def test_build_site_row_keeps_only_the_expected_columns() -> None:
+    site = make_site()
+
+    site["unexpected"] = "valeur hostile"
+    site["capacity_kw"] = -5.0
+    site["status"] = 12
+
+    row = build_site_row(site)
+
+    assert set(row) == {
+        "site_id",
+        "site_type",
+        "site_name",
+        "location",
+        "capacity_kw",
+        "status",
+    }
+
+    assert row["capacity_kw"] is None
+    assert row["status"] is None
+
+
+async def test_upsert_sites_sends_only_the_expected_columns() -> None:
+    connection = AsyncMock()
+
+    site = make_site()
+    site["unexpected"] = "valeur hostile"
+
+    await upsert_sites(
+        connection,
+        [site],
+    )
+
+    rows = connection.execute.await_args.args[1]
+
+    assert "unexpected" not in rows[0]
+    assert rows[0]["site_id"] == "SITE001"
+
+
+@pytest.mark.integration
+async def test_reading_insert_is_idempotent(
+    session: AsyncSession,
+) -> None:
+    reading = make_reading()
+    row = build_reading_row(reading)
+
+    connection = await session.connection()
+
+    await upsert_sites(
+        connection,
+        [make_site()],
+    )
+
+    await session.execute(
+        READING_INSERT,
+        [row],
+    )
+
+    await session.execute(
+        READING_INSERT,
+        [row],
+    )
+
+    result = await session.execute(
+        text(
+            """
+            SELECT COUNT(*)
+            FROM reading
+            WHERE site_id = :site_id
+              AND timestamp = :timestamp
+              AND source = :source
+            """
+        ),
+        {
+            "site_id": row["site_id"],
+            "timestamp": row["timestamp"],
+            "source": row["source"],
+        },
+    )
+
+    assert result.scalar_one() == 1
+
+    await session.rollback()
+
+
+@pytest.mark.integration
+async def test_out_of_bounds_reading_is_stored_neutralised(
+    session: AsyncSession,
+) -> None:
+    reading = make_reading()
+    reading["power_factor"] = 42.0
+
+    row = build_reading_row(reading)
+
+    connection = await session.connection()
+
+    await upsert_sites(
+        connection,
+        [make_site()],
+    )
+
+    await session.execute(
+        READING_INSERT,
+        [row],
+    )
+
+    result = await session.execute(
+        text(
+            """
+            SELECT power_factor, data_quality, null_reasons, raw_data ->> 'power_factor'
+            FROM reading
+            WHERE site_id = :site_id
+              AND timestamp = :timestamp
+              AND source = :source
+            """
+        ),
+        {
+            "site_id": row["site_id"],
+            "timestamp": row["timestamp"],
+            "source": row["source"],
+        },
+    )
+
+    stored = result.one()
+
+    await session.rollback()
+
+    assert stored[0] is None
+    assert stored[1] == "degraded"
+    assert stored[2] == ["out_of_physical_bounds:power_factor"]
+    assert stored[3] == "42.0"

apps/backend/tests/repositories/test_alert.py

@@ -89,3 +89,60 @@ async def test_list_all_returns_an_empty_list_when_there_is_nothing(
     alertes = await depot.list_all(site_id=identifiant_site())
 
     assert list(alertes) == []
+
+
+def _alerte_a_inserer(*, site_id: str, source_alert_id: str) -> Alert:
+    return Alert(
+        source_alert_id=source_alert_id,
+        site_id=site_id,
+        source="enervision",
+        timestamp=datetime(2026, 9, 16, tzinfo=UTC),
+        type="threshold",
+        severity="high",
+        message="Dépassement du seuil configuré",
+        value=812.5,
+        threshold=720.0,
+        metric="consumption_kw",
+        prediction_id=None,
+        raw_data={},
+    )
+
+
+async def test_create_many_inserts_every_alert(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = AlertRepository(session)
+
+    creees = await depot.create_many(
+        [
+            _alerte_a_inserer(site_id=site.site_id, source_alert_id="threshold:a"),
+            _alerte_a_inserer(site_id=site.site_id, source_alert_id="threshold:b"),
+        ]
+    )
+    identifiants = [a.alert_id for a in creees]
+    await session.rollback()
+
+    assert len(identifiants) == 2
+    assert all(identifiant is not None for identifiant in identifiants)
+
+
+async def test_create_many_skips_a_duplicate_source_alert_id(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = AlertRepository(session)
+    await depot.create_many(
+        [_alerte_a_inserer(site_id=site.site_id, source_alert_id="threshold:rejouee")]
+    )
+
+    rejouees = await depot.create_many(
+        [_alerte_a_inserer(site_id=site.site_id, source_alert_id="threshold:rejouee")]
+    )
+    await session.rollback()
+
+    assert rejouees == []
+
+
+async def test_create_many_does_nothing_for_an_empty_list(session: AsyncSession) -> None:
+    depot = AlertRepository(session)
+
+    creees = await depot.create_many([])
+
+    assert creees == []

apps/backend/tests/repositories/test_audit_log.py

@@ -49,8 +49,10 @@ async def test_the_database_refuses_to_mutate_the_audit_log(
 ) -> None:
     await une_ligne(session)
 
+    requete = text(instruction)
+
     with pytest.raises(DBAPIError, match="ajout seul"):
-        await session.execute(text(instruction))
+        await session.execute(requete)
     await session.rollback()
 
 

apps/backend/tests/repositories/test_password_reset_token.py

@@ -0,0 +1,145 @@
+# Le premier test démontre l'atomicité de `consume()` : sur un double, deux soumissions
+# concurrentes du même lien réussiraient toutes les deux.
+
+import uuid
+from datetime import UTC, datetime, timedelta
+
+import pytest
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.core.roles import Role
+from app.core.security import fingerprint_refresh, generate_refresh_secret
+from app.repositories.password_reset_token import PasswordResetTokenRepository
+from app.repositories.user import UserRepository
+
+pytestmark = pytest.mark.integration
+
+DUREE = timedelta(minutes=15)
+
+
+async def un_compte(session: AsyncSession) -> uuid.UUID:
+    compte = await UserRepository(session).create(
+        email=f"reset-{uuid.uuid4().hex[:12]}@enervision.fr",
+        password_hash="$argon2id$x",
+        role=Role.LECTEUR,
+    )
+    return compte.id
+
+
+async def un_jeton(
+    depot: PasswordResetTokenRepository, user_id: uuid.UUID, *, duree: timedelta = DUREE
+) -> str:
+    secret = generate_refresh_secret()
+    await depot.create(
+        user_id=user_id,
+        token_hash=fingerprint_refresh(secret),
+        expires_at=datetime.now(UTC) + duree,
+        client_ip="203.0.113.10",
+        user_agent="pytest",
+    )
+    return secret
+
+
+async def test_consume_only_succeeds_once(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+    secret = await un_jeton(depot, await un_compte(session))
+
+    premier = await depot.consume(fingerprint_refresh(secret))
+    second = await depot.consume(fingerprint_refresh(secret))
+    await session.rollback()
+
+    assert premier is not None
+    assert second is None
+
+
+async def test_consume_refuses_an_expired_token(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+    secret = await un_jeton(depot, await un_compte(session), duree=-timedelta(minutes=1))
+
+    revendique = await depot.consume(fingerprint_refresh(secret))
+    await session.rollback()
+
+    assert revendique is None
+
+
+async def test_consume_returns_nothing_for_an_unknown_fingerprint(
+    session: AsyncSession,
+) -> None:
+    revendique = await PasswordResetTokenRepository(session).consume(
+        fingerprint_refresh(generate_refresh_secret())
+    )
+
+    assert revendique is None
+
+
+async def test_invalidate_all_for_user_only_touches_living_tokens(
+    session: AsyncSession,
+) -> None:
+    depot = PasswordResetTokenRepository(session)
+    compte = await un_compte(session)
+    await un_jeton(depot, compte)
+    await un_jeton(depot, compte)
+
+    invalides = await depot.invalidate_all_for_user(compte)
+    second_passage = await depot.invalidate_all_for_user(compte)
+    await session.rollback()
+
+    assert invalides == 2
+    assert second_passage == 0
+
+
+async def test_exists_valid_is_true_for_a_living_token(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+    secret = await un_jeton(depot, await un_compte(session))
+
+    assert await depot.exists_valid(fingerprint_refresh(secret)) is True
+
+
+async def test_exists_valid_is_false_for_an_expired_token(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+    secret = await un_jeton(depot, await un_compte(session), duree=-timedelta(minutes=1))
+
+    assert await depot.exists_valid(fingerprint_refresh(secret)) is False
+
+
+async def test_exists_valid_is_false_once_the_token_is_consumed(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+    secret = await un_jeton(depot, await un_compte(session))
+    await depot.consume(fingerprint_refresh(secret))
+
+    assert await depot.exists_valid(fingerprint_refresh(secret)) is False
+
+
+async def test_exists_valid_is_false_for_an_unknown_fingerprint(session: AsyncSession) -> None:
+    depot = PasswordResetTokenRepository(session)
+
+    assert await depot.exists_valid(fingerprint_refresh(generate_refresh_secret())) is False
+
+
+async def test_the_database_refuses_two_tokens_sharing_a_fingerprint(
+    session: AsyncSession,
+) -> None:
+    depot = PasswordResetTokenRepository(session)
+    compte = await un_compte(session)
+    secret = generate_refresh_secret()
+    await depot.create(
+        user_id=compte,
+        token_hash=fingerprint_refresh(secret),
+        expires_at=datetime.now(UTC) + DUREE,
+        client_ip=None,
+        user_agent=None,
+    )
+
+    empreinte = fingerprint_refresh(secret)
+    expiration = datetime.now(UTC) + DUREE
+
+    with pytest.raises(IntegrityError):
+        await depot.create(
+            user_id=compte,
+            token_hash=empreinte,
+            expires_at=expiration,
+            client_ip=None,
+            user_agent=None,
+        )
+    await session.rollback()

apps/backend/tests/repositories/test_prediction.py

@@ -0,0 +1,171 @@
+from datetime import UTC, datetime
+
+import pytest
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models.energy import Prediction
+from app.repositories.prediction import PredictionRepository
+from tests.repositories.test_site import creer as creer_site
+from tests.repositories.test_site import identifiant as identifiant_site
+
+pytestmark = pytest.mark.integration
+
+
+async def creer_prediction(
+    session: AsyncSession, *, site_id: str, **overrides: object
+) -> Prediction:
+    prediction = Prediction(
+        site_id=site_id,
+        target_at=overrides.get("target_at", datetime(2026, 9, 16, tzinfo=UTC)),
+        target_metric=overrides.get("target_metric", "consumption_kwh"),
+        period_minutes=overrides.get("period_minutes", 60),
+        predicted_value=overrides.get("predicted_value", 42.0),
+        model_reference=overrides.get("model_reference", "lightgbm-test"),
+        status=overrides.get("status", "available"),
+        failure_reason=overrides.get("failure_reason"),
+    )
+    session.add(prediction)
+    await session.flush()
+    return prediction
+
+
+async def test_list_since_excludes_predictions_before_the_cutoff(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = PredictionRepository(session)
+    dedans = await creer_prediction(
+        session, site_id=site.site_id, target_at=datetime(2026, 9, 16, tzinfo=UTC)
+    )
+    await creer_prediction(
+        session, site_id=site.site_id, target_at=datetime(2026, 9, 1, tzinfo=UTC)
+    )
+
+    resultats = await depot.list_since(
+        since=datetime(2026, 9, 10, tzinfo=UTC), site_id=site.site_id
+    )
+    identifiants = [p.prediction_id for p in resultats]
+    await session.rollback()
+
+    assert identifiants == [dedans.prediction_id]
+
+
+async def test_list_since_excludes_predictions_that_are_not_available(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    depot = PredictionRepository(session)
+    await creer_prediction(
+        session,
+        site_id=site.site_id,
+        target_at=datetime(2026, 9, 16, tzinfo=UTC),
+        status="insufficient_data",
+        predicted_value=None,
+        failure_reason="pas assez d'historique",
+    )
+
+    resultats = await depot.list_since(since=datetime(2026, 9, 1, tzinfo=UTC), site_id=site.site_id)
+    await session.rollback()
+
+    assert list(resultats) == []
+
+
+async def test_list_since_breaks_a_target_at_tie_by_ascending_prediction_id(
+    session: AsyncSession,
+) -> None:
+    # `prediction` n'a pas d'unicité sur `(site_id, target_at)` : deux runs de scoring sans
+    # nouvelle lecture entre-temps produisent deux lignes `available` à la même cible. Sans ce
+    # départage, `_detect_anomaly` retiendrait une ligne au hasard plutôt que le run le plus
+    # récent.
+    site = await creer_site(session)
+    depot = PredictionRepository(session)
+    cible = datetime(2026, 9, 16, tzinfo=UTC)
+    premier_run = await creer_prediction(
+        session, site_id=site.site_id, target_at=cible, predicted_value=10.0
+    )
+    second_run = await creer_prediction(
+        session, site_id=site.site_id, target_at=cible, predicted_value=20.0
+    )
+
+    resultats = await depot.list_since(since=datetime(2026, 9, 1, tzinfo=UTC), site_id=site.site_id)
+    identifiants = [p.prediction_id for p in resultats]
+    await session.rollback()
+
+    assert identifiants == [premier_run.prediction_id, second_run.prediction_id]
+
+
+async def test_list_since_filters_by_site_id(session: AsyncSession) -> None:
+    premier = await creer_site(session)
+    second = await creer_site(session)
+    depot = PredictionRepository(session)
+    voulue = await creer_prediction(session, site_id=premier.site_id)
+    await creer_prediction(session, site_id=second.site_id)
+
+    resultats = await depot.list_since(
+        since=datetime(2026, 8, 1, tzinfo=UTC), site_id=premier.site_id
+    )
+    identifiants = [p.prediction_id for p in resultats]
+    await session.rollback()
+
+    assert identifiants == [voulue.prediction_id]
+
+
+async def test_latest_by_site_keeps_only_the_most_recent_target(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = PredictionRepository(session)
+    ancienne = await creer_prediction(
+        session, site_id=site.site_id, target_at=datetime(2026, 9, 1, tzinfo=UTC)
+    )
+    recente = await creer_prediction(
+        session, site_id=site.site_id, target_at=datetime(2026, 9, 15, tzinfo=UTC)
+    )
+
+    resultats = await depot.latest_by_site()
+    identifiants = [
+        p.prediction_id
+        for p in resultats
+        if p.prediction_id in (ancienne.prediction_id, recente.prediction_id)
+    ]
+    await session.rollback()
+
+    assert identifiants == [recente.prediction_id]
+
+
+async def test_latest_by_site_returns_one_row_per_site(session: AsyncSession) -> None:
+    premier = await creer_site(session)
+    second = await creer_site(session)
+    depot = PredictionRepository(session)
+    voulue_premier = await creer_prediction(session, site_id=premier.site_id)
+    voulue_second = await creer_prediction(session, site_id=second.site_id)
+
+    resultats = await depot.latest_by_site()
+    identifiants = {p.site_id for p in resultats if p.site_id in (premier.site_id, second.site_id)}
+    await session.rollback()
+
+    assert identifiants == {voulue_premier.site_id, voulue_second.site_id}
+
+
+async def test_latest_by_site_keeps_an_insufficient_data_prediction(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = PredictionRepository(session)
+    voulue = await creer_prediction(
+        session,
+        site_id=site.site_id,
+        status="insufficient_data",
+        predicted_value=None,
+        failure_reason="pas assez d'historique",
+    )
+
+    resultats = await depot.latest_by_site()
+    identifiants = [p.prediction_id for p in resultats if p.site_id == site.site_id]
+    await session.rollback()
+
+    assert identifiants == [voulue.prediction_id]
+
+
+async def test_latest_by_site_returns_an_empty_list_when_there_is_nothing(
+    session: AsyncSession,
+) -> None:
+    depot = PredictionRepository(session)
+
+    resultats = [p for p in await depot.latest_by_site() if p.site_id == identifiant_site()]
+
+    assert resultats == []

apps/backend/tests/repositories/test_reading.py

@@ -6,6 +6,8 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Reading, Site
 from app.repositories.reading import ReadingRepository
+from tests.repositories.test_site import creer as creer_site
+from tests.repositories.test_site import identifiant as identifiant_site
 
 pytestmark = pytest.mark.integration
 
@@ -25,6 +27,20 @@ def lecture(site_id: str, *, timestamp: datetime, consumption_kw: float) -> Read
     )
 
 
+async def creer_lecture(session: AsyncSession, *, site_id: str, **overrides: object) -> Reading:
+    reading = Reading(
+        site_id=site_id,
+        timestamp=overrides.get("timestamp", datetime(2026, 9, 16, tzinfo=UTC)),
+        source=overrides.get("source", "api_current"),
+        consumption_kw=overrides.get("consumption_kw", 10.0),
+        data_quality=overrides.get("data_quality", "good"),
+        raw_data=overrides.get("raw_data", {}),
+    )
+    session.add(reading)
+    await session.flush()
+    return reading
+
+
 async def test_latest_by_site_keeps_only_the_most_recent_reading(session: AsyncSession) -> None:
     site_id = identifiant()
     maintenant = datetime.now(UTC)
@@ -70,3 +86,250 @@ async def test_latest_by_site_returns_one_row_per_site(session: AsyncSession) ->
     await session.rollback()
 
     assert identifiants == {premier, second}
+
+
+async def test_latest_by_site_breaks_a_timestamp_tie_on_the_last_written_reading(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    horodatage = datetime(2026, 9, 15, tzinfo=UTC)
+    await creer_lecture(
+        session, site_id=site.site_id, timestamp=horodatage, source="api_history", consumption_kw=10
+    )
+    derniere = await creer_lecture(
+        session, site_id=site.site_id, timestamp=horodatage, source="api_current", consumption_kw=42
+    )
+
+    resultats = await depot.latest_by_site()
+    retenues = [r.reading_id for r in resultats if r.site_id == site.site_id]
+    await session.rollback()
+
+    assert retenues == [derniere.reading_id]
+
+
+async def test_latest_for_site_returns_the_most_recent_reading(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    await creer_lecture(session, site_id=site.site_id, timestamp=datetime(2026, 9, 1, tzinfo=UTC))
+    recente = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 15, tzinfo=UTC)
+    )
+
+    trouvee = await depot.latest_for_site(site.site_id)
+    reading_id = trouvee.reading_id if trouvee else None
+    await session.rollback()
+
+    assert reading_id == recente.reading_id
+
+
+async def test_latest_for_site_breaks_a_timestamp_tie_on_the_last_written_reading(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    horodatage = datetime(2026, 9, 15, tzinfo=UTC)
+    await creer_lecture(session, site_id=site.site_id, timestamp=horodatage, source="api_history")
+    derniere = await creer_lecture(
+        session, site_id=site.site_id, timestamp=horodatage, source="api_current"
+    )
+
+    trouvee = await depot.latest_for_site(site.site_id)
+    reading_id = trouvee.reading_id if trouvee else None
+    await session.rollback()
+
+    assert reading_id == derniere.reading_id
+
+
+async def test_latest_for_site_ignores_the_readings_of_the_other_sites(
+    session: AsyncSession,
+) -> None:
+    sans_lecture = await creer_site(session)
+    autre = await creer_site(session)
+    depot = ReadingRepository(session)
+    await creer_lecture(session, site_id=autre.site_id)
+
+    trouvee = await depot.latest_for_site(sans_lecture.site_id)
+    await session.rollback()
+
+    assert trouvee is None
+
+
+async def test_list_since_orders_by_site_then_by_time_ascending(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    plus_recente = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 16, tzinfo=UTC)
+    )
+    plus_ancienne = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 15, tzinfo=UTC)
+    )
+
+    resultats = await depot.list_since(since=datetime(2026, 9, 1, tzinfo=UTC), site_id=site.site_id)
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [plus_ancienne.reading_id, plus_recente.reading_id]
+
+
+async def test_list_since_excludes_readings_before_the_cutoff(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    dedans = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 16, tzinfo=UTC)
+    )
+    await creer_lecture(session, site_id=site.site_id, timestamp=datetime(2026, 9, 1, tzinfo=UTC))
+
+    resultats = await depot.list_since(
+        since=datetime(2026, 9, 10, tzinfo=UTC), site_id=site.site_id
+    )
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [dedans.reading_id]
+
+
+async def test_list_since_breaks_a_timestamp_tie_by_ascending_reading_id(
+    session: AsyncSession,
+) -> None:
+    # `uq_reading_source` autorise deux lignes au même `site_id`+`timestamp` quand la `source`
+    # diffère (même piège que `latest_for_site`). Sans ce départage, `_detect_spike` traiterait
+    # cette paire comme une variation réelle selon un ordre non garanti par le plan d'exécution.
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    horodatage = datetime(2026, 9, 16, tzinfo=UTC)
+    premiere = await creer_lecture(
+        session, site_id=site.site_id, timestamp=horodatage, source="api_history", consumption_kw=10
+    )
+    seconde = await creer_lecture(
+        session, site_id=site.site_id, timestamp=horodatage, source="api_current", consumption_kw=42
+    )
+
+    resultats = await depot.list_since(since=datetime(2026, 9, 1, tzinfo=UTC), site_id=site.site_id)
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [premiere.reading_id, seconde.reading_id]
+
+
+async def test_list_since_filters_by_site_id(session: AsyncSession) -> None:
+    premier = await creer_site(session)
+    second = await creer_site(session)
+    depot = ReadingRepository(session)
+    voulue = await creer_lecture(session, site_id=premier.site_id)
+    await creer_lecture(session, site_id=second.site_id)
+
+    resultats = await depot.list_since(
+        since=datetime(2026, 8, 1, tzinfo=UTC), site_id=premier.site_id
+    )
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [voulue.reading_id]
+
+
+async def test_list_history_orders_the_readings_by_timestamp_descending(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    ancienne = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 1, tzinfo=UTC)
+    )
+    recente = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 15, tzinfo=UTC)
+    )
+
+    resultats = await depot.list_history(
+        start=datetime(2026, 8, 1, tzinfo=UTC),
+        end=datetime(2026, 10, 1, tzinfo=UTC),
+        limit=100,
+        offset=0,
+    )
+    identifiants = [
+        r.reading_id for r in resultats if r.reading_id in (ancienne.reading_id, recente.reading_id)
+    ]
+    await session.rollback()
+
+    assert identifiants == [recente.reading_id, ancienne.reading_id]
+
+
+async def test_list_history_filters_by_site_id(session: AsyncSession) -> None:
+    premier = await creer_site(session)
+    second = await creer_site(session)
+    depot = ReadingRepository(session)
+    voulue = await creer_lecture(session, site_id=premier.site_id)
+    await creer_lecture(session, site_id=second.site_id)
+
+    resultats = await depot.list_history(
+        site_id=premier.site_id,
+        start=datetime(2026, 8, 1, tzinfo=UTC),
+        end=datetime(2026, 10, 1, tzinfo=UTC),
+        limit=100,
+        offset=0,
+    )
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [voulue.reading_id]
+
+
+async def test_list_history_excludes_readings_outside_the_window(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    dedans = await creer_lecture(
+        session, site_id=site.site_id, timestamp=datetime(2026, 9, 10, tzinfo=UTC)
+    )
+    await creer_lecture(session, site_id=site.site_id, timestamp=datetime(2026, 8, 1, tzinfo=UTC))
+    await creer_lecture(session, site_id=site.site_id, timestamp=datetime(2026, 10, 1, tzinfo=UTC))
+
+    resultats = await depot.list_history(
+        site_id=site.site_id,
+        start=datetime(2026, 9, 1, tzinfo=UTC),
+        end=datetime(2026, 9, 30, tzinfo=UTC),
+        limit=100,
+        offset=0,
+    )
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [dedans.reading_id]
+
+
+async def test_list_history_respects_limit_and_offset(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    depot = ReadingRepository(session)
+    lectures = [
+        await creer_lecture(
+            session, site_id=site.site_id, timestamp=datetime(2026, 9, jour, tzinfo=UTC)
+        )
+        for jour in (1, 2, 3)
+    ]
+
+    resultats = await depot.list_history(
+        site_id=site.site_id,
+        start=datetime(2026, 8, 1, tzinfo=UTC),
+        end=datetime(2026, 10, 1, tzinfo=UTC),
+        limit=1,
+        offset=1,
+    )
+    identifiants = [r.reading_id for r in resultats]
+    await session.rollback()
+
+    assert identifiants == [lectures[1].reading_id]
+
+
+async def test_list_history_returns_an_empty_list_when_there_is_nothing(
+    session: AsyncSession,
+) -> None:
+    depot = ReadingRepository(session)
+
+    resultats = await depot.list_history(
+        site_id=identifiant_site(),
+        start=datetime(2026, 8, 1, tzinfo=UTC),
+        end=datetime(2026, 10, 1, tzinfo=UTC),
+        limit=100,
+        offset=0,
+    )
+
+    assert list(resultats) == []

apps/backend/tests/repositories/test_recommendation.py

@@ -5,7 +5,8 @@ import pytest
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Alert, Recommendation, Site
-from app.repositories.recommendation import RecommendationRepository
+from app.repositories import recommendation as module_recommendation
+from app.repositories.recommendation import NouvelleRecommandation, RecommendationRepository
 
 pytestmark = pytest.mark.integration
 
@@ -83,3 +84,59 @@ async def test_list_all_returns_the_recommendations_sorted_by_identifier(
     await session.rollback()
 
     assert identifiants == sorted(identifiants)
+
+
+def nouvelle(alert_id: int, reference: str = "spike-delestage-v1") -> NouvelleRecommandation:
+    return NouvelleRecommandation(
+        alert_id=alert_id,
+        action="Délester les équipements non prioritaires",
+        explanation="Pic de consommation signalé.",
+        rule_reference=reference,
+    )
+
+
+async def test_create_missing_inserts_the_proposals(session: AsyncSession) -> None:
+    depot = RecommendationRepository(session)
+    alert_id = await creer_alerte(session)
+
+    creees = await depot.create_missing(
+        [nouvelle(alert_id), nouvelle(alert_id, "escalade-astreinte-v1")]
+    )
+    await session.rollback()
+
+    assert creees == 2
+
+
+async def test_create_missing_ignores_a_rule_already_held_for_the_alert(
+    session: AsyncSession,
+) -> None:
+    depot = RecommendationRepository(session)
+    alert_id = await creer_alerte(session)
+    await depot.create_missing([nouvelle(alert_id)])
+
+    creees = await depot.create_missing([nouvelle(alert_id)])
+    await session.rollback()
+
+    assert creees == 0
+
+
+async def test_create_missing_returns_zero_without_any_proposal(session: AsyncSession) -> None:
+    creees = await RecommendationRepository(session).create_missing([])
+
+    assert creees == 0
+
+
+async def test_create_missing_inserts_every_proposal_across_several_batches(
+    session: AsyncSession, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setattr(module_recommendation, "TAILLE_DE_LOT", 2)
+    depot = RecommendationRepository(session)
+    alert_id = await creer_alerte(session)
+    propositions = [nouvelle(alert_id, f"regle-{index}-v1") for index in range(5)]
+
+    creees = await depot.create_missing(propositions)
+    enregistrees = [r for r in await depot.list_all() if r.alert_id == alert_id]
+    await session.rollback()
+
+    assert creees == 5
+    assert len(enregistrees) == 5

apps/backend/tests/repositories/test_refresh_token.py

@@ -178,12 +178,16 @@ async def test_the_database_refuses_two_tokens_sharing_a_fingerprint(
         user_agent=None,
     )
 
+    famille = uuid.uuid4()
+    empreinte = fingerprint_refresh(secret)
+    expiration = datetime.now(UTC) + DUREE
+
     with pytest.raises(IntegrityError):
         await depot.create(
             user_id=compte,
-            family_id=uuid.uuid4(),
-            token_hash=fingerprint_refresh(secret),
-            expires_at=datetime.now(UTC) + DUREE,
+            family_id=famille,
+            token_hash=empreinte,
+            expires_at=expiration,
             client_ip=None,
             user_agent=None,
         )

apps/backend/tests/repositories/test_user.py

@@ -31,14 +31,12 @@ async def test_the_database_refuses_an_email_written_in_upper_case(
 ) -> None:
     saisie = adresse().upper()
 
+    requete = text(
+        "insert into app_user (email, password_hash, role) values (:e, '$argon2id$x', 'lecteur')"
+    )
+
     with pytest.raises(IntegrityError):
-        await session.execute(
-            text(
-                "insert into app_user (email, password_hash, role) "
-                "values (:e, '$argon2id$x', 'lecteur')"
-            ),
-            {"e": saisie},
-        )
+        await session.execute(requete, {"e": saisie})
     await session.rollback()
 
 

apps/backend/tests/schemas/test_auth.py

@@ -0,0 +1,61 @@
+import pytest
+from pydantic import ValidationError
+
+from app.schemas.auth import PasswordChangeRequest, valide_complexite
+
+MOT_DE_PASSE_VALIDE = "Un-mot-de-passe1!"
+
+
+def test_password_change_request_accepts_a_password_covering_the_four_classes() -> None:
+    requete = PasswordChangeRequest(
+        current_password="peu-importe", new_password=MOT_DE_PASSE_VALIDE
+    )
+
+    assert requete.new_password == MOT_DE_PASSE_VALIDE
+
+
+@pytest.mark.parametrize(
+    "new_password",
+    [
+        "un-mot-de-passe1!",
+        "UN-MOT-DE-PASSE1!",
+        "Un-mot-de-passe!",
+        "Un mot de passe 1",
+    ],
+    ids=["sans_majuscule", "sans_minuscule", "sans_chiffre", "sans_caractere_special"],
+)
+def test_password_change_request_rejects_a_password_missing_a_character_class(
+    new_password: str,
+) -> None:
+    with pytest.raises(ValidationError):
+        PasswordChangeRequest(current_password="peu-importe", new_password=new_password)
+
+
+def test_password_change_request_rejects_a_password_below_the_minimum_length() -> None:
+    with pytest.raises(ValidationError):
+        PasswordChangeRequest(current_password="peu-importe", new_password="Ab1!")
+
+
+def test_valide_complexite_names_every_missing_class_in_the_error() -> None:
+    with pytest.raises(ValueError, match=r"majuscule.*chiffre|chiffre.*majuscule"):
+        valide_complexite("minuscules-seulement")
+
+
+def test_valide_complexite_accepts_an_accented_password() -> None:
+    assert valide_complexite("Sécurité1!") == "Sécurité1!"
+
+
+@pytest.mark.parametrize("mot_de_passe", ["abcdefg1×", "abcdefg1÷"])  # noqa: RUF001
+def test_valide_complexite_rejects_a_password_without_uppercase_despite_times_or_divide(
+    mot_de_passe: str,
+) -> None:
+    with pytest.raises(ValueError, match="majuscule"):
+        valide_complexite(mot_de_passe)
+
+
+@pytest.mark.parametrize("mot_de_passe", ["ABCDEFG1×", "ABCDEFG1÷"])  # noqa: RUF001
+def test_valide_complexite_rejects_a_password_without_lowercase_despite_times_or_divide(
+    mot_de_passe: str,
+) -> None:
+    with pytest.raises(ValueError, match="minuscule"):
+        valide_complexite(mot_de_passe)

apps/backend/tests/services/test_alert.py

@@ -1,7 +1,10 @@
-from datetime import UTC, datetime
+from dataclasses import dataclass, field
+from datetime import UTC, datetime, timedelta
 
 from app.models.energy import Alert
-from app.services.alert import AlertService
+from app.services.alert import OUTAGE_THRESHOLD, AlertService, _severity_from_ratio
+
+NOW = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
 
 
 def alert(
@@ -26,10 +29,36 @@ def alert(
     )
 
 
+@dataclass
+class FauxSite:
+    site_id: str
+    capacity_kw: float | None = None
+
+
+@dataclass
+class FauxLecture:
+    site_id: str
+    timestamp: datetime
+    consumption_kw: float | None = None
+    consumption_kwh: float | None = None
+    data_quality: str | None = None
+    null_reasons: list[str] | None = None
+
+
+@dataclass
+class FauxPrediction:
+    site_id: str
+    target_at: datetime
+    predicted_value: float | None
+    target_metric: str = "consumption_kwh"
+    prediction_id: int = 1
+
+
 class FakeRepository:
     def __init__(self, alerts: list[Alert]) -> None:
         self._alerts = alerts
         self.appels: list[tuple[str | None, str | None]] = []
+        self.crees: list[Alert] = []
 
     async def list_all(
         self, *, site_id: str | None = None, severity: str | None = None
@@ -37,19 +66,392 @@ class FakeRepository:
         self.appels.append((site_id, severity))
         return self._alerts
 
+    async def create_many(self, alerts: list[Alert]) -> list[Alert]:
+        self.crees = list(alerts)
+        return self.crees
+
+
+@dataclass
+class FauxDepotLectures:
+    depuis: list[FauxLecture] = field(default_factory=list)
+    dernieres: list[FauxLecture] = field(default_factory=list)
+
+    async def list_since(self, *, since: datetime, site_id: str | None = None) -> list[FauxLecture]:
+        return [lecture for lecture in self.depuis if site_id is None or lecture.site_id == site_id]
+
+    async def latest_by_site(self) -> list[FauxLecture]:
+        return self.dernieres
+
+
+@dataclass
+class FauxDepotPredictions:
+    predictions: list[FauxPrediction] = field(default_factory=list)
+
+    async def list_since(
+        self, *, since: datetime, site_id: str | None = None
+    ) -> list[FauxPrediction]:
+        return [p for p in self.predictions if site_id is None or p.site_id == site_id]
+
+
+@dataclass
+class FauxDepotSites:
+    sites: list[FauxSite]
+
+    async def list_all(self) -> list[FauxSite]:
+        return self.sites
+
+
+def service(
+    *,
+    sites: list[FauxSite],
+    lectures: list[FauxLecture] | None = None,
+    dernieres: list[FauxLecture] | None = None,
+    predictions: list[FauxPrediction] | None = None,
+    alerts: FakeRepository | None = None,
+) -> tuple[AlertService, FakeRepository]:
+    depot_alertes = alerts or FakeRepository([])
+    dernieres_lectures = dernieres if dernieres is not None else (lectures or [])
+    return (
+        AlertService(
+            alerts=depot_alertes,  # type: ignore[arg-type]
+            readings=FauxDepotLectures(depuis=lectures or [], dernieres=dernieres_lectures),  # type: ignore[arg-type]
+            predictions=FauxDepotPredictions(predictions or []),  # type: ignore[arg-type]
+            sites=FauxDepotSites(sites),  # type: ignore[arg-type]
+        ),
+        depot_alertes,
+    )
+
 
 async def test_list_all_returns_the_repository_alerts() -> None:
-    service = AlertService(alerts=FakeRepository([alert(1), alert(2)]))
+    svc, _ = service(sites=[], alerts=FakeRepository([alert(1), alert(2)]))
 
-    alertes = await service.list_all()
+    alertes = await svc.list_all()
 
     assert [a.alert_id for a in alertes] == [1, 2]
 
 
 async def test_list_all_relays_the_filters_to_the_repository() -> None:
     depot = FakeRepository([])
-    service = AlertService(alerts=depot)
+    svc, _ = service(sites=[], alerts=depot)
 
-    await service.list_all(site_id="site-1", severity="critical")
+    await svc.list_all(site_id="site-1", severity="critical")
 
     assert depot.appels == [("site-1", "critical")]
+
+
+async def test_detect_raises_a_threshold_alert_above_site_capacity() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A", capacity_kw=100.0)],
+        lectures=[FauxLecture("A", NOW, consumption_kw=150.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = depot.crees
+    assert candidate.type == "threshold"
+    assert candidate.severity == "high"
+    assert candidate.value == 150.0
+    assert candidate.threshold == 100.0
+    assert candidate.metric == "consumption_kw"
+
+
+async def test_detect_ignores_a_reading_within_capacity() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A", capacity_kw=100.0)],
+        lectures=[FauxLecture("A", NOW, consumption_kw=80.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert depot.crees == []
+
+
+async def test_detect_ignores_threshold_when_the_site_has_no_declared_capacity() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A", capacity_kw=None)],
+        lectures=[FauxLecture("A", NOW, consumption_kw=9999.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert depot.crees == []
+
+
+async def test_detect_raises_a_spike_alert_on_a_brutal_consecutive_variation() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=100.0),
+            FauxLecture("A", NOW, consumption_kw=160.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "spike"]
+    assert candidate.value == 160.0
+    assert candidate.threshold == 100.0
+    assert candidate.timestamp == NOW
+
+
+async def test_detect_ignores_a_moderate_consecutive_variation() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=100.0),
+            FauxLecture("A", NOW, consumption_kw=110.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "spike"] == []
+
+
+async def test_detect_never_compares_consecutive_readings_across_two_sites() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A"), FauxSite("B")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=10.0),
+            FauxLecture("B", NOW, consumption_kw=1000.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "spike"] == []
+
+
+async def test_detect_raises_an_anomaly_alert_far_from_the_matching_prediction() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, consumption_kwh=100.0)],
+        predictions=[FauxPrediction("A", target_at=NOW, predicted_value=70.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "anomaly"]
+    assert candidate.value == 100.0
+    assert candidate.threshold == 70.0
+    assert candidate.metric == "consumption_kwh"
+    assert candidate.prediction_id == 1
+
+
+async def test_detect_ignores_a_reading_close_to_its_prediction() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, consumption_kwh=100.0)],
+        predictions=[FauxPrediction("A", target_at=NOW, predicted_value=95.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "anomaly"] == []
+
+
+async def test_detect_ignores_a_prediction_whose_target_at_does_not_match_the_reading() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, consumption_kwh=100.0)],
+        predictions=[FauxPrediction("A", target_at=NOW - timedelta(hours=1), predicted_value=1.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "anomaly"] == []
+
+
+async def test_detect_keeps_the_most_recent_run_when_two_predictions_share_the_same_target() -> (
+    None
+):
+    # `PredictionRepository.list_since` départage les égalités de `target_at` par `prediction_id`
+    # croissant : le repository fait donc déjà passer le run le plus récent en dernier dans la
+    # liste, et c'est ce dernier que le dict de `_detect_anomaly` doit retenir.
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, consumption_kwh=100.0)],
+        predictions=[
+            FauxPrediction("A", target_at=NOW, predicted_value=100.0, prediction_id=1),
+            FauxPrediction("A", target_at=NOW, predicted_value=70.0, prediction_id=2),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "anomaly"]
+    assert candidate.threshold == 70.0
+    assert candidate.prediction_id == 2
+
+
+async def test_detect_raises_an_outage_alert_past_the_threshold() -> None:
+    derniere = NOW - OUTAGE_THRESHOLD - timedelta(minutes=1)
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[],
+        dernieres=[FauxLecture("A", derniere)],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "outage"]
+    assert candidate.severity in {"low", "medium", "high", "critical"}
+
+
+async def test_detect_ignores_a_site_still_within_the_outage_threshold() -> None:
+    derniere = NOW - OUTAGE_THRESHOLD + timedelta(minutes=1)
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[],
+        dernieres=[FauxLecture("A", derniere)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "outage"] == []
+
+
+async def test_detect_raises_a_critical_outage_alert_for_a_site_never_read() -> None:
+    svc, depot = service(sites=[FauxSite("A")], lectures=[], dernieres=[])
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "outage"]
+    assert candidate.severity == "critical"
+    assert candidate.source_alert_id == "outage:jamais"
+
+
+async def test_detect_raises_a_sensor_alert_on_a_degraded_reading() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, data_quality="critical", null_reasons=["missing:x"])],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "sensor"]
+    assert candidate.severity == "critical"
+
+
+async def test_detect_ignores_a_good_quality_reading_for_the_sensor_rule() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, data_quality="good")],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "sensor"] == []
+
+
+async def test_detect_scopes_to_a_single_site_when_asked() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A", capacity_kw=100.0), FauxSite("B", capacity_kw=100.0)],
+        lectures=[
+            FauxLecture("A", NOW, consumption_kw=150.0),
+            FauxLecture("B", NOW, consumption_kw=150.0),
+        ],
+    )
+
+    await svc.detect(now=NOW, site_id="A")
+
+    assert {a.site_id for a in depot.crees} == {"A"}
+
+
+async def test_detect_returns_early_when_there_is_no_site() -> None:
+    svc, depot = service(sites=[])
+
+    resultat = await svc.detect(now=NOW)
+
+    assert resultat == []
+    assert depot.crees == []
+
+
+async def test_detect_ignores_a_spike_pair_with_a_missing_measurement() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=None),
+            FauxLecture("A", NOW, consumption_kw=160.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "spike"] == []
+
+
+async def test_detect_ignores_a_reading_still_at_zero_after_a_previous_zero() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=0.0),
+            FauxLecture("A", NOW, consumption_kw=0.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "spike"] == []
+
+
+async def test_detect_raises_a_critical_spike_when_a_site_restarts_from_zero() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW - timedelta(hours=1), consumption_kw=0.0),
+            FauxLecture("A", NOW, consumption_kw=50.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    (candidate,) = [a for a in depot.crees if a.type == "spike"]
+    assert candidate.severity == "critical"
+    assert candidate.value == 50.0
+    assert candidate.threshold == 0.0
+
+
+async def test_detect_ignores_a_spike_pair_sharing_the_same_timestamp() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[
+            FauxLecture("A", NOW, consumption_kw=100.0),
+            FauxLecture("A", NOW, consumption_kw=160.0),
+        ],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "spike"] == []
+
+
+async def test_detect_ignores_an_anomaly_when_the_prediction_is_near_zero() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A")],
+        lectures=[FauxLecture("A", NOW, consumption_kwh=5.0)],
+        predictions=[FauxPrediction("A", target_at=NOW, predicted_value=0.0)],
+    )
+
+    await svc.detect(now=NOW)
+
+    assert [a for a in depot.crees if a.type == "anomaly"] == []
+
+
+def test_severity_from_ratio_covers_every_band() -> None:
+    assert _severity_from_ratio(1.0) == "low"
+    assert _severity_from_ratio(1.2) == "medium"
+    assert _severity_from_ratio(1.5) == "high"
+    assert _severity_from_ratio(2.0) == "critical"
+
+
+async def test_detect_does_not_call_create_many_when_nothing_triggers() -> None:
+    svc, depot = service(
+        sites=[FauxSite("A", capacity_kw=100.0)],
+        lectures=[FauxLecture("A", NOW, consumption_kw=10.0, data_quality="good")],
+    )
+
+    resultat = await svc.detect(now=NOW)
+
+    assert resultat == []
+    assert depot.crees == []

apps/backend/tests/services/test_auth.py

@@ -5,6 +5,7 @@ from typing import Any
 from uuid import UUID, uuid4
 
 import pytest
+from fastapi import BackgroundTasks
 
 from app.core.principal import Principal
 from app.core.roles import AccountKind, Role
@@ -16,11 +17,15 @@ from app.core.security import (
 from app.models.login_attempt import LoginOutcome
 from app.models.refresh_token import RevocationReason
 from app.repositories.login_attempt import FailureCounts
+from app.repositories.password_reset_attempt import ResetRequestCounts
+from app.repositories.password_reset_token import ConsumedResetToken
 from app.repositories.refresh_token import ClaimedToken
 from app.services.auth import (
     AuthService,
     InvalidCredentialsError,
+    InvalidOrExpiredResetTokenError,
     LoginPolicy,
+    PasswordResetPolicy,
     RateLimitedError,
     SessionRejectedError,
 )
@@ -37,6 +42,13 @@ POLITIQUE_CONNEXION = LoginPolicy(
     max_failures_per_ip=20,
     max_failures_per_identifier=50,
 )
+POLITIQUE_RESET = PasswordResetPolicy(
+    window_seconds=900,
+    max_requests_per_identifier=3,
+    max_requests_per_ip=10,
+    token_ttl=timedelta(minutes=15),
+    frontend_reset_url="http://localhost:4200/reset-password",
+)
 
 
 @dataclass
@@ -168,6 +180,49 @@ class FausseTransaction:
         self.validations += 1
 
 
+class FauxDepotJetonsReset:
+    def __init__(
+        self, revendique: ConsumedResetToken | None = None, *, valide: bool = False
+    ) -> None:
+        self.revendique = revendique
+        self.valide = valide
+        self.crees: list[UUID] = []
+        self.invalidations: list[UUID] = []
+
+    async def create(self, *, user_id: UUID, **_: object) -> None:
+        self.crees.append(user_id)
+
+    async def consume(self, token_hash: bytes) -> ConsumedResetToken | None:
+        return self.revendique
+
+    async def exists_valid(self, token_hash: bytes) -> bool:
+        return self.valide
+
+    async def invalidate_all_for_user(self, user_id: UUID) -> int:
+        self.invalidations.append(user_id)
+        return len(self.invalidations)
+
+
+class FauxDepotTentativesReset:
+    def __init__(self, compteurs: ResetRequestCounts | None = None) -> None:
+        self.compteurs = compteurs or ResetRequestCounts(0, 0)
+        self.enregistrees: list[str] = []
+
+    async def count_recent(self, **_: object) -> ResetRequestCounts:
+        return self.compteurs
+
+    async def record(self, *, email: str, **_: object) -> None:
+        self.enregistrees.append(email)
+
+
+class FauxMailer:
+    def __init__(self) -> None:
+        self.envois: list[tuple[str, str]] = []
+
+    async def send_password_reset_email(self, *, to: str, reset_url: str) -> None:
+        self.envois.append((to, reset_url))
+
+
 @dataclass
 class Attirail:
     service: AuthService
@@ -176,6 +231,9 @@ class Attirail:
     jetons: FauxDepotJetons
     audit: FauxDepotAudit
     hacheur: FauxHacheur
+    jetons_reset: FauxDepotJetonsReset
+    tentatives_reset: FauxDepotTentativesReset
+    mailer: FauxMailer
 
 
 def fabrique_service(
@@ -184,12 +242,17 @@ def fabrique_service(
     compteurs: FailureCounts | None = None,
     hacheur: FauxHacheur | None = None,
     jetons: FauxDepotJetons | None = None,
+    jetons_reset: FauxDepotJetonsReset | None = None,
+    compteurs_reset: ResetRequestCounts | None = None,
 ) -> Attirail:
     comptes = FauxDepotComptes(compte)
     tentatives = FauxDepotTentatives(compteurs)
     depot_jetons = jetons or FauxDepotJetons()
     audit = FauxDepotAudit()
     hacheur = hacheur or FauxHacheur()
+    depot_jetons_reset = jetons_reset or FauxDepotJetonsReset()
+    tentatives_reset = FauxDepotTentativesReset(compteurs_reset)
+    mailer = FauxMailer()
     service = AuthService(
         users=comptes,  # type: ignore[arg-type]
         attempts=tentatives,  # type: ignore[arg-type]
@@ -200,8 +263,22 @@ def fabrique_service(
         token_policy=POLITIQUE_JETON,
         login_policy=POLITIQUE_CONNEXION,
         refresh_ttl=timedelta(days=7),
+        reset_tokens=depot_jetons_reset,  # type: ignore[arg-type]
+        reset_attempts=tentatives_reset,  # type: ignore[arg-type]
+        reset_policy=POLITIQUE_RESET,
+        mailer=mailer,  # type: ignore[arg-type]
+    )
+    return Attirail(
+        service,
+        comptes,
+        tentatives,
+        depot_jetons,
+        audit,
+        hacheur,
+        depot_jetons_reset,
+        tentatives_reset,
+        mailer,
     )
-    return Attirail(service, comptes, tentatives, depot_jetons, audit, hacheur)
 
 
 async def connecte(service: AuthService, mot_de_passe: str = "un-mot-de-passe-valide") -> object:
@@ -493,3 +570,148 @@ async def test_change_password_refuses_a_wrong_current_password() -> None:
 
     assert attirail.jetons.revocations_par_compte == []
     assert attirail.jetons.crees == []
+
+
+async def test_request_password_reset_emails_a_link_when_the_account_exists() -> None:
+    compte = FauxCompte()
+    attirail = fabrique_service(compte=compte)
+    taches = BackgroundTasks()
+
+    await attirail.service.request_password_reset(
+        email=compte.email, client_ip="203.0.113.10", user_agent="pytest", background_tasks=taches
+    )
+
+    assert attirail.jetons_reset.invalidations == [compte.id]
+    assert attirail.jetons_reset.crees == [compte.id]
+    assert attirail.mailer.envois == [], "l'envoi doit être différé, pas fait dans la réponse"
+    await taches()
+    assert len(attirail.mailer.envois) == 1
+    assert attirail.mailer.envois[0][0] == compte.email
+    assert "auth.password_reset_requested" in attirail.audit.lignes[0][0]
+
+
+async def test_request_password_reset_stays_silent_when_the_account_is_unknown() -> None:
+    attirail = fabrique_service(compte=None)
+    taches = BackgroundTasks()
+
+    await attirail.service.request_password_reset(
+        email="inconnu@enervision.fr",
+        client_ip="203.0.113.10",
+        user_agent="pytest",
+        background_tasks=taches,
+    )
+    await taches()
+
+    assert attirail.jetons_reset.crees == []
+    assert attirail.mailer.envois == []
+    assert attirail.hacheur.verifications == 1, "le hachage factice doit tout de même tourner"
+
+
+async def test_request_password_reset_stays_silent_when_the_account_is_inactive() -> None:
+    compte = FauxCompte(is_active=False)
+    attirail = fabrique_service(compte=compte)
+    taches = BackgroundTasks()
+
+    await attirail.service.request_password_reset(
+        email=compte.email, client_ip="203.0.113.10", user_agent="pytest", background_tasks=taches
+    )
+    await taches()
+
+    assert attirail.jetons_reset.crees == []
+    assert attirail.mailer.envois == []
+
+
+async def test_request_password_reset_raises_when_the_rate_limit_is_reached() -> None:
+    attirail = fabrique_service(compteurs_reset=ResetRequestCounts(per_identifier=3, per_ip=0))
+    taches = BackgroundTasks()
+
+    with pytest.raises(RateLimitedError):
+        await attirail.service.request_password_reset(
+            email="operateur@enervision.fr",
+            client_ip="203.0.113.10",
+            user_agent="pytest",
+            background_tasks=taches,
+        )
+
+    await taches()
+    assert attirail.mailer.envois == []
+
+
+async def test_request_password_reset_logs_instead_of_raising_when_the_mailer_fails() -> None:
+    compte = FauxCompte()
+    attirail = fabrique_service(compte=compte)
+    taches = BackgroundTasks()
+
+    async def echoue(*, to: str, reset_url: str) -> None:
+        raise RuntimeError("relais SMTP indisponible")
+
+    attirail.mailer.send_password_reset_email = echoue  # type: ignore[method-assign]
+
+    await attirail.service.request_password_reset(
+        email=compte.email, client_ip="203.0.113.10", user_agent="pytest", background_tasks=taches
+    )
+
+    await taches()
+
+
+async def test_confirm_password_reset_revokes_every_session_then_reopens_the_current_one() -> None:
+    compte = FauxCompte()
+    jetons_reset = FauxDepotJetonsReset(
+        revendique=ConsumedResetToken(id=uuid4(), user_id=compte.id)
+    )
+    attirail = fabrique_service(compte=compte, jetons_reset=jetons_reset)
+
+    session = await attirail.service.confirm_password_reset(
+        token="un-secret-opaque",
+        new_password="Un-nouveau-mot-de-passe1!",
+        client_ip="203.0.113.10",
+        user_agent="pytest",
+    )
+
+    assert attirail.jetons.revocations_par_compte == [
+        (compte.id, RevocationReason.CHANGEMENT_MOT_DE_PASSE.value)
+    ]
+    assert len(attirail.jetons.crees) == 1
+    assert session.refresh_secret
+    assert "auth.password_reset_self_service" in attirail.audit.lignes[0][0]
+
+
+async def test_is_reset_token_valid_reflects_the_repository() -> None:
+    attirail_valide = fabrique_service(jetons_reset=FauxDepotJetonsReset(valide=True))
+    attirail_invalide = fabrique_service(jetons_reset=FauxDepotJetonsReset(valide=False))
+
+    assert await attirail_valide.service.is_reset_token_valid("un-secret-opaque") is True
+    assert await attirail_invalide.service.is_reset_token_valid("un-secret-opaque") is False
+
+
+async def test_confirm_password_reset_rejects_a_token_for_an_account_disabled_since() -> None:
+    compte = FauxCompte(is_active=False)
+    jetons_reset = FauxDepotJetonsReset(
+        revendique=ConsumedResetToken(id=uuid4(), user_id=compte.id)
+    )
+    attirail = fabrique_service(compte=compte, jetons_reset=jetons_reset)
+
+    with pytest.raises(InvalidOrExpiredResetTokenError):
+        await attirail.service.confirm_password_reset(
+            token="un-secret-opaque",
+            new_password="Un-nouveau-mot-de-passe1!",
+            client_ip="203.0.113.10",
+            user_agent="pytest",
+        )
+
+    assert attirail.comptes.mots_de_passe_changes == 0
+    assert attirail.jetons.revocations_par_compte == []
+
+
+async def test_confirm_password_reset_rejects_an_invalid_or_expired_token() -> None:
+    attirail = fabrique_service(jetons_reset=FauxDepotJetonsReset(revendique=None))
+
+    with pytest.raises(InvalidOrExpiredResetTokenError):
+        await attirail.service.confirm_password_reset(
+            token="un-secret-invalide",
+            new_password="Un-nouveau-mot-de-passe1!",
+            client_ip=None,
+            user_agent=None,
+        )
+
+    assert attirail.jetons.revocations_par_compte == []

apps/backend/tests/services/test_prediction.py

@@ -0,0 +1,121 @@
+from dataclasses import dataclass
+from datetime import UTC, datetime
+
+from app.services.prediction import PredictionService
+
+TARGET_AT = datetime(2026, 9, 16, 13, 0, tzinfo=UTC)
+CREATED_AT = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
+
+
+@dataclass
+class FauxSite:
+    site_id: str
+    site_name: str
+
+
+@dataclass
+class FauxPrediction:
+    site_id: str
+    target_at: datetime
+    target_metric: str
+    period_minutes: int | None
+    predicted_value: float | None
+    status: str
+    failure_reason: str | None
+    model_reference: str
+    created_at: datetime
+
+
+class FauxDepotSites:
+    def __init__(self, sites: list[FauxSite]) -> None:
+        self._sites = sites
+
+    async def list_all(self) -> list[FauxSite]:
+        return self._sites
+
+
+class FauxDepotPredictions:
+    def __init__(self, predictions: list[FauxPrediction]) -> None:
+        self._predictions = predictions
+
+    async def latest_by_site(self) -> list[FauxPrediction]:
+        return self._predictions
+
+
+def prediction_disponible(site_id: str = "A") -> FauxPrediction:
+    return FauxPrediction(
+        site_id=site_id,
+        target_at=TARGET_AT,
+        target_metric="consumption_kwh",
+        period_minutes=60,
+        predicted_value=812.5,
+        status="available",
+        failure_reason=None,
+        model_reference="lightgbm-abc123",
+        created_at=CREATED_AT,
+    )
+
+
+async def test_summary_attaches_the_latest_prediction_to_its_site() -> None:
+    service = PredictionService(
+        sites=FauxDepotSites([FauxSite("A", "Site A")]),  # type: ignore[arg-type]
+        predictions=FauxDepotPredictions([prediction_disponible("A")]),  # type: ignore[arg-type]
+    )
+
+    resume = await service.summary()
+
+    site = resume.sites[0]
+    assert site.site_id == "A"
+    assert site.prediction is not None
+    assert site.prediction.predicted_value == 812.5
+    assert site.prediction.status == "available"
+
+
+async def test_summary_leaves_prediction_none_for_a_site_never_scored() -> None:
+    service = PredictionService(
+        sites=FauxDepotSites([FauxSite("A", "Site A")]),  # type: ignore[arg-type]
+        predictions=FauxDepotPredictions([]),  # type: ignore[arg-type]
+    )
+
+    resume = await service.summary()
+
+    assert resume.sites[0].prediction is None
+
+
+async def test_summary_carries_an_insufficient_data_prediction_without_a_value() -> None:
+    insuffisante = FauxPrediction(
+        site_id="A",
+        target_at=TARGET_AT,
+        target_metric="consumption_kwh",
+        period_minutes=60,
+        predicted_value=None,
+        status="insufficient_data",
+        failure_reason="pas assez d'historique",
+        model_reference="lightgbm-abc123",
+        created_at=CREATED_AT,
+    )
+    service = PredictionService(
+        sites=FauxDepotSites([FauxSite("A", "Site A")]),  # type: ignore[arg-type]
+        predictions=FauxDepotPredictions([insuffisante]),  # type: ignore[arg-type]
+    )
+
+    resume = await service.summary()
+
+    site = resume.sites[0]
+    assert site.prediction is not None
+    assert site.prediction.status == "insufficient_data"
+    assert site.prediction.predicted_value is None
+    assert site.prediction.failure_reason == "pas assez d'historique"
+
+
+async def test_summary_covers_every_site_even_with_a_single_prediction_in_the_repository() -> None:
+    service = PredictionService(
+        sites=FauxDepotSites([FauxSite("A", "Site A"), FauxSite("B", "Site B")]),  # type: ignore[arg-type]
+        predictions=FauxDepotPredictions([prediction_disponible("A")]),  # type: ignore[arg-type]
+    )
+
+    resume = await service.summary()
+
+    par_site = {site.site_id: site for site in resume.sites}
+    assert par_site["A"].prediction is not None
+    assert par_site["B"].prediction is None

apps/backend/tests/services/test_reading.py

@@ -0,0 +1,151 @@
+from datetime import UTC, datetime, timedelta
+
+import pytest
+
+from app.models.energy import Reading
+from app.services.reading import (
+    FENETRE_MAXIMALE,
+    FENETRE_PAR_DEFAUT,
+    FenetreInverseeError,
+    FenetreTropLargeError,
+    ReadingService,
+)
+
+
+def reading(reading_id: int = 1, site_id: str = "site-1") -> Reading:
+    return Reading(
+        reading_id=reading_id,
+        site_id=site_id,
+        timestamp=datetime(2026, 9, 16, tzinfo=UTC),
+        source="api_current",
+        consumption_kw=10.0,
+        data_quality="good",
+        raw_data={},
+    )
+
+
+class FakeRepository:
+    def __init__(self, readings: list[Reading]) -> None:
+        self._readings = readings
+        self.appels: list[tuple[str | None, datetime, datetime, int, int]] = []
+
+    async def list_history(
+        self,
+        *,
+        start: datetime,
+        end: datetime,
+        site_id: str | None = None,
+        limit: int,
+        offset: int,
+    ) -> list[Reading]:
+        self.appels.append((site_id, start, end, limit, offset))
+        return self._readings
+
+
+async def test_list_history_returns_the_repository_readings() -> None:
+    service = ReadingService(readings=FakeRepository([reading(1), reading(2)]))
+
+    lectures = await service.list_history(limit=500, offset=0)
+
+    assert [r.reading_id for r in lectures] == [1, 2]
+
+
+async def test_list_history_relays_the_site_id_limit_and_offset() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+    debut = datetime(2026, 9, 1, tzinfo=UTC)
+    fin = datetime(2026, 9, 2, tzinfo=UTC)
+
+    await service.list_history(site_id="site-1", start=debut, end=fin, limit=50, offset=10)
+
+    assert depot.appels == [("site-1", debut, fin, 50, 10)]
+
+
+async def test_list_history_defaults_to_the_last_24_hours_when_no_window_is_given() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+    avant = datetime.now(UTC)
+
+    await service.list_history(limit=500, offset=0)
+
+    apres = datetime.now(UTC)
+    _, debut, fin, _, _ = depot.appels[0]
+    assert avant <= fin <= apres
+    assert fin - debut == FENETRE_PAR_DEFAUT
+
+
+async def test_list_history_defaults_end_to_now_when_only_start_is_given() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+    debut = datetime.now(UTC) - timedelta(hours=1)
+    avant = datetime.now(UTC)
+
+    await service.list_history(start=debut, limit=500, offset=0)
+
+    apres = datetime.now(UTC)
+    _, debut_transmis, fin, _, _ = depot.appels[0]
+    assert debut_transmis == debut
+    assert avant <= fin <= apres
+
+
+async def test_list_history_defaults_start_to_24_hours_before_end_when_only_end_is_given() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+    fin = datetime(2026, 9, 16, tzinfo=UTC)
+
+    await service.list_history(end=fin, limit=500, offset=0)
+
+    _, debut, fin_transmise, _, _ = depot.appels[0]
+    assert fin_transmise == fin
+    assert debut == fin - FENETRE_PAR_DEFAUT
+
+
+async def test_list_history_normalizes_naive_datetimes_to_utc() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+
+    await service.list_history(
+        start=datetime(2026, 9, 1), end=datetime(2026, 9, 2), limit=500, offset=0
+    )
+
+    _, debut, fin, _, _ = depot.appels[0]
+    assert debut == datetime(2026, 9, 1, tzinfo=UTC)
+    assert fin == datetime(2026, 9, 2, tzinfo=UTC)
+
+
+async def test_list_history_raises_when_start_is_after_end() -> None:
+    service = ReadingService(readings=FakeRepository([]))
+
+    debut = datetime(2026, 9, 2, tzinfo=UTC)
+    fin = datetime(2026, 9, 1, tzinfo=UTC)
+
+    with pytest.raises(FenetreInverseeError):
+        await service.list_history(start=debut, end=fin, limit=500, offset=0)
+
+
+async def test_list_history_raises_when_start_equals_end() -> None:
+    service = ReadingService(readings=FakeRepository([]))
+    instant = datetime(2026, 9, 1, tzinfo=UTC)
+
+    with pytest.raises(FenetreInverseeError):
+        await service.list_history(start=instant, end=instant, limit=500, offset=0)
+
+
+async def test_list_history_raises_when_the_window_exceeds_the_maximum_span() -> None:
+    service = ReadingService(readings=FakeRepository([]))
+    debut = datetime(2026, 1, 1, tzinfo=UTC)
+    fin = debut + FENETRE_MAXIMALE + timedelta(seconds=1)
+
+    with pytest.raises(FenetreTropLargeError):
+        await service.list_history(start=debut, end=fin, limit=500, offset=0)
+
+
+async def test_list_history_accepts_a_window_exactly_at_the_maximum_span() -> None:
+    depot = FakeRepository([])
+    service = ReadingService(readings=depot)
+    debut = datetime(2026, 1, 1, tzinfo=UTC)
+    fin = debut + FENETRE_MAXIMALE
+
+    await service.list_history(start=debut, end=fin, limit=500, offset=0)
+
+    assert depot.appels == [(None, debut, fin, 500, 0)]

apps/backend/tests/services/test_recommendation.py

@@ -1,10 +1,14 @@
+from collections.abc import Sequence
 from datetime import UTC, datetime
 
 import pytest
 
-from app.models.energy import Recommendation
+from app.models.energy import Alert, Recommendation
+from app.repositories.recommendation import NouvelleRecommandation
 from app.services.recommendation import RecommendationNotFoundError, RecommendationService
 
+MOMENT = datetime(2024, 1, 1, tzinfo=UTC)
+
 
 def recommendation(recommendation_id: int = 1) -> Recommendation:
     return Recommendation(
@@ -13,13 +17,33 @@ def recommendation(recommendation_id: int = 1) -> Recommendation:
         action="Vérifier la consommation",
         explanation="Pic détecté",
         rule_reference="spike-v1",
-        created_at=datetime(2024, 1, 1, tzinfo=UTC),
+        created_at=MOMENT,
+    )
+
+
+def alerte(alert_id: int = 1, site_id: str = "SITE001", severity: str = "high") -> Alert:
+    return Alert(
+        alert_id=alert_id,
+        source_alert_id=f"ALR-{alert_id}",
+        site_id=site_id,
+        source="api_mock",
+        timestamp=MOMENT,
+        type="spike",
+        severity=severity,
+        message="Pic de consommation",
+        value=None,
+        threshold=None,
+        metric=None,
+        prediction_id=None,
+        raw_data={},
     )
 
 
 class FakeRepository:
-    def __init__(self, recommendations: list[Recommendation]) -> None:
+    def __init__(self, recommendations: list[Recommendation], creees: int | None = None) -> None:
         self._recommendations = recommendations
+        self._creees = creees
+        self.recues: list[NouvelleRecommandation] = []
 
     async def list_all(self) -> list[Recommendation]:
         return self._recommendations
@@ -29,27 +53,111 @@ class FakeRepository:
             (r for r in self._recommendations if r.recommendation_id == recommendation_id), None
         )
 
+    async def create_missing(self, nouvelles: Sequence[NouvelleRecommandation]) -> int:
+        self.recues = list(nouvelles)
+        return len(self.recues) if self._creees is None else self._creees
 
-async def test_list_all_returns_the_repository_recommendations() -> None:
-    service = RecommendationService(
-        recommendations=FakeRepository([recommendation(1), recommendation(2)])
+
+class FakeAlertRepository:
+    def __init__(self, alertes: list[Alert]) -> None:
+        self._alertes = alertes
+        self.site_demande: str | None = None
+
+    async def list_all(
+        self, *, site_id: str | None = None, severity: str | None = None
+    ) -> list[Alert]:
+        self.site_demande = site_id
+        if site_id is None:
+            return self._alertes
+        return [a for a in self._alertes if a.site_id == site_id]
+
+
+class FakeTransaction:
+    def __init__(self) -> None:
+        self.commits = 0
+
+    async def commit(self) -> None:
+        self.commits += 1
+
+
+def service(
+    recommendations: FakeRepository | None = None,
+    alerts: FakeAlertRepository | None = None,
+    transaction: FakeTransaction | None = None,
+) -> RecommendationService:
+    return RecommendationService(
+        recommendations=recommendations or FakeRepository([]),
+        alerts=alerts or FakeAlertRepository([]),
+        transaction=transaction or FakeTransaction(),
     )
 
-    recommendations = await service.list_all()
+
+async def test_list_all_returns_the_repository_recommendations() -> None:
+    depot = FakeRepository([recommendation(1), recommendation(2)])
+
+    recommendations = await service(recommendations=depot).list_all()
 
     assert [r.recommendation_id for r in recommendations] == [1, 2]
 
 
 async def test_get_by_id_returns_the_matching_recommendation() -> None:
-    service = RecommendationService(recommendations=FakeRepository([recommendation(1)]))
-
-    trouve = await service.get_by_id(1)
+    trouve = await service(recommendations=FakeRepository([recommendation(1)])).get_by_id(1)
 
     assert trouve.recommendation_id == 1
 
 
 async def test_get_by_id_raises_when_the_recommendation_is_unknown() -> None:
-    service = RecommendationService(recommendations=FakeRepository([]))
-
     with pytest.raises(RecommendationNotFoundError):
-        await service.get_by_id(404)
+        await service().get_by_id(404)
+
+
+async def test_generate_persists_one_proposal_per_triggered_rule() -> None:
+    depot = FakeRepository([])
+
+    rapport = await service(
+        recommendations=depot, alerts=FakeAlertRepository([alerte(severity="critical")])
+    ).generate()
+
+    assert {n.rule_reference for n in depot.recues} == {
+        "spike-delestage-v1",
+        "escalade-astreinte-v1",
+    }
+    assert rapport.recommandations_creees == 2
+
+
+async def test_generate_commits_once() -> None:
+    transaction = FakeTransaction()
+
+    await service(alerts=FakeAlertRepository([alerte()]), transaction=transaction).generate()
+
+    assert transaction.commits == 1
+
+
+async def test_generate_restricts_the_alerts_to_the_requested_site() -> None:
+    alertes = FakeAlertRepository([alerte(1, site_id="SITE001"), alerte(2, site_id="SITE002")])
+    depot = FakeRepository([])
+
+    rapport = await service(recommendations=depot, alerts=alertes).generate(site_id="SITE002")
+
+    assert alertes.site_demande == "SITE002"
+    assert rapport.alertes_examinees == 1
+    assert {n.alert_id for n in depot.recues} == {2}
+
+
+async def test_generate_reports_nothing_when_no_alert_matches() -> None:
+    rapport = await service().generate()
+
+    assert rapport.alertes_examinees == 0
+    assert rapport.recommandations_creees == 0
+    assert rapport.deja_presentes == 0
+
+
+async def test_generate_counts_the_proposals_the_database_already_held() -> None:
+    depot = FakeRepository([], creees=0)
+
+    rapport = await service(
+        recommendations=depot, alerts=FakeAlertRepository([alerte()])
+    ).generate()
+
+    assert rapport.recommandations_creees == 0
+    assert rapport.deja_presentes == 1

apps/backend/tests/services/test_recommendation_rules.py

@@ -0,0 +1,142 @@
+from datetime import UTC, datetime
+
+import pytest
+
+from app.models.energy import Alert
+from app.services.recommendation_rules import FACTEUR_DEPASSEMENT_MAJEUR, applique_les_regles
+
+MOMENT = datetime(2024, 1, 1, tzinfo=UTC)
+
+
+def alerte(
+    *,
+    alert_id: int = 1,
+    type_alerte: str = "spike",
+    severity: str = "high",
+    value: float | None = None,
+    threshold: float | None = None,
+    metric: str | None = None,
+    site_id: str = "SITE001",
+) -> Alert:
+    return Alert(
+        alert_id=alert_id,
+        source_alert_id=f"ALR-{alert_id}",
+        site_id=site_id,
+        source="api_mock",
+        timestamp=MOMENT,
+        type=type_alerte,
+        severity=severity,
+        message="Alerte de test",
+        value=value,
+        threshold=threshold,
+        metric=metric,
+        prediction_id=None,
+        raw_data={},
+    )
+
+
+@pytest.mark.parametrize(
+    ("type_alerte", "attendue"),
+    [
+        ("spike", "spike-delestage-v1"),
+        ("threshold", "threshold-reduction-v1"),
+        ("outage", "outage-secours-v1"),
+        ("sensor", "sensor-maintenance-v1"),
+        ("anomaly", "anomaly-verification-v1"),
+    ],
+    ids=["pic", "seuil", "coupure", "capteur", "anomalie"],
+)
+def test_each_alert_type_yields_its_own_rule(type_alerte: str, attendue: str) -> None:
+    proposees = applique_les_regles(alerte(type_alerte=type_alerte))
+
+    assert [p.rule_reference for p in proposees] == [attendue]
+
+
+def test_a_critical_alert_adds_the_escalation_rule() -> None:
+    proposees = applique_les_regles(alerte(severity="critical"))
+
+    assert "escalade-astreinte-v1" in {p.rule_reference for p in proposees}
+
+
+@pytest.mark.parametrize("severity", ["low", "medium", "high"], ids=["faible", "moyenne", "haute"])
+def test_a_non_critical_alert_does_not_escalate(severity: str) -> None:
+    proposees = applique_les_regles(alerte(severity=severity))
+
+    assert "escalade-astreinte-v1" not in {p.rule_reference for p in proposees}
+
+
+def test_a_large_overshoot_adds_the_contract_rule() -> None:
+    proposees = applique_les_regles(
+        alerte(value=720.0 * FACTEUR_DEPASSEMENT_MAJEUR, threshold=720.0)
+    )
+
+    assert "contrat-puissance-v1" in {p.rule_reference for p in proposees}
+
+
+def test_an_overshoot_below_the_factor_does_not_add_the_contract_rule() -> None:
+    proposees = applique_les_regles(alerte(value=800.0, threshold=720.0))
+
+    assert "contrat-puissance-v1" not in {p.rule_reference for p in proposees}
+
+
+@pytest.mark.parametrize(
+    ("value", "threshold"),
+    [(None, 720.0), (900.0, None), (900.0, 0.0), (900.0, -10.0)],
+    ids=["sans mesure", "sans seuil", "seuil nul", "seuil negatif"],
+)
+def test_the_contract_rule_stays_silent_without_an_exploitable_threshold(
+    value: float | None, threshold: float | None
+) -> None:
+    proposees = applique_les_regles(alerte(value=value, threshold=threshold))
+
+    assert "contrat-puissance-v1" not in {p.rule_reference for p in proposees}
+
+
+def test_the_explanation_quotes_the_measure_and_the_threshold() -> None:
+    proposees = applique_les_regles(alerte(value=812.5, threshold=720.0, metric="consumption_kw"))
+
+    assert "(consumption_kw mesurée à 812.5, seuil 720.0)" in proposees[0].explanation
+
+
+def test_the_explanation_quotes_the_measure_alone_when_no_threshold_is_known() -> None:
+    proposees = applique_les_regles(alerte(value=812.5, metric="consumption_kw"))
+
+    assert "(consumption_kw mesurée à 812.5)" in proposees[0].explanation
+
+
+def test_the_explanation_omits_the_measure_when_the_alert_carries_none() -> None:
+    proposees = applique_les_regles(alerte())
+
+    assert "(" not in proposees[0].explanation
+
+
+def test_the_explanation_names_the_site() -> None:
+    proposees = applique_les_regles(alerte(site_id="SITE042"))
+
+    assert "SITE042" in proposees[0].explanation
+
+
+def test_every_proposal_carries_the_alert_identifier() -> None:
+    proposees = applique_les_regles(alerte(alert_id=77, severity="critical"))
+
+    assert {p.alert_id for p in proposees} == {77}
+
+
+def test_an_alert_never_yields_the_same_rule_twice() -> None:
+    proposees = applique_les_regles(
+        alerte(severity="critical", value=900.0, threshold=720.0, metric="consumption_kw")
+    )
+
+    assert len(proposees) == len({p.rule_reference for p in proposees})
+
+
+def test_a_critical_alert_over_the_threshold_yields_the_three_rules() -> None:
+    proposees = applique_les_regles(
+        alerte(severity="critical", value=900.0, threshold=720.0, metric="consumption_kw")
+    )
+
+    assert {p.rule_reference for p in proposees} == {
+        "spike-delestage-v1",
+        "escalade-astreinte-v1",
+        "contrat-puissance-v1",
+    }

apps/backend/tests/services/test_site.py

@@ -1,8 +1,13 @@
+from dataclasses import dataclass, field
+from datetime import UTC, datetime
+
 import pytest
 
 from app.models.energy import Site
 from app.services.site import SiteNotFoundError, SiteService
 
+TIMESTAMP = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
+
 
 def site(site_id: str = "site-1") -> Site:
     return Site(
@@ -15,6 +20,21 @@ def site(site_id: str = "site-1") -> Site:
     )
 
 
+@dataclass
+class FauxLecture:
+    site_id: str
+    timestamp: datetime = TIMESTAMP
+    consumption_kw: float | None = 87.34
+    consumption_kwh: float | None = 87.34
+    voltage_v: float | None = 401.2
+    current_a: float | None = 132.5
+    power_factor: float | None = 0.923
+    temperature_celsius: float | None = 22.1
+    humidity_percent: float | None = 58.4
+    null_reasons: list[str] | None = field(default_factory=list)
+    data_quality: str | None = "good"
+
+
 class FakeRepository:
     def __init__(self, sites: list[Site]) -> None:
         self._sites = sites
@@ -26,24 +46,77 @@ class FakeRepository:
         return next((s for s in self._sites if s.site_id == site_id), None)
 
 
-async def test_list_all_returns_the_repository_sites() -> None:
-    service = SiteService(sites=FakeRepository([site("a"), site("b")]))
+class FauxDepotLectures:
+    def __init__(self, lectures: dict[str, FauxLecture]) -> None:
+        self._lectures = lectures
 
-    sites = await service.list_all()
+    async def latest_for_site(self, site_id: str) -> FauxLecture | None:
+        return self._lectures.get(site_id)
+
+
+def service(sites: list[Site], lectures: dict[str, FauxLecture] | None = None) -> SiteService:
+    return SiteService(
+        sites=FakeRepository(sites),  # type: ignore[arg-type]
+        readings=FauxDepotLectures(lectures or {}),  # type: ignore[arg-type]
+    )
+
+
+async def test_list_all_returns_the_repository_sites() -> None:
+    svc = service([site("a"), site("b")])
+
+    sites = await svc.list_all()
 
     assert [s.site_id for s in sites] == ["a", "b"]
 
 
 async def test_get_by_id_returns_the_matching_site() -> None:
-    service = SiteService(sites=FakeRepository([site("a")]))
+    svc = service([site("a")])
 
-    trouve = await service.get_by_id("a")
+    trouve = await svc.get_by_id("a")
 
     assert trouve.site_id == "a"
 
 
 async def test_get_by_id_raises_when_the_site_is_unknown() -> None:
-    service = SiteService(sites=FakeRepository([]))
+    svc = service([])
 
     with pytest.raises(SiteNotFoundError):
-        await service.get_by_id("inconnu")
+        await svc.get_by_id("inconnu")
+
+
+async def test_current_raises_when_the_site_is_unknown() -> None:
+    svc = service([])
+
+    with pytest.raises(SiteNotFoundError):
+        await svc.current("inconnu")
+
+
+async def test_current_returns_every_field_as_null_when_the_site_has_no_reading() -> None:
+    svc = service([site("a")])
+
+    actuel = await svc.current("a")
+
+    assert actuel.timestamp is None
+    assert actuel.consumption_kw is None
+    assert actuel.data_quality == "critical"
+    assert actuel.null_reasons == []
+
+
+async def test_current_copies_every_field_from_the_latest_reading() -> None:
+    svc = service([site("a")], {"a": FauxLecture(site_id="a")})
+
+    actuel = await svc.current("a")
+
+    assert actuel.timestamp == TIMESTAMP
+    assert actuel.site_type == "industriel"
+    assert actuel.consumption_kw == 87.34
+    assert actuel.voltage_v == 401.2
+    assert actuel.data_quality == "good"
+
+
+async def test_current_treats_an_unknown_data_quality_as_critical() -> None:
+    svc = service([site("a")], {"a": FauxLecture(site_id="a", data_quality=None)})
+
+    actuel = await svc.current("a")
+
+    assert actuel.data_quality == "critical"

apps/backend/tests/services/test_user.py

@@ -235,5 +235,7 @@ async def test_every_operation_refuses_an_unknown_account(action: str) -> None:
     if action == "set_active":
         arguments["is_active"] = False
 
+    methode = getattr(attirail.service, action)
+
     with pytest.raises(UserNotFoundError):
-        await getattr(attirail.service, action)(**arguments)
+        await methode(**arguments)

apps/backend/tests/test_cli.py

@@ -4,6 +4,7 @@ from pathlib import Path
 import pytest
 
 from app import cli
+from app.schemas.auth import valide_complexite
 
 
 def test_build_parser_reads_the_create_admin_arguments() -> None:
@@ -18,13 +19,17 @@ def test_build_parser_reads_the_create_admin_arguments() -> None:
 
 
 def test_build_parser_requires_a_subcommand() -> None:
+    parser = cli.build_parser()
+
     with pytest.raises(SystemExit):
-        cli.build_parser().parse_args([])
+        parser.parse_args([])
 
 
 def test_build_parser_requires_an_email() -> None:
+    parser = cli.build_parser()
+
     with pytest.raises(SystemExit):
-        cli.build_parser().parse_args(["create-admin"])
+        parser.parse_args(["create-admin"])
 
 
 def test_read_password_generates_a_long_secret_when_asked(
@@ -34,26 +39,36 @@ def test_read_password_generates_a_long_secret_when_asked(
 
     assert len(mot_de_passe) >= cli.LONGUEUR_MOT_DE_PASSE_GENERE
     assert mot_de_passe in capsys.readouterr().out
+    valide_complexite(mot_de_passe)
 
 
 def test_read_password_accepts_two_matching_entries(monkeypatch: pytest.MonkeyPatch) -> None:
-    saisies = iter(["un-mot-de-passe-valide", "un-mot-de-passe-valide"])
+    saisies = iter(["Un-mot-de-passe-valide1", "Un-mot-de-passe-valide1"])
     monkeypatch.setattr(cli, "getpass", lambda _: next(saisies))
 
-    assert cli.read_password(generate=False) == "un-mot-de-passe-valide"
+    assert cli.read_password(generate=False) == "Un-mot-de-passe-valide1"
 
 
 def test_read_password_refuses_a_password_below_the_minimum_length(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
-    monkeypatch.setattr(cli, "getpass", lambda _: "court")
+    monkeypatch.setattr(cli, "getpass", lambda _: "Court1!")
+
+    with pytest.raises(SystemExit):
+        cli.read_password(generate=False)
+
+
+def test_read_password_refuses_a_password_missing_a_character_class(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(cli, "getpass", lambda _: "un-mot-de-passe-sans-majuscule-ni-chiffre")
 
     with pytest.raises(SystemExit):
         cli.read_password(generate=False)
 
 
 def test_read_password_refuses_two_different_entries(monkeypatch: pytest.MonkeyPatch) -> None:
-    saisies = iter(["un-mot-de-passe-valide", "un-autre-mot-de-passe"])
+    saisies = iter(["Un-mot-de-passe-valide1", "Un-autre-mot-de-passe2"])
     monkeypatch.setattr(cli, "getpass", lambda _: next(saisies))
 
     with pytest.raises(SystemExit):
@@ -107,3 +122,30 @@ def test_main_exports_the_contract_without_asking_for_a_password(
     assert code == 0
     assert destination.exists()
     assert str(destination) in capsys.readouterr().out
+
+
+def test_build_parser_reads_the_generate_recommendations_arguments() -> None:
+    arguments = cli.build_parser().parse_args(["generate-recommendations", "--site-id", "SITE002"])
+
+    assert arguments.commande == "generate-recommendations"
+    assert arguments.site_id == "SITE002"
+
+
+def test_build_parser_defaults_the_generation_to_every_site() -> None:
+    arguments = cli.build_parser().parse_args(["generate-recommendations"])
+
+    assert arguments.site_id is None
+
+
+def test_main_generates_the_recommendations_without_asking_for_a_password(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    async def fausse_generation(*, site_id: str | None) -> str:
+        return f"génération lancée pour {site_id}"
+
+    monkeypatch.setattr(cli, "generate_recommendations", fausse_generation)
+
+    code = cli.main(["generate-recommendations", "--site-id", "SITE002"])
+
+    assert code == 0
+    assert "SITE002" in capsys.readouterr().out

apps/backend/tests/test_internal_alerts.py

@@ -0,0 +1,87 @@
+from datetime import UTC, datetime
+
+import pytest
+from sqlalchemy import text
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.db.session import get_session_factory
+from app.detection import internal_alerts
+from app.repositories.alert import AlertRepository
+from tests.repositories.test_reading import creer_lecture
+from tests.repositories.test_site import creer as creer_site
+
+
+def test_parse_args_defaults_to_no_site_and_no_instant() -> None:
+    arguments = internal_alerts.parse_args([])
+
+    assert arguments.site_id is None
+    assert arguments.now is None
+
+
+def test_parse_args_reads_the_site_id() -> None:
+    arguments = internal_alerts.parse_args(["--site-id", "site-1"])
+
+    assert arguments.site_id == "site-1"
+
+
+def test_parse_args_parses_the_instant_option() -> None:
+    arguments = internal_alerts.parse_args(["--now", "2026-09-16T12:00:00+00:00"])
+
+    assert arguments.now == datetime(2026, 9, 16, 12, tzinfo=UTC)
+
+
+def test_parse_instant_treats_a_naive_datetime_as_utc() -> None:
+    assert internal_alerts._parse_instant("2026-09-16T12:00:00") == datetime(
+        2026, 9, 16, 12, tzinfo=UTC
+    )
+
+
+def test_main_prints_how_many_alerts_were_recorded(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    async def fausse_execution(*, now: datetime | None, site_id: str | None) -> int:
+        return 3
+
+    monkeypatch.setattr(internal_alerts, "run_detection", fausse_execution)
+
+    code = internal_alerts.main([])
+
+    assert code == 0
+    assert "3 nouvelle" in capsys.readouterr().out
+
+
+@pytest.mark.integration
+async def test_run_detection_writes_a_threshold_alert_end_to_end(session: AsyncSession) -> None:
+    # `run_detection` ouvre sa propre session et commite : `session.rollback()` seul ne défait
+    # rien ici (contrairement au reste de la suite), d'où le nettoyage explicite ci-dessous, sur
+    # le modèle de `tests/api/test_matrice_acces.py`.
+    site = await creer_site(session, capacity_kw=100.0)
+    site_id = site.site_id
+    instant = datetime(2026, 9, 16, 12, tzinfo=UTC)
+    await creer_lecture(session, site_id=site_id, timestamp=instant, consumption_kw=150.0)
+    await session.commit()
+
+    try:
+        nombre = await internal_alerts.run_detection(now=instant, site_id=site_id)
+
+        alertes = await AlertRepository(session).list_all(site_id=site_id)
+        types = [a.type for a in alertes]
+        await session.rollback()
+
+        assert nombre == 1
+        assert types == ["threshold"]
+    finally:
+        # `site.site_id` n'est plus sûr après `session.rollback()` : le rollback expire tous les
+        # objets de la session (indépendamment d'`expire_on_commit`), et y accéder ici relance une
+        # requête hors contexte async. D'où `site_id`, capturé avant.
+        async with get_session_factory()() as nettoyage:
+            await nettoyage.execute(
+                text("delete from alert where site_id = :site_id"), {"site_id": site_id}
+            )
+            await nettoyage.execute(
+                text("delete from reading where site_id = :site_id"), {"site_id": site_id}
+            )
+            await nettoyage.execute(
+                text("delete from site where site_id = :site_id"), {"site_id": site_id}
+            )
+            await nettoyage.commit()

apps/backend/tests/test_static_assets.py

@@ -0,0 +1,13 @@
+# Piège : le logo est committé indépendamment à deux endroits (`app/static/`, servi par
+# `/docs`/`/redoc`, et `apps/frontend/public/`, servi au front) faute d'étape de build partagée.
+# Sans ce test, une mise à jour d'un seul des deux fichiers dérive silencieusement : rien en CI
+# ne le détecte.
+
+from pathlib import Path
+
+BACKEND_LOGO = Path(__file__).parent.parent / "app" / "static" / "logo-icon.png"
+FRONTEND_LOGO = Path(__file__).parent.parent.parent / "frontend" / "public" / "logo-icon.png"
+
+
+def test_the_backend_logo_stays_in_sync_with_the_frontend_one() -> None:
+    assert BACKEND_LOGO.read_bytes() == FRONTEND_LOGO.read_bytes()

apps/backend/uv.lock

@@ -1,6 +1,20 @@
 version = 1
 revision = 3
 requires-python = "==3.14.*"
+resolution-markers = [
+    "sys_platform == 'win32'",
+    "sys_platform == 'emscripten'",
+    "sys_platform != 'emscripten' and sys_platform != 'win32'",
+]
+
+[[package]]
+name = "aiosmtplib"
+version = "5.1.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/9b/5c/9cabc5db6d607616e81ba6d8f1f231cd5a75955807a308c1090a59072d6d/aiosmtplib-5.1.3.tar.gz", hash = "sha256:ac2b418d3260ba62d9cfd0fe7359726e9dc009a4e8e8d9909fdfae332f522a7c", size = 77010, upload-time = "2026-09-08T02:11:20.532Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/9c/0a/b56ab8163d54960337fdca475d3dfd56c8badf6172e79cf2ad00d5335dc1/aiosmtplib-5.1.3-py3-none-any.whl", hash = "sha256:f7d76ce3d4995a65a178c1f11e1bd1607706b921d00cb768e7a2c7f7ef5517a8", size = 30116, upload-time = "2026-09-08T02:11:19.352Z" },
+]
 
 [[package]]
 name = "alembic"
@@ -306,11 +320,14 @@ name = "enervision-backend"
 version = "0.1.0"
 source = { editable = "." }
 dependencies = [
+    { name = "aiosmtplib" },
     { name = "alembic" },
     { name = "anyio" },
     { name = "argon2-cffi" },
     { name = "asyncpg" },
     { name = "fastapi" },
+    { name = "httpx" },
+    { name = "pandas" },
     { name = "prometheus-fastapi-instrumentator" },
     { name = "pydantic", extra = ["email"] },
     { name = "pydantic-settings" },
@@ -322,8 +339,8 @@ dependencies = [
 
 [package.dev-dependencies]
 dev = [
-    { name = "httpx" },
     { name = "mypy" },
+    { name = "pandas-stubs" },
     { name = "pytest" },
     { name = "pytest-asyncio" },
     { name = "pytest-cov" },
@@ -332,11 +349,14 @@ dev = [
 
 [package.metadata]
 requires-dist = [
+    { name = "aiosmtplib", specifier = ">=5.1.3" },
     { name = "alembic", specifier = ">=1.20.0" },
     { name = "anyio", specifier = ">=4.0" },
     { name = "argon2-cffi", specifier = ">=23.1" },
     { name = "asyncpg", specifier = ">=0.31.0" },
     { name = "fastapi", specifier = ">=0.141.1" },
+    { name = "httpx", specifier = ">=0.28.1" },
+    { name = "pandas", specifier = ">=3.0.5" },
     { name = "prometheus-fastapi-instrumentator", specifier = ">=8.1.0" },
     { name = "pydantic", extras = ["email"], specifier = ">=2.13.5" },
     { name = "pydantic-settings", specifier = ">=2.15.0" },
@@ -348,8 +368,8 @@ requires-dist = [
 
 [package.metadata.requires-dev]
 dev = [
-    { name = "httpx", specifier = ">=0.28.1" },
     { name = "mypy", specifier = ">=2.3.1" },
+    { name = "pandas-stubs", specifier = ">=3.0.5.260914" },
     { name = "pytest", specifier = ">=9.1.1" },
     { name = "pytest-asyncio", specifier = ">=1.4.0" },
     { name = "pytest-cov", specifier = ">=7.1.0" },
@@ -595,6 +615,35 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/79/7b/2c79738432f5c924bef5071f933bcc9efd0473bac3b4aa584a6f7c1c8df8/mypy_extensions-1.1.0-py3-none-any.whl", hash = "sha256:1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505", size = 4963, upload-time = "2025-04-22T14:54:22.983Z" },
 ]
 
+[[package]]
+name = "numpy"
+version = "2.5.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/13/01/11703282db468b85f6f7b8c7f22d058de5970d5c7e60a3a8aaa313c3de36/numpy-2.5.3.tar.gz", hash = "sha256:df2d5874ff183595a4ba404edd04f6bd9b5505c1d7708573f6a6c17489a67563", size = 20791231, upload-time = "2026-09-06T16:27:47.073Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/70/78/cf416f15dc29375a229d9dfebf8db6e313f291580b39fa1a568b6052bb07/numpy-2.5.3-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:350ba9783ce969cf9f7ce6e6a9a58e1a6e2a19ca025b7ee448c4db727706212a", size = 16998686, upload-time = "2026-09-06T16:25:33.171Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/59/abcc2d8def4fd60eec7d87f92d27c13448ffd9ab14339bcc63a0d7a2fdea/numpy-2.5.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:012e66aca395d795496446e52aeeb5866312a5d4d3f27da270e5a0b43f70dc5c", size = 12013862, upload-time = "2026-09-06T16:25:36.748Z" },
+    { url = "https://files.pythonhosted.org/packages/94/75/4640d2d6e4b64a049e48425a82728a41ef4adb61332d2cba68055774878b/numpy-2.5.3-cp314-cp314-macosx_14_0_arm64.whl", hash = "sha256:adc1ada2662f8a5f960b8a10d9986897e7499ef07e06d4cfe7197f8cce923c07", size = 5449793, upload-time = "2026-09-06T16:25:39.476Z" },
+    { url = "https://files.pythonhosted.org/packages/96/cd/625b57ae33d4ca560f32cc0b47b4a5922146d9beb998ddf773900d440a73/numpy-2.5.3-cp314-cp314-macosx_14_0_x86_64.whl", hash = "sha256:54a115e5a73b8fc44f0cebef486365a1894b5c9760685d4558b72b7c3eb846e0", size = 6785176, upload-time = "2026-09-06T16:25:42.069Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/72/12918652e7912ef9751e8694c88820fcd1908e0618cb23f5f3caa6004b7b/numpy-2.5.3-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:be5a8381859b6da607c84f4f7d6847725f1cf1853ef8a2c9e115b7d58bef47dc", size = 15703377, upload-time = "2026-09-06T16:25:45.135Z" },
+    { url = "https://files.pythonhosted.org/packages/45/8f/9beacf79ca7c650688ad0baa80931adb988fe6e6e5d5903c23cc3dbd70eb/numpy-2.5.3-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b0521d0f4aebb6e06189451025fa17a913287b13c03d5fe05c017333b654ea5b", size = 16711928, upload-time = "2026-09-06T16:25:48.461Z" },
+    { url = "https://files.pythonhosted.org/packages/09/8d/41d0a56e1ac4c87495c897a211b1368691b7237aadabec8b3b8f3a74d48f/numpy-2.5.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:9deb49575e5b0b94ed72c8a64ec4d033381adc27e9060ae842971f697ba96104", size = 17059507, upload-time = "2026-09-06T16:25:51.873Z" },
+    { url = "https://files.pythonhosted.org/packages/08/1e/0dfbc5cc251d54e2af790f254d24ec38637fa97ec7d5d11de7ffed787098/numpy-2.5.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:b00eefbcf0f292945c4b4dec2ae845389ef5bcdcd596e6e4328051db5b5ba694", size = 18471002, upload-time = "2026-09-06T16:25:55.233Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/2c/dfa40f6991f8185c8c30ffd023dfcbb11888e823cfab9557b920f3bb7bed/numpy-2.5.3-cp314-cp314-win32.whl", hash = "sha256:c2381f82999704f818e2c987a865050e285ec3621262c66d40f5a96c8f899f8e", size = 6180485, upload-time = "2026-09-06T16:25:58.157Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/73/d2c08231e4fde7e415501fd02c715d96e98599b2d8384445933944152984/numpy-2.5.3-cp314-cp314-win_amd64.whl", hash = "sha256:2c25dfa72943e4336ddb6b0ee4277b47a0c85bede0807530ec68103bf58e2c10", size = 12698179, upload-time = "2026-09-06T16:26:00.789Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/e9/dcdcc9b95cf5f49815055573aee1b11cfbf5299f38a180e437ded050810f/numpy-2.5.3-cp314-cp314-win_arm64.whl", hash = "sha256:15aa985ac73a8db02db7663381aa109510449d3819d37206caed27b33a65a8a6", size = 10769383, upload-time = "2026-09-06T16:26:04.011Z" },
+    { url = "https://files.pythonhosted.org/packages/49/c4/af8bc08a7ef4e1529a7c0cf24969accce316b783999802089a581ec99272/numpy-2.5.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:ac7bb1c52d445bd4f8f7f97fefe6abc3a084dc4d63df50d79b17fa2b78e89297", size = 12132668, upload-time = "2026-09-06T16:26:07.138Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/ae/0f15eb56d4ec5e13c1f7ff04ff407f997d1acbadb45d3e1f2e2645a8f43c/numpy-2.5.3-cp314-cp314t-macosx_14_0_arm64.whl", hash = "sha256:e6ab667ba76450084eb64013762c438ea76d9d29cc676dcd6c2e9892ba37f841", size = 5568580, upload-time = "2026-09-06T16:26:09.828Z" },
+    { url = "https://files.pythonhosted.org/packages/23/fb/c72a8f25d4b6e96c354e7ab45ace3b27dc11e5d6a13b6c7d0cd6b08bf112/numpy-2.5.3-cp314-cp314t-macosx_14_0_x86_64.whl", hash = "sha256:f7fabeb6cea87d65f3b926de33d03fb016cfdc29314c90974383b5582ae72891", size = 6882634, upload-time = "2026-09-06T16:26:12.524Z" },
+    { url = "https://files.pythonhosted.org/packages/07/a9/968c90ed2ab15060c338e8137f1215b5a60756ae07328e0a60d1c6734df4/numpy-2.5.3-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1fb6f8fb9ff0b3a69f52c66ce397b0246583e9f28616231b0e32ca49259a5fa6", size = 15748923, upload-time = "2026-09-06T16:26:15.092Z" },
+    { url = "https://files.pythonhosted.org/packages/59/08/9df04103947b95e3b6b1f2ed1a70521f325647a31b82da6a2aae3a485508/numpy-2.5.3-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:93e1f5447e2b1e479d7bd74701e84746b86450cff1fc368b132d195e2b8f8211", size = 16746748, upload-time = "2026-09-06T16:26:18.43Z" },
+    { url = "https://files.pythonhosted.org/packages/41/a0/14c8d5fe5b53a334aabb653deb391c0fef49558f491880ea300ed6785224/numpy-2.5.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:c00abe94c1a69d75d827dcf1c025b25c8a45d230b3bcd77a9020883a1b047653", size = 17111561, upload-time = "2026-09-06T16:26:22.113Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/a6/d7e96e42f01522e154c32489640f16dfc4f6181d165d05fc3bec8c2c4999/numpy-2.5.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:536f963710a4e63934d80ac0dc4f478804a83e9a84b6828018f25d09953ada33", size = 18513945, upload-time = "2026-09-06T16:26:25.401Z" },
+    { url = "https://files.pythonhosted.org/packages/25/39/3453afb7119d0449ef11c886874120ff180e2c337760e0e2d88f70f1a945/numpy-2.5.3-cp314-cp314t-win32.whl", hash = "sha256:4c8a6d2ebce6305fd82fbefca827775437147052a976ee7c94b36a0c1b52ac6c", size = 6335421, upload-time = "2026-09-06T16:26:28.175Z" },
+    { url = "https://files.pythonhosted.org/packages/99/01/22815d2b19a1a746b1d45205cffebb3fe511a18acb75fba6c88491fc9894/numpy-2.5.3-cp314-cp314t-win_amd64.whl", hash = "sha256:9a37475425b431b4d060f23b4f52cd2f3aef6bc7c654bd760adf0040eec9d435", size = 12896420, upload-time = "2026-09-06T16:26:31.265Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/ee/a7cbba67eeaff038dc29ca8b98a88396c8b0cc9c89d4924f4a27a5c9150b/numpy-2.5.3-cp314-cp314t-win_arm64.whl", hash = "sha256:2d8240cb4c16fd831074aa2b2cf9fc54664d826341d61c372245b96a74a49a9a", size = 10857177, upload-time = "2026-09-06T16:26:34.167Z" },
+]
+
 [[package]]
 name = "packaging"
 version = "26.3"
@@ -604,6 +653,47 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/63/34/ba1c580383c9eada3711951fef0795c80b829a078d72188184bcab9dd527/packaging-26.3-py3-none-any.whl", hash = "sha256:d7193f7c8e4e93f444fde0262bf90af30e16fa0ad0ad44cb553c87339b23cd1c", size = 129956, upload-time = "2026-08-04T18:15:27.159Z" },
 ]
 
+[[package]]
+name = "pandas"
+version = "3.0.5"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy" },
+    { name = "python-dateutil" },
+    { name = "tzdata", marker = "sys_platform == 'emscripten' or sys_platform == 'win32'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/be/4f/5f3422a2afec5ffc46308b79e53291365a93748b498ac2e58bead0197916/pandas-3.0.5.tar.gz", hash = "sha256:dca3734d6ab7c906e6730f0788b0a1dbb9f2467731f9711f77995c8e9d62d712", size = 4658219, upload-time = "2026-07-22T22:19:28.819Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/51/2f/cf6aae281264f4463f0875bcbb15fd2bb6d291cc535187dad1732475e4a9/pandas-3.0.5-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:2f264fc46911cc8131a7322a16199bbf8e353d27c10bb211f5bd0c814324dc36", size = 10390034, upload-time = "2026-07-22T22:18:49.818Z" },
+    { url = "https://files.pythonhosted.org/packages/06/ec/5189518c7a7659c4bdcc6b1eb32c46c6f3c86b0661ffd84143d1112c7732/pandas-3.0.5-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:53730687fcd161883b24e10411c06d6a4c0f2275d2faf3bb2bc25deb4ba8007c", size = 9980065, upload-time = "2026-07-22T22:18:52.249Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/f1/598503ce8d7e3c35601e0747ba288c7864baae66380725bc12f13f884dfe/pandas-3.0.5-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:960d3ebcf249f75206899fcd2c6de53f736b7265759ced0d3e559df0b8b709b0", size = 10545532, upload-time = "2026-07-22T22:18:54.813Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/de/ceae2adf7034e07e9910299fe412e1819c4f0dd520700a888bcb03625448/pandas-3.0.5-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9e94c2c5ca43bd3ca32bf64d32308887b65e5f9bfd8023ea52755107a999f93b", size = 10963120, upload-time = "2026-07-22T22:18:57.42Z" },
+    { url = "https://files.pythonhosted.org/packages/66/25/86e0f4451874eb79e688deeebe3c451fec4557f8952005818d800ee8ac7e/pandas-3.0.5-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:e819dd5f62966b481a8cb649d3299ebd886a1ea91ed5a99bf7ce77c98d18ab94", size = 11563178, upload-time = "2026-07-22T22:18:59.729Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/45/8643daa3b4147e433adfcccefdd0380d3aad79d86b15d8999730fe1944d5/pandas-3.0.5-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:3c5ed2e7c06e91d340dfd091d7934f9bc82e4a36b95f647f090b9d1c9ac649da", size = 12028708, upload-time = "2026-07-22T22:19:02.164Z" },
+    { url = "https://files.pythonhosted.org/packages/96/58/ad979ae617615576e8aafd569c9d4b62f1191d896e38f51d66ba06f3b89a/pandas-3.0.5-cp314-cp314-win_amd64.whl", hash = "sha256:cd8f7c6dc98527058ee6264219343f5392240a6f1bfa654fc5d79023020d0c92", size = 9951806, upload-time = "2026-07-22T22:19:04.596Z" },
+    { url = "https://files.pythonhosted.org/packages/69/32/7ac03886b304049a9d2625ee88f59af760d8a93bd30ed9239bce7b9869a8/pandas-3.0.5-cp314-cp314-win_arm64.whl", hash = "sha256:5183427f5a8156d480f30333777bc978be93650a49a7c01db26adffe95b31e85", size = 9238297, upload-time = "2026-07-22T22:19:06.836Z" },
+    { url = "https://files.pythonhosted.org/packages/be/ed/1d1f2ee5547d5167face2376d11c8b2a4c7bfff5a416ee7a9046891fab1e/pandas-3.0.5-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:303da736987d481074ca720ada325f8bd80c64ebc2d45ed79b29df3aaa4a26ca", size = 10849690, upload-time = "2026-07-22T22:19:09.391Z" },
+    { url = "https://files.pythonhosted.org/packages/57/55/17e17152e98fbb0c4b1e562bc65387a2f20a80db0f4a86bf8d3a0e4248d4/pandas-3.0.5-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:3b2801bbb049d0136f6c213eae02b5fca969384fc2064dd728d8620552aa49da", size = 10509945, upload-time = "2026-07-22T22:19:11.773Z" },
+    { url = "https://files.pythonhosted.org/packages/88/90/817d44dbf83facf9556f33576d9af0a241981e7bb5c00606c0bcb5df8dda/pandas-3.0.5-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:cce3a9d11d2b1f82c69a27ec1f4948a170e2c403c4bbfa8cca62e3fdebe2ef3a", size = 10392197, upload-time = "2026-07-22T22:19:14.024Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/da/889f00c0a6f5aa1545add70abbf01502dff87ab577adb855bd631c54d2f2/pandas-3.0.5-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ef01af4d8dc6cd2c8d6c7736f149574ef93fe043811eeb5e445f2647154b5040", size = 10862726, upload-time = "2026-07-22T22:19:16.351Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/98/f1e934fb3c98fce859c6147c6785816c7b5b9ab7821115c5d8c4de9842b9/pandas-3.0.5-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:e2759e890db96dfcffdbd9b86c3c2cb6afaf58def482820317e06163ec1066cd", size = 11414864, upload-time = "2026-07-22T22:19:18.981Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/be/d448af7d657d82e1888dd8551f79c6d6fb161080b5b9752d84d910ec2319/pandas-3.0.5-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:b58b1b39d46a5862e3fb18f50d1a201398619d16a0f9f73f57eea5583cf0e63c", size = 11925105, upload-time = "2026-07-22T22:19:21.515Z" },
+    { url = "https://files.pythonhosted.org/packages/29/c1/ccb4238212c8c4f496c584f3044d94e0c030ed8e1d68999db46c91c2242f/pandas-3.0.5-cp314-cp314t-win_amd64.whl", hash = "sha256:1c10461f6eeb35d8f05b6184c65c8b9991663b66c46b1d559b682cb34ae7c6ea", size = 10387612, upload-time = "2026-07-22T22:19:24.257Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/cf/6a51b2c38980e04c279fd2fa908a1b0982064e860444acfca4ec2e2c8359/pandas-3.0.5-cp314-cp314t-win_arm64.whl", hash = "sha256:3c5015fd1730fbf883647e88068176c839c102cea883ba1769a6f4593bfc1f8c", size = 9509776, upload-time = "2026-07-22T22:19:26.694Z" },
+]
+
+[[package]]
+name = "pandas-stubs"
+version = "3.0.5.260914"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c1/93/8948ae6c1e1e3d6833596fd266f7be2d27c1451b8be094975ad42c5e842e/pandas_stubs-3.0.5.260914.tar.gz", hash = "sha256:3f6fc1f147f68fd89c007105e7c94a948acb4ecd7eb20dc1c02e153c4ed5c250", size = 117622, upload-time = "2026-09-14T16:42:35.065Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/9a/cb/5ad79e02a556cc23fed5816de0109fa8af660c66cfa5f4af74c3e8d4cd26/pandas_stubs-3.0.5.260914-py3-none-any.whl", hash = "sha256:39a1300c5c5c55fdf609e3476805decce5d5015539a4dcb683449f8feaeee2fb", size = 177344, upload-time = "2026-09-14T16:42:33.771Z" },
+]
+
 [[package]]
 name = "pathspec"
 version = "1.1.1"
@@ -788,6 +878,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9d/7a/d968e294073affff457b041c2be9868a40c1c71f4a35fcc1e45e5493067b/pytest_cov-7.1.0-py3-none-any.whl", hash = "sha256:a0461110b7865f9a271aa1b51e516c9a95de9d696734a2f71e3e78f46e1d4678", size = 22876, upload-time = "2026-03-21T20:11:14.438Z" },
 ]
 
+[[package]]
+name = "python-dateutil"
+version = "2.9.0.post0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "six" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/66/c0/0c8b6ad9f17a802ee498c46e004a0eb49bc148f2fd230864601a86dcf6db/python-dateutil-2.9.0.post0.tar.gz", hash = "sha256:37dd54208da7e1cd875388217d5e00ebd4179249f90fb72437e91a35459a0ad3", size = 342432, upload-time = "2024-03-01T18:36:20.211Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ec/57/56b9bcc3c9c6a792fcbaf139543cee77261f3651ca9da0c93f5c1221264b/python_dateutil-2.9.0.post0-py2.py3-none-any.whl", hash = "sha256:a8b2bc7bffae282281c8140a97d3aa9c14da0b136dfe83f850eea9a5f7470427", size = 229892, upload-time = "2024-03-01T18:36:18.57Z" },
+]
+
 [[package]]
 name = "python-dotenv"
 version = "1.2.3"
@@ -857,6 +959,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/8b/4b/51327018d056f0dad2c2238f26d1fb0f53707a9d91b75dea6d1b3039f136/ruff-0.16.7-py3-none-win_arm64.whl", hash = "sha256:aab7f39e2c9df6c596216070f98eef1207b94f8516cca20c808826974971855b", size = 10412401, upload-time = "2026-09-10T18:04:04.098Z" },
 ]
 
+[[package]]
+name = "six"
+version = "1.17.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/94/e7/b2c673351809dca68a0e064b6af791aa332cf192da575fd474ed7d6f16a2/six-1.17.0.tar.gz", hash = "sha256:ff70335d468e7eb6ec65b95b99d3a2836546063f63acc5171de367e834932a81", size = 34031, upload-time = "2024-12-04T17:35:28.174Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b7/ce/149a00dd41f10bc29e5921b496af8b574d8413afcd5e30dfa0ed46c2cc5e/six-1.17.0-py2.py3-none-any.whl", hash = "sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274", size = 11050, upload-time = "2024-12-04T17:35:26.475Z" },
+]
+
 [[package]]
 name = "sqlalchemy"
 version = "2.0.52"
@@ -916,6 +1027,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/67/81/4add07e5172b7ac40d8ed5ff580409a7801a4fe26d529bdd915401dabfbe/typing_inspection-0.4.4-py3-none-any.whl", hash = "sha256:65b8397ba37ccbce054456aaccddfc91e6e3083c92824df348d96ca832f3f147", size = 14750, upload-time = "2026-08-12T12:37:24.648Z" },
 ]
 
+[[package]]
+name = "tzdata"
+version = "2026.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/e4/31/3d74fa778a63b98b7374323befcc0be5ab3bd94afd4096a0124e7379152c/tzdata-2026.4.tar.gz", hash = "sha256:f1b8bd365d8d210c55353f4d7f8d6d8561c0ba50d704b700d195a9424bba0d79", size = 199350, upload-time = "2026-09-12T12:56:03.251Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f9/bc/8737e8d54cf51106118039b83f485a4783112fab49ea9d044b234978a46e/tzdata-2026.4-py2.py3-none-any.whl", hash = "sha256:c2169a8b0a7a5e9674da5a135ccdfb2b3e671b333ed9fed17b41f73c34476e81", size = 347494, upload-time = "2026-09-12T12:56:01.67Z" },
+]
+
 [[package]]
 name = "uvicorn"
 version = "0.53.0"

apps/frontend/README.md

@@ -76,6 +76,13 @@ Points à vérifier après toute regénération :
    côté backend. Le `docker-compose.yml` n'a aucun service frontend.
 4. Ajouter le `Dockerfile` multi-stage (build Angular puis service statique nginx).
 
+## Design système
+
+Tokens (couleurs, typo, espacements) et composants partagés (`ev-button`, `ev-card`,
+`ev-alert`, `ev-badge`) sont documentés dans
+[`docs/architecture/32-design-systeme-frontend.md`](../../docs/architecture/32-design-systeme-frontend.md).
+Toute nouvelle page doit les réutiliser plutôt que définir ses propres valeurs.
+
 ## Additional Resources
 
 For more information on using the Angular CLI, including detailed command references, visit the [Angular CLI Overview and Command Reference](https://angular.dev/tools/cli) page.

apps/frontend/TESTING.md

@@ -1,4 +1,4 @@
-# Conventions de tests unitaires — Frontend
+# Conventions de tests unitaires : Frontend
 
 ## Outil
 Vitest (intégré nativement à Angular CLI, pas d'installation à faire).
@@ -83,3 +83,6 @@ describe('MonComposant', () => {
 ## Lancer les tests
 - Développement (mode watch) : `npm test`
 - Rapport de couverture (CI) : `npm run test:ci -- --coverage`, puis ouvrir `coverage/index.html`
+- Un fichier ou un dossier seulement :
+  `npx ng test --watch=false --coverage=false --include=src/app/core/services/alerts.service.spec.ts`
+  (répéter `--include` pour plusieurs cibles ; un dossier joue tous ses specs)

apps/frontend/angular.json

@@ -34,6 +34,11 @@
           },
           "configurations": {
             "production": {
+              "optimization": {
+                "styles": {
+                  "inlineCritical": false
+                }
+              },
               "budgets": [
                 {
                   "type": "initial",
@@ -86,15 +91,6 @@
               "text-summary",
               "lcov",
               "html"
-            ],
-            "reporters": [
-              "default",
-              [
-                "junit",
-                {
-                  "outputFile": "test-results/junit.xml"
-                }
-              ]
             ]
           }
         }