fix(backend,ci): repare trois angles morts de la surveillance de derive

Revue de la branche : trois defauts empechaient la surveillance de tenir ce qu'elle annonce.

- `evaluate()` gardait les microsecondes de `now()` dans `window_end`, la cle de
  `uq_drift_report_window`. Deux executions ne collidaient donc jamais et l'index ne
  dedoublonnait rien, contrairement a ce qu'affirmaient l'ADR 0011, 20-backend et le docstring
  du DAG. L'instant de reference est desormais tronque a l'heure.
- Un site qui cessait d'etre score disparaissait du rapport : la liste des sites ne venait que
  de la fenetre recente. La panne que cette surveillance existe pour dire etait exactement
  celle qu'elle taisait. La fenetre de reference entre maintenant dans l'union, et le site
  recoit sa ligne `indetermine` a zero observation.
- Sans fenetre de reference, `_plafond` rendait `None` et le verdict tombait sur `stable`, une
  affirmation que la donnee ne portait pas. C'est `indetermine` desormais.

`ml.yml` ecoute `apps/backend/app/**` et non les seuls modeles : ce workflow est le seul a
jouer `-m chaine`, or la chaine traverse les endpoints, les services et les schemas jusqu'a
`GET /predictions`. Une PR touchant `predictions.py` ne declenchait pas le test qui l'assert.

Hygiene de tests : le nettoyage des fixtures API connait `drift_report` (cle etrangere RESTRICT
vers `site`), le test sans rapport rend ses overrides en teardown, `test_chaine_ml_api` compare
les `created_at` strictement (un `>=` passait aussi quand l'API resservait la premiere ligne),
et `test_data_integration` filtre sur le site seme au lieu de juger tout le contenu d'une
fenetre dans une base partagee.

Docs remises d'aplomb : sept revisions Alembic et non six, `derive.py` dans l'inventaire de
etl/README, et le diagramme de 20-backend gagne DriftService, le depot drift et sa treizieme
table.

.env.example

@@ -29,22 +29,48 @@ 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
+# Airflow (api-server + scheduler + dag-processor, 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
+# Clé de session de l'api-server Airflow, distincte de la précédente. Générer la
 # vôtre : python -c "import secrets; print(secrets.token_urlsafe(48))"
-AIRFLOW_WEBSERVER_SECRET_KEY=change_me
+AIRFLOW_API_SECRET_KEY=change_me
+# Secret des jetons JWT entre scheduler, tâches et api-server (conteneurs distincts, le secret
+# doit être partagé). Même générateur que ci-dessus.
+AIRFLOW_JWT_SECRET=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=
+
+# Deux environnements sur la même machine (ADR 0009) : un dossier, un `.env` et un projet Compose
+# chacun. Le nom de projet préfixe volumes, réseau et conteneurs et l'emporte sur `name:`.
+# Vide sur un poste de développement : le projet reste `enervision`.
+COMPOSE_PROJECT_NAME=
+# Origine publique, avec le port si le proxy HTTPS n'écoute pas 443. Vide : https://PUBLIC_HOST.
+# Recette : PUBLIC_HOST=rec.enervision.local et PUBLIC_ORIGIN=https://rec.enervision.local:8443.
+PUBLIC_ORIGIN=
+# Ports publiés par le proxy. Vides : 80 et 443. Recette : PROXY_HTTPS_PORT=8443 et
+# PROXY_HTTP_PORT=127.0.0.1:8081, la redirection vers 443 n'ayant pas à être joignable de
+# l'extérieur. Décaler aussi POSTGRES_PORT, MAILPIT_UI_PORT et AIRFLOW_PORT (5434, 8026, 8082).
+PROXY_HTTP_PORT=
+PROXY_HTTPS_PORT=
+# Réglages mémoire de la stack déployée. Sans eux, timescaledb-tune réserve 25 % de la RAM de la
+# machine à chaque base au premier démarrage. L'api-server Airflow 3 n'a rien à régler ici : son
+# nombre de workers vaut 1 par défaut, contre 4 pour le webserver d'Airflow 2.
+TS_TUNE_MEMORY=2GB
+TS_TUNE_NUM_CPUS=2

.github/dependabot.yml

@@ -10,6 +10,15 @@ updates:
       frontend-dependencies:
         patterns:
           - "*"
+    ignore:
+      # Pourquoi : @angular/build épingle typescript (>=6.0 <6.1) et vitest (^4). Une majeure
+      # de l'un ou l'autre casse `npm ci` tant qu'Angular ne suit pas.
+      - dependency-name: "typescript"
+        update-types: ["version-update:semver-major"]
+      - dependency-name: "vitest"
+        update-types: ["version-update:semver-major"]
+      - dependency-name: "@vitest/coverage-v8"
+        update-types: ["version-update:semver-major"]
 
   # Backend — uv (lit pyproject.toml / uv.lock)
   - package-ecosystem: "uv"

.github/workflows/airflow.yml

@@ -1,11 +1,13 @@
 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
+# (contrairement à backend.yml et ml.yml) : celui de l'image apache/airflow retenue, et les tests
-# ml/ ne vit que dans l'image Docker, dans son propre environnement (cf. etl/airflow/Dockerfile).
+# doivent tourner sur le même interpréteur qu'elle. 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/. Une modification de ml/
+# Piège : l'image COPY les fichiers de dépendances et le code de ml/ et de apps/backend/. Une
-# peut donc casser sa construction, d'où ces chemins dans les déclencheurs.
+# 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:
@@ -14,6 +16,9 @@ on:
       - "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:
@@ -21,6 +26,9 @@ on:
       - "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:
@@ -40,10 +48,10 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: etl/airflow/uv.lock
@@ -71,9 +79,9 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
-      - name: Construit l'image (contexte à la racine, elle COPY ml/)
+      - 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
@@ -82,3 +90,14 @@ jobs:
         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 des modules prouve que l'environnement /opt/backend est complet.
+      # Les deux commandes du DAG `alertes` et la commande du DAG historique sont couvertes.
+      - name: Vérifie que les trois commandes backend 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
+          && env -u VIRTUAL_ENV uv run --no-sync python -m app.etl.historical_import --help"

.github/workflows/backend.yml

@@ -30,10 +30,10 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: apps/backend/uv.lock
@@ -90,10 +90,10 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: apps/backend/uv.lock
@@ -126,10 +126,10 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: apps/backend/uv.lock
@@ -140,3 +140,29 @@ jobs:
         # 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@v7
+
+      # 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@v7
+
+      # 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/deploy.yml

@@ -0,0 +1,58 @@
+# Pourquoi : le runner tourne sur la VM ENI, adresse privée que les runners hébergés par GitHub
+# ne joignent pas, et travaille dans un dossier stable par environnement plutôt que dans son
+# espace de travail : `.env`, certificats et volumes y survivent d'un déploiement à l'autre.
+# Piège : jamais de déclencheur `pull_request` ici. Sur un dépôt public, une PR de fork
+# exécuterait son code sur la machine de production (ADR 0009) - job deploy.
+
+name: Déploiement
+
+on:
+  push:
+    branches: [dev, main]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: deploy-${{ github.ref_name }}
+  cancel-in-progress: false
+
+jobs:
+  deploy:
+    runs-on: [self-hosted, linux, eni-g3]
+    timeout-minutes: 30
+    environment:
+      name: ${{ github.ref_name == 'main' && 'prod' || 'rec' }}
+      url: ${{ github.ref_name == 'main' && 'https://enervision.local' || 'https://rec.enervision.local:8443' }}
+    env:
+      ENVIRONNEMENT: ${{ github.ref_name == 'main' && 'prod' || 'rec' }}
+      PORT_HTTPS: ${{ github.ref_name == 'main' && '443' || '8443' }}
+    steps:
+      - name: Aligner le dossier de l'environnement sur la branche poussée
+        run: |
+          cd "/srv/enervision/${ENVIRONNEMENT}"
+          git fetch --quiet origin "${GITHUB_REF_NAME}"
+          git checkout --quiet "${GITHUB_REF_NAME}"
+          git reset --quiet --hard "origin/${GITHUB_REF_NAME}"
+          git log -1 --format='%h %s'
+
+      - name: Reconstruire et redémarrer la stack
+        run: |
+          cd "/srv/enervision/${ENVIRONNEMENT}"
+          make stack-up
+
+      - name: Attendre que l'API réponde derrière le proxy
+        run: |
+          for tentative in $(seq 1 36); do
+            if curl --fail --silent --insecure "https://localhost:${PORT_HTTPS}/api/v1/health/ready"; then
+              exit 0
+            fi
+            sleep 5
+          done
+          echo "L'API ne répond pas après 3 minutes" >&2
+          cd "/srv/enervision/${ENVIRONNEMENT}"
+          compose="docker compose -f docker-compose.yml -f docker-compose.prod.yml"
+          $compose ps
+          $compose logs --tail=50 backend proxy
+          exit 1

.github/workflows/frontend.yml

@@ -17,10 +17,10 @@ jobs:
   build:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
         with:
-          node-version: 24
+          node-version: 26
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
 
@@ -33,10 +33,10 @@ jobs:
     name: Audit des dépendances
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
         with:
-          node-version: 24
+          node-version: 26
       # 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
@@ -45,10 +45,10 @@ jobs:
     needs: build
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
         with:
-          node-version: 24
+          node-version: 26
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
       - name : Installation des dépendances (Front)
@@ -58,7 +58,7 @@ jobs:
         run: npm test --watch=false --code-coverage --coverageReporters=lcov
         working-directory: apps/frontend
       - name: Upload coverage
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: frontend-coverage
           path: apps/frontend/coverage/frontend/lcov.info

.github/workflows/infra.yml

@@ -0,0 +1,52 @@
+name: Infra
+
+# Pourquoi : le Terraform du dépôt est resté cassé sans que rien ne le dise, faute de job qui le
+# joue. Ce workflow n'applique rien : il vérifie le formatage et la validité de chaque racine.
+# Piège : la boucle parcourt `environments/*`, pour qu'une racine ajoutée soit couverte sans
+# toucher à ce fichier.
+
+on:
+  push:
+    paths:
+      - "infra/terraform/**"
+      - ".github/workflows/infra.yml"
+  pull_request:
+    paths:
+      - "infra/terraform/**"
+      - ".github/workflows/infra.yml"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: infra-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  terraform:
+    name: Formatage et validation Terraform
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v7
+
+      # Action tierce, donc epinglee sur un SHA de commit et pas sur un tag mobile : un tag se
+      # redeplace, et ce workflow tourne avec les droits du depot (regle Sonar githubactions:S7637).
+      - name: Installe Terraform
+        uses: hashicorp/setup-terraform@dfe3c3f87815947d99a8997f908cb6525fc44e9e # v4.0.1
+        with:
+          terraform_version: 1.16.3
+          terraform_wrapper: false
+
+      - name: Vérifie le formatage
+        run: terraform fmt -check -recursive infra/terraform
+
+      - name: Valide chaque racine
+        run: |
+          for racine in infra/terraform/environments/*/; do
+            echo "::group::${racine}"
+            terraform -chdir="${racine}" init -backend=false -input=false
+            terraform -chdir="${racine}" validate
+            echo "::endgroup::"
+          done

.github/workflows/ml.yml

@@ -8,10 +8,28 @@ on:
     paths:
       - "ml/**"
       - ".github/workflows/ml.yml"
+      # Piege : `backend.yml` ne joue jamais `-m chaine`, ce job est le seul. Le test de chaine
+      # traverse tout `apps/backend/app` jusqu'a GET /predictions : restreindre le filtre aux
+      # migrations et aux modeles le laisserait muet sur la PR meme qui le casse. Meme raisonnement
+      # que le filtre d'airflow.yml, qui inclut deja des chemins de ml/ et de apps/backend/.
+      - "apps/backend/alembic/**"
+      - "apps/backend/app/**"
+      - "apps/backend/tests/test_chaine_ml_api.py"
+      - "apps/backend/pyproject.toml"
+      - "apps/backend/uv.lock"
   pull_request:
     paths:
       - "ml/**"
       - ".github/workflows/ml.yml"
+      # Piege : `backend.yml` ne joue jamais `-m chaine`, ce job est le seul. Le test de chaine
+      # traverse tout `apps/backend/app` jusqu'a GET /predictions : restreindre le filtre aux
+      # migrations et aux modeles le laisserait muet sur la PR meme qui le casse. Meme raisonnement
+      # que le filtre d'airflow.yml, qui inclut deja des chemins de ml/ et de apps/backend/.
+      - "apps/backend/alembic/**"
+      - "apps/backend/app/**"
+      - "apps/backend/tests/test_chaine_ml_api.py"
+      - "apps/backend/pyproject.toml"
+      - "apps/backend/uv.lock"
 
 permissions:
   contents: read
@@ -30,10 +48,10 @@ jobs:
 
     steps:
       - name: Récupère le dépôt
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: ml/uv.lock
@@ -53,7 +71,110 @@ jobs:
       - name: Typage
         run: uv run mypy enervision_ml tests
 
-      # Aucun test ne touche PostgreSQL ni MLflow distant : tout tourne sur donnees
+      # Les tests exigeant une base portent le marqueur `integration`, ecarte par defaut et
-      # synthetiques ou un magasin SQLite local jetable (cf. ml/tests/test_train.py).
+      # joue par le job `integration` ci-dessous.
       - name: Tests
         run: uv run pytest
+
+  # Le seul job du depot qui dispose a la fois des deux environnements uv et d'une base. Piege :
+  # le schema de la base ML est celui du backend (apps/backend/alembic, proprietaire du schema).
+  # Le reconstruire ici a la main rendrait ce job vert sur une base qui n'est pas la notre.
+  integration:
+    name: ML - DB et chaîne ML - DB - API
+    runs-on: ubuntu-latest
+
+    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:
+      # Deux variables, deux dialectes : Alembic et l'API parlent asyncpg, le pipeline ML parle
+      # psycopg en synchrone. Cf. docs/ML-START.md, section 1.
+      DATABASE_URL: postgresql+asyncpg://enervision:change_me@localhost:5433/enervision_test
+      ML_DATABASE_URL: postgresql+psycopg://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@v7
+
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          cache-dependency-glob: |
+            ml/uv.lock
+            apps/backend/uv.lock
+
+      - name: Installe l'interpréteur déclaré par .python-version
+        working-directory: ml
+        run: uv python install
+
+      - name: Synchronise le pipeline ML sans dévier du verrou
+        working-directory: ml
+        run: uv sync --all-groups --frozen
+
+      # Le backend est installé ici parce qu'il porte les migrations, seule source du schéma, et
+      # le test de chaîne, qui interroge l'API.
+      - name: Synchronise le backend sans dévier du verrou
+        working-directory: apps/backend
+        run: uv sync --all-groups --frozen
+
+      # db/init/110-test-database.sql n'est pas monté ici, et sans l'extension la première
+      # révision Alembic refuse de s'appliquer.
+      - 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 du backend, propriétaire du schéma
+        working-directory: apps/backend
+        run: uv run alembic upgrade head
+
+      # `-m` en ligne de commande écrase celui d'addopts. Couverture désactivée : ce job ne joue
+      # qu'une partie de la suite, son taux n'aurait pas de sens (même raison que backend.yml).
+      - name: Tests ML exigeant une base
+        working-directory: ml
+        run: uv run pytest -m integration --no-cov
+
+      # Lance les vrais binaires enervision_ml.train et .score en sous-processus, comme les DAGs
+      # ml_train et ml_score, puis relit le résultat par GET /api/v1/predictions.
+      - name: Chaîne complète ML vers DB vers API
+        working-directory: apps/backend
+        env:
+          ML_PYTHON: ${{ github.workspace }}/ml/.venv/bin/python
+        run: uv run pytest -m chaine --no-cov
+
+  sast:
+    name: Analyse statique de sécurité
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ml
+
+    steps:
+      - name: Récupère le dépôt
+        uses: actions/checkout@v7
+
+      # 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@v7
+
+      - 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

@@ -5,11 +5,15 @@ on:
     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"
 
 
@@ -20,10 +24,10 @@ jobs:
   build-front:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
         with:
-          node-version: 24
+          node-version: 26
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
       
@@ -36,10 +40,10 @@ jobs:
     needs: build-front
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@v7
         with:
-          node-version: 24
+          node-version: 26
           cache: npm
           cache-dependency-path: apps/frontend/package-lock.json
 
@@ -52,7 +56,7 @@ jobs:
         working-directory: apps/frontend
 
       - name: Upload coverage
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: frontend-coverage
           path: apps/frontend/coverage/frontend/lcov.info
@@ -60,9 +64,9 @@ jobs:
   build-back:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: apps/backend/uv.lock
@@ -91,9 +95,9 @@ jobs:
     needs: build-back
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
       - name: Installe uv
-        uses: astral-sh/setup-uv@v5
+        uses: astral-sh/setup-uv@v7
         with:
           enable-cache: true
           cache-dependency-glob: apps/backend/uv.lock
@@ -103,29 +107,65 @@ jobs:
         working-directory: apps/backend
 
       - name: Upload coverage
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: backend-coverage
           path: apps/backend/coverage.xml
 
-  sonarqube:
+  test-ml:
-    needs: [build-front, build-back, test-front, test-back]
-    name: SonarQube
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
+      - name: Installe uv
+        uses: astral-sh/setup-uv@v7
+        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@v7
+        with:
+          name: ml-coverage
+          path: ml/coverage.xml
+
+  sonarqube:
+    needs: [build-front, build-back, test-front, test-back, test-ml]
+    name: SonarQube
+    # Pourquoi : GitHub ne fournit pas les secrets aux workflows lancés par dependabot[bot].
+    # Sans SONAR_TOKEN le scan échoue sans rien analyser ; build et tests restent joués.
+    if: github.actor != 'dependabot[bot]'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
         with:
           fetch-depth: 0
       - name: Téléchargement du rapport de couverture (Front)
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v8
         with:
           name: frontend-coverage
           path: apps/frontend/coverage/frontend
       - name: Téléchargement du rapport de couverture (Back)
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v8
         with:
           name: backend-coverage
           path: apps/backend
+      - name: Téléchargement du rapport de couverture (ML)
+        uses: actions/download-artifact@v8
+        with:
+          name: ml-coverage
+          path: ml
       - name: SonarQube Scan
         uses: SonarSource/sonarqube-scan-action@v8
         env:

.gitignore

@@ -40,7 +40,6 @@ kubeconfig
 # Airflow
 etl/airflow/logs/
 airflow.db
-airflow-webserver.pid
 standalone_admin_password.txt
 
 # Environnement et secrets

Makefile

@@ -13,12 +13,37 @@ ifdef ACME_EMAIL
 export ACME_EMAIL
 endif
 
+# Piege : make ne lit pas `.env`, que seul docker compose interpole. Les cibles hors conteneur
+# (ml-*, demo-data, db-wait) joignent la base par le port publie et ont besoin de ces valeurs.
+env-val = $(shell sed -n 's/^$(1)=//p' .env 2>/dev/null | tail -1)
+PG_USER := $(or $(strip $(call env-val,POSTGRES_USER)),enervision)
+PG_PASSWORD := $(or $(strip $(call env-val,POSTGRES_PASSWORD)),change_me)
+PG_DB := $(or $(strip $(call env-val,POSTGRES_DB)),enervision)
+PG_PORT := $(or $(strip $(call env-val,POSTGRES_PORT)),5433)
+AIRFLOW_PORT := $(or $(strip $(call env-val,AIRFLOW_PORT)),8080)
+MAILPIT_UI_PORT := $(or $(strip $(call env-val,MAILPIT_UI_PORT)),8025)
+ML_DATABASE_URL ?= postgresql+psycopg://$(PG_USER):$(PG_PASSWORD)@localhost:$(PG_PORT)/$(PG_DB)
+export ML_DATABASE_URL
+
+# Piege : la base des tests d'integration n'est pas la base de developpement. Ces tests ecrivent
+# et suppriment des lignes, et leurs fixtures refusent de demarrer ailleurs que sur
+# `enervision_test` (garde sur le nom, cf. ml/tests/conftest.py).
+PG_TEST_DB ?= enervision_test
+TEST_DATABASE_URL ?= postgresql+asyncpg://$(PG_USER):$(PG_PASSWORD)@localhost:$(PG_PORT)/$(PG_TEST_DB)
+ML_TEST_DATABASE_URL ?= postgresql+psycopg://$(PG_USER):$(PG_PASSWORD)@localhost:$(PG_PORT)/$(PG_TEST_DB)
+
+# Le jeu historique s'arrete au 31/12/2024 : score et detection ancres a l'horloge reelle ne
+# verraient qu'un parc muet depuis des mois. Cf. `--now` de enervision_ml.score.
+DEMO_NOW ?= 2024-12-31T00:00:00Z
+
 .DEFAULT_GOAL := help
 .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 \
+        lint format typecheck test test-cov test-integration ml-test-integration \
-        openapi docker-build db-up db-down db-reset db-logs db-psql migrate bootstrap-admin \
+        test-chaine check \
-        ml-lint ml-typecheck ml-test ml-check ml-train ml-score recommendations \
+        openapi docker-build db-up db-down db-reset db-logs db-psql db-wait db-ensure-airflow \
+        migrate migrate-test bootstrap-admin services-up demo-data demo-data-force \
+        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
 
@@ -39,12 +64,19 @@ install-ml: ## Installe les dépendances du pipeline ML
 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
+dev: services-up migrate demo-data ## Lance toute la stack : base, Mailpit, Airflow, puis backend et frontend
+	@echo "airflow  -> http://localhost:$(AIRFLOW_PORT)  mailpit -> http://localhost:$(MAILPIT_UI_PORT)"
 	@trap 'kill 0' EXIT INT TERM; \
 	$(MAKE) --no-print-directory dev-backend & \
 	$(MAKE) --no-print-directory dev-frontend & \
 	wait
 
+services-up: ## Démarre les services conteneurisés dont `make dev` dépend (base, Mailpit, Airflow)
+	docker compose up -d db mailpit
+	@$(MAKE) --no-print-directory db-wait
+	@$(MAKE) --no-print-directory db-ensure-airflow
+	docker compose up -d airflow-init airflow-apiserver airflow-scheduler airflow-dag-processor
+
 dev-backend: ## Lance l'API seule en rechargement à chaud
 	@echo "backend  -> http://localhost:8000 (docs sur /docs)"
 	cd $(BACKEND) && uv run uvicorn app.main:create_app --factory --reload --host 0.0.0.0 --port 8000
@@ -88,11 +120,24 @@ ml-test: ## Exécute les tests du pipeline ML (donnees synthetiques, sans base n
 
 ml-check: ml-lint ml-typecheck ml-test ## Chaîne de vérification complète du pipeline ML
 
+# La cible surcharge ML_DATABASE_URL, que ce Makefile exporte vers la base de développement : la
+# garde du conftest ferait échouer la cible sans cette surcharge.
+ml-test-integration: ML_DATABASE_URL := $(ML_TEST_DATABASE_URL)
+ml-test-integration: ## Tests ML exigeant une base migrée. Faire `make db-up migrate-test` avant
+	cd $(ML) && uv run pytest -m integration --no-cov
+
+test-chaine: ## Chaîne ML -> DB -> API, vrais binaires. Exige les deux environnements uv
+	cd $(BACKEND) && DATABASE_URL=$(TEST_DATABASE_URL) ML_PYTHON=$(CURDIR)/$(ML)/.venv/bin/python \
+		uv run pytest -m chaine --no-cov
+
 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
+ml-score: ## Score le prochain pas horaire et l'ecrit dans `prediction`. CSV= et NOW= optionnels
-	cd $(ML) && uv run python -m enervision_ml.score $(if $(CSV),--csv $(CSV),)
+	cd $(ML) && uv run python -m enervision_ml.score $(if $(CSV),--csv $(CSV),) $(if $(NOW),--now $(NOW),)
+
+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),)
@@ -105,12 +150,12 @@ airflow-test: ## Verifie que les DAGs s'importent sans erreur et ont la structur
 
 airflow-check: airflow-lint airflow-test ## Chaîne de vérification complète des DAGs Airflow
 
-airflow-up: ## Démarre Airflow (webserver + scheduler, LocalExecutor). db-up requis avant.
+airflow-up: db-ensure-airflow ## Démarre Airflow (api-server + scheduler + dag-processor, LocalExecutor). db-up requis avant.
-	docker compose up -d airflow-init airflow-webserver airflow-scheduler
+	docker compose up -d airflow-init airflow-apiserver airflow-scheduler airflow-dag-processor
 	@echo "airflow  -> http://localhost:$${AIRFLOW_PORT:-8080}"
 
-airflow-down: ## Arrête le webserver et le scheduler Airflow
+airflow-down: ## Arrête l'api-server, le scheduler et le dag-processor Airflow
-	docker compose stop airflow-webserver airflow-scheduler
+	docker compose stop airflow-apiserver airflow-scheduler airflow-dag-processor
 
 airflow-logs: ## Suit les journaux du scheduler Airflow (où tournent les tâches, LocalExecutor)
 	docker compose logs -f airflow-scheduler
@@ -121,12 +166,15 @@ docker-build: ## Construit l'image du 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
+# Piège : l'image backend ne migre pas au démarrage, et `/health/ready` ne teste que la connexion
+# et l'extension. Sans `alembic upgrade head`, la stack démarre verte sur une base sans schéma.
+stack-up: ## Démarre la stack derrière le reverse proxy, puis migre la base. 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
+	$(COMPOSE_PROD) exec -T backend alembic upgrade head
 
 stack-down: ## Arrête la stack complète en conservant les données
 	$(COMPOSE_PROD) stop
@@ -162,8 +210,39 @@ db-logs: ## Suit les journaux de la base
 db-psql: ## Ouvre une session psql sur la base applicative
 	docker compose exec db psql -U $${POSTGRES_USER:-enervision} -d $${POSTGRES_DB:-enervision}
 
+db-wait: ## Attend que la base accepte les connexions
+	@for _ in $$(seq 1 60); do \
+		docker compose exec -T db pg_isready -U $(PG_USER) -d $(PG_DB) >/dev/null 2>&1 && exit 0; \
+		sleep 1; \
+	done; \
+	echo "La base n'accepte toujours pas de connexion apres 60s"; exit 1
+
+# Piege : db/init ne rejoue qu'a la premiere initialisation du volume. Un `pgdata` cree avant
+# db/init/120-airflow-database.sql n'a pas de base `airflow`, et airflow-init boucle dessus.
+db-ensure-airflow: ## Crée la base de métadonnées Airflow si le volume pgdata est antérieur à db/init/120
+	@docker compose exec -T db psql -U $(PG_USER) -d postgres -tAc \
+		"SELECT 1 FROM pg_database WHERE datname = 'airflow'" | grep -q 1 \
+		|| docker compose exec -T db psql -U $(PG_USER) -d postgres -c "CREATE DATABASE airflow"
+
 migrate: ## Applique les migrations Alembic
 	cd $(BACKEND) && uv run alembic upgrade head
 
+migrate-test: ## Applique les migrations sur enervision_test, la base des tests d'intégration
+	cd $(BACKEND) && DATABASE_URL=$(TEST_DATABASE_URL) uv run alembic upgrade head
+
 bootstrap-admin: ## Crée le premier administrateur, mot de passe saisi au clavier
 	cd $(BACKEND) && uv run python -m app.cli create-admin --email $${EMAIL:?EMAIL=... requis}
+
+demo-data: ## Renseigne prédictions, alertes et recommandations si elles manquent. NOW= optionnel
+	@nombre=$$(docker compose exec -T db psql -U $(PG_USER) -d $(PG_DB) -tAc 'SELECT count(*) FROM alert') \
+		|| { echo "demo-data : base injoignable ou migrations non appliquees"; exit 1; }; \
+	if [ "$$nombre" = 0 ]; then \
+		$(MAKE) --no-print-directory demo-data-force; \
+	else \
+		echo "demo-data : $$nombre alerte(s) deja en base (make demo-data-force pour rejouer)"; \
+	fi
+
+demo-data-force: ## Rejoue le peuplement sans regarder l'existant. Les trois etapes sont idempotentes
+	$(MAKE) --no-print-directory ml-score NOW=$(DEMO_NOW)
+	$(MAKE) --no-print-directory detect-alerts NOW=$(DEMO_NOW)
+	$(MAKE) --no-print-directory recommendations

README.md

@@ -9,8 +9,10 @@ series temporelles energetiques, deployee sur une machine on-premise.
 |-------|----------------------------------------------------------|
 | J1    | Valider la préparation de l'environnement et du repo     |
 | J2    | Valider le périmètre retenu et les choix technologiques  |
-| J3    | Valider l'architecture et la gestion de la sécurité      |
+| J3    | Ingestion & backend                                      |
-| J4    | Valider la robustesse et assurer les livrables           |
+| J4    | Architecture, sécurité & frontend                        |
+| J5    | Valider la robustesse et assurer les livrables           |
+| J6    | Amélioration possible                                    |
 
 Ce que la documentation apporte à chacun : [docs/architecture/00-vue-ensemble.md](docs/architecture/00-vue-ensemble.md).
 
@@ -18,15 +20,15 @@ Ce que la documentation apporte à chacun : [docs/architecture/00-vue-ensemble.m
 
 | Domaine    | Technologie                         | Emplacement         | Etat          |
 |------------|-------------------------------------|---------------------|---------------|
-| Backend    | FastAPI, Python 3.14                | `apps/backend`      | Initialise    |
+| Backend    | FastAPI, Python 3.14                | `apps/backend`      | En place    |
-| Frontend   | Angular 22, Node 24 LTS             | `apps/frontend`     | Tableau de bord |
+| Frontend   | Angular 22, Node 26                 | `apps/frontend`     | En place |
-| Base       | PostgreSQL 17 + TimescaleDB         | `db`                | Initialise    |
+| Base       | PostgreSQL 17 + TimescaleDB         | `db`                | En place    |
-| ETL        | Apache Airflow                      | `etl/airflow`       | A initialiser |
+| ETL        | Apache Airflow                      | `etl/airflow`       | Cinq 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 |
+| CI/CD      | GitHub Actions                      | `.github/workflows` | En place |
 | Monitoring | Prometheus, Grafana, Alertmanager   | `monitoring`        | A initialiser |
-| ML         | LightGBM, MLflow                    | `ml`                | Entrainement initialise |
+| ML         | LightGBM, MLflow                    | `ml`                | En place |
 
 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
@@ -48,7 +50,7 @@ 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, import, dérive)
 │   ├── plugins/        Operateurs et hooks maison
 │   ├── include/        Requetes SQL et ressources des DAGs
 │   └── tests/          Tests d'integrite des DAGs
@@ -75,25 +77,53 @@ Prerequis : uv, Docker, Node 24 LTS (npm fourni). Le poste doit disposer de Pyth
 cp .env.example .env                               # variables de docker-compose
 cp apps/backend/.env.example apps/backend/.env     # variables du backend hors conteneur
 
-make db-up     # PostgreSQL + TimescaleDB, publie sur le port 5433
+make install   # dependances du backend, du frontend, du ML et des DAGs
-make install   # dependances du backend et du frontend
+make dev       # toute la stack, voir ci-dessous
-make migrate   # applique les migrations Alembic
-make dev       # backend sur http://localhost:8000 (docs sur /docs), frontend sur http://localhost:4200
 make check     # lint + typage + tests
 ```
 
+`make dev` enchaine tout : demarrage des services conteneurises (base sur le port 5433, Mailpit,
+Airflow), migrations Alembic, peuplement de demonstration si les alertes manquent, puis backend
+et frontend en rechargement a chaud sur le poste.
+
+| Service | Adresse |
+|---|---|
+| Backend | <http://localhost:8000> (documentation sur `/docs`) |
+| Frontend | <http://localhost:4200> |
+| Airflow | <http://localhost:8080> (`AIRFLOW_ADMIN_USERNAME` / `AIRFLOW_ADMIN_PASSWORD` du `.env`) |
+| Mailpit | <http://localhost:8025> |
+
+Le `.env` doit porter les cles Airflow avant le premier `make dev` : `AIRFLOW_FERNET_KEY`,
+`AIRFLOW_API_SECRET_KEY`, `AIRFLOW_JWT_SECRET`, `AIRFLOW_APP_SECRET_KEY` et
+`AIRFLOW_ADMIN_PASSWORD`. Sans elles `airflow-init` refuse de demarrer, et `airflow-apiserver`,
+`airflow-scheduler` et `airflow-dag-processor` avec lui.
+
+Les cibles d'origine restent disponibles pour ne demarrer qu'une partie : `make db-up`,
+`make airflow-up`, `make dev-backend`, `make dev-frontend`.
+
 `make help` liste les cibles disponibles.
 
 Deux fichiers d'environnement, deux usages : `.env` a la racine alimente `docker-compose.yml`,
 `apps/backend/.env` alimente le backend lance sur le poste. Le port 5433 est publie plutot que
 5432, souvent deja pris par une autre base.
 
-La boucle de developpement est `make db-up` puis `make dev` : seule la base tourne en
+Le backend et le frontend tournent sur le poste, lances ensemble par `make dev` (logs
-conteneur, le backend et le frontend tournent tous les deux sur le poste, lances ensemble par
+entrelaces dans le meme terminal, Ctrl+C arrete les deux) ; la base, Mailpit et Airflow tournent
-`make dev` (logs entrelaces dans le meme terminal, Ctrl+C arrete les deux). `make dev-backend`
+en conteneur. Le service `backend` du `docker-compose.yml` sert la stack complete et la recette,
-et `make dev-frontend` restent disponibles pour lancer un seul des deux. Le service `backend`
+et n'embarque pas le source, donc toute modification y demande un
-du `docker-compose.yml` sert la stack complete et la recette, et n'embarque pas le source, donc
+`docker compose up -d --build backend`.
-toute modification y demande un `docker compose up -d --build backend`.
+
+### Donnees de demonstration
+
+Le jeu historique s'arrete au 31/12/2024. `make demo-data` renseigne les tables que les vues
+alertes, recommandations et previsions lisent, en ancrant le scoring et la detection a cette
+date (`DEMO_NOW`) plutot qu'a l'horloge reelle, qui ne verrait qu'un parc muet depuis des mois.
+La cible ne fait rien si des alertes existent deja ; `make demo-data-force` rejoue les trois
+etapes, toutes idempotentes en base.
+
+Un volume `pgdata` cree avant `db/init/120-airflow-database.sql` n'a pas de base `airflow` :
+`db/init` ne rejoue qu'a la premiere initialisation. `make db-ensure-airflow`, appelee par
+`make dev` et `make airflow-up`, la cree au besoin, sans detruire les donnees applicatives.
 
 Verifier que la base repond et que l'extension est chargee :
 
@@ -116,6 +146,12 @@ nom de domaine public ne résout vers la machine. Routage, mode ACME et renouvel
 [`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).
 
+Sur la VM ENI, deux environnements cohabitent, recette sur `dev` et production sur `main`,
+chacun dans son dossier et son projet Compose : `scripts/provision-host.sh` les prépare, le
+workflow `deploy.yml` les redéploie à chaque push par un runner auto-hébergé. Ports, noms
+d'hôte et garde-fous dans [`docs/architecture/10-infra.md`](docs/architecture/10-infra.md) et
+[l'ADR 0009](docs/adr/0009-deux-environnements-compose-sur-la-vm-eni.md).
+
 ## Conventions
 
 - Branches : `feat/`, `fix/`, `chore/`, `docs/`, `test/` suivi d'un libelle court.

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/alembic/versions/d3f1a2b7c904_rapports_de_derive.py

@@ -0,0 +1,81 @@
+"""rapports de derive du modele de prevision
+
+Revision ID: d3f1a2b7c904
+Revises: c0adab96238c
+Create Date: 2026-09-22 14:40:00.000000
+
+`site_id` est nullable, et c'est le coeur du schema : une ligne par site, plus une ligne
+globale tous sites confondus, que `NULL` designe. Un seul site qui derive est invisible dans
+une moyenne d'ensemble, et une derive d'ensemble sans rupture par site signale un changement
+de modele ou de saison, pas une panne.
+
+L'unicite passe par un index a `coalesce` et non par une `UniqueConstraint` : deux lignes
+globales successives ont toutes deux `site_id` a NULL, et NULL n'est egal a aucune valeur, pas
+meme a lui-meme. Meme forme que `uq_reading_source`.
+
+Les trois `CHECK` sont portees par la base, comme `ck_prediction_status` : un verdict sans
+motif, ou un statut inconnu, ne doit pas dependre de la vigilance de l'appelant.
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+revision: str = "d3f1a2b7c904"
+down_revision: str | Sequence[str] | None = "c0adab96238c"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "drift_report",
+        sa.Column("drift_report_id", sa.BigInteger(), autoincrement=True, nullable=False),
+        sa.Column(
+            "computed_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column("site_id", sa.Text(), nullable=True),
+        sa.Column("window_start", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("window_end", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("reference_start", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("reference_end", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("n_observations", sa.Integer(), nullable=False),
+        sa.Column("mae", sa.Double(), nullable=True),
+        sa.Column("mape", sa.Double(), nullable=True),
+        sa.Column("bias", sa.Double(), nullable=True),
+        sa.Column("reference_mae", sa.Double(), nullable=True),
+        sa.Column("coverage_ratio", sa.Double(), nullable=True),
+        sa.Column("insufficient_data_ratio", sa.Double(), nullable=True),
+        sa.Column("model_references", postgresql.ARRAY(sa.Text()), nullable=False),
+        sa.Column("status", sa.Text(), nullable=False),
+        sa.Column("reason", sa.Text(), nullable=True),
+        sa.CheckConstraint(
+            "status IN ('stable', 'derive', 'indetermine')", name="ck_drift_report_status"
+        ),
+        sa.CheckConstraint(
+            "status = 'stable' OR reason IS NOT NULL", name="ck_drift_report_reason"
+        ),
+        sa.CheckConstraint("n_observations >= 0", name="ck_drift_report_observations"),
+        sa.ForeignKeyConstraint(
+            ["site_id"], ["site.site_id"], name="fk_drift_report_site", ondelete="RESTRICT"
+        ),
+        sa.PrimaryKeyConstraint("drift_report_id"),
+    )
+    op.create_index(
+        "ix_drift_report_site_computed", "drift_report", ["site_id", "computed_at"], unique=False
+    )
+    op.create_index(
+        "uq_drift_report_window",
+        "drift_report",
+        ["window_end", sa.literal_column("coalesce(site_id, '')")],
+        unique=True,
+    )
+
+
+def downgrade() -> None:
+    op.drop_table("drift_report")

apps/backend/app/api/deps.py

@@ -24,6 +24,7 @@ from app.core.security import decode_access_token as decode_token
 from app.db.session import get_session
 from app.repositories.alert import AlertRepository
 from app.repositories.audit_log import AuditLogRepository
+from app.repositories.drift import DriftRepository
 from app.repositories.login_attempt import LoginAttemptRepository
 from app.repositories.password_reset_attempt import PasswordResetAttemptRepository
 from app.repositories.password_reset_token import PasswordResetTokenRepository
@@ -35,6 +36,7 @@ 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, PasswordResetPolicy
+from app.services.drift import DriftService
 from app.services.prediction import PredictionService
 from app.services.reading import ReadingService
 from app.services.recommendation import RecommendationService
@@ -232,6 +234,13 @@ def get_prediction_service(session: SessionDep) -> PredictionService:
 PredictionServiceDep = Annotated[PredictionService, Depends(get_prediction_service)]
 
 
+def get_drift_service(session: SessionDep) -> DriftService:
+    return DriftService(DriftRepository(session))
+
+
+DriftServiceDep = Annotated[DriftService, Depends(get_drift_service)]
+
+
 async def get_current_principal(
     credentials: CredentialsDep,
     session: SessionDep,

apps/backend/app/api/openapi.py

@@ -90,6 +90,14 @@ TAGS: Final[list[dict[str, Any]]] = [
             "de scoring (`ml/`) et simplement lue ici. Accessible à partir du rôle `lecteur`."
         ),
     },
+    {
+        "name": "monitoring",
+        "description": (
+            "Surveillance de la dérive du modèle : écart entre les prévisions déjà écrites et "
+            "les lectures réellement arrivées, par site et tous sites confondus. Réservé à "
+            "partir du rôle `operateur`, qui agit sur un pipeline dégradé."
+        ),
+    },
 ]
 
 cookie_de_rafraichissement = APIKeyCookie(
@@ -153,6 +161,17 @@ REPONSES_ADMIN: Final[Reponses] = {
     },
 }
 
+REPONSES_OPERATEUR: Final[Reponses] = {
+    **REPONSES_AUTHENTIFIEES,
+    403: {
+        "model": ErrorResponse,
+        "description": (
+            "Droits insuffisants, ou mot de passe provisoire à changer quand `detail` vaut "
+            "`password_change_required`."
+        ),
+    },
+}
+
 # `lecteur` est le rôle minimum : `require_role` n'y refuse jamais un 403 pour droits
 # insuffisants, seulement pour le mot de passe provisoire.
 REPONSES_LECTEUR: Final[Reponses] = {

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

@@ -0,0 +1,20 @@
+from fastapi import APIRouter
+
+from app.api.deps import DriftServiceDep, OperateurDep
+from app.api.openapi import REPONSE_VALIDATION
+from app.schemas.drift import DriftReportResponse
+
+router = APIRouter()
+
+
+@router.get(
+    "/drift",
+    response_model=list[DriftReportResponse],
+    summary="Dernier rapport de dérive par site, plus la ligne globale",
+    responses=REPONSE_VALIDATION,
+)
+async def get_drift(
+    _: OperateurDep, service: DriftServiceDep, site_id: str | None = None
+) -> list[DriftReportResponse]:
+    rapports = await service.derniers(site_id=site_id)
+    return [DriftReportResponse.model_validate(rapport) for rapport in rapports]

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

@@ -1,10 +1,16 @@
 from fastapi import APIRouter
 
-from app.api.openapi import REPONSE_SERVEUR, REPONSES_ADMIN, REPONSES_LECTEUR
+from app.api.openapi import (
+    REPONSE_SERVEUR,
+    REPONSES_ADMIN,
+    REPONSES_LECTEUR,
+    REPONSES_OPERATEUR,
+)
 from app.api.v1.endpoints import (
     alerts,
     auth,
     health,
+    monitoring,
     predictions,
     readings,
     recommendations,
@@ -38,3 +44,6 @@ api_router.include_router(
 api_router.include_router(
     predictions.router, prefix="/predictions", tags=["predictions"], responses=REPONSES_LECTEUR
 )
+api_router.include_router(
+    monitoring.router, prefix="/monitoring", tags=["monitoring"], responses=REPONSES_OPERATEUR
+)

apps/backend/app/models/__init__.py

@@ -2,7 +2,15 @@
 # --autogenerate`, qui générerait alors un drop de sa table.
 
 from app.models.audit_log import AuditLog
-from app.models.energy import Alert, Dataset, Prediction, Reading, Recommendation, Site
+from app.models.energy import (
+    Alert,
+    Dataset,
+    DriftReport,
+    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
@@ -14,6 +22,7 @@ __all__ = [
     "AppUser",
     "AuditLog",
     "Dataset",
+    "DriftReport",
     "LoginAttempt",
     "PasswordResetAttempt",
     "PasswordResetToken",

apps/backend/app/models/energy.py

@@ -208,3 +208,49 @@ class Recommendation(Base):
     explanation: Mapped[str] = mapped_column(Text)
     rule_reference: Mapped[str] = mapped_column(Text)
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
+
+
+class DriftReport(Base):
+    __tablename__ = "drift_report"
+    __table_args__ = (
+        CheckConstraint(
+            "status IN ('stable', 'derive', 'indetermine')", name="ck_drift_report_status"
+        ),
+        CheckConstraint("status = 'stable' OR reason IS NOT NULL", name="ck_drift_report_reason"),
+        CheckConstraint("n_observations >= 0", name="ck_drift_report_observations"),
+        Index("ix_drift_report_site_computed", "site_id", "computed_at"),
+    )
+
+    drift_report_id: Mapped[int] = mapped_column(BigInteger, primary_key=True, autoincrement=True)
+    computed_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), server_default=func.now()
+    )
+    # `NULL` porte la ligne globale, tous sites confondus : une derive d'ensemble et la derive
+    # d'un seul site ne se lisent pas dans le meme chiffre.
+    site_id: Mapped[str | None] = mapped_column(
+        Text, ForeignKey("site.site_id", name="fk_drift_report_site", ondelete="RESTRICT")
+    )
+    window_start: Mapped[datetime] = mapped_column(DateTime(timezone=True))
+    window_end: Mapped[datetime] = mapped_column(DateTime(timezone=True))
+    reference_start: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
+    reference_end: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
+    n_observations: Mapped[int] = mapped_column(Integer)
+    mae: Mapped[float | None] = mapped_column(Double)
+    mape: Mapped[float | None] = mapped_column(Double)
+    bias: Mapped[float | None] = mapped_column(Double)
+    reference_mae: Mapped[float | None] = mapped_column(Double)
+    coverage_ratio: Mapped[float | None] = mapped_column(Double)
+    insufficient_data_ratio: Mapped[float | None] = mapped_column(Double)
+    model_references: Mapped[list[str]] = mapped_column(ARRAY(Text))
+    status: Mapped[str] = mapped_column(Text)
+    reason: Mapped[str | None] = mapped_column(Text)
+
+
+# Piège : une `UniqueConstraint` ne dédoublonnerait pas les lignes globales, dont `site_id` est
+# NULL et qu'aucune n'est égale à une autre. Même forme que `uq_reading_source`.
+Index(
+    "uq_drift_report_window",
+    DriftReport.window_end,
+    func.coalesce(DriftReport.site_id, text("''")),
+    unique=True,
+)

apps/backend/app/monitoring/drift.py

@@ -0,0 +1,105 @@
+# Surveillance de dérive du modèle de prévision (EC06, issue #45) : même gabarit que
+# `app.detection.internal_alerts`, ordonnancé par le DAG `derive`.
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import sys
+from datetime import UTC, datetime, timedelta
+
+from app.core.config import get_settings
+from app.db.session import get_session_factory
+from app.repositories.drift import DriftRepository, NouveauRapportDerive
+from app.services.drift import STATUT_DERIVE, DriftService, Seuils
+
+
+async def run_drift(
+    *, now: datetime | None = None, site_id: str | None = None, seuils: Seuils | None = None
+) -> list[NouveauRapportDerive]:
+    """Calcule les rapports de la fenêtre et les enregistre. Rend ce qui a été calculé, que la
+    ligne ait été écrite ou ignorée par l'index d'idempotence."""
+    async with get_session_factory()() as session:
+        depot = DriftRepository(session)
+        rapports = await DriftService(depot, seuils=seuils).evaluate(now=now, site_id=site_id)
+        await depot.enregistre(rapports)
+        await session.commit()
+    return rapports
+
+
+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:
+    defauts = Seuils()
+    parser = argparse.ArgumentParser(
+        prog="python -m app.monitoring.drift",
+        description="Surveillance de dérive du modèle de prévision EnerVision",
+    )
+    parser.add_argument("--site-id", default=None, help="Limite le calcul 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."
+        ),
+    )
+    parser.add_argument(
+        "--window-hours",
+        type=int,
+        default=int(defauts.fenetre.total_seconds() // 3600),
+        help="Durée de la fenêtre récente, et de la fenêtre de référence qui la précède.",
+    )
+    parser.add_argument(
+        "--grace-hours",
+        type=int,
+        default=int(defauts.grace.total_seconds() // 3600),
+        help="Délai laissé à l'ingestion avant qu'une prévision soit jugée vérifiable.",
+    )
+    parser.add_argument(
+        "--min-observations",
+        type=int,
+        default=defauts.min_observations,
+        help="En deçà, le verdict est `indetermine` plutôt qu'un chiffre trompeur.",
+    )
+    parser.add_argument(
+        "--fail-on-drift",
+        action="store_true",
+        help="Sort en code non nul si une dérive est constatée, pour que la tâche rougisse.",
+    )
+    return parser.parse_args(argv)
+
+
+def seuils_depuis(args: argparse.Namespace) -> Seuils:
+    return Seuils(
+        fenetre=timedelta(hours=args.window_hours),
+        grace=timedelta(hours=args.grace_hours),
+        min_observations=args.min_observations,
+    )
+
+
+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()
+    rapports = asyncio.run(
+        run_drift(now=args.now, site_id=args.site_id, seuils=seuils_depuis(args))
+    )
+
+    for rapport in rapports:
+        cible = rapport.site_id or "TOUS SITES"
+        mae = f"{rapport.mae:.2f}" if rapport.mae is not None else "-"
+        print(
+            f"{cible} : {rapport.status}, MAE {mae} kWh sur {rapport.n_observations} prévision(s)"
+            f"{' : ' + rapport.reason if rapport.reason else ''}"
+        )
+
+    derive = any(rapport.status == STATUT_DERIVE for rapport in rapports)
+    return 1 if derive and args.fail_on_drift else 0
+
+
+if __name__ == "__main__":  # pragma: no cover
+    sys.exit(main())

apps/backend/app/repositories/alert.py

@@ -6,6 +6,10 @@ from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Alert
 
+# Douze colonnes par alerte, contre quatre pour une recommandation : le plafond asyncpg de
+# 32 767 parametres tombe a 2 730 lignes, d'ou un lot plus petit que `recommendation.py`.
+TAILLE_DE_LOT = 1000
+
 
 class AlertRepository:
     def __init__(self, session: AsyncSession) -> None:
@@ -44,12 +48,17 @@ class AlertRepository:
             }
             for alerte in alerts
         ]
+        creees: list[Alert] = []
+        # Piège : asyncpg plafonne une requête à 32 767 paramètres. Une détection sur une fenêtre
+        # chargée dépasse ce seuil, et l'`INSERT` d'un seul tenant échouerait.
+        for debut in range(0, len(valeurs), TAILLE_DE_LOT):
             requete = (
                 insert(Alert)
-            .values(valeurs)
+                .values(valeurs[debut : debut + TAILLE_DE_LOT])
                 .on_conflict_do_nothing(constraint="uq_alert_source_reference")
                 .returning(Alert)
             )
             resultat = await self._session.execute(requete)
+            creees.extend(resultat.scalars().all())
         await self._session.flush()
-        return resultat.scalars().all()
+        return creees

apps/backend/app/repositories/drift.py

@@ -0,0 +1,188 @@
+"""Piège : deux dédoublonnages, pas un - DriftRepository.paires()
+
+`prediction` n'a pas d'unicité sur `(site_id, target_at)` : chaque run de scoring empile une
+ligne de plus. `uq_reading_source` autorise de son côté deux lectures au même instant quand la
+`source` diffère. Joindre les deux tables sans `DISTINCT ON` des deux côtés compterait donc la
+même heure plusieurs fois, et la moyenne d'erreur pèserait ces sites en double.
+
+On retient la prédiction du run le plus récent, celle que sert `GET /api/v1/predictions`, avec
+`prediction_id` en départage : `created_at` vaut l'heure de début de transaction et ne
+distingue pas deux lignes du même run.
+
+Côté lectures, le départage est `reading_id` décroissant, la règle même de
+`GET /api/v1/sites/{site_id}/current`. La dérive se mesure donc contre le réalisé que l'API
+affiche, et non contre une source élue ici et nulle part ailleurs.
+"""
+
+from collections.abc import Sequence
+from dataclasses import asdict, dataclass
+from datetime import datetime
+
+from sqlalchemy import Subquery, func, select
+from sqlalchemy.dialects.postgresql import insert
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models.energy import DriftReport, Prediction, Reading
+
+TARGET_METRIC = "consumption_kwh"
+STATUT_DISPONIBLE = "available"
+
+
+@dataclass(frozen=True, slots=True)
+class PaireDerive:
+    site_id: str
+    target_at: datetime
+    predicted_value: float
+    actual_value: float
+    model_reference: str
+
+
+@dataclass(frozen=True, slots=True)
+class NouveauRapportDerive:
+    site_id: str | None
+    window_start: datetime
+    window_end: datetime
+    reference_start: datetime | None
+    reference_end: datetime | None
+    n_observations: int
+    mae: float | None
+    mape: float | None
+    bias: float | None
+    reference_mae: float | None
+    coverage_ratio: float | None
+    insufficient_data_ratio: float | None
+    model_references: list[str]
+    status: str
+    reason: str | None
+
+
+@dataclass(frozen=True, slots=True)
+class ComptageStatut:
+    site_id: str
+    status: str
+    nombre: int
+
+
+def _predictions_retenues(*, debut: datetime, fin: datetime, site_id: str | None) -> Subquery:
+    requete = (
+        select(
+            Prediction.site_id,
+            Prediction.target_at,
+            Prediction.predicted_value,
+            Prediction.model_reference,
+            Prediction.status,
+        )
+        .distinct(Prediction.site_id, Prediction.target_at)
+        .where(
+            Prediction.target_metric == TARGET_METRIC,
+            Prediction.target_at >= debut,
+            Prediction.target_at < fin,
+        )
+        .order_by(Prediction.site_id, Prediction.target_at, Prediction.prediction_id.desc())
+    )
+    if site_id is not None:
+        requete = requete.where(Prediction.site_id == site_id)
+    return requete.subquery()
+
+
+def _lectures_retenues(*, debut: datetime, fin: datetime, site_id: str | None) -> Subquery:
+    requete = (
+        select(Reading.site_id, Reading.timestamp, Reading.consumption_kwh)
+        .distinct(Reading.site_id, Reading.timestamp)
+        .where(
+            Reading.timestamp >= debut,
+            Reading.timestamp < fin,
+            Reading.consumption_kwh.is_not(None),
+        )
+        .order_by(Reading.site_id, Reading.timestamp, Reading.reading_id.desc())
+    )
+    if site_id is not None:
+        requete = requete.where(Reading.site_id == site_id)
+    return requete.subquery()
+
+
+class DriftRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def paires(
+        self, *, debut: datetime, fin: datetime, site_id: str | None = None
+    ) -> Sequence[PaireDerive]:
+        predictions = _predictions_retenues(debut=debut, fin=fin, site_id=site_id)
+        lectures = _lectures_retenues(debut=debut, fin=fin, site_id=site_id)
+        requete = (
+            select(
+                predictions.c.site_id,
+                predictions.c.target_at,
+                predictions.c.predicted_value,
+                lectures.c.consumption_kwh,
+                predictions.c.model_reference,
+            )
+            .select_from(predictions)
+            .join(
+                lectures,
+                (lectures.c.site_id == predictions.c.site_id)
+                & (lectures.c.timestamp == predictions.c.target_at),
+            )
+            .where(predictions.c.status == STATUT_DISPONIBLE)
+            .order_by(predictions.c.site_id, predictions.c.target_at)
+        )
+
+        lignes = await self._session.execute(requete)
+        return [
+            PaireDerive(
+                site_id=ligne[0],
+                target_at=ligne[1],
+                predicted_value=ligne[2],
+                actual_value=ligne[3],
+                model_reference=ligne[4],
+            )
+            for ligne in lignes
+        ]
+
+    async def comptages(
+        self, *, debut: datetime, fin: datetime, site_id: str | None = None
+    ) -> Sequence[ComptageStatut]:
+        predictions = _predictions_retenues(debut=debut, fin=fin, site_id=site_id)
+        requete = (
+            select(predictions.c.site_id, predictions.c.status, func.count())
+            .select_from(predictions)
+            .group_by(predictions.c.site_id, predictions.c.status)
+        )
+
+        lignes = await self._session.execute(requete)
+        return [
+            ComptageStatut(site_id=ligne[0], status=ligne[1], nombre=ligne[2]) for ligne in lignes
+        ]
+
+    # Pourquoi : l'idempotence est déléguée à `uq_drift_report_window` plutôt qu'à une lecture
+    # préalable, comme pour les recommandations. Rejouer la commande sur la même fenêtre ne
+    # duplique donc rien.
+    async def enregistre(self, rapports: Sequence[NouveauRapportDerive]) -> int:
+        if not rapports:
+            return 0
+
+        valeurs = [asdict(rapport) for rapport in rapports]
+        requete = (
+            insert(DriftReport)
+            .values(valeurs)
+            .on_conflict_do_nothing(
+                index_elements=[DriftReport.window_end, func.coalesce(DriftReport.site_id, "")]
+            )
+            .returning(DriftReport.drift_report_id)
+        )
+        return len((await self._session.scalars(requete)).all())
+
+    async def derniers(self, *, site_id: str | None = None) -> Sequence[DriftReport]:
+        requete = (
+            select(DriftReport)
+            .distinct(DriftReport.site_id)
+            .order_by(
+                DriftReport.site_id,
+                DriftReport.computed_at.desc(),
+                DriftReport.drift_report_id.desc(),
+            )
+        )
+        if site_id is not None:
+            requete = requete.where(DriftReport.site_id == site_id)
+        return (await self._session.scalars(requete)).all()

apps/backend/app/schemas/drift.py

@@ -0,0 +1,31 @@
+from datetime import datetime
+from enum import StrEnum
+
+from pydantic import BaseModel, ConfigDict
+
+
+class DriftStatus(StrEnum):
+    STABLE = "stable"
+    DERIVE = "derive"
+    INDETERMINE = "indetermine"
+
+
+class DriftReportResponse(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    site_id: str | None
+    computed_at: datetime
+    window_start: datetime
+    window_end: datetime
+    reference_start: datetime | None
+    reference_end: datetime | None
+    n_observations: int
+    mae: float | None
+    mape: float | None
+    bias: float | None
+    reference_mae: float | None
+    coverage_ratio: float | None
+    insufficient_data_ratio: float | None
+    model_references: list[str]
+    status: DriftStatus
+    reason: str | None

apps/backend/app/services/drift.py

@@ -0,0 +1,245 @@
+"""Contrainte : la dérive se mesure sur ce qui a déjà eu lieu - DriftService.evaluate()
+
+Une prévision ne devient vérifiable que quand la lecture de son instant cible est ingérée. La
+fenêtre est donc fermée à droite par un délai de grâce : sans lui, la dernière heure ferait
+chuter le taux de couverture à chaque exécution, et le verdict dirait « dérive » alors que
+seule l'ingestion n'avait pas fini son tour.
+
+La comparaison se fait entre deux fenêtres vives de même durée, pas contre la métrique de
+référence du modèle journalisée à l'entraînement. Ce ne sont pas les mêmes grandeurs :
+l'entraînement mesure un backtest où la météo de l'heure cible est connue, le scoring prévoit
+une heure future dont la météo ne l'est pas. Les comparer classerait le modèle « en dérive »
+dès le premier jour, ce qui ne prouverait rien.
+"""
+
+from collections.abc import Sequence
+from dataclasses import dataclass, replace
+from datetime import UTC, datetime, timedelta
+
+from app.models.energy import DriftReport
+from app.repositories.drift import (
+    ComptageStatut,
+    DriftRepository,
+    NouveauRapportDerive,
+    PaireDerive,
+)
+
+STATUT_STABLE = "stable"
+STATUT_DERIVE = "derive"
+STATUT_INDETERMINE = "indetermine"
+
+STATUT_INSUFFISANT = "insufficient_data"
+STATUT_DISPONIBLE = "available"
+
+
+@dataclass(frozen=True, slots=True)
+class Seuils:
+    # 168 h, la saisonnalité hebdomadaire que le modèle apprend par son lag principal : une
+    # fenêtre plus courte comparerait un week-end à une semaine ouvrée.
+    fenetre: timedelta = timedelta(hours=168)
+    grace: timedelta = timedelta(hours=2)
+    min_observations: int = 24
+    ratio_derive: float = 1.25
+    mae_plancher: float = 0.0
+    seuil_biais: float = 0.0
+    seuil_couverture: float = 0.8
+
+
+@dataclass(frozen=True, slots=True)
+class Metriques:
+    n_observations: int
+    mae: float | None
+    mape: float | None
+    bias: float | None
+    model_references: list[str]
+
+
+def mesure(paires: Sequence[PaireDerive]) -> Metriques:
+    if not paires:
+        return Metriques(n_observations=0, mae=None, mape=None, bias=None, model_references=[])
+
+    ecarts = [paire.predicted_value - paire.actual_value for paire in paires]
+    # Le MAPE diverge sur une consommation nulle : les sites à l'arrêt sortent de ce seul
+    # rapport, jamais des autres métriques.
+    ratios = [
+        abs(ecart / paire.actual_value)
+        for ecart, paire in zip(ecarts, paires, strict=True)
+        if paire.actual_value != 0
+    ]
+
+    return Metriques(
+        n_observations=len(paires),
+        mae=sum(abs(ecart) for ecart in ecarts) / len(ecarts),
+        mape=(sum(ratios) / len(ratios) * 100) if ratios else None,
+        bias=sum(ecarts) / len(ecarts),
+        model_references=sorted({paire.model_reference for paire in paires}),
+    )
+
+
+@dataclass(frozen=True, slots=True)
+class Verdict:
+    status: str
+    reason: str | None
+
+
+class DriftService:
+    def __init__(self, depot: DriftRepository, *, seuils: Seuils | None = None) -> None:
+        self._depot = depot
+        self._seuils = seuils or Seuils()
+
+    async def derniers(self, *, site_id: str | None = None) -> Sequence[DriftReport]:
+        """Ce que sert l'API : le dernier rapport de chaque site, plus la ligne globale."""
+        return await self._depot.derniers(site_id=site_id)
+
+    async def evaluate(
+        self, *, now: datetime | None = None, site_id: str | None = None
+    ) -> list[NouveauRapportDerive]:
+        """Une ligne par site, plus une ligne globale dont le `site_id` est nul."""
+        # Piège : `window_end` est la clé d'idempotence de `uq_drift_report_window`. Sans
+        # troncature à l'heure, deux exécutions ne collident jamais, aux microsecondes près.
+        instant = (now or datetime.now(UTC)).replace(minute=0, second=0, microsecond=0)
+        fin = instant - self._seuils.grace
+        debut = fin - self._seuils.fenetre
+        reference_fin = debut
+        reference_debut = reference_fin - self._seuils.fenetre
+
+        recentes = await self._depot.paires(debut=debut, fin=fin, site_id=site_id)
+        anciennes = await self._depot.paires(
+            debut=reference_debut, fin=reference_fin, site_id=site_id
+        )
+        comptages = await self._depot.comptages(debut=debut, fin=fin, site_id=site_id)
+
+        gabarit = NouveauRapportDerive(
+            site_id=None,
+            window_start=debut,
+            window_end=fin,
+            reference_start=reference_debut,
+            reference_end=reference_fin,
+            n_observations=0,
+            mae=None,
+            mape=None,
+            bias=None,
+            reference_mae=None,
+            coverage_ratio=None,
+            insufficient_data_ratio=None,
+            model_references=[],
+            status=STATUT_INDETERMINE,
+            reason=None,
+        )
+
+        rapports = [
+            self._rapport(
+                gabarit,
+                site=site,
+                recentes=[p for p in recentes if p.site_id == site],
+                anciennes=[p for p in anciennes if p.site_id == site],
+                comptages=[c for c in comptages if c.site_id == site],
+            )
+            # Piège : un site présent dans la référence et absent de la fenêtre récente a cessé
+            # d'être scoré. C'est la panne à crier, pas une ligne à omettre.
+            for site in sorted(
+                {paire.site_id for paire in recentes}
+                | {paire.site_id for paire in anciennes}
+                | {c.site_id for c in comptages}
+            )
+        ]
+        rapports.append(
+            self._rapport(
+                gabarit, site=None, recentes=recentes, anciennes=anciennes, comptages=comptages
+            )
+        )
+        return rapports
+
+    def _rapport(
+        self,
+        gabarit: NouveauRapportDerive,
+        *,
+        site: str | None,
+        recentes: Sequence[PaireDerive],
+        anciennes: Sequence[PaireDerive],
+        comptages: Sequence[ComptageStatut],
+    ) -> NouveauRapportDerive:
+        metriques = mesure(recentes)
+        reference = mesure(anciennes)
+        couverture = _couverture(len(recentes), comptages)
+        verdict = self._verdict(metriques, reference_mae=reference.mae, couverture=couverture)
+
+        return replace(
+            gabarit,
+            site_id=site,
+            n_observations=metriques.n_observations,
+            mae=metriques.mae,
+            mape=metriques.mape,
+            bias=metriques.bias,
+            reference_mae=reference.mae,
+            coverage_ratio=couverture,
+            insufficient_data_ratio=_part_insuffisante(comptages),
+            model_references=metriques.model_references,
+            status=verdict.status,
+            reason=verdict.reason,
+        )
+
+    def _verdict(
+        self, metriques: Metriques, *, reference_mae: float | None, couverture: float | None
+    ) -> Verdict:
+        seuils = self._seuils
+        if metriques.n_observations < seuils.min_observations:
+            return Verdict(
+                STATUT_INDETERMINE,
+                f"{metriques.n_observations} prévision(s) vérifiée(s) sur la fenêtre, "
+                f"minimum {seuils.min_observations}.",
+            )
+
+        if couverture is not None and couverture < seuils.seuil_couverture:
+            return Verdict(
+                STATUT_DERIVE,
+                f"Couverture de {couverture:.0%}, sous le seuil de {seuils.seuil_couverture:.0%} : "
+                "le pipeline, pas le modèle.",
+            )
+
+        plafond = _plafond(reference_mae, ratio=seuils.ratio_derive, plancher=seuils.mae_plancher)
+        if plafond is None:
+            return Verdict(
+                STATUT_INDETERMINE,
+                "Fenêtre de référence sans paire vérifiée, et aucun plancher de MAE : "
+                "rien à quoi comparer cette fenêtre.",
+            )
+
+        if metriques.mae is not None and metriques.mae > plafond:
+            return Verdict(
+                STATUT_DERIVE,
+                f"MAE de {metriques.mae:.2f} kWh au-delà de {plafond:.2f} kWh, "
+                "seuil dérivé de la fenêtre de référence.",
+            )
+
+        if (
+            seuils.seuil_biais > 0
+            and metriques.bias is not None
+            and abs(metriques.bias) > seuils.seuil_biais
+        ):
+            return Verdict(
+                STATUT_DERIVE,
+                f"Biais de {metriques.bias:+.2f} kWh : le modèle se trompe toujours du même côté.",
+            )
+
+        return Verdict(STATUT_STABLE, None)
+
+
+def _plafond(reference_mae: float | None, *, ratio: float, plancher: float) -> float | None:
+    if reference_mae is None:
+        return plancher or None
+    return max(plancher, reference_mae * ratio)
+
+
+def _couverture(apparie: int, comptages: Sequence[ComptageStatut]) -> float | None:
+    """Part des prévisions disponibles qui ont trouvé leur réalisé. Mesure l'ingestion et
+    l'ordonnancement, pas la qualité du modèle."""
+    disponibles = sum(c.nombre for c in comptages if c.status == STATUT_DISPONIBLE)
+    return apparie / disponibles if disponibles else None
+
+
+def _part_insuffisante(comptages: Sequence[ComptageStatut]) -> float | None:
+    total = sum(c.nombre for c in comptages)
+    if not total:
+        return None
+    return sum(c.nombre for c in comptages if c.status == STATUT_INSUFFISANT) / total

apps/backend/openapi.json

@@ -1859,6 +1859,94 @@
           }
         ]
       }
+    },
+    "/api/v1/monitoring/drift": {
+      "get": {
+        "tags": [
+          "monitoring"
+        ],
+        "summary": "Dernier rapport de dérive par site, plus la ligne globale",
+        "operationId": "get_drift_api_v1_monitoring_drift_get",
+        "security": [
+          {
+            "Jeton d'accès": []
+          }
+        ],
+        "parameters": [
+          {
+            "name": "site_id",
+            "in": "query",
+            "required": false,
+            "schema": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "title": "Site Id"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "type": "array",
+                  "items": {
+                    "$ref": "#/components/schemas/DriftReportResponse"
+                  },
+                  "title": "Response Get Drift Api V1 Monitoring Drift Get"
+                }
+              }
+            }
+          },
+          "500": {
+            "description": "Erreur interne. `correlation` identifie la trace côté serveur, qui n'est pas renvoyée au client.",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/InternalErrorResponse"
+                }
+              }
+            }
+          },
+          "401": {
+            "description": "Jeton absent, illisible, périmé, ou rendu caduc par un changement de rôle ou une désactivation. L'en-tête `WWW-Authenticate` porte la cause dans `error=`.",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "403": {
+            "description": "Droits insuffisants, ou mot de passe provisoire à changer quand `detail` vaut `password_change_required`.",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "422": {
+            "description": "Corps invalide. Le détail nomme le champ fautif et le type d'erreur, jamais la valeur envoyée.",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ValidationErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
     }
   },
   "components": {
@@ -1977,6 +2065,180 @@
         ],
         "title": "AlertType"
       },
+      "DriftReportResponse": {
+        "properties": {
+          "site_id": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Site Id"
+          },
+          "computed_at": {
+            "type": "string",
+            "format": "date-time",
+            "title": "Computed At"
+          },
+          "window_start": {
+            "type": "string",
+            "format": "date-time",
+            "title": "Window Start"
+          },
+          "window_end": {
+            "type": "string",
+            "format": "date-time",
+            "title": "Window End"
+          },
+          "reference_start": {
+            "anyOf": [
+              {
+                "type": "string",
+                "format": "date-time"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Reference Start"
+          },
+          "reference_end": {
+            "anyOf": [
+              {
+                "type": "string",
+                "format": "date-time"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Reference End"
+          },
+          "n_observations": {
+            "type": "integer",
+            "title": "N Observations"
+          },
+          "mae": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Mae"
+          },
+          "mape": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Mape"
+          },
+          "bias": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Bias"
+          },
+          "reference_mae": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Reference Mae"
+          },
+          "coverage_ratio": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Coverage Ratio"
+          },
+          "insufficient_data_ratio": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Insufficient Data Ratio"
+          },
+          "model_references": {
+            "items": {
+              "type": "string"
+            },
+            "type": "array",
+            "title": "Model References"
+          },
+          "status": {
+            "$ref": "#/components/schemas/DriftStatus"
+          },
+          "reason": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Reason"
+          }
+        },
+        "type": "object",
+        "required": [
+          "site_id",
+          "computed_at",
+          "window_start",
+          "window_end",
+          "reference_start",
+          "reference_end",
+          "n_observations",
+          "mae",
+          "mape",
+          "bias",
+          "reference_mae",
+          "coverage_ratio",
+          "insufficient_data_ratio",
+          "model_references",
+          "status",
+          "reason"
+        ],
+        "title": "DriftReportResponse"
+      },
+      "DriftStatus": {
+        "type": "string",
+        "enum": [
+          "stable",
+          "derive",
+          "indetermine"
+        ],
+        "title": "DriftStatus"
+      },
       "ErrorResponse": {
         "properties": {
           "detail": {
@@ -3277,6 +3539,10 @@
     {
       "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`."
+    },
+    {
+      "name": "monitoring",
+      "description": "Surveillance de la dérive du modèle : écart entre les prévisions déjà écrites et les lectures réellement arrivées, par site et tous sites confondus. Réservé à partir du rôle `operateur`, qui agit sur un pipeline dégradé."
     }
   ]
 }

apps/backend/pyproject.toml

@@ -87,8 +87,11 @@ disallow_untyped_defs = false
 testpaths = ["tests"]
 asyncio_mode = "auto"
 asyncio_default_fixture_loop_scope = "function"
-addopts = "-q --strict-markers -m 'not integration' --cov=app --cov-report=term-missing"
+addopts = "-q --strict-markers -m 'not integration and not chaine' --cov=app --cov-report=term-missing"
-markers = ["integration: requiert une base PostgreSQL joignable, hors `make test`"]
+markers = [
+    "integration: requiert une base PostgreSQL joignable, hors `make test`",
+    "chaine: requiert en plus l'environnement uv de ml/, hors `make test` et hors `-m integration`",
+]
 
 [tool.coverage.run]
 source = ["app"]

apps/backend/tests/api/acces.py

@@ -56,6 +56,7 @@ ROLE_MINIMUM: Final[dict[Route, Role]] = {
     ("GET", "/api/v1/readings"): Role.LECTEUR,
     ("GET", "/api/v1/predictions"): Role.LECTEUR,
     ("GET", "/api/v1/sensors/status"): Role.ADMIN,
+    ("GET", "/api/v1/monitoring/drift"): Role.OPERATEUR,
     ("GET", "/api/v1/users"): Role.ADMIN,
     ("POST", "/api/v1/users"): Role.ADMIN,
     ("PATCH", "/api/v1/users/{user_id}"): Role.ADMIN,

apps/backend/tests/api/conftest.py

@@ -0,0 +1,132 @@
+"""Piège : ces fixtures valident leurs écritures, contrairement à celles de tests/repositories.
+
+Un endpoint ouvre sa propre session par `get_session` : il ne verrait pas une ligne semée dans
+une transaction en cours. Lui passer la session de la fixture par `dependency_overrides`
+supprimerait justement ce que ces tests prouvent, et `RecommendationService.generate` valide de
+toute façon lui-même. L'isolation vient donc de la marque portée par chaque `site_id`, et le
+nettoyage est explicite, dans l'ordre imposé par les clés étrangères `RESTRICT`.
+
+Contrainte : toutes ces fixtures sont à portée fonction. `engine_per_test` vide le cache du
+moteur après chaque test ; une fixture de module verrait un moteur déjà fermé à son démontage,
+et ses lignes resteraient en base.
+"""
+
+from collections.abc import AsyncIterator, Callable, Iterator
+from dataclasses import dataclass
+from datetime import UTC, datetime, timedelta
+from uuid import uuid4
+
+import pytest
+from fastapi import FastAPI
+from sqlalchemy import delete, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.api.deps import get_current_principal
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role
+from app.db.session import get_session_factory
+from app.models.energy import Alert, DriftReport, Prediction, Reading, Recommendation, Site
+from tests.repositories.test_alert import creer_alerte
+from tests.repositories.test_prediction import creer_prediction
+from tests.repositories.test_reading import creer_lecture
+from tests.repositories.test_site import creer as creer_site
+
+INSTANT = datetime(2026, 9, 16, 12, 0, tzinfo=UTC)
+
+
+@dataclass(frozen=True)
+class JeuMetier:
+    """Identifiants seuls, jamais d'instance ORM : un attribut relu sur une session fermée
+    déclenche un `MissingGreenlet`."""
+
+    site_id: str
+    site_voisin: str
+    alert_id: int
+    prediction_id: int
+    instant: datetime
+
+
+async def _supprime(session: AsyncSession, sites: list[str]) -> None:
+    # La suppression des recommandations est inconditionnelle : `POST /generate` en cree hors du
+    # controle de la fixture, et `alert` les retient par une cle etrangere `RESTRICT`.
+    alertes = select(Alert.alert_id).where(Alert.site_id.in_(sites))
+    await session.execute(delete(Recommendation).where(Recommendation.alert_id.in_(alertes)))
+    await session.execute(delete(Alert).where(Alert.site_id.in_(sites)))
+    await session.execute(delete(DriftReport).where(DriftReport.site_id.in_(sites)))
+    await session.execute(delete(Prediction).where(Prediction.site_id.in_(sites)))
+    await session.execute(delete(Reading).where(Reading.site_id.in_(sites)))
+    await session.execute(delete(Site).where(Site.site_id.in_(sites)))
+    await session.commit()
+
+
+@pytest.fixture
+def marque() -> str:
+    return uuid4().hex[:12]
+
+
+@pytest.fixture
+async def jeu_metier(marque: str) -> AsyncIterator[JeuMetier]:
+    """Un site instrumenté, un site voisin, trois lectures horaires, une prédiction, une alerte.
+
+    Le voisin existe pour que les tests de filtre prouvent qu'ils écartent quelque chose.
+    """
+    site_id = f"SITE-{marque}"
+    voisin = f"SITE-{marque}-VOISIN"
+
+    async with get_session_factory()() as session:
+        await creer_site(session, site_id=site_id, capacity_kw=100.0)
+        await creer_site(session, site_id=voisin, capacity_kw=100.0)
+        for decalage in range(3):
+            await creer_lecture(
+                session,
+                site_id=site_id,
+                timestamp=INSTANT - timedelta(hours=decalage),
+                consumption_kw=10.0 + decalage,
+            )
+        prediction = await creer_prediction(session, site_id=site_id, target_at=INSTANT)
+        alerte = await creer_alerte(session, site_id=site_id, timestamp=INSTANT)
+        jeu = JeuMetier(
+            site_id=site_id,
+            site_voisin=voisin,
+            alert_id=alerte.alert_id,
+            prediction_id=prediction.prediction_id,
+            instant=INSTANT,
+        )
+        await session.commit()
+
+    try:
+        yield jeu
+    finally:
+        async with get_session_factory()() as session:
+            await _supprime(session, [site_id, voisin])
+
+
+@pytest.fixture
+async def site_nu(marque: str) -> AsyncIterator[str]:
+    """Un site sans lecture ni prédiction : le cas que seul un vrai `LEFT JOIN` distingue."""
+    site_id = f"SITE-{marque}-NU"
+
+    async with get_session_factory()() as session:
+        await creer_site(session, site_id=site_id, capacity_kw=100.0)
+        await session.commit()
+
+    try:
+        yield site_id
+    finally:
+        async with get_session_factory()() as session:
+            await _supprime(session, [site_id])
+
+
+@pytest.fixture
+def principal_injecte(app: FastAPI) -> Iterator[Callable[[Role], None]]:
+    def installe(role: Role = Role.LECTEUR) -> None:
+        app.dependency_overrides[get_current_principal] = lambda: Principal(
+            id=uuid4(),
+            email="parcours@enervision.fr",
+            role=role,
+            kind=AccountKind.HUMAIN,
+            must_change_password=False,
+        )
+
+    yield installe
+    app.dependency_overrides.pop(get_current_principal, None)

apps/backend/tests/api/test_monitoring.py

@@ -0,0 +1,105 @@
+from collections.abc import Iterator, Sequence
+from datetime import UTC, datetime, timedelta
+from uuid import uuid4
+
+import pytest
+from fastapi import FastAPI
+from httpx import AsyncClient
+
+from app.api.deps import get_current_principal, get_drift_service
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role
+from app.models.energy import DriftReport
+
+INSTANT = datetime(2026, 9, 22, 12, tzinfo=UTC)
+
+
+def operateur() -> Principal:
+    return Principal(
+        id=uuid4(),
+        email="operateur@enervision.fr",
+        role=Role.OPERATEUR,
+        kind=AccountKind.HUMAIN,
+        must_change_password=False,
+    )
+
+
+def rapport(*, site_id: str | None) -> DriftReport:
+    return DriftReport(
+        drift_report_id=1,
+        computed_at=INSTANT,
+        site_id=site_id,
+        window_start=INSTANT - timedelta(hours=168),
+        window_end=INSTANT,
+        reference_start=None,
+        reference_end=None,
+        n_observations=48,
+        mae=1.5,
+        mape=12.0,
+        bias=0.3,
+        reference_mae=1.2,
+        coverage_ratio=0.95,
+        insufficient_data_ratio=0.0,
+        model_references=["lightgbm-aaa"],
+        status="stable",
+        reason=None,
+    )
+
+
+class FauxService:
+    def __init__(self, rapports: Sequence[DriftReport]) -> None:
+        self.rapports = list(rapports)
+        self.site_demande: str | None = None
+
+    async def derniers(self, *, site_id: str | None = None) -> Sequence[DriftReport]:
+        self.site_demande = site_id
+        return self.rapports
+
+
+@pytest.fixture
+def servi(app: FastAPI) -> Iterator[list[DriftReport]]:
+    rapports = [rapport(site_id="SITE001"), rapport(site_id=None)]
+    service = FauxService(rapports)
+    app.dependency_overrides[get_current_principal] = operateur
+    app.dependency_overrides[get_drift_service] = lambda: service
+    yield rapports
+    app.dependency_overrides.pop(get_current_principal, None)
+    app.dependency_overrides.pop(get_drift_service, None)
+
+
+async def test_drift_returns_the_latest_report_of_every_site(
+    servi: list[DriftReport], client: AsyncClient
+) -> None:
+    reponse = await client.get("/api/v1/monitoring/drift")
+
+    assert reponse.status_code == 200
+    assert [ligne["site_id"] for ligne in reponse.json()] == ["SITE001", None]
+
+
+async def test_drift_exposes_the_metrics_of_the_stored_report(
+    servi: list[DriftReport], client: AsyncClient
+) -> None:
+    reponse = await client.get("/api/v1/monitoring/drift")
+
+    premier = reponse.json()[0]
+    assert premier["status"] == "stable"
+    assert premier["mae"] == 1.5
+    assert premier["model_references"] == ["lightgbm-aaa"]
+
+
+@pytest.fixture
+def sans_rapport(app: FastAPI) -> Iterator[None]:
+    app.dependency_overrides[get_current_principal] = operateur
+    app.dependency_overrides[get_drift_service] = lambda: FauxService([])
+    yield
+    app.dependency_overrides.pop(get_current_principal, None)
+    app.dependency_overrides.pop(get_drift_service, None)
+
+
+async def test_drift_returns_an_empty_list_when_no_report_exists(
+    sans_rapport: None, client: AsyncClient
+) -> None:
+    reponse = await client.get("/api/v1/monitoring/drift")
+
+    assert reponse.status_code == 200
+    assert reponse.json() == []

apps/backend/tests/api/test_parcours_recommandations.py

@@ -0,0 +1,92 @@
+from collections.abc import Callable
+
+import pytest
+from httpx import AsyncClient
+
+from app.core.roles import Role
+from tests.api.conftest import JeuMetier
+
+pytestmark = pytest.mark.integration
+
+
+async def genere(client: AsyncClient, site_id: str) -> dict[str, int]:
+    # Toujours borne a un site : sans `site_id`, le service examine toutes les alertes de la
+    # base, y compris celles d'un autre test, et le rapport cesse d'etre deterministe.
+    reponse = await client.post(f"/api/v1/recommendations/generate?site_id={site_id}")
+
+    assert reponse.status_code == 200
+    return dict(reponse.json())
+
+
+async def test_generate_creates_a_recommendation_for_the_alert_of_the_requested_site(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.ADMIN)
+
+    rapport = await genere(client, jeu_metier.site_id)
+
+    assert rapport["alerts_examined"] == 1
+    assert rapport["recommendations_created"] >= 1
+    assert rapport["already_present"] == 0
+
+
+async def test_generate_creates_nothing_more_when_it_runs_twice_on_the_same_alerts(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.ADMIN)
+    premier = await genere(client, jeu_metier.site_id)
+
+    second = await genere(client, jeu_metier.site_id)
+
+    assert second["recommendations_created"] == 0
+    assert second["already_present"] == premier["recommendations_created"]
+
+
+async def test_generate_examines_no_alert_when_the_requested_site_has_none(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.ADMIN)
+
+    rapport = await genere(client, jeu_metier.site_voisin)
+
+    assert rapport["alerts_examined"] == 0
+    assert rapport["recommendations_created"] == 0
+
+
+async def test_list_recommendations_returns_what_generate_persisted_in_another_session(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.ADMIN)
+    await genere(client, jeu_metier.site_id)
+
+    reponse = await client.get("/api/v1/recommendations")
+
+    assert reponse.status_code == 200
+    miennes = [r for r in reponse.json() if r["alert_id"] == jeu_metier.alert_id]
+    assert miennes != []
+    assert all(r["rule_reference"] for r in miennes)
+
+
+async def test_get_recommendation_returns_the_row_created_by_generate(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.ADMIN)
+    await genere(client, jeu_metier.site_id)
+    liste = await client.get("/api/v1/recommendations")
+    creee = next(r for r in liste.json() if r["alert_id"] == jeu_metier.alert_id)
+
+    reponse = await client.get(f"/api/v1/recommendations/{creee['recommendation_id']}")
+
+    assert reponse.status_code == 200
+    assert reponse.json() == creee
+
+
+async def test_get_recommendation_returns_404_when_the_identifier_is_unknown(
+    principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+
+    reponse = await client.get("/api/v1/recommendations/9999999")
+
+    assert reponse.status_code == 404
+    assert reponse.json()["detail"] == "Recommandation introuvable"

apps/backend/tests/api/test_parcours_sites.py

@@ -0,0 +1,80 @@
+from collections.abc import Callable
+
+import pytest
+from httpx import AsyncClient
+
+from app.core.roles import Role
+from app.db.session import get_session_factory
+from tests.api.conftest import JeuMetier
+from tests.repositories.test_reading import creer_lecture
+
+pytestmark = pytest.mark.integration
+
+
+async def test_list_sites_returns_the_seeded_site_with_its_stored_attributes(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+
+    reponse = await client.get("/api/v1/sites")
+
+    assert reponse.status_code == 200
+    mien = next(site for site in reponse.json() if site["site_id"] == jeu_metier.site_id)
+    assert mien["capacity_kw"] == 100.0
+    assert mien["site_name"] == "Site de test"
+
+
+async def test_get_site_returns_404_when_the_identifier_is_absent_from_the_database(
+    principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+
+    reponse = await client.get("/api/v1/sites/SITE-JAMAIS-INSERE")
+
+    assert reponse.status_code == 404
+    assert reponse.json()["detail"] == "Site introuvable"
+
+
+async def test_get_current_returns_the_most_recent_reading_when_several_hours_are_stored(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+
+    reponse = await client.get(f"/api/v1/sites/{jeu_metier.site_id}/current")
+
+    assert reponse.status_code == 200
+    corps = reponse.json()
+    assert corps["consumption_kw"] == 10.0
+    assert corps["timestamp"].startswith("2026-09-16T12:00")
+
+
+async def test_get_current_keeps_the_highest_reading_id_when_two_sources_share_the_timestamp(
+    jeu_metier: JeuMetier, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+    async with get_session_factory()() as session:
+        await creer_lecture(
+            session,
+            site_id=jeu_metier.site_id,
+            timestamp=jeu_metier.instant,
+            source="api_history",
+            consumption_kw=999.0,
+        )
+        await session.commit()
+
+    reponse = await client.get(f"/api/v1/sites/{jeu_metier.site_id}/current")
+
+    assert reponse.json()["consumption_kw"] == 999.0
+
+
+async def test_get_current_reports_a_critical_quality_when_the_site_has_no_reading(
+    site_nu: str, principal_injecte: Callable[[Role], None], client: AsyncClient
+) -> None:
+    principal_injecte(Role.LECTEUR)
+
+    reponse = await client.get(f"/api/v1/sites/{site_nu}/current")
+
+    assert reponse.status_code == 200
+    corps = reponse.json()
+    assert corps["timestamp"] is None
+    assert corps["data_quality"] == "critical"

apps/backend/tests/db/test_data_schema.py

@@ -1,5 +1,5 @@
 from collections.abc import AsyncIterator
-from datetime import UTC, datetime
+from datetime import UTC, datetime, timedelta
 from uuid import uuid4
 
 import pytest
@@ -9,7 +9,15 @@ from sqlalchemy.exc import IntegrityError
 from sqlalchemy.ext.asyncio import AsyncConnection, create_async_engine
 
 from app.core.config import get_settings
-from app.models.energy import Alert, Dataset, Prediction, Reading, Recommendation, Site
+from app.models.energy import (
+    Alert,
+    Dataset,
+    DriftReport,
+    Prediction,
+    Reading,
+    Recommendation,
+    Site,
+)
 
 pytestmark = pytest.mark.integration
 MOMENT = datetime(2024, 1, 1, tzinfo=UTC)
@@ -112,8 +120,10 @@ 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 +157,12 @@ 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():
+        async with savepoint:
-            await data_connection.execute(insert(Reading).values(**values))
+            await data_connection.execute(statement)
 
 
 async def test_prediction_requires_period_when_energy_is_predicted(
@@ -164,8 +177,10 @@ 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)
 
 
@@ -212,10 +227,7 @@ async def test_alert_rejects_prediction_when_site_differs(
         )
     ).scalar_one()
 
-    with pytest.raises(IntegrityError):
+    statement = insert(Alert).values(
-        async with data_connection.begin_nested():
-            await data_connection.execute(
-                insert(Alert).values(
         source_alert_id=str(uuid4()),
         site_id=other_site,
         source="enervision",
@@ -226,7 +238,11 @@ async def test_alert_rejects_prediction_when_site_differs(
         prediction_id=prediction_id,
         raw_data={},
     )
-            )
+    savepoint = data_connection.begin_nested()
+
+    with pytest.raises(IntegrityError):
+        async with savepoint:
+            await data_connection.execute(statement)
 
 
 async def test_recommendation_is_unique_when_alert_and_rule_match(
@@ -256,6 +272,67 @@ 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)
+
+
+def _rapport(**remplacements: object) -> dict[str, object]:
+    defauts: dict[str, object] = {
+        "site_id": None,
+        "window_start": MOMENT,
+        "window_end": MOMENT,
+        "n_observations": 12,
+        "model_references": ["lightgbm-aaa"],
+        "status": "stable",
+        "reason": None,
+    }
+    return {**defauts, **remplacements}
+
+
+async def test_drift_report_rejects_an_unknown_status(data_connection: AsyncConnection) -> None:
+    statement = insert(DriftReport).values(**_rapport(status="douteux", reason="x"))
+    savepoint = data_connection.begin_nested()
+
+    with pytest.raises(IntegrityError):
+        async with savepoint:
+            await data_connection.execute(statement)
+
+
+async def test_drift_report_rejects_a_drift_without_a_reason(
+    data_connection: AsyncConnection,
+) -> None:
+    statement = insert(DriftReport).values(**_rapport(status="derive"))
+    savepoint = data_connection.begin_nested()
+
+    with pytest.raises(IntegrityError):
+        async with savepoint:
+            await data_connection.execute(statement)
+
+
+async def test_drift_report_accepts_one_global_row_without_a_site(
+    data_connection: AsyncConnection,
+) -> None:
+    identifiant = (
+        await data_connection.execute(
+            insert(DriftReport).values(**_rapport()).returning(DriftReport.drift_report_id)
+        )
+    ).scalar_one()
+
+    assert identifiant is not None
+
+
+async def test_drift_report_is_unique_when_window_and_site_match(
+    data_connection: AsyncConnection,
+) -> None:
+    fenetre = MOMENT + timedelta(days=1)
+    statement = insert(DriftReport).values(**_rapport(window_end=fenetre))
+    await data_connection.execute(statement)
+
+    savepoint = data_connection.begin_nested()
+
+    with pytest.raises(IntegrityError):
+        async with savepoint:
             await data_connection.execute(statement)

apps/backend/tests/etl/test_historical_import.py

@@ -106,13 +106,15 @@ def test_validate_source_accepts_valid_dataset():
 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,
-            make_metadata(),
+            metadata,
         )
 
 
@@ -124,13 +126,15 @@ def test_validate_source_rejects_duplicates():
         "timestamp",
     ]
 
+    metadata = make_metadata()
+
     with pytest.raises(
         ValueError,
         match="doublons",
     ):
         validate_source(
             frame,
-            make_metadata(),
+            metadata,
         )
 
 
@@ -139,13 +143,15 @@ def test_validate_source_rejects_unknown_site():
 
     frame.loc[1, "site_id"] = "SITE999"
 
+    metadata = make_metadata()
+
     with pytest.raises(
         ValueError,
         match="Sites incohérents",
     ):
         validate_source(
             frame,
-            make_metadata(),
+            metadata,
         )
 
 

apps/backend/tests/repositories/test_alert.py

@@ -5,6 +5,7 @@ import pytest
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models.energy import Alert
+from app.repositories import alert as module_alert
 from app.repositories.alert import AlertRepository
 from app.schemas.alert import AlertSeverity
 from tests.repositories.test_site import creer as creer_site
@@ -146,3 +147,20 @@ async def test_create_many_does_nothing_for_an_empty_list(session: AsyncSession)
     creees = await depot.create_many([])
 
     assert creees == []
+
+
+async def test_create_many_inserts_every_alert_across_several_batches(
+    session: AsyncSession, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setattr(module_alert, "TAILLE_DE_LOT", 2)
+    site = await creer_site(session)
+    depot = AlertRepository(session)
+    a_inserer = [
+        _alerte_a_inserer(site_id=site.site_id, source_alert_id=f"threshold:lot-{index}")
+        for index in range(5)
+    ]
+
+    creees = await depot.create_many(a_inserer)
+    await session.rollback()
+
+    assert len(creees) == 5

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_drift.py

@@ -0,0 +1,187 @@
+from datetime import UTC, datetime, timedelta
+
+import pytest
+from sqlalchemy.dialects import postgresql
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy.sql import ClauseElement
+
+from app.repositories.drift import (
+    DriftRepository,
+    NouveauRapportDerive,
+    _lectures_retenues,
+    _predictions_retenues,
+)
+from tests.repositories.test_prediction import creer_prediction
+from tests.repositories.test_reading import creer_lecture
+from tests.repositories.test_site import creer as creer_site
+
+DEBUT = datetime(2026, 9, 15, tzinfo=UTC)
+FIN = datetime(2026, 9, 22, tzinfo=UTC)
+CIBLE = datetime(2026, 9, 16, 12, tzinfo=UTC)
+
+
+def sql(requete: ClauseElement) -> str:
+    return str(requete.compile(dialect=postgresql.dialect()))  # type: ignore[no-untyped-call]
+
+
+def rapport(**remplacements: object) -> NouveauRapportDerive:
+    defauts: dict[str, object] = {
+        "site_id": None,
+        "window_start": DEBUT,
+        "window_end": FIN,
+        "reference_start": None,
+        "reference_end": None,
+        "n_observations": 10,
+        "mae": 1.0,
+        "mape": 5.0,
+        "bias": 0.1,
+        "reference_mae": None,
+        "coverage_ratio": 1.0,
+        "insufficient_data_ratio": 0.0,
+        "model_references": ["lightgbm-aaa"],
+        "status": "stable",
+        "reason": None,
+    }
+    return NouveauRapportDerive(**{**defauts, **remplacements})  # type: ignore[arg-type]
+
+
+def test_predictions_keep_one_row_per_site_and_target_in_sql() -> None:
+    requete = sql(_predictions_retenues(debut=DEBUT, fin=FIN, site_id=None).element)
+
+    assert "DISTINCT ON (prediction.site_id, prediction.target_at)" in requete
+    assert "prediction.prediction_id DESC" in requete
+
+
+def test_readings_keep_one_row_per_site_and_instant_in_sql() -> None:
+    requete = sql(_lectures_retenues(debut=DEBUT, fin=FIN, site_id=None).element)
+
+    assert "DISTINCT ON (reading.site_id, reading.timestamp)" in requete
+    assert "reading.reading_id DESC" in requete
+
+
+def test_predictions_restrict_themselves_to_the_requested_site_in_sql() -> None:
+    requete = sql(_predictions_retenues(debut=DEBUT, fin=FIN, site_id="SITE001").element)
+
+    assert requete.count("prediction.site_id = ") == 1
+
+
+def test_readings_ignore_a_missing_consumption_in_sql() -> None:
+    requete = sql(_lectures_retenues(debut=DEBUT, fin=FIN, site_id=None).element)
+
+    assert "reading.consumption_kwh IS NOT NULL" in requete
+
+
+@pytest.mark.integration
+async def test_repository_pairs_a_prediction_with_the_reading_of_the_same_instant(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    await creer_prediction(session, site_id=site.site_id, target_at=CIBLE, predicted_value=12.0)
+    await creer_lecture(session, site_id=site.site_id, timestamp=CIBLE, consumption_kwh=10.0)
+
+    paires = await DriftRepository(session).paires(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    await session.rollback()
+
+    assert [(p.predicted_value, p.actual_value) for p in paires] == [(12.0, 10.0)]
+
+
+@pytest.mark.integration
+async def test_repository_keeps_the_latest_run_when_several_predictions_share_a_target(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    await creer_prediction(session, site_id=site.site_id, target_at=CIBLE, predicted_value=12.0)
+    await creer_prediction(session, site_id=site.site_id, target_at=CIBLE, predicted_value=99.0)
+    await creer_lecture(session, site_id=site.site_id, timestamp=CIBLE, consumption_kwh=10.0)
+
+    paires = await DriftRepository(session).paires(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    await session.rollback()
+
+    assert [p.predicted_value for p in paires] == [99.0]
+
+
+@pytest.mark.integration
+async def test_repository_keeps_one_reading_per_instant_when_two_sources_wrote_the_same_hour(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    await creer_prediction(session, site_id=site.site_id, target_at=CIBLE, predicted_value=12.0)
+    await creer_lecture(
+        session, site_id=site.site_id, timestamp=CIBLE, source="api_current", consumption_kwh=10.0
+    )
+    await creer_lecture(
+        session, site_id=site.site_id, timestamp=CIBLE, source="api_history", consumption_kwh=20.0
+    )
+
+    paires = await DriftRepository(session).paires(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    await session.rollback()
+
+    assert [p.actual_value for p in paires] == [20.0]
+
+
+@pytest.mark.integration
+async def test_repository_excludes_an_insufficient_data_prediction_from_the_pairs(
+    session: AsyncSession,
+) -> None:
+    site = await creer_site(session)
+    await creer_prediction(
+        session,
+        site_id=site.site_id,
+        target_at=CIBLE,
+        predicted_value=None,
+        status="insufficient_data",
+        failure_reason="historique trop court",
+    )
+    await creer_lecture(session, site_id=site.site_id, timestamp=CIBLE, consumption_kwh=10.0)
+
+    depot = DriftRepository(session)
+    paires = await depot.paires(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    comptages = await depot.comptages(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    await session.rollback()
+
+    assert paires == []
+    assert [(c.status, c.nombre) for c in comptages] == [("insufficient_data", 1)]
+
+
+@pytest.mark.integration
+async def test_repository_excludes_a_target_outside_the_window(session: AsyncSession) -> None:
+    site = await creer_site(session)
+    hors_fenetre = FIN + timedelta(hours=1)
+    await creer_prediction(
+        session, site_id=site.site_id, target_at=hors_fenetre, predicted_value=12.0
+    )
+    await creer_lecture(session, site_id=site.site_id, timestamp=hors_fenetre, consumption_kwh=10.0)
+
+    paires = await DriftRepository(session).paires(debut=DEBUT, fin=FIN, site_id=site.site_id)
+    await session.rollback()
+
+    assert paires == []
+
+
+@pytest.mark.integration
+async def test_repository_reads_back_the_global_report_it_wrote(session: AsyncSession) -> None:
+    depot = DriftRepository(session)
+    fenetre = datetime(2035, 3, 1, tzinfo=UTC)
+
+    ecrites = await depot.enregistre([rapport(window_end=fenetre)])
+    derniers = await depot.derniers()
+    globaux = [r for r in derniers if r.site_id is None and r.window_end == fenetre]
+    await session.rollback()
+
+    assert ecrites == 1
+    assert len(globaux) == 1
+
+
+@pytest.mark.integration
+async def test_repository_ignores_a_second_report_for_the_same_window_and_site(
+    session: AsyncSession,
+) -> None:
+    depot = DriftRepository(session)
+    fenetre = datetime(2035, 4, 1, tzinfo=UTC)
+
+    premiere = await depot.enregistre([rapport(window_end=fenetre)])
+    seconde = await depot.enregistre([rapport(window_end=fenetre, status="derive", reason="x")])
+    await session.rollback()
+
+    assert premiere == 1
+    assert seconde == 0

apps/backend/tests/repositories/test_password_reset_token.py

@@ -131,11 +131,14 @@ async def test_the_database_refuses_two_tokens_sharing_a_fingerprint(
         user_agent=None,
     )
 
+    empreinte = fingerprint_refresh(secret)
+    expiration = datetime.now(UTC) + DUREE
+
     with pytest.raises(IntegrityError):
         await depot.create(
             user_id=compte,
-            token_hash=fingerprint_refresh(secret),
+            token_hash=empreinte,
-            expires_at=datetime.now(UTC) + DUREE,
+            expires_at=expiration,
             client_ip=None,
             user_agent=None,
         )

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(),
+            family_id=famille,
-            token_hash=fingerprint_refresh(secret),
+            token_hash=empreinte,
-            expires_at=datetime.now(UTC) + DUREE,
+            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()
 
-    with pytest.raises(IntegrityError):
+    requete = text(
-        await session.execute(
+        "insert into app_user (email, password_hash, role) values (:e, '$argon2id$x', 'lecteur')"
-            text(
-                "insert into app_user (email, password_hash, role) "
-                "values (:e, '$argon2id$x', 'lecteur')"
-            ),
-            {"e": saisie},
     )
+
+    with pytest.raises(IntegrityError):
+        await session.execute(requete, {"e": saisie})
     await session.rollback()
 
 

apps/backend/tests/services/test_drift.py

@@ -0,0 +1,260 @@
+from collections.abc import Sequence
+from datetime import UTC, datetime, timedelta
+
+import pytest
+
+from app.repositories.drift import ComptageStatut, PaireDerive
+from app.services.drift import (
+    STATUT_DERIVE,
+    STATUT_INDETERMINE,
+    STATUT_STABLE,
+    DriftService,
+    Seuils,
+    mesure,
+)
+
+INSTANT = datetime(2026, 9, 22, 12, 0, tzinfo=UTC)
+
+
+def paire(
+    *, site_id: str = "SITE001", prevu: float, reel: float, reference: str = "lightgbm-aaa"
+) -> PaireDerive:
+    return PaireDerive(
+        site_id=site_id,
+        target_at=INSTANT,
+        predicted_value=prevu,
+        actual_value=reel,
+        model_reference=reference,
+    )
+
+
+def paires(
+    *, site_id: str = "SITE001", nombre: int, prevu: float, reel: float
+) -> list[PaireDerive]:
+    return [paire(site_id=site_id, prevu=prevu, reel=reel) for _ in range(nombre)]
+
+
+class FauxDepot:
+    def __init__(
+        self,
+        *,
+        recentes: Sequence[PaireDerive] = (),
+        anciennes: Sequence[PaireDerive] = (),
+        comptages: Sequence[ComptageStatut] = (),
+    ) -> None:
+        self.recentes = list(recentes)
+        self.anciennes = list(anciennes)
+        self._comptages = list(comptages)
+        self.fenetres: list[tuple[datetime, datetime]] = []
+
+    async def paires(
+        self, *, debut: datetime, fin: datetime, site_id: str | None = None
+    ) -> Sequence[PaireDerive]:
+        self.fenetres.append((debut, fin))
+        return self.recentes if len(self.fenetres) == 1 else self.anciennes
+
+    async def comptages(
+        self, *, debut: datetime, fin: datetime, site_id: str | None = None
+    ) -> Sequence[ComptageStatut]:
+        return self._comptages
+
+
+def service(depot: FauxDepot, **surcharges: object) -> DriftService:
+    return DriftService(depot, seuils=Seuils(**surcharges))  # type: ignore[arg-type]
+
+
+def test_drift_averages_the_absolute_gap_between_forecast_and_actual() -> None:
+    metriques = mesure([paire(prevu=12.0, reel=10.0), paire(prevu=8.0, reel=10.0)])
+
+    assert metriques.mae == 2.0
+    assert metriques.n_observations == 2
+
+
+def test_drift_computes_a_signed_bias_when_the_model_overforecasts() -> None:
+    metriques = mesure([paire(prevu=12.0, reel=10.0), paire(prevu=14.0, reel=10.0)])
+
+    assert metriques.bias == 3.0
+
+
+def test_drift_computes_a_negative_bias_when_the_model_underforecasts() -> None:
+    metriques = mesure([paire(prevu=8.0, reel=10.0), paire(prevu=6.0, reel=10.0)])
+
+    assert metriques.bias == -3.0
+
+
+def test_drift_excludes_a_zero_actual_from_the_mape_only() -> None:
+    metriques = mesure([paire(prevu=11.0, reel=10.0), paire(prevu=5.0, reel=0.0)])
+
+    assert metriques.mape == 10.0
+    assert metriques.n_observations == 2
+    assert metriques.mae == 3.0
+
+
+def test_drift_reports_no_mape_when_every_actual_is_zero() -> None:
+    metriques = mesure([paire(prevu=1.0, reel=0.0)])
+
+    assert metriques.mape is None
+
+
+def test_drift_lists_every_model_reference_seen_in_the_window() -> None:
+    metriques = mesure(
+        [paire(prevu=10.0, reel=10.0, reference="lightgbm-bbb"), paire(prevu=10.0, reel=10.0)]
+    )
+
+    assert metriques.model_references == ["lightgbm-aaa", "lightgbm-bbb"]
+
+
+async def test_drift_reports_indetermine_when_the_window_holds_too_few_observations() -> None:
+    depot = FauxDepot(recentes=paires(nombre=3, prevu=10.0, reel=10.0))
+
+    rapports = await service(depot, min_observations=24).evaluate(now=INSTANT)
+
+    assert {rapport.status for rapport in rapports} == {STATUT_INDETERMINE}
+    assert all(rapport.reason for rapport in rapports)
+
+
+async def test_drift_reports_derive_when_the_recent_mae_exceeds_the_reference_ratio() -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=14.0, reel=10.0),
+        anciennes=paires(nombre=30, prevu=11.0, reel=10.0),
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=30)],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    global_ = next(rapport for rapport in rapports if rapport.site_id is None)
+    assert global_.status == STATUT_DERIVE
+    assert global_.mae == 4.0
+    assert global_.reference_mae == 1.0
+
+
+async def test_drift_reports_stable_when_the_recent_mae_stays_close_to_the_reference() -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=11.0, reel=10.0),
+        anciennes=paires(nombre=30, prevu=11.0, reel=10.0),
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=30)],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    global_ = next(rapport for rapport in rapports if rapport.site_id is None)
+    assert global_.status == STATUT_STABLE
+    assert global_.reason is None
+
+
+async def test_drift_reports_indetermine_when_the_reference_window_is_empty() -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=14.0, reel=10.0),
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=30)],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    global_ = next(rapport for rapport in rapports if rapport.site_id is None)
+    assert global_.status == STATUT_INDETERMINE
+    assert global_.reference_mae is None
+
+
+async def test_drift_reports_derive_when_the_coverage_ratio_falls_under_the_threshold() -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=10.0, reel=10.0),
+        anciennes=paires(nombre=30, prevu=10.0, reel=10.0),
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=100)],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    global_ = next(rapport for rapport in rapports if rapport.site_id is None)
+    assert global_.status == STATUT_DERIVE
+    assert global_.coverage_ratio == 0.3
+
+
+async def test_drift_reports_one_line_per_site_and_one_global_line() -> None:
+    depot = FauxDepot(
+        recentes=[
+            *paires(site_id="SITE001", nombre=12, prevu=10.0, reel=10.0),
+            *paires(site_id="SITE002", nombre=12, prevu=10.0, reel=10.0),
+        ],
+        comptages=[
+            ComptageStatut(site_id="SITE001", status="available", nombre=12),
+            ComptageStatut(site_id="SITE002", status="available", nombre=12),
+        ],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    assert [rapport.site_id for rapport in rapports] == ["SITE001", "SITE002", None]
+    assert next(r for r in rapports if r.site_id is None).n_observations == 24
+
+
+async def test_drift_still_reports_a_site_that_stopped_being_scored() -> None:
+    depot = FauxDepot(
+        recentes=paires(site_id="SITE001", nombre=30, prevu=10.0, reel=10.0),
+        anciennes=[
+            *paires(site_id="SITE001", nombre=30, prevu=10.0, reel=10.0),
+            *paires(site_id="SITE002", nombre=30, prevu=10.0, reel=10.0),
+        ],
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=30)],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    disparu = next(rapport for rapport in rapports if rapport.site_id == "SITE002")
+    assert disparu.status == STATUT_INDETERMINE
+    assert disparu.n_observations == 0
+
+
+async def test_drift_measures_the_share_of_sites_left_without_enough_history() -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=10.0, reel=10.0),
+        comptages=[
+            ComptageStatut(site_id="SITE001", status="available", nombre=30),
+            ComptageStatut(site_id="SITE001", status="insufficient_data", nombre=10),
+        ],
+    )
+
+    rapports = await service(depot, min_observations=10).evaluate(now=INSTANT)
+
+    assert next(r for r in rapports if r.site_id is None).insufficient_data_ratio == 0.25
+
+
+async def test_drift_closes_the_window_before_the_grace_delay() -> None:
+    depot = FauxDepot()
+
+    await service(depot, grace=timedelta(hours=2), fenetre=timedelta(hours=168)).evaluate(
+        now=INSTANT
+    )
+
+    recente, reference = depot.fenetres
+    assert recente[1] == INSTANT - timedelta(hours=2)
+    assert recente[0] == INSTANT - timedelta(hours=170)
+    assert reference[1] == recente[0]
+
+
+async def test_drift_truncates_the_reference_instant_to_the_hour() -> None:
+    premier = FauxDepot()
+    second = FauxDepot()
+
+    await service(premier).evaluate(now=INSTANT + timedelta(minutes=17, microseconds=3))
+    await service(second).evaluate(now=INSTANT + timedelta(minutes=48))
+
+    assert premier.fenetres[0] == second.fenetres[0]
+
+
+@pytest.mark.parametrize(
+    ("prevu", "attendu"),
+    [(10.0, STATUT_STABLE), (30.0, STATUT_DERIVE)],
+    ids=["mae_stable", "mae_triplee"],
+)
+async def test_drift_compares_the_recent_window_to_the_reference_one(
+    prevu: float, attendu: str
+) -> None:
+    depot = FauxDepot(
+        recentes=paires(nombre=30, prevu=prevu, reel=10.0),
+        anciennes=paires(nombre=30, prevu=10.0, reel=10.0),
+        comptages=[ComptageStatut(site_id="SITE001", status="available", nombre=30)],
+    )
+
+    rapports = await service(depot, min_observations=10, mae_plancher=1.0).evaluate(now=INSTANT)
+
+    assert next(r for r in rapports if r.site_id is None).status == attendu

apps/backend/tests/services/test_reading.py

@@ -116,13 +116,11 @@ async def test_list_history_normalizes_naive_datetimes_to_utc() -> None:
 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(
+        await service.list_history(start=debut, end=fin, limit=500, offset=0)
-            start=datetime(2026, 9, 2, tzinfo=UTC),
-            end=datetime(2026, 9, 1, tzinfo=UTC),
-            limit=500,
-            offset=0,
-        )
 
 
 async def test_list_history_raises_when_start_equals_end() -> None:

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_chaine_ml_api.py

@@ -0,0 +1,223 @@
+"""Piege : ce fichier porte le marqueur `chaine`, pas `integration` - test_the_ml_binaries...()
+
+Il lance les vrais binaires `enervision_ml.train` et `enervision_ml.score` dans l'environnement
+uv de `ml/`, que le job `integration` de `backend.yml` n'installe pas. Un marqueur distinct evite
+que ce job, et `make test`, ne le selectionnent et n'echouent faute de `ml/.venv`.
+"""
+
+import math
+import os
+import subprocess
+from collections.abc import AsyncIterator, Iterator
+from dataclasses import dataclass, field
+from datetime import UTC, datetime, timedelta
+from functools import partial
+from pathlib import Path
+from typing import Any
+from uuid import uuid4
+
+import anyio
+import pytest
+from fastapi import FastAPI
+from httpx import AsyncClient
+from sqlalchemy import delete, insert, make_url
+
+from app.api.deps import get_current_principal
+from app.core.config import get_settings
+from app.core.principal import Principal
+from app.core.roles import AccountKind, Role
+from app.db.session import get_session_factory
+from app.models.energy import Prediction, Reading, Site
+
+pytestmark = pytest.mark.chaine
+
+RACINE = Path(__file__).resolve().parents[3]
+ML = RACINE / "ml"
+PYTHON_ML = Path(os.environ.get("ML_PYTHON", ML / ".venv" / "bin" / "python"))
+
+HEURES_COMPLETES = 400
+HEURES_INSUFFISANTES = 100
+
+
+def lecteur() -> Principal:
+    return Principal(
+        id=uuid4(),
+        email="lecteur@enervision.fr",
+        role=Role.LECTEUR,
+        kind=AccountKind.HUMAIN,
+        must_change_password=False,
+    )
+
+
+def url_ml() -> str:
+    """Derive la chaine du pipeline de celle du backend plutot que de la recopier : les deux
+    cotes visent ainsi la meme base, dans leur dialecte respectif."""
+    return (
+        make_url(get_settings().database_url)
+        .set(drivername="postgresql+psycopg")
+        .render_as_string(hide_password=False)
+    )
+
+
+def lance_ml(module: str, *arguments: str, journal: Path) -> subprocess.CompletedProcess[str]:
+    if not PYTHON_ML.exists():
+        pytest.fail(
+            f"Environnement ml/ absent ({PYTHON_ML}). Lancer `cd ml && uv sync --all-groups`."
+        )
+
+    return subprocess.run(  # noqa: S603 -- argv en liste, sans shell, binaire resolu dans le depot
+        [str(PYTHON_ML), "-m", module, *arguments],
+        cwd=ML,
+        text=True,
+        capture_output=True,
+        timeout=600,
+        check=False,
+        env={
+            **os.environ,
+            "ML_DATABASE_URL": url_ml(),
+            "MLFLOW_TRACKING_URI": f"sqlite:///{journal}/mlflow.db",
+        },
+    )
+
+
+async def executer(module: str, *arguments: str, journal: Path) -> subprocess.CompletedProcess[str]:
+    resultat = await anyio.to_thread.run_sync(
+        partial(lance_ml, module, *arguments, journal=journal)
+    )
+    assert resultat.returncode == 0, resultat.stderr
+    return resultat
+
+
+@dataclass
+class Parc:
+    sites: list[str] = field(default_factory=list)
+
+
+def lignes_horaires(site_id: str, *, heures: int, fin: datetime) -> list[dict[str, Any]]:
+    return [
+        {
+            "site_id": site_id,
+            "timestamp": fin - timedelta(hours=decalage),
+            "source": "api_history",
+            "consumption_kwh": 50.0 + math.sin(decalage / 12.0) * 10.0,
+            "temperature_celsius": 15.0,
+            "humidity_percent": 50.0,
+            "solar_irradiance_wm2": 0.0,
+            "is_working_hours": True,
+            "raw_data": {},
+        }
+        for decalage in reversed(range(heures))
+    ]
+
+
+@pytest.fixture
+async def parc() -> AsyncIterator[Parc]:
+    """Deux sites dotes d'un historique complet, un troisieme qui n'atteint pas le lag de 168 h.
+
+    Les ecritures sont validees : les binaires ML ouvrent leur propre connexion et ne verraient
+    pas une transaction en cours.
+    """
+    fin = datetime.now(UTC).replace(minute=0, second=0, microsecond=0) - timedelta(hours=1)
+    marque = uuid4().hex[:12]
+    complets = [f"TEST-{marque}-A", f"TEST-{marque}-B"]
+    partiel = f"TEST-{marque}-C"
+    parc = Parc(sites=[*complets, partiel])
+
+    async with get_session_factory()() as session:
+        await session.execute(
+            insert(Site),
+            [
+                {
+                    "site_id": site_id,
+                    "site_name": f"Site {site_id}",
+                    "site_type": "office",
+                    "capacity_kw": 100.0,
+                }
+                for site_id in parc.sites
+            ],
+        )
+        for site_id in complets:
+            await session.execute(
+                insert(Reading), lignes_horaires(site_id, heures=HEURES_COMPLETES, fin=fin)
+            )
+        await session.execute(
+            insert(Reading), lignes_horaires(partiel, heures=HEURES_INSUFFISANTES, fin=fin)
+        )
+        await session.commit()
+
+    try:
+        yield parc
+    finally:
+        async with get_session_factory()() as session:
+            await session.execute(delete(Prediction).where(Prediction.site_id.in_(parc.sites)))
+            await session.execute(delete(Reading).where(Reading.site_id.in_(parc.sites)))
+            await session.execute(delete(Site).where(Site.site_id.in_(parc.sites)))
+            await session.commit()
+
+
+@pytest.fixture
+def principal_lecteur(app: FastAPI) -> Iterator[None]:
+    app.dependency_overrides[get_current_principal] = lecteur
+    yield
+    app.dependency_overrides.pop(get_current_principal, None)
+
+
+async def resume_du_site(client: AsyncClient, site_id: str) -> dict[str, Any]:
+    reponse = await client.get("/api/v1/predictions")
+
+    assert reponse.status_code == 200
+    sites = reponse.json()["sites"]
+    return next(site for site in sites if site["site_id"] == site_id)
+
+
+async def entraine_et_score(parc: Parc, tmp_path: Path, *arguments: str) -> Path:
+    modele = tmp_path / "lightgbm-consumption.txt"
+
+    await executer(
+        "enervision_ml.train",
+        "--model-output",
+        str(modele),
+        "--mlflow-tracking-uri",
+        f"sqlite:///{tmp_path}/mlflow.db",
+        journal=tmp_path,
+    )
+    await executer("enervision_ml.score", "--model", str(modele), *arguments, journal=tmp_path)
+
+    return modele
+
+
+async def test_the_ml_binaries_produce_a_prediction_that_the_api_serves(
+    parc: Parc, tmp_path: Path, client: AsyncClient, principal_lecteur: None
+) -> None:
+    await entraine_et_score(parc, tmp_path)
+
+    servi = await resume_du_site(client, parc.sites[0])
+
+    assert servi["prediction"]["status"] == "available"
+    assert servi["prediction"]["predicted_value"] is not None
+    assert servi["prediction"]["target_metric"] == "consumption_kwh"
+
+
+async def test_the_api_exposes_the_failure_reason_of_a_site_without_enough_history(
+    parc: Parc, tmp_path: Path, client: AsyncClient, principal_lecteur: None
+) -> None:
+    await entraine_et_score(parc, tmp_path)
+
+    servi = await resume_du_site(client, parc.sites[-1])
+
+    assert servi["prediction"]["status"] == "insufficient_data"
+    assert servi["prediction"]["predicted_value"] is None
+    assert servi["prediction"]["failure_reason"] is not None
+
+
+async def test_the_api_serves_the_latest_run_when_the_score_cli_runs_twice(
+    parc: Parc, tmp_path: Path, client: AsyncClient, principal_lecteur: None
+) -> None:
+    modele = await entraine_et_score(parc, tmp_path)
+    premier = await resume_du_site(client, parc.sites[0])
+
+    await executer("enervision_ml.score", "--model", str(modele), journal=tmp_path)
+
+    second = await resume_du_site(client, parc.sites[0])
+    assert second["prediction"]["created_at"] > premier["prediction"]["created_at"]
+    assert second["prediction"]["model_reference"] == premier["prediction"]["model_reference"]

apps/backend/tests/test_cli.py

@@ -19,13 +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(

apps/backend/tests/test_drift_cli.py

@@ -0,0 +1,108 @@
+from datetime import UTC, datetime, timedelta
+
+import pytest
+
+from app.monitoring import drift as cli
+from app.repositories.drift import NouveauRapportDerive
+from app.services.drift import STATUT_DERIVE, STATUT_STABLE, Seuils
+
+INSTANT = datetime(2026, 9, 22, 12, tzinfo=UTC)
+
+
+def rapport(*, site_id: str | None, status: str, reason: str | None = None) -> NouveauRapportDerive:
+    return NouveauRapportDerive(
+        site_id=site_id,
+        window_start=INSTANT - timedelta(hours=168),
+        window_end=INSTANT,
+        reference_start=None,
+        reference_end=None,
+        n_observations=48,
+        mae=1.5,
+        mape=12.0,
+        bias=0.3,
+        reference_mae=1.2,
+        coverage_ratio=1.0,
+        insufficient_data_ratio=0.0,
+        model_references=["lightgbm-aaa"],
+        status=status,
+        reason=reason,
+    )
+
+
+def installe(monkeypatch: pytest.MonkeyPatch, rapports: list[NouveauRapportDerive]) -> None:
+    async def fausse_execution(
+        *, now: datetime | None, site_id: str | None, seuils: Seuils | None
+    ) -> list[NouveauRapportDerive]:
+        return rapports
+
+    monkeypatch.setattr(cli, "run_drift", fausse_execution)
+
+
+def test_parse_args_defaults_to_the_standard_window() -> None:
+    arguments = cli.parse_args([])
+
+    assert arguments.window_hours == 168
+    assert arguments.grace_hours == 2
+    assert arguments.fail_on_drift is False
+
+
+def test_parse_args_reads_the_site_id() -> None:
+    assert cli.parse_args(["--site-id", "SITE001"]).site_id == "SITE001"
+
+
+def test_parse_args_parses_the_instant_option() -> None:
+    arguments = cli.parse_args(["--now", "2026-09-22T12:00:00+00:00"])
+
+    assert arguments.now == INSTANT
+
+
+def test_parse_instant_treats_a_naive_datetime_as_utc() -> None:
+    assert cli._parse_instant("2026-09-22T12:00:00") == INSTANT
+
+
+def test_seuils_depuis_translates_the_hour_options_into_durations() -> None:
+    seuils = cli.seuils_depuis(cli.parse_args(["--window-hours", "24", "--grace-hours", "1"]))
+
+    assert seuils.fenetre == timedelta(hours=24)
+    assert seuils.grace == timedelta(hours=1)
+
+
+def test_main_prints_the_verdict_of_every_line(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    installe(
+        monkeypatch,
+        [
+            rapport(site_id="SITE001", status=STATUT_STABLE),
+            rapport(site_id=None, status=STATUT_STABLE),
+        ],
+    )
+
+    code = cli.main([])
+
+    sortie = capsys.readouterr().out
+    assert code == 0
+    assert "SITE001" in sortie
+    assert "TOUS SITES" in sortie
+
+
+def test_main_exits_non_zero_when_drift_is_detected_and_the_flag_is_set(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    installe(monkeypatch, [rapport(site_id=None, status=STATUT_DERIVE, reason="MAE doublée")])
+
+    code = cli.main(["--fail-on-drift"])
+
+    assert code == 1
+    assert "MAE doublée" in capsys.readouterr().out
+
+
+def test_main_exits_zero_when_drift_is_detected_without_the_flag(
+    monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
+) -> None:
+    installe(monkeypatch, [rapport(site_id=None, status=STATUT_DERIVE, reason="MAE doublée")])
+
+    code = cli.main([])
+
+    assert code == 0
+    assert capsys.readouterr().out != ""

apps/backend/uv.lock

@@ -655,31 +655,31 @@ wheels = [
 
 [[package]]
 name = "pandas"
-version = "3.0.5"
+version = "3.0.6"
 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" }
+sdist = { url = "https://files.pythonhosted.org/packages/e2/17/d7b106e05bfa642e8694451e7d3d759c6a241c5386a5d962e4f66c047e06/pandas-3.0.6.tar.gz", hash = "sha256:66b07ef7315a31bfe1089cd3d71a7de781c9dca986762d0b4fe7c0ef17465d10", size = 4667686, upload-time = "2026-09-17T23:23:18.345Z" }
 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/75/55/1a8875395b05ccd572cbca0b9255dcd2db6e6508e632a558c1a6884b39ad/pandas-3.0.6-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:ee913a91669056c1de1a6b733fbfeab711de9e54e3bee2dfa5fe79d9457247d1", size = 10492213, upload-time = "2026-09-17T23:21:40.746Z" },
-    { 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/35/61/47ae13476995cc8a40cd609e93e7cf11f273d8692925c2903cb6d38aa0d1/pandas-3.0.6-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:ff51a4459ed036e93d1eb1bb5e6e7b28685d3cb6b7c12b91c05b31024e234729", size = 10156618, upload-time = "2026-09-17T23:21:44.142Z" },
-    { 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/bc/f2/cc5f2adb8d6e86a85d9fb5128f8cf205a61189336f70d1f7faf0d1b53ec9/pandas-3.0.6-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:654aae059295dbba6ecd2328ca12712a2cf1676214c8699f1c29213f7ccf9c34", size = 10375479, upload-time = "2026-09-17T23:21:47.159Z" },
-    { 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/ca/ba/ffdcb19be4ff6bfe7d969e7cef2c567c633df5a3a1cc1053394ad053bca8/pandas-3.0.6-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:62f51d7f651c8054c5e82a69265c98082e795d1442df7ca6edc3a545d61214b1", size = 10783244, upload-time = "2026-09-17T23:21:50.367Z" },
-    { 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/77/5b/e150075b2c6eb69fae896f2d9239bc6ed07db97735971d53d66de6553460/pandas-3.0.6-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:22172a92e7ee678ec0140c7af4fc9366b55413834a1cd86af78b3caa0b0574de", size = 11382223, upload-time = "2026-09-17T23:21:53.355Z" },
-    { 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/d6/8a/b441c587dc7355bf6e1f68a91b4f76a6c29740f0c23be3acc5d4ebbeea6d/pandas-3.0.6-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:583be68728a31d0d750d5b8d9e00f02b153df0d4655f858bde93cb84cfc4227c", size = 11854881, upload-time = "2026-09-17T23:21:56.342Z" },
-    { 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/b7/e9/f43410fada510b43fec09993c08f552086c3d247d3ee801a678f3cb10ea5/pandas-3.0.6-cp314-cp314-win_amd64.whl", hash = "sha256:77ccbe5057aece6fc172b9b77f19c04335af6882bc2e10c8f3ee4e6bfb3da553", size = 9791672, upload-time = "2026-09-17T23:21:59.332Z" },
-    { 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/8b/9e/db14c059c21f9baa1907d436f8bf30e0c76c6288225c5e8b79a08ba8b2c5/pandas-3.0.6-cp314-cp314-win_arm64.whl", hash = "sha256:fb625f426b375bcc96e3a04c5d5d266cd7be6ae5d6866e0e703382ab5164068c", size = 9121246, upload-time = "2026-09-17T23:22:02.123Z" },
-    { 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/67/ba/bad0f8dac020ab38a8637fddab01a57a82da7a496a6e6f19590aad53ab62/pandas-3.0.6-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:9e492cd4bdba6778de4fe0df7f4590c012161ebcf9902dce01b01dc683105514", size = 10920612, upload-time = "2026-09-17T23:22:05.404Z" },
-    { 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/c4/a9/b500982e9aac6d52a58da4ad3f11e14168a315b06906b3f397c427878065/pandas-3.0.6-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:d7dcd21238cbb4828ff148481ba01cac8946dc5121457b5aeba28636f8f99a60", size = 10570213, upload-time = "2026-09-17T23:22:08.44Z" },
-    { 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/4b/fa/e6ecd0073c98be8f840ac3125b955272835d7d9fd69f5944383b164deb5e/pandas-3.0.6-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6ff482fa91fa2bafd92e8fe66ce3645c851824310f295c1f0a2f96e928fc4541", size = 10252356, upload-time = "2026-09-17T23:22:11.302Z" },
-    { 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/04/f5/001e230a7a7803590d9275a1a3f7e1bb605e3a495cfe5e8d3a532090621b/pandas-3.0.6-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:db7ec631f26223beee8e5c9e0b8f23c24d8197bbd1d982421d4e3188bea51965", size = 10655952, upload-time = "2026-09-17T23:22:14.283Z" },
-    { 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/ca/ab/bab587148a3852c96aae26c4b5f9e04ce2221801ad94e166b4fbf969ede0/pandas-3.0.6-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:bd75ed0c840f709fc2ae26ddd9534ac77ca1a48ac0cce521a74acaa85f3340a7", size = 11274177, upload-time = "2026-09-17T23:22:17.352Z" },
-    { 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/f3/32/74b48d87df2b80892d713c149abfe36d5db4de41b4eccb042a2bc07dafc1/pandas-3.0.6-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:ef738d71d1059245b6bb03e312be06d8b3821326a83486c1ad03b9aba3710e44", size = 11719041, upload-time = "2026-09-17T23:22:20.227Z" },
-    { 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/f6/c6/d64b72d64d7eb0fad9fe424d34138e45dee70ddbea1360dcd0adf30e28f6/pandas-3.0.6-cp314-cp314t-win_amd64.whl", hash = "sha256:429d9df32731ab01383ed98f2baa7a60368090d1a94fc06019a12062510e8630", size = 10191388, upload-time = "2026-09-17T23:22:23.524Z" },
-    { 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" },
+    { url = "https://files.pythonhosted.org/packages/7a/30/5e5b2ccabeca73ae2b03fc82bca3eabb7466cf43737f05ac08d591665d47/pandas-3.0.6-cp314-cp314t-win_arm64.whl", hash = "sha256:a4dbd4dc65cbe645b92b8785d0f96dd7311010dc6606cf620e51b07b8788a12a", size = 9387796, upload-time = "2026-09-17T23:22:26.64Z" },
 ]
 
 [[package]]
@@ -936,27 +936,27 @@ wheels = [
 
 [[package]]
 name = "ruff"
-version = "0.16.7"
+version = "0.16.8"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/82/bb/5a449b9162e49b139d72f61672bd3ac1d790221f796d3304e2241fff4c58/ruff-0.16.7.tar.gz", hash = "sha256:5f71d004ac1263b22fa39462ac5ae618a4b77d58981af2cc79bf79a29c12b1a6", size = 4924184, upload-time = "2026-09-10T18:04:06.336Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/ba/78/449cb84790bd5cc3823b2652ee405a4558856e5c4195aee3a16bf7b3eb5d/ruff-0.16.8.tar.gz", hash = "sha256:9247bf92b5f04d825c8639a4fe423ec2e4222acd9222e58412b0dab7e442798b", size = 4938814, upload-time = "2026-09-16T15:54:46.688Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e3/b2/c80aeeb7f9e469c0d63a85d2f1ab6e1ebfbe10ea7a8d2438b7e09e3ff09e/ruff-0.16.7-py3-none-linux_armv6l.whl", hash = "sha256:727307773e7c7f9181d3ed3a2484186e56c1fa1874255911c74585eb2c7c19f9", size = 10048917, upload-time = "2026-09-10T18:03:30.28Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/25/6071aabc530e9be7e2c195e8fe3f7aea2735405b6cf447212832d7811831/ruff-0.16.8-py3-none-linux_armv6l.whl", hash = "sha256:6ffbd6d87383c1edf5f6fa890f10200950240d7c1a16052a19a09d3a2307dd38", size = 10048966, upload-time = "2026-09-16T15:53:57.605Z" },
-    { url = "https://files.pythonhosted.org/packages/7b/96/20bb7bcae008004df52afcb7ac83432d4a467f2c17b672fe46d26be231c5/ruff-0.16.7-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:9d61c258deabf58f34c67bd4bb4d939c7f2e6b5f0e59c1cdd1cf771b11cde929", size = 10242929, upload-time = "2026-09-10T18:03:32.706Z" },
+    { url = "https://files.pythonhosted.org/packages/54/98/07f90ecbc74dd5fb5764f11f2bc774d6a7cffef92d2ff5f5b4e9e23c754e/ruff-0.16.8-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:42ed6b878ed61e3acca92f2730a17acff39286944ea82398544696366a6f925e", size = 10165498, upload-time = "2026-09-16T15:54:01.14Z" },
-    { url = "https://files.pythonhosted.org/packages/90/b2/f184b0d5abec02db69cfd7e49b688ae0237554528ca777136c613bf36bee/ruff-0.16.7-py3-none-macosx_11_0_arm64.whl", hash = "sha256:7ab81118df8945e0193d0240712aa4496573595b75185c3636ed825592a0f728", size = 9847245, upload-time = "2026-09-10T18:03:34.509Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/1f/e6a712e3b47cad4a40600134105ed193cb773f618a42eb7ba323cb812cc0/ruff-0.16.8-py3-none-macosx_11_0_arm64.whl", hash = "sha256:7ea781c7f2afba8c6a505ea0fb3f994020249e0c450635f5381286fea6b46170", size = 9830004, upload-time = "2026-09-16T15:54:03.998Z" },
-    { url = "https://files.pythonhosted.org/packages/eb/2d/db1633a641866ed801e34cc6b60ef236c5e16f9b2124ab1d49cc24a5fe4f/ruff-0.16.7-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4c196c968874fc8019da8e7163de7a1a370f111e2309b4b7dfea0fce950198d0", size = 9961780, upload-time = "2026-09-10T18:03:36.618Z" },
+    { url = "https://files.pythonhosted.org/packages/23/f2/311a08776d75d81c7676e20b6b020ae63cbe881fcdc7a8dd64e6e18bdd93/ruff-0.16.8-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8efeae3bbe414a5efefda11a792dfb51ef90ac48d50c4830de2f644caf3e8659", size = 9986558, upload-time = "2026-09-16T15:54:06.804Z" },
-    { url = "https://files.pythonhosted.org/packages/4d/98/edea21e1a3e38dbbc3bf6bb068b863b3b06184cf8533a4c7dbbe208a89d5/ruff-0.16.7-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:ac8c3bd0a7e10ad31e6ce51e7a99f3cb772e69aecdd6b9ea7e99b362f62a62c0", size = 9866337, upload-time = "2026-09-10T18:03:38.805Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/ed/37b6cb3d3ba8c73e68ae3eb1d502383beb5aa05a582bb7bb3a922f929f54/ruff-0.16.8-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:3a79b795469fef7fc6e908b218eed2eb17332afd85031db6480dc864560e69b2", size = 9877332, upload-time = "2026-09-16T15:54:09.552Z" },
-    { url = "https://files.pythonhosted.org/packages/0b/11/a15e60d4c87b214646f116ca9d204475bf993ee1047459bc9a360fd4d6d1/ruff-0.16.7-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:398d3988edde000b5c75dc1b3f584708da9bc990de069c18909142580fec1af9", size = 10562512, upload-time = "2026-09-10T18:03:40.71Z" },
+    { url = "https://files.pythonhosted.org/packages/22/cc/40873a8f36ad084cc540d55fcca7077264d5b13b24659e9180c176fb2b08/ruff-0.16.8-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3fdc5563cdc50555e6fba39322850860e9267c1b3d12c26a74729d8604c3c812", size = 10507125, upload-time = "2026-09-16T15:54:12.152Z" },
-    { url = "https://files.pythonhosted.org/packages/29/42/eaff4c9b6d0c7cdf56df313a17e89ae854f5bbc0b0c8f9cce19be0ab7a8f/ruff-0.16.7-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ce05b62b770a8217c4646a9c4139fca00efe8fe5d71f87df2b243ff20d4584d1", size = 11302938, upload-time = "2026-09-10T18:03:42.607Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/e4/fc91a642b78ccbab6b9477720f3644ae7a10a9bcce69a934679cd64f62bc/ruff-0.16.8-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:34508983c70665578dab88f5223d8e6228307e1135398ca8bfc8b7e9501e282b", size = 11336694, upload-time = "2026-09-16T15:54:15.489Z" },
-    { url = "https://files.pythonhosted.org/packages/5d/43/c75aa59a4ec181fe2ec06cab30e198c1c6d107229a9f008ae3a7c16cabd8/ruff-0.16.7-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:af1b576fddb9d9ef2ececfb5fadcd6a624b25070ed85e3cfcfe449fc3ff6a7b9", size = 10840857, upload-time = "2026-09-10T18:03:44.604Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/3d/bbd2a9a600a4e73dc3e7548a249c8d1671273464b55822c6fae50f602dff/ruff-0.16.8-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:644bb578569e0ffc575741232bd385dacdd6fbe123f1a729e7a225f54aa3957f", size = 10774448, upload-time = "2026-09-16T15:54:18.16Z" },
-    { url = "https://files.pythonhosted.org/packages/21/33/81f3da371942ea031105ba679d8d6e28ec1660ccd690a45f42d381161356/ruff-0.16.7-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9ce7f8f22df67c93ed96c717f9128eadb797144ac2bad475cf536f31d6100c55", size = 10370001, upload-time = "2026-09-10T18:03:46.706Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/41/d83af9879a7b6e8bf5fe16b1da0b134049d2f5d3afac12defb0897cb84bd/ruff-0.16.8-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:15e7d226246961db9235098333caa13063906d3851136b84c2900b82f5daa1df", size = 10323796, upload-time = "2026-09-16T15:54:20.743Z" },
-    { url = "https://files.pythonhosted.org/packages/fa/0b/6345fb4dbf6dd0ed1cfe5d18391dc9c3f59cc81622a7b0a65b84b3e730ba/ruff-0.16.7-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:06d0e93d04f392996435ebd600c153f65b47d73fbec2415aa99c5ee5756b3a5f", size = 10548735, upload-time = "2026-09-10T18:03:48.658Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/2c/cefd07bfe914b84943ea769ade8d607bd22750b965d3228eefd7cebd15d0/ruff-0.16.8-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:a2bf6bc3e9ebdd4449abc6f06cf64b98051a2c61cf94d2fe9596518c881f1a1e", size = 10514115, upload-time = "2026-09-16T15:54:23.497Z" },
-    { url = "https://files.pythonhosted.org/packages/3f/4d/c5576adf511f92a328e5569dda190ecdd430da51f1a649f3a4a2fd73e21e/ruff-0.16.7-py3-none-musllinux_1_2_aarch64.whl", hash = "sha256:142151a5e7b93c1b11111337142f89dd2fbfee92161225c99a97222f22e32656", size = 10108496, upload-time = "2026-09-10T18:03:50.563Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/9d/76a2e26c79a23be6e6e3664c57bec9e9fc8de155cfb9e4b67ea91b64f9d7/ruff-0.16.8-py3-none-musllinux_1_2_aarch64.whl", hash = "sha256:6ca111ba0849539165e9e59d2b442542f3c1e8060ebbdea82494f1ffbccb1e1f", size = 10072582, upload-time = "2026-09-16T15:54:26.185Z" },
-    { url = "https://files.pythonhosted.org/packages/ff/8c/667d83c16199a17a56adc6b0bd4c3beb5b767a2babcd16a56f76f9be7fd6/ruff-0.16.7-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:e6651f97a342d8b35d54d8991544ca22169b86dc54111cb604666940c431b750", size = 9860136, upload-time = "2026-09-10T18:03:52.621Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/d4/f42edddb39668af1a559ceafa3823aedd65633a48dc9768e775485faa2c1/ruff-0.16.8-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:359a1e5b495448ee1e91018064382ebc86f90e8aac2fed222c7d0e4e8df85fd2", size = 9879644, upload-time = "2026-09-16T15:54:29.278Z" },
-    { url = "https://files.pythonhosted.org/packages/99/75/78d401106731999a1dd20cc5a6961e37e1eb9397a3b589f73f3a5ce146a3/ruff-0.16.7-py3-none-musllinux_1_2_i686.whl", hash = "sha256:ef140c6eb935fa9a84c9c607dfb2cb1b85843c192e79265b0c54f35f557ea8e5", size = 10286290, upload-time = "2026-09-10T18:03:55.207Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/d4/913e3195d95e0378786c6656945c865f534a3560e29139da4882aff630d1/ruff-0.16.8-py3-none-musllinux_1_2_i686.whl", hash = "sha256:59e8f5681349474110b24d62e93cfda6593f5fa3473446ca3705200cac1a08b9", size = 10231569, upload-time = "2026-09-16T15:54:32.036Z" },
-    { url = "https://files.pythonhosted.org/packages/68/49/56f9c3a8b755df93a0ad318b2147bf4ef5dae9a7e5ec61c460109c67957f/ruff-0.16.7-py3-none-musllinux_1_2_x86_64.whl", hash = "sha256:53e39506a730fadeee0d998ed5946f30671f0db240c6c7c73bdabbe33604bb6f", size = 10745048, upload-time = "2026-09-10T18:03:57.299Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/c4/8aa6ea0bdcedbd1bf87397e2fc4ed8406448ea5842f8660bc6e5f163039d/ruff-0.16.8-py3-none-musllinux_1_2_x86_64.whl", hash = "sha256:efa3e7a16d1baaa79957888dfdf8be9ef2e44db81cb032af06d76632ab59e773", size = 10663666, upload-time = "2026-09-16T15:54:34.838Z" },
-    { url = "https://files.pythonhosted.org/packages/5f/ea/7f9b938a63ece4bec677ad7f9f7fa02df3383db1949ed93a382441c09a87/ruff-0.16.7-py3-none-win32.whl", hash = "sha256:2ea3470fcebcbc5df2fb0c6f3b90333fa9084c534e0111c038fa4a6ab9f1c4b7", size = 10059082, upload-time = "2026-09-10T18:03:59.632Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/02/7f10ef4700bc223c30a3fdd10631a29830c45524b810a3c7ed947af64591/ruff-0.16.8-py3-none-win32.whl", hash = "sha256:55793ba85c69921e89be061426d91a78652d6e50317c962240922747a4eb713f", size = 10093472, upload-time = "2026-09-16T15:54:37.47Z" },
-    { url = "https://files.pythonhosted.org/packages/39/11/480a6973a927aa653e1cead6a6416008640e03a99d05b34c0434b8c6c366/ruff-0.16.7-py3-none-win_amd64.whl", hash = "sha256:7ac26aca826e9e21d0f1cb25b54ac660760a9fdd094d3e4df9848232be98cfc6", size = 10593368, upload-time = "2026-09-10T18:04:01.999Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/5d/a509c07d714b6da88f2c518b4637cf6f1d46b074be8f0f1e5fb9ff5126fe/ruff-0.16.8-py3-none-win_amd64.whl", hash = "sha256:a6b85621fd3c81e31fc5f5add09c9c078b430db3595ca632efafdec9e64ebfaa", size = 10586899, upload-time = "2026-09-16T15:54:40.488Z" },
-    { 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" },
+    { url = "https://files.pythonhosted.org/packages/fe/a0/50787329e4f20bf9dc9f6230015d46ec69c51a97ace5bc202dae4755365d/ruff-0.16.8-py3-none-win_arm64.whl", hash = "sha256:d075e820af612102ce217f07cc93e69f9490b10ec13ea85fa87bd03d996cef8a", size = 10386316, upload-time = "2026-09-16T15:54:43.332Z" },
 ]
 
 [[package]]

apps/frontend/Dockerfile

@@ -3,7 +3,7 @@
 # ==================
 
 # Image pour frontend
-FROM node:24-alpine3.22 AS builder
+FROM node:26-alpine3.22 AS builder
 
 WORKDIR /app
 
@@ -23,7 +23,7 @@ RUN npm run build
 # ==================
 
 
-FROM dhi.io/nginx:1.28.0-alpine3.21-dev AS runner
+FROM nginx:1.31-alpine AS runner
 
 # Copie de la configuration de nginx
 COPY --chown=root:root --chmod=755 nginx.conf /etc/nginx/nginx.conf

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/package-lock.json

@@ -8,12 +8,12 @@
       "name": "frontend",
       "version": "0.0.0",
       "dependencies": {
-        "@angular/common": "^22.1.0",
+        "@angular/common": "^22.1.7",
-        "@angular/compiler": "^22.1.0",
+        "@angular/compiler": "^22.1.7",
-        "@angular/core": "^22.1.0",
+        "@angular/core": "^22.1.7",
-        "@angular/forms": "^22.1.0",
+        "@angular/forms": "^22.1.7",
-        "@angular/platform-browser": "^22.1.0",
+        "@angular/platform-browser": "^22.1.7",
-        "@angular/router": "^22.1.0",
+        "@angular/router": "^22.1.7",
         "chart.js": "^4.5.1",
         "rxjs": "~7.8.0",
         "tslib": "^2.3.0"
@@ -21,21 +21,14 @@
       "devDependencies": {
         "@angular/build": "^22.1.8",
         "@angular/cli": "^22.1.8",
-        "@angular/compiler-cli": "^22.1.0",
+        "@angular/compiler-cli": "^22.1.7",
         "@vitest/coverage-v8": "^4.1.11",
-        "jsdom": "^28.0.0",
+        "jsdom": "^30.1.0",
-        "prettier": "^3.8.1",
+        "prettier": "^3.9.8",
         "typescript": "~6.0.2",
-        "vitest": "^4.0.8"
+        "vitest": "^4.1.11"
       }
     },
-    "node_modules/@acemir/cssom": {
-      "version": "0.9.31",
-      "resolved": "https://registry.npmjs.org/@acemir/cssom/-/cssom-0.9.31.tgz",
-      "integrity": "sha512-ZnR3GSaH+/vJ0YlHau21FjfLYjMpYVIzTD8M8vIEQvIGxeOXyXdzCI140rrCY862p/C/BbzWsjc1dgnM9mkoTA==",
-      "dev": true,
-      "license": "MIT"
-    },
     "node_modules/@ampproject/remapping": {
       "version": "2.3.0",
       "resolved": "https://registry.npmjs.org/@ampproject/remapping/-/remapping-2.3.0.tgz",
@@ -255,9 +248,9 @@
       }
     },
     "node_modules/@angular/common": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/common/-/common-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/common/-/common-22.1.7.tgz",
-      "integrity": "sha512-giuH+jJvo6YbBxbKofJCXvq6k8g1Z/xCAvh4piNFSSS6/toXTLCeZ+snr6Stw5b2wRbArM5Q5nDeuto3NqVPnQ==",
+      "integrity": "sha512-SLW4AdH2Y8G9SM/iFRNpCQtkmvq55RcIQWsADT9ypZ7FxxcnJRsr4Q/WnbUBbuCslsCkrgzU1i9WDiW0m7gcnw==",
       "license": "MIT",
       "dependencies": {
         "tslib": "^2.3.0"
@@ -266,14 +259,14 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/core": "22.1.6",
+        "@angular/core": "22.1.7",
         "rxjs": "^6.5.3 || ^7.4.0"
       }
     },
     "node_modules/@angular/compiler": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/compiler/-/compiler-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/compiler/-/compiler-22.1.7.tgz",
-      "integrity": "sha512-JjOUm/qD338+fGfZvxSNn/vTUiVqNwOiPzacInVUq1eVp7Jev+cnvEwXA2cFJkYZoy3Imz6wHoMvTUZNN8cbKQ==",
+      "integrity": "sha512-qhoHAByh805nT4Xy4Wd2kE5FvkTc7wcg8Qps+EH6+z4KRfwGiL5cwdgooGWjTAldFCeqTgJ0BO/0dB6GHZjdkQ==",
       "license": "MIT",
       "dependencies": {
         "tslib": "^2.3.0"
@@ -283,9 +276,9 @@
       }
     },
     "node_modules/@angular/compiler-cli": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/compiler-cli/-/compiler-cli-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/compiler-cli/-/compiler-cli-22.1.7.tgz",
-      "integrity": "sha512-C1fQuaSLnibhfbb7Im/vurdBEcfQk+/GqPkY4+dgEKd4EPleO0xWlD+k5DwyNFX8a9w7HebWfo0Zl66HuaEwOQ==",
+      "integrity": "sha512-VGg8Kyt0YymQfnUUFk79vFaM+x5xmIMHy+BZh3P2Lh/uxqCXTWhJJH8kOHKtIGGH5/wKhBEmSy2CezdcIuM9xg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -306,7 +299,7 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/compiler": "22.1.6",
+        "@angular/compiler": "22.1.7",
         "typescript": ">=6.0 <6.1"
       },
       "peerDependenciesMeta": {
@@ -316,9 +309,9 @@
       }
     },
     "node_modules/@angular/core": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/core/-/core-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/core/-/core-22.1.7.tgz",
-      "integrity": "sha512-3Ln9YYOhsaU2vPufnpcu6C4dlmX4e/nJTlggVcKMT7bGZH5KlEtw3h0uh9YfANv8YhQCEk1AcuRI7KpBnh5ing==",
+      "integrity": "sha512-T9bt1PIgoN7n+TxGHTUgu0EsomlGMQ84YMXbMB+N6HhD8C1qZxn0uCbnAtCUjcecg1am8ioKxNTO0eII9ouioQ==",
       "license": "MIT",
       "dependencies": {
         "tslib": "^2.3.0"
@@ -327,7 +320,7 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/compiler": "22.1.6",
+        "@angular/compiler": "22.1.7",
         "rxjs": "^6.5.3 || ^7.4.0",
         "zone.js": "~0.15.0 || ~0.16.0"
       },
@@ -341,9 +334,9 @@
       }
     },
     "node_modules/@angular/forms": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/forms/-/forms-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/forms/-/forms-22.1.7.tgz",
-      "integrity": "sha512-rfV4G4UB4l69yXSRvhaHPzTMIruvBlRO+ak9NtTcYMnsoj8O5ZCyPv0ledGIrLTBwGSI3X1j4nr9aoV+Rj/n+Q==",
+      "integrity": "sha512-oc0DT39C3ZboJpDx8xKUCltgn56LHi0kbv+ThNK/dCSZLgjF2nd+muMMRYF+amdljQp2q7+hm2ORmhON3CNFog==",
       "license": "MIT",
       "dependencies": {
         "@standard-schema/spec": "^1.0.0",
@@ -354,16 +347,16 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/common": "22.1.6",
+        "@angular/common": "22.1.7",
-        "@angular/core": "22.1.6",
+        "@angular/core": "22.1.7",
-        "@angular/platform-browser": "22.1.6",
+        "@angular/platform-browser": "22.1.7",
         "rxjs": "^6.5.3 || ^7.4.0"
       }
     },
     "node_modules/@angular/platform-browser": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/platform-browser/-/platform-browser-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/platform-browser/-/platform-browser-22.1.7.tgz",
-      "integrity": "sha512-jrRi6zpdz+jOle5l0OW7QL0a8xPgdnxWT5FrJabbiNKfYzQfqKcaAu02l8uJoJxtGb8fLQ8pbjkPpZEvKO2z+w==",
+      "integrity": "sha512-4dT3qGw3ZHVQUcHKALzMx9MTo+pxAMzHIPxc8HUESklqXfq6GRaIIfFMvE0RsquI9iqu6admD6lOAEzLWS8DxA==",
       "license": "MIT",
       "dependencies": {
         "tslib": "^2.3.0"
@@ -372,9 +365,9 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/animations": "22.1.6",
+        "@angular/animations": "22.1.7",
-        "@angular/common": "22.1.6",
+        "@angular/common": "22.1.7",
-        "@angular/core": "22.1.6"
+        "@angular/core": "22.1.7"
       },
       "peerDependenciesMeta": {
         "@angular/animations": {
@@ -383,9 +376,9 @@
       }
     },
     "node_modules/@angular/router": {
-      "version": "22.1.6",
+      "version": "22.1.7",
-      "resolved": "https://registry.npmjs.org/@angular/router/-/router-22.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@angular/router/-/router-22.1.7.tgz",
-      "integrity": "sha512-ex0vrkcVyJJdn7NzZXun+5ctaPPLTZJ/gE7A3g1dun7BM3g4Z6zT738mODFeJgtPtJNf4bTs2gpgvyegvzWH7w==",
+      "integrity": "sha512-CxYa2Ym5L0WmqToHU2I0Z4VEpocdryLlvx5svLfo9M3uOaP3/9icB/gjS/vXPbbQxV8/GZ59E/7/LF14anIoRg==",
       "license": "MIT",
       "dependencies": {
         "tslib": "^2.3.0"
@@ -394,60 +387,45 @@
         "node": "^22.22.3 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "@angular/common": "22.1.6",
+        "@angular/common": "22.1.7",
-        "@angular/core": "22.1.6",
+        "@angular/core": "22.1.7",
-        "@angular/platform-browser": "22.1.6",
+        "@angular/platform-browser": "22.1.7",
         "rxjs": "^6.5.3 || ^7.4.0"
       }
     },
     "node_modules/@asamuzakjp/css-color": {
-      "version": "5.1.11",
+      "version": "7.0.0",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/css-color/-/css-color-5.1.11.tgz",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/css-color/-/css-color-7.0.0.tgz",
-      "integrity": "sha512-KVw6qIiCTUQhByfTd78h2yD1/00waTmm9uy/R7Ck/ctUyAPj+AEDLkQIdJW0T8+qGgj3j5bpNKK7Q3G+LedJWg==",
+      "integrity": "sha512-IxddtGnnoidVSO6O/aiTueptbTThHvi2qGHu9/AWuD1J7i86Ox+jl+Y5SXcDH1RixzQlP2cjH/PjnlImZiAurQ==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@asamuzakjp/generational-cache": "^1.0.1",
+        "@csstools/css-calc": "^3.3.0",
-        "@csstools/css-calc": "^3.2.0",
+        "@csstools/css-color-parser": "^4.2.0",
-        "@csstools/css-color-parser": "^4.1.0",
         "@csstools/css-parser-algorithms": "^4.0.0",
-        "@csstools/css-tokenizer": "^4.0.0"
+        "@csstools/css-tokenizer": "^4.0.0",
+        "lru-cache": "^11.5.2"
       },
       "engines": {
-        "node": "^20.19.0 || ^22.12.0 || >=24.0.0"
+        "node": "^22.22.2 || ^24.15.0 || >=26.0.0"
       }
     },
     "node_modules/@asamuzakjp/dom-selector": {
-      "version": "6.8.1",
+      "version": "9.1.4",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-6.8.1.tgz",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-9.1.4.tgz",
-      "integrity": "sha512-MvRz1nCqW0fsy8Qz4dnLIvhOlMzqDVBabZx6lH+YywFDdjXhMY37SmpV1XFX3JzG5GWHn63j6HX6QPr3lZXHvQ==",
+      "integrity": "sha512-ynDE7RDZM1z+YuSU+iGhRp8WcSXHqK9+G32ZEzsL35TKSoy72fYR7VfrJCQkWe3ciiIWGSTAIBc9XB5jVpqyTw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@asamuzakjp/nwsapi": "^2.3.9",
+        "bidi-js": "^1.1.0",
-        "bidi-js": "^1.0.3",
+        "css-tree": "^3.2.1",
-        "css-tree": "^3.1.0",
         "is-potential-custom-element-name": "^1.0.1",
-        "lru-cache": "^11.2.6"
+        "lru-cache": "^11.5.2"
-      }
       },
-    "node_modules/@asamuzakjp/generational-cache": {
-      "version": "1.0.1",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/generational-cache/-/generational-cache-1.0.1.tgz",
-      "integrity": "sha512-wajfB8KqzMCN2KGNFdLkReeHncd0AslUSrvHVvvYWuU8ghncRJoA50kT3zP9MVL0+9g4/67H+cdvBskj9THPzg==",
-      "dev": true,
-      "license": "MIT",
       "engines": {
-        "node": "^20.19.0 || ^22.12.0 || >=24.0.0"
+        "node": "^22.22.2 || ^24.15.0 || >=26.0.0"
       }
     },
-    "node_modules/@asamuzakjp/nwsapi": {
-      "version": "2.3.9",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/nwsapi/-/nwsapi-2.3.9.tgz",
-      "integrity": "sha512-n8GuYSrI9bF7FFZ/SjhwevlHc8xaVlb/7HmHelnc/PZXBD2ZR49NnN9sMMuDdEGPeeRQ5d0hqlSlEpgCX3Wl0Q==",
-      "dev": true,
-      "license": "MIT"
-    },
     "node_modules/@babel/code-frame": {
       "version": "8.0.0",
       "resolved": "https://registry.npmjs.org/@babel/code-frame/-/code-frame-8.0.0.tgz",
@@ -854,9 +832,9 @@
       }
     },
     "node_modules/@csstools/css-syntax-patches-for-csstree": {
-      "version": "1.1.13",
+      "version": "1.1.14",
-      "resolved": "https://registry.npmjs.org/@csstools/css-syntax-patches-for-csstree/-/css-syntax-patches-for-csstree-1.1.13.tgz",
+      "resolved": "https://registry.npmjs.org/@csstools/css-syntax-patches-for-csstree/-/css-syntax-patches-for-csstree-1.1.14.tgz",
-      "integrity": "sha512-i9ZylF5QNhmNfPA9l0vHAWK4kPrbIp6g9lKgaiIFsIBz2F/WNB7OLrzlNNcCOm+h42bkaSD2v1PG+IBPHhc3ZA==",
+      "integrity": "sha512-HpbVXyrofRXpHpgkNIjU/3EWR4WJvOkO3emNK/L6X/mTJU7bGUI3AkkpoTNXznQLp0KRjLHELTGeKI5dIkI9JQ==",
       "dev": true,
       "funding": [
         {
@@ -879,9 +857,9 @@
       }
     },
     "node_modules/@csstools/css-tokenizer": {
-      "version": "4.0.0",
+      "version": "4.0.1",
-      "resolved": "https://registry.npmjs.org/@csstools/css-tokenizer/-/css-tokenizer-4.0.0.tgz",
+      "resolved": "https://registry.npmjs.org/@csstools/css-tokenizer/-/css-tokenizer-4.0.1.tgz",
-      "integrity": "sha512-QxULHAm7cNu72w97JUNCBFODFaXpbDg+dP8b/oWFAZ2MTRppA3U00Y2L1HqaS4J6yBqxwa/Y3nMBaxVKbB/NsA==",
+      "integrity": "sha512-bPlN9S9O1A0euCpEWE4qnvB5YDuyYVsUTrxSgmAM1Is0j4tICHoVyOVAXfWMP/kS9ZrjvyIXWV2PmomiAXXqOw==",
       "dev": true,
       "funding": [
         {
@@ -4490,22 +4468,6 @@
         "url": "https://github.com/sponsors/fb55"
       }
     },
-    "node_modules/cssstyle": {
-      "version": "6.2.0",
-      "resolved": "https://registry.npmjs.org/cssstyle/-/cssstyle-6.2.0.tgz",
-      "integrity": "sha512-Fm5NvhYathRnXNVndkUsCCuR63DCLVVwGOOwQw782coXFi5HhkXdu289l59HlXZBawsyNccXfWRYvLzcDCdDig==",
-      "dev": true,
-      "license": "MIT",
-      "dependencies": {
-        "@asamuzakjp/css-color": "^5.0.1",
-        "@csstools/css-syntax-patches-for-csstree": "^1.0.28",
-        "css-tree": "^3.1.0",
-        "lru-cache": "^11.2.6"
-      },
-      "engines": {
-        "node": ">=20"
-      }
-    },
     "node_modules/data-urls": {
       "version": "7.0.0",
       "resolved": "https://registry.npmjs.org/data-urls/-/data-urls-7.0.0.tgz",
@@ -5283,30 +5245,6 @@
         "url": "https://opencollective.com/express"
       }
     },
-    "node_modules/http-proxy-agent": {
-      "version": "7.0.2",
-      "resolved": "https://registry.npmjs.org/http-proxy-agent/-/http-proxy-agent-7.0.2.tgz",
-      "integrity": "sha512-T1gkAiYYDWYx3V5Bmyu7HcfcvL7mUrTWiM6yOfa3PIphViJ/gFPbvidQ+veqSOHci/PxBcDabeUNCzpOODJZig==",
-      "dev": true,
-      "license": "MIT",
-      "dependencies": {
-        "agent-base": "^7.1.0",
-        "debug": "^4.3.4"
-      },
-      "engines": {
-        "node": ">= 14"
-      }
-    },
-    "node_modules/http-proxy-agent/node_modules/agent-base": {
-      "version": "7.1.4",
-      "resolved": "https://registry.npmjs.org/agent-base/-/agent-base-7.1.4.tgz",
-      "integrity": "sha512-MnA+YT8fwfJPgBx3m60MNqakm30XOkyIoH1y6huTQvC0PwZG7ki8NacLBcrPbNoo8vEZy7Jpuk7+jMO+CUovTQ==",
-      "dev": true,
-      "license": "MIT",
-      "engines": {
-        "node": ">= 14"
-      }
-    },
     "node_modules/https-proxy-agent": {
       "version": "9.1.0",
       "resolved": "https://registry.npmjs.org/https-proxy-agent/-/https-proxy-agent-9.1.0.tgz",
@@ -5529,39 +5467,38 @@
       "license": "MIT"
     },
     "node_modules/jsdom": {
-      "version": "28.1.0",
+      "version": "30.1.0",
-      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-28.1.0.tgz",
+      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.0.tgz",
-      "integrity": "sha512-0+MoQNYyr2rBHqO1xilltfDjV9G7ymYGlAUazgcDLQaUf8JDHbuGwsxN6U9qWaElZ4w1B2r7yEGIL3GdeW3Rug==",
+      "integrity": "sha512-h/Q+H+KmlGf49ITw3TrykSVbg04S75S2O7ZtJc9gMGOWBrQRZmx5AJIKLtJ01e6Kszl+/tpFthCVqxNifxhoqg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@acemir/cssom": "^0.9.31",
+        "@asamuzakjp/css-color": "^7.0.0",
-        "@asamuzakjp/dom-selector": "^6.8.1",
+        "@asamuzakjp/dom-selector": "^9.1.2",
         "@bramus/specificity": "^2.4.2",
-        "@exodus/bytes": "^1.11.0",
+        "@csstools/css-syntax-patches-for-csstree": "^1.1.13",
-        "cssstyle": "^6.0.1",
+        "@exodus/bytes": "^1.15.1",
+        "css-tree": "^3.2.1",
         "data-urls": "^7.0.0",
         "decimal.js": "^10.6.0",
         "html-encoding-sniffer": "^6.0.0",
-        "http-proxy-agent": "^7.0.2",
-        "https-proxy-agent": "^7.0.6",
         "is-potential-custom-element-name": "^1.0.1",
-        "parse5": "^8.0.0",
+        "lru-cache": "^11.5.2",
+        "parse5": "^8.0.1",
         "saxes": "^6.0.0",
-        "symbol-tree": "^3.2.4",
+        "tough-cookie": "^6.0.2",
-        "tough-cookie": "^6.0.0",
+        "undici": "^8.10.2",
-        "undici": "^7.21.0",
         "w3c-xmlserializer": "^5.0.0",
         "webidl-conversions": "^8.0.1",
         "whatwg-mimetype": "^5.0.0",
-        "whatwg-url": "^16.0.0",
+        "whatwg-url": "^17.1.1",
         "xml-name-validator": "^5.0.0"
       },
       "engines": {
-        "node": "^20.19.0 || ^22.12.0 || >=24.0.0"
+        "node": "^22.22.2 || ^24.15.0 || >=26.0.0"
       },
       "peerDependencies": {
-        "canvas": "^3.0.0"
+        "canvas": "^3.2.3"
       },
       "peerDependenciesMeta": {
         "canvas": {
@@ -5569,28 +5506,19 @@
         }
       }
     },
-    "node_modules/jsdom/node_modules/agent-base": {
+    "node_modules/jsdom/node_modules/whatwg-url": {
-      "version": "7.1.4",
+      "version": "17.1.1",
-      "resolved": "https://registry.npmjs.org/agent-base/-/agent-base-7.1.4.tgz",
+      "resolved": "https://registry.npmjs.org/whatwg-url/-/whatwg-url-17.1.1.tgz",
-      "integrity": "sha512-MnA+YT8fwfJPgBx3m60MNqakm30XOkyIoH1y6huTQvC0PwZG7ki8NacLBcrPbNoo8vEZy7Jpuk7+jMO+CUovTQ==",
+      "integrity": "sha512-ohjk1mdUebJVadRt3bAhQhx8lSnISq+GDttK79LFl8EHQkAPvzwctoasC4hs8tBt6kLAncBWWyq1N52qEfKvDw==",
-      "dev": true,
-      "license": "MIT",
-      "engines": {
-        "node": ">= 14"
-      }
-    },
-    "node_modules/jsdom/node_modules/https-proxy-agent": {
-      "version": "7.0.6",
-      "resolved": "https://registry.npmjs.org/https-proxy-agent/-/https-proxy-agent-7.0.6.tgz",
-      "integrity": "sha512-vK9P5/iUfdl95AI+JVyUuIcVtd4ofvtrOr3HNtM2yxC9bnMbEdp3x01OhQNnjb8IJYi38VlTE3mBXwcfvywuSw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "agent-base": "^7.1.2",
+        "@exodus/bytes": "^1.15.1",
-        "debug": "4"
+        "tr46": "^6.0.0",
+        "webidl-conversions": "^8.0.1"
       },
       "engines": {
-        "node": ">= 14"
+        "node": "^22.14.0 || >=24.0.0"
       }
     },
     "node_modules/jsesc": {
@@ -6708,9 +6636,9 @@
       }
     },
     "node_modules/prettier": {
-      "version": "3.9.6",
+      "version": "3.9.8",
-      "resolved": "https://registry.npmjs.org/prettier/-/prettier-3.9.6.tgz",
+      "resolved": "https://registry.npmjs.org/prettier/-/prettier-3.9.8.tgz",
-      "integrity": "sha512-OpN0zzVdiaiAhxpuuj5efpIS4sY9j7bY6uR5mnj5yPzGkdkjNKSJeUThPb60Jw29QuAZgA4o+/iB49kFiaBX6g==",
+      "integrity": "sha512-WRFq3Wn3WId7LLROfMLdH7xaFr2jR62wU8nLO6rQUOLOxNZUviyJQs1M0iIhLexSFy+L+w0ch66wtoO2jRjG0A==",
       "dev": true,
       "license": "MIT",
       "bin": {
@@ -7308,13 +7236,6 @@
         "node": ">=8"
       }
     },
-    "node_modules/symbol-tree": {
-      "version": "3.2.4",
-      "resolved": "https://registry.npmjs.org/symbol-tree/-/symbol-tree-3.2.4.tgz",
-      "integrity": "sha512-9QNk5KwDF+Bvz+PyObkmSYjI5ksVUYtjW7AU22r2NKcfLJcXp96hkDWU3+XndOsUb+AQ9QhfzfCT2O+CNWT5Tw==",
-      "dev": true,
-      "license": "MIT"
-    },
     "node_modules/tinybench": {
       "version": "2.9.0",
       "resolved": "https://registry.npmjs.org/tinybench/-/tinybench-2.9.0.tgz",
@@ -7469,13 +7390,13 @@
       }
     },
     "node_modules/undici": {
-      "version": "7.29.1",
+      "version": "8.10.2",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-7.29.1.tgz",
+      "resolved": "https://registry.npmjs.org/undici/-/undici-8.10.2.tgz",
-      "integrity": "sha512-RYONW2MeafgYlkVOKYKkA/Ag7BmXqgIWCa8t1m0JcxrQg9pI9lEqRhAOruOBCbAohOa/gkCF+iPi9hrgvTzu6Q==",
+      "integrity": "sha512-/y4/bH9YNU5hi9NIrpOuvGXFcxrj3CMrV+/AYpowAYTpHn8gX/XPFjNy766FPoYY0miQhdW977JFWKGNhBdwyQ==",
       "dev": true,
       "license": "MIT",
       "engines": {
-        "node": ">=20.18.1"
+        "node": ">=22.19.0"
       }
     },
     "node_modules/unpipe": {

apps/frontend/package.json

@@ -12,12 +12,12 @@
   "private": true,
   "packageManager": "npm@11.19.0",
   "dependencies": {
-    "@angular/common": "^22.1.0",
+    "@angular/common": "^22.1.7",
-    "@angular/compiler": "^22.1.0",
+    "@angular/compiler": "^22.1.7",
-    "@angular/core": "^22.1.0",
+    "@angular/core": "^22.1.7",
-    "@angular/forms": "^22.1.0",
+    "@angular/forms": "^22.1.7",
-    "@angular/platform-browser": "^22.1.0",
+    "@angular/platform-browser": "^22.1.7",
-    "@angular/router": "^22.1.0",
+    "@angular/router": "^22.1.7",
     "chart.js": "^4.5.1",
     "rxjs": "~7.8.0",
     "tslib": "^2.3.0"
@@ -25,11 +25,11 @@
   "devDependencies": {
     "@angular/build": "^22.1.8",
     "@angular/cli": "^22.1.8",
-    "@angular/compiler-cli": "^22.1.0",
+    "@angular/compiler-cli": "^22.1.7",
     "@vitest/coverage-v8": "^4.1.11",
-    "jsdom": "^28.0.0",
+    "jsdom": "^30.1.0",
-    "prettier": "^3.8.1",
+    "prettier": "^3.9.8",
     "typescript": "~6.0.2",
-    "vitest": "^4.0.8"
+    "vitest": "^4.1.11"
   }
 }

apps/frontend/src/app/app.routes.ts

@@ -1,21 +1,36 @@
 import { Routes } from '@angular/router';
-import {authGuard} from './core/guards/auth-guard';
+import { authGuard } from './core/guards/auth-guard';
 
 export const routes: Routes = [
   { path: '', redirectTo: 'dashboard', pathMatch: 'full' },
-  { path: 'login', loadComponent: () => import('./features/auth/login/login').then(m => m.Login) },
+  {
-  { path: 'change-password', loadComponent: () => import('./features/auth/change-password/change-password').then(m => m.ChangePassword) },
+    path: 'login',
-  { path: 'forgot-password', loadComponent: () => import('./features/auth/forgot-password/forgot-password').then(m => m.ForgotPassword) },
+    loadComponent: () => import('./features/auth/login/login').then((m) => m.Login),
-  { path: 'reset-password', loadComponent: () => import('./features/auth/reset-password/reset-password').then(m => m.ResetPassword) },
+  },
+  {
+    path: 'change-password',
+    loadComponent: () =>
+      import('./features/auth/change-password/change-password').then((m) => m.ChangePassword),
+  },
+  {
+    path: 'forgot-password',
+    loadComponent: () =>
+      import('./features/auth/forgot-password/forgot-password').then((m) => m.ForgotPassword),
+  },
+  {
+    path: 'reset-password',
+    loadComponent: () =>
+      import('./features/auth/reset-password/reset-password').then((m) => m.ResetPassword),
+  },
   {
     path: 'dashboard',
     canActivate: [authGuard],
-    loadComponent: () => import('./features/dashboard/dashboard').then(m => m.Dashboard),
+    loadComponent: () => import('./features/dashboard/dashboard').then((m) => m.Dashboard),
   },
   {
     path: 'sites',
     canActivate: [authGuard],
-    loadComponent: () => import('./features/sites/site-list/site-list').then(m => m.SiteList),
+    loadComponent: () => import('./features/sites/site-list/site-list').then((m) => m.SiteList),
   },
   {
     path: 'sites/:siteId',
@@ -23,6 +38,12 @@ export const routes: Routes = [
     loadComponent: () =>
       import('./features/sites/site-detail/site-detail').then((m) => m.SiteDetail),
   },
+  {
+    path: 'recommendations',
+    canActivate: [authGuard],
+    loadComponent: () =>
+      import('./features/recommendations/recommendations').then((m) => m.RecommendationsView),
+  },
   {
     path: 'monitoring/sensors',
     canActivate: [authGuard],

apps/frontend/src/app/core/interceptors/mock-api-interceptor.spec.ts

@@ -8,8 +8,10 @@ import { STATS_SUMMARY_FIXTURE } from '../mocks/stats-summary.fixture';
 describe('mockApiInterceptor', () => {
   let http: HttpClient;
   let httpMock: HttpTestingController;
+  let useMockFixturesInitial: boolean;
 
   beforeEach(() => {
+    useMockFixturesInitial = environment.useMockFixtures;
     TestBed.configureTestingModule({
       providers: [
         provideHttpClient(withInterceptors([mockApiInterceptor])),
@@ -21,7 +23,7 @@ describe('mockApiInterceptor', () => {
   });
 
   afterEach(() => {
-    environment.useMockFixtures = true;
+    environment.useMockFixtures = useMockFixturesInitial;
     httpMock.verify();
   });
 

apps/frontend/src/app/core/mocks/alerts.fixture.ts

@@ -2,53 +2,63 @@ import { Alert } from '../../shared/models/alert.model';
 
 export const ALERTS_FIXTURE: Alert[] = [
   {
-    alert_id: 'ALR-SITE002-1718458320',
+    alert_id: 5,
-    timestamp: '2026-09-15T11:12:00',
     site_id: 'SITE002',
+    timestamp: '2026-09-15T11:12:00Z',
+    type: 'threshold',
     severity: 'critical',
-    type: 'outage',
+    message: 'Puissance appelée 812.5 kW au-dessus de la capacité du site (720.0 kW)',
-    message: 'Risque de surcharge sur Usine Lyon Vénissieux',
     value: 812.5,
     threshold: 720.0,
+    metric: 'consumption_kw',
+    prediction_id: null,
   },
   {
-    alert_id: 'ALR-SITE003-1718458321',
+    alert_id: 4,
-    timestamp: '2026-09-15T11:05:00',
     site_id: 'SITE003',
+    timestamp: '2026-09-15T11:05:00Z',
+    type: 'outage',
     severity: 'critical',
-    type: 'sensor',
+    message: 'Aucune lecture depuis 5:00:00 (dernière lecture : 2026-09-15T06:05:00+00:00)',
-    message: 'Perte réseau totale sur Data Center Marseille',
+    value: null,
-    value: 0,
+    threshold: null,
-    threshold: 0,
+    metric: null,
+    prediction_id: null,
   },
   {
-    alert_id: 'ALR-SITE005-1718458322',
+    alert_id: 3,
-    timestamp: '2026-09-15T10:47:00',
     site_id: 'SITE005',
+    timestamp: '2026-09-15T10:47:00Z',
+    type: 'spike',
     severity: 'high',
-    type: 'threshold',
+    message: 'Variation brutale entre deux lectures consécutives (260.0 kW -> 410.0 kW)',
-    message: 'Usine Toulouse approche de son seuil de capacité',
     value: 410.0,
-    threshold: 480.0,
+    threshold: 260.0,
+    metric: 'consumption_kw',
+    prediction_id: null,
   },
   {
-    alert_id: 'ALR-SITE006-1718458323',
+    alert_id: 2,
-    timestamp: '2026-09-15T10:30:00',
     site_id: 'SITE006',
-    severity: 'medium',
+    timestamp: '2026-09-15T10:30:00Z',
     type: 'sensor',
-    message: 'Capteur de température défaillant sur Bureau Lille',
+    severity: 'medium',
-    value: 0,
+    message: 'Qualité de mesure degraded (capteur hors ligne, valeur nulle)',
-    threshold: 0,
+    value: null,
+    threshold: null,
+    metric: null,
+    prediction_id: null,
   },
   {
-    alert_id: 'ALR-SITE004-1718458324',
+    alert_id: 1,
-    timestamp: '2026-09-15T09:58:00',
     site_id: 'SITE004',
-    severity: 'low',
+    timestamp: '2026-09-15T09:58:00Z',
     type: 'anomaly',
-    message: 'Comportement de consommation inhabituel sur Bureau Bordeaux',
+    severity: 'low',
+    message: 'Écart de 13% entre la consommation mesurée (62.0 kWh) et la prévision (55.0 kWh)',
     value: 62.0,
     threshold: 55.0,
+    metric: 'consumption_kwh',
+    prediction_id: 42,
   },
 ];

apps/frontend/src/app/core/services/alerts.service.spec.ts

@@ -3,6 +3,20 @@ import { provideHttpClient } from '@angular/common/http';
 import { provideHttpClientTesting, HttpTestingController } from '@angular/common/http/testing';
 import { AlertsService } from './alerts.service';
 import { environment } from '../../../environments/environment';
+import { Alert } from '../../shared/models/alert.model';
+
+const ALERT_API: Alert = {
+  alert_id: 1,
+  site_id: 'site-1',
+  timestamp: '2026-09-16T00:00:00Z',
+  type: 'threshold',
+  severity: 'high',
+  message: 'Dépassement du seuil configuré',
+  value: 812.5,
+  threshold: 720.0,
+  metric: 'consumption_kw',
+  prediction_id: null,
+};
 
 describe('AlertsService', () => {
   let service: AlertsService;
@@ -18,26 +32,35 @@ describe('AlertsService', () => {
 
   afterEach(() => httpMock.verify());
 
-  it("appelle le bon endpoint et retourne un tableau d'alertes", () => {
+  it("appelle le bon endpoint sans paramètre et retourne un tableau d'alertes", () => {
-    let result: unknown;
+    let result: Alert[] = [];
     service.getAlerts().subscribe((r) => (result = r));
 
-    const req = httpMock.expectOne(`${environment.apiUrl}/alerts`);
+    const req = httpMock.expectOne(
-    expect(req.request.method).toBe('GET');
+      (r) => r.url === `${environment.apiUrl}/alerts` && r.method === 'GET',
+    );
+    expect(req.request.params.keys()).toEqual([]);
+    req.flush([ALERT_API]);
 
-    req.flush([
+    expect(result.length).toBe(1);
-      {
+    expect(result[0].alert_id).toBe(1);
-        alert_id: 'ALR-TEST-1',
+    expect(result[0].prediction_id).toBeNull();
-        timestamp: '2026-09-15T12:00:00',
+  });
-        site_id: 'SITE001',
-        severity: 'high',
-        type: 'threshold',
-        message: 'Test',
-        value: 100,
-        threshold: 90,
-      },
-    ]);
 
-    expect((result as unknown[]).length).toBe(1);
+  it('transmet les filtres site_id et severity en paramètres de requête', () => {
+    service.getAlerts({ site_id: 'SITE001', severity: 'high' }).subscribe();
+
+    const req = httpMock.expectOne((r) => r.url === `${environment.apiUrl}/alerts`);
+    expect(req.request.params.get('site_id')).toBe('SITE001');
+    expect(req.request.params.get('severity')).toBe('high');
+    req.flush([]);
+  });
+
+  it('ne pose pas de paramètre pour un filtre omis', () => {
+    service.getAlerts({ site_id: 'SITE001' }).subscribe();
+
+    const req = httpMock.expectOne((r) => r.url === `${environment.apiUrl}/alerts`);
+    expect(req.request.params.has('severity')).toBe(false);
+    req.flush([]);
   });
 });

apps/frontend/src/app/core/services/alerts.service.ts

@@ -1,13 +1,25 @@
 import { Service, inject } from '@angular/core';
-import { HttpClient } from '@angular/common/http';
+import { HttpClient, HttpParams } from '@angular/common/http';
 import { environment } from '../../../environments/environment';
-import { Alert } from '../../shared/models/alert.model';
+import { Alert, AlertSeverity } from '../../shared/models/alert.model';
+
+export interface AlertFilters {
+  site_id?: string;
+  severity?: AlertSeverity;
+}
 
 @Service()
 export class AlertsService {
   private http = inject(HttpClient);
 
-  getAlerts() {
+  getAlerts(filters: AlertFilters = {}) {
-    return this.http.get<Alert[]>(`${environment.apiUrl}/alerts`);
+    let params = new HttpParams();
+    if (filters.site_id) {
+      params = params.set('site_id', filters.site_id);
+    }
+    if (filters.severity) {
+      params = params.set('severity', filters.severity);
+    }
+    return this.http.get<Alert[]>(`${environment.apiUrl}/alerts`, { params });
   }
 }

apps/frontend/src/app/core/services/recommendations.service.spec.ts

@@ -0,0 +1,74 @@
+import { TestBed } from '@angular/core/testing';
+import { provideHttpClient } from '@angular/common/http';
+import { provideHttpClientTesting, HttpTestingController } from '@angular/common/http/testing';
+import { RecommendationsService } from './recommendations.service';
+import { environment } from '../../../environments/environment';
+import { Recommendation } from '../../shared/models/recommendation.model';
+
+const RECOMMANDATION_API: Recommendation = {
+  recommendation_id: 1,
+  alert_id: 1,
+  action: 'Vérifier la consommation',
+  explanation: 'Pic détecté',
+  rule_reference: 'spike-v1',
+  created_at: '2024-01-01T00:00:00Z',
+};
+
+describe('RecommendationsService', () => {
+  let service: RecommendationsService;
+  let httpMock: HttpTestingController;
+
+  beforeEach(() => {
+    TestBed.configureTestingModule({
+      providers: [provideHttpClient(), provideHttpClientTesting()],
+    });
+    service = TestBed.inject(RecommendationsService);
+    httpMock = TestBed.inject(HttpTestingController);
+  });
+
+  afterEach(() => httpMock.verify());
+
+  it('liste les recommandations depuis le bon endpoint', () => {
+    let result: Recommendation[] = [];
+    service.getRecommendations().subscribe((r) => (result = r));
+
+    const req = httpMock.expectOne(`${environment.apiUrl}/recommendations`);
+    expect(req.request.method).toBe('GET');
+    req.flush([RECOMMANDATION_API]);
+
+    expect(result.length).toBe(1);
+    expect(result[0].alert_id).toBe(1);
+  });
+
+  it('décrit une recommandation par son identifiant', () => {
+    service.getRecommendation(42).subscribe();
+
+    const req = httpMock.expectOne(`${environment.apiUrl}/recommendations/42`);
+    expect(req.request.method).toBe('GET');
+    req.flush({ ...RECOMMANDATION_API, recommendation_id: 42 });
+  });
+
+  it('déclenche la génération en POST avec le site en paramètre de requête', () => {
+    let result: unknown;
+    service.generate('SITE001').subscribe((r) => (result = r));
+
+    const req = httpMock.expectOne(
+      (r) => r.url === `${environment.apiUrl}/recommendations/generate` && r.method === 'POST',
+    );
+    expect(req.request.params.get('site_id')).toBe('SITE001');
+    expect(req.request.body).toBeNull();
+    req.flush({ alerts_examined: 2, recommendations_created: 3, already_present: 1 });
+
+    expect(result).toEqual({ alerts_examined: 2, recommendations_created: 3, already_present: 1 });
+  });
+
+  it('génère pour tout le parc quand aucun site n’est donné', () => {
+    service.generate().subscribe();
+
+    const req = httpMock.expectOne(
+      (r) => r.url === `${environment.apiUrl}/recommendations/generate` && r.method === 'POST',
+    );
+    expect(req.request.params.has('site_id')).toBe(false);
+    req.flush({ alerts_examined: 0, recommendations_created: 0, already_present: 0 });
+  });
+});

apps/frontend/src/app/core/services/recommendations.service.ts

@@ -0,0 +1,34 @@
+import { Service, inject } from '@angular/core';
+import { HttpClient, HttpParams } from '@angular/common/http';
+import { environment } from '../../../environments/environment';
+import {
+  Recommendation,
+  RecommendationGenerationReport,
+} from '../../shared/models/recommendation.model';
+
+@Service()
+export class RecommendationsService {
+  private http = inject(HttpClient);
+
+  getRecommendations() {
+    return this.http.get<Recommendation[]>(`${environment.apiUrl}/recommendations`);
+  }
+
+  getRecommendation(recommendationId: number) {
+    return this.http.get<Recommendation>(
+      `${environment.apiUrl}/recommendations/${recommendationId}`,
+    );
+  }
+
+  generate(siteId?: string) {
+    let params = new HttpParams();
+    if (siteId) {
+      params = params.set('site_id', siteId);
+    }
+    return this.http.post<RecommendationGenerationReport>(
+      `${environment.apiUrl}/recommendations/generate`,
+      null,
+      { params },
+    );
+  }
+}

apps/frontend/src/app/features/dashboard/dashboard.html

@@ -14,6 +14,7 @@
         <a routerLink="/monitoring/sensors" class="ev-link">Supervision des capteurs</a>
       }
       <a routerLink="/sites" class="ev-link">Voir les sites</a>
+      <a routerLink="/recommendations" class="ev-link">Recommandations</a>
       <ev-button
         class="logout-button"
         variant="secondary"
@@ -27,87 +28,120 @@
   @if (statsError(); as message) {
     <ev-alert severity="danger" class="banner-error">{{ message }}</ev-alert>
   }
-  @if (alertsError(); as message) {
-    <ev-alert severity="danger" class="banner-error">{{ message }}</ev-alert>
-  }
   @if (predictionsError(); as message) {
     <ev-alert severity="danger" class="banner-error">{{ message }}</ev-alert>
   }
 
   @if (stats(); as s) {
-    <section class="overview">
+    <p class="dashboard__status">
-      <ev-card class="card card--gauge">
+      <span class="dashboard__pulse" aria-hidden="true"></span>
-        <span class="card__label">Consommation vs capacité</span>
+      Actualisé à {{ s.timestamp | date: 'HH:mm:ss' }} · {{ s.total_sites }} sites suivis
+    </p>
+
+    <section class="overview" aria-label="Indicateurs du parc">
+      <ev-card class="kpi kpi--gauge">
+        <span class="kpi__label">Consommation vs capacité</span>
         <app-consumption-gauge
           [consumption]="s.total_consumption_kw"
           [capacity]="s.total_capacity_kw"
         />
-        <span class="card__value"
+        <span class="kpi__value">
-          >{{ s.total_consumption_kw | number: '1.0-1' }} /
+          {{ s.total_consumption_kw | number: '1.0-1' }}
-          {{ s.total_capacity_kw | number }} kW</span
+          <small>/ {{ s.total_capacity_kw | number }} kW</small>
+        </span>
+      </ev-card>
+
+      <ev-card class="kpi">
+        <span class="kpi__label">Charge moyenne du parc</span>
+        <span class="kpi__value"
+          >{{ s.average_load_percent | number: '1.0-0' }} <small>%</small></span
         >
-      </ev-card>
+        <div
-
+          class="progress-bar"
-      <ev-card class="card">
+          role="progressbar"
-        <span class="card__label">Charge moyenne du parc</span>
+          aria-valuemin="0"
-        <span class="card__value">{{ s.average_load_percent }} %</span>
+          aria-valuemax="100"
-        <div class="progress-bar">
+          [attr.aria-valuenow]="s.average_load_percent"
-          <div class="progress-bar__fill" [style.width.%]="s.average_load_percent"></div>
+        >
+          <div
+            class="progress-bar__fill"
+            [class]="'progress-bar__fill--' + loadTone(s.average_load_percent)"
+            [style.width.%]="s.average_load_percent"
+          ></div>
         </div>
+        <span class="kpi__hint">{{ loadHint(s.average_load_percent) }}</span>
       </ev-card>
 
-      <ev-card class="card">
+      <ev-card class="kpi">
-        <span class="card__label">Sites suivis</span>
+        <span class="kpi__label">Sites suivis</span>
-        <span class="card__value">{{ s.total_sites }}</span>
+        <span class="kpi__value">{{ s.total_sites }}</span>
+        <a routerLink="/sites" class="ev-link kpi__link">Voir la liste des sites</a>
       </ev-card>
     </section>
+  }
 
+  <div class="dashboard__grid">
+    <div class="dashboard__main">
+      @if (stats(); as s) {
         <section class="chart-section">
-      <h2>Charge et alerte visuelle par site</h2>
+          <h2>Charge par site</h2>
+          <ev-card class="chart-card">
             <app-site-load-chart [sites]="s.sites" />
+          </ev-card>
         </section>
       }
 
-  @if (alerts().length > 0) {
-    <section class="alerts-section">
-      <h2>Alertes actives</h2>
-      <ul class="alerts-list">
-        @for (alert of alerts(); track alert.alert_id) {
-          <li class="alert-item">
-            <ev-badge [tone]="badgeToneForSeverity(alert.severity)">{{ alert.severity }}</ev-badge>
-            <span class="alert-item__message">{{ alert.message }}</span>
-          </li>
-        }
-      </ul>
-    </section>
-  }
-
-  @if (predictions().length > 0) {
       <section class="predictions-section">
         <h2>Prévisions de consommation</h2>
-      <ul class="predictions-list">
+        @if (predictions().length > 0) {
+          <ev-card class="ev-table-card">
+            <table class="ev-table">
+              <thead>
+                <tr>
+                  <th>Site</th>
+                  <th>Prévision</th>
+                  <th>Échéance</th>
+                  <th></th>
+                </tr>
+              </thead>
+              <tbody>
                 @for (site of predictions(); track site.site_id) {
-          <li class="prediction-item">
+                  <tr>
-            <span class="prediction-item__site">{{ site.site_name }}</span>
+                    <td>{{ site.site_name }}</td>
                     @if (site.prediction; as prediction) {
                       @if (prediction.status === 'available') {
-                <span class="prediction-item__value">
+                        <td class="ev-table__number">
                           {{ prediction.predicted_value | number: '1.0-1' }} kWh
-                  <span class="prediction-item__target"
+                        </td>
-                    >{{ prediction.target_at | date: "dd/MM 'à' HH:mm" }}</span
+                        <td>{{ prediction.target_at | date: "dd/MM 'à' HH:mm" }}</td>
-                  >
-                </span>
                       } @else {
+                        <td>
                           <ev-badge [tone]="badgeToneForPredictionStatus(prediction.status)">{{
-                  prediction.status === 'insufficient_data' ? 'Historique insuffisant' : 'Erreur'
+                            prediction.status === 'insufficient_data'
+                              ? 'Historique insuffisant'
+                              : 'Erreur'
                           }}</ev-badge>
+                        </td>
+                        <td class="ev-table__muted">-</td>
                       }
                     } @else {
-              <ev-badge tone="neutral">Pas encore de prévision</ev-badge>
+                      <td><ev-badge tone="neutral">Pas encore de prévision</ev-badge></td>
+                      <td class="ev-table__muted">-</td>
                     }
-          </li>
+                    <td><a [routerLink]="['/sites', site.site_id]" class="ev-link">Détail</a></td>
+                  </tr>
+                }
+              </tbody>
+            </table>
+          </ev-card>
+        } @else if (!predictionsError()) {
+          <p class="dashboard__empty">Aucune prévision disponible pour le moment.</p>
         }
-      </ul>
       </section>
-  }
+    </div>
+
+    <aside class="dashboard__side" aria-label="Alertes actives">
+      <app-alert-feed />
+    </aside>
+  </div>
 </div>

apps/frontend/src/app/features/dashboard/dashboard.scss

@@ -8,9 +8,11 @@
 
 .dashboard__header {
   display: flex;
+  flex-wrap: wrap;
   align-items: flex-start;
   justify-content: space-between;
-  margin-bottom: 2rem;
+  gap: var(--space-3);
+  margin-bottom: var(--space-3);
 }
 
 .dashboard__brand {
@@ -20,8 +22,9 @@
 
   h1 {
     margin: 0;
-    font-size: 1.75rem;
+    font-size: var(--font-size-xl);
     font-weight: 700;
+    letter-spacing: -0.01em;
   }
 }
 
@@ -31,19 +34,69 @@
 
 .dashboard__subtitle {
   margin: 0.25rem 0 0;
+  font-size: var(--font-size-sm);
   color: var(--color-text-muted);
 }
 
 .dashboard__actions {
   display: flex;
+  flex-wrap: wrap;
   align-items: center;
-  gap: 1rem;
+  gap: var(--space-2);
+
+  .ev-link {
+    padding: 0.45rem 0.9rem;
+    border-radius: var(--radius-pill);
+    background: var(--color-primary-light);
+    color: var(--color-primary-hover);
+    font-size: var(--font-size-sm);
+    transition: background 0.15s ease;
+
+    &:hover {
+      background: var(--color-primary);
+      color: var(--color-text-inverse);
+      text-decoration: none;
+    }
+  }
+}
+
+.dashboard__status {
+  display: flex;
+  align-items: center;
+  gap: var(--space-2);
+  margin: 0 0 var(--space-4);
+  font-size: var(--font-size-sm);
+  color: var(--color-text-muted);
+}
+
+.dashboard__pulse {
+  width: 0.6rem;
+  height: 0.6rem;
+  border-radius: 50%;
+  background: var(--color-success);
+  animation: pulse 2s ease-out infinite;
+}
+
+@keyframes pulse {
+  0% {
+    box-shadow: 0 0 0 0 rgba(22, 163, 74, 0.45);
+  }
+
+  100% {
+    box-shadow: 0 0 0 8px rgba(22, 163, 74, 0);
+  }
+}
+
+@media (prefers-reduced-motion: reduce) {
+  .dashboard__pulse {
+    animation: none;
+  }
 }
 
 h2 {
-  font-size: 1.1rem;
+  margin: 0 0 var(--space-3);
+  font-size: var(--font-size-lg);
   font-weight: 600;
-  margin: 0 0 1rem;
 }
 
 .banner-error {
@@ -54,116 +107,124 @@ h2 {
 .overview {
   display: grid;
   grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
-  gap: 1rem;
+  gap: var(--space-3);
-  margin-bottom: 2.5rem;
+  margin-bottom: var(--space-5);
 }
 
-.card {
+.kpi {
-  padding: 1.25rem;
+  position: relative;
-  gap: 0.35rem;
+  overflow: hidden;
+  padding: var(--space-4);
+  gap: var(--space-1);
+  transition:
+    box-shadow 0.15s ease,
+    transform 0.15s ease;
+
+  &::before {
+    content: '';
+    position: absolute;
+    inset: 0 0 auto 0;
+    height: 3px;
+    background: var(--color-primary);
+  }
+
+  &:hover {
+    box-shadow: var(--shadow-card-hover);
+    transform: translateY(-1px);
+  }
 }
 
-.card--gauge {
+.kpi--gauge {
   align-items: center;
   text-align: center;
 }
 
-.card--link {
+.kpi__label {
-  cursor: pointer;
+  font-size: var(--font-size-xs);
-  transition: border-color 0.15s ease;
+  font-weight: 600;
+  color: var(--color-text-muted);
+  text-transform: uppercase;
+  letter-spacing: 0.06em;
+}
 
-  &:hover {
+.kpi__value {
-    border-color: var(--color-primary);
+  font-size: var(--font-size-2xl);
+  font-weight: 700;
+  line-height: 1.1;
+  font-variant-numeric: tabular-nums;
+
+  small {
+    font-size: var(--font-size-sm);
+    font-weight: 500;
+    color: var(--color-text-muted);
   }
 }
 
-.card__label {
+.kpi__hint {
-  font-size: 0.8rem;
+  font-size: var(--font-size-xs);
   color: var(--color-text-muted);
-  text-transform: uppercase;
-  letter-spacing: 0.02em;
 }
 
-.card__value {
+.kpi__link {
-  font-size: 1.6rem;
+  margin-top: auto;
-  font-weight: 700;
+  font-size: var(--font-size-sm);
 }
 
 .progress-bar {
-  height: 6px;
+  height: 8px;
+  margin: var(--space-1) 0;
   background: var(--color-border-light);
   border-radius: var(--radius-pill);
   overflow: hidden;
-  margin-top: 0.25rem;
 }
 
 .progress-bar__fill {
   height: 100%;
-  background: var(--color-primary);
   border-radius: var(--radius-pill);
+  background: var(--color-success);
   transition: width 0.3s ease;
 }
 
+.progress-bar__fill--warning {
+  background: var(--color-warning);
+}
+
+.progress-bar__fill--danger {
+  background: var(--color-danger);
+}
+
+.dashboard__grid {
+  display: grid;
+  grid-template-columns: minmax(0, 1fr) minmax(320px, 380px);
+  gap: var(--space-4);
+  align-items: start;
+}
+
+.dashboard__side {
+  position: sticky;
+  top: var(--space-3);
+}
+
 .chart-section {
-  margin-bottom: 2.5rem;
+  margin-bottom: var(--space-5);
 }
 
-.alerts-list {
+.chart-card {
-  list-style: none;
+  padding: var(--space-3);
+}
+
+.dashboard__empty {
   margin: 0;
-  padding: 0;
+  font-size: var(--font-size-sm);
-  display: flex;
-  flex-direction: column;
-  gap: 0.5rem;
-}
-
-.alert-item {
-  display: flex;
-  align-items: center;
-  gap: 0.75rem;
-  padding: 0.7rem 1rem;
-  border-radius: var(--radius-md);
-  background: var(--color-danger-bg);
-  border: 1px solid var(--color-danger-border);
-}
-
-.alert-item__message {
-  font-size: 0.9rem;
-}
-
-.predictions-list {
-  list-style: none;
-  margin: 0;
-  padding: 0;
-  display: flex;
-  flex-direction: column;
-  gap: 0.5rem;
-}
-
-.prediction-item {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  gap: 0.75rem;
-  padding: 0.7rem 1rem;
-  border-radius: var(--radius-md);
-  background: var(--color-surface);
-  border: 1px solid var(--color-border-light);
-}
-
-.prediction-item__site {
-  font-size: 0.9rem;
-  font-weight: 600;
-}
-
-.prediction-item__value {
-  font-size: 0.9rem;
-  font-weight: 600;
-}
-
-.prediction-item__target {
-  margin-left: 0.35rem;
-  font-size: 0.8rem;
-  font-weight: 400;
   color: var(--color-text-muted);
 }
+
+@media (max-width: 900px) {
+  .dashboard__grid {
+    grid-template-columns: 1fr;
+  }
+
+  .dashboard__side {
+    position: static;
+  }
+}

apps/frontend/src/app/features/dashboard/dashboard.spec.ts

@@ -1,12 +1,13 @@
 import { TestBed } from '@angular/core/testing';
 import { vi } from 'vitest';
-import { of, throwError } from 'rxjs';
+import { Observable, of, throwError } from 'rxjs';
+import { Router, provideRouter } from '@angular/router';
 import { Dashboard } from './dashboard';
 import { StatsService } from '../../core/services/stats.service';
 import { AlertsService } from '../../core/services/alerts.service';
+import { SitesService } from '../../core/services/sites.service';
 import { PredictionsService } from '../../core/services/predictions.service';
-import {AuthService} from '../../core/services/auth.service';
+import { AuthService } from '../../core/services/auth.service';
-import {Router, provideRouter} from '@angular/router';
 
 vi.mock('chart.js', () => {
   class ChartMock {
@@ -18,66 +19,73 @@ vi.mock('chart.js', () => {
   return { Chart: ChartMock, registerables: [] };
 });
 
+const STATS = { total_sites: 7, sites: [] };
+
 function predictionsMock(sites: unknown[] = []) {
-  return { getPredictions: vi.fn().mockReturnValue(of({ timestamp: '2026-09-18T09:00:00Z', sites })) };
+  return {
+    getPredictions: vi.fn().mockReturnValue(of({ timestamp: '2026-09-18T09:00:00Z', sites })),
+  };
+}
+
+function setup(
+  options: {
+    stats?: Observable<unknown>;
+    predictions?: { getPredictions: ReturnType<typeof vi.fn> };
+    auth?: Record<string, unknown>;
+  } = {},
+) {
+  const statsMock = { getSummary: vi.fn().mockReturnValue(options.stats ?? of(STATS)) };
+  const predictions = options.predictions ?? predictionsMock();
+  TestBed.configureTestingModule({
+    imports: [Dashboard],
+    providers: [
+      { provide: StatsService, useValue: statsMock },
+      { provide: AlertsService, useValue: { getAlerts: vi.fn().mockReturnValue(of([])) } },
+      { provide: SitesService, useValue: { getSites: vi.fn().mockReturnValue(of([])) } },
+      { provide: PredictionsService, useValue: predictions },
+      ...(options.auth ? [{ provide: AuthService, useValue: options.auth }] : []),
+      provideRouter([]),
+    ],
+  });
+  return { fixture: TestBed.createComponent(Dashboard), statsMock, predictions };
 }
 
 describe('Dashboard', () => {
   afterEach(() => vi.useRealTimers());
 
-  it('charge les stats, les alertes et les prévisions au démarrage', async () => {
+  it('charge les stats et les prévisions au démarrage', async () => {
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
+    const { fixture, statsMock, predictions } = setup({
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([{ alert_id: 'A1' }])) };
+      predictions: predictionsMock([{ site_id: 'SITE001', site_name: 'Test', prediction: null }]),
-    const predictions = predictionsMock([{ site_id: 'SITE001', site_name: 'Test', prediction: null }]);
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictions },
-        provideRouter([]),
-      ],
     });
 
-    const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
-
-    // laisse le timer(0, ...) se déclencher avant de vérifier
     await new Promise((resolve) => setTimeout(resolve, 0));
     fixture.detectChanges();
 
     expect(statsMock.getSummary).toHaveBeenCalled();
-    expect(alertsMock.getAlerts).toHaveBeenCalled();
     expect(predictions.getPredictions).toHaveBeenCalled();
-    expect(fixture.componentInstance.alerts().length).toBe(1);
     expect(fixture.componentInstance.predictions().length).toBe(1);
     expect(fixture.componentInstance.statsError()).toBeNull();
-    expect(fixture.componentInstance.alertsError()).toBeNull();
     expect(fixture.componentInstance.predictionsError()).toBeNull();
   });
 
+  it('délègue les alertes au widget app-alert-feed', () => {
+    const { fixture } = setup();
+
+    fixture.detectChanges();
+
+    expect(fixture.nativeElement.querySelector('app-alert-feed')).not.toBeNull();
+  });
+
   it("signale l'indisponibilité puis repart au rafraîchissement suivant", () => {
     vi.useFakeTimers();
-    const statsMock = {
+    const { fixture, statsMock } = setup({
-      getSummary: vi
+      stats: throwError(() => new Error('API injoignable')),
-        .fn()
-        .mockReturnValueOnce(throwError(() => new Error('API injoignable')))
-        .mockReturnValue(of({ total_sites: 7, sites: [] })),
-    };
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictionsMock() },
-        provideRouter([]),
-      ],
     });
+    statsMock.getSummary
+      .mockReturnValueOnce(throwError(() => new Error('API injoignable')))
+      .mockReturnValue(of(STATS));
 
-    const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
 
     vi.advanceTimersByTime(1);
@@ -91,45 +99,11 @@ describe('Dashboard', () => {
     expect(fixture.componentInstance.statsError()).toBeNull();
   });
 
-  it("n'interrompt pas la page quand le chargement des alertes échoue", () => {
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(throwError(() => new Error('nope'))) };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictionsMock() },
-        provideRouter([]),
-      ],
-    });
-
-    const fixture = TestBed.createComponent(Dashboard);
-    fixture.detectChanges();
-
-    expect(fixture.componentInstance.alerts().length).toBe(0);
-    expect(fixture.componentInstance.alertsError()).not.toBeNull();
-  });
-
   it("n'interrompt pas la page quand le chargement des prévisions échoue", () => {
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
+    const { fixture } = setup({
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
+      predictions: { getPredictions: vi.fn().mockReturnValue(throwError(() => new Error('nope'))) },
-    const predictions = {
-      getPredictions: vi.fn().mockReturnValue(throwError(() => new Error('nope'))),
-    };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictions },
-        provideRouter([]),
-      ],
     });
 
-    const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
 
     expect(fixture.componentInstance.predictions().length).toBe(0);
@@ -138,25 +112,11 @@ describe('Dashboard', () => {
 
   it("un rafraîchissement de stats n'efface pas une erreur de prévisions en attente", () => {
     vi.useFakeTimers();
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
+    const { fixture } = setup({
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
+      predictions: { getPredictions: vi.fn().mockReturnValue(throwError(() => new Error('nope'))) },
-    const predictions = {
-      getPredictions: vi.fn().mockReturnValue(throwError(() => new Error('nope'))),
-    };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictions },
-        provideRouter([]),
-      ],
     });
 
-    const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
-
     expect(fixture.componentInstance.predictionsError()).not.toBeNull();
 
     // Plusieurs cycles de `timer(0, 10_000)` (stats) plus tard, l'erreur des prévisions doit
@@ -168,109 +128,106 @@ describe('Dashboard', () => {
   });
 
   it('appelle logout et redirige vers /login au clic sur le bouton de déconnexion', () => {
-  const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
-  const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
     const authMock = {
       logout: vi.fn().mockReturnValue(of(undefined)),
       clearSession: vi.fn(),
       principal: vi.fn().mockReturnValue({ role: 'admin' }),
     };
-  TestBed.configureTestingModule({
+    const { fixture } = setup({ auth: authMock });
-    imports: [Dashboard],
-    providers: [
-      { provide: StatsService, useValue: statsMock },
-      { provide: AlertsService, useValue: alertsMock },
-      { provide: PredictionsService, useValue: predictionsMock() },
-      { provide: AuthService, useValue: authMock },
-      provideRouter([]),
-    ],
-  });
-
-  const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
-
     const router = TestBed.inject(Router);
     const navigateSpy = vi.spyOn(router, 'navigate').mockResolvedValue(true);
 
-  const button = fixture.nativeElement.querySelector('.logout-button');
+    fixture.nativeElement.querySelector('.logout-button').click();
-  button.click();
 
     expect(authMock.logout).toHaveBeenCalled();
     expect(navigateSpy).toHaveBeenCalledWith(['/login']);
   });
+
   it('déconnecte localement et redirige vers /login même si logout échoue côté réseau', () => {
-  const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
-  const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
     const authMock = {
       logout: vi.fn().mockReturnValue(throwError(() => new Error('réseau indisponible'))),
       clearSession: vi.fn(),
       principal: vi.fn().mockReturnValue({ role: 'admin' }),
     };
-  TestBed.configureTestingModule({
+    const { fixture } = setup({ auth: authMock });
-    imports: [Dashboard],
-    providers: [
-      { provide: StatsService, useValue: statsMock },
-      { provide: AlertsService, useValue: alertsMock },
-      { provide: PredictionsService, useValue: predictionsMock() },
-      { provide: AuthService, useValue: authMock },
-      provideRouter([]),
-    ],
-  });
-
-  const fixture = TestBed.createComponent(Dashboard);
     fixture.detectChanges();
-
     const router = TestBed.inject(Router);
     const navigateSpy = vi.spyOn(router, 'navigate').mockResolvedValue(true);
 
-  const button = fixture.nativeElement.querySelector('.logout-button');
+    fixture.nativeElement.querySelector('.logout-button').click();
-  button.click();
 
     expect(authMock.clearSession).toHaveBeenCalled();
     expect(navigateSpy).toHaveBeenCalledWith(['/login']);
-});
-
-  it('distingue le ton des sévérités high et critical', () => {
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictionsMock() },
-        provideRouter([]),
-      ],
   });
 
-    const fixture = TestBed.createComponent(Dashboard);
+  it('affiche l’heure du dernier relevé et les indicateurs du parc', () => {
+    vi.useFakeTimers();
+    const { fixture } = setup({
+      stats: of({
+        timestamp: '2026-09-18T09:00:00Z',
+        total_sites: 7,
+        total_consumption_kw: 1234.5,
+        total_capacity_kw: 5000,
+        average_load_percent: 24.7,
+        sites: [],
+      }),
+    });
+
+    fixture.detectChanges();
+    vi.advanceTimersByTime(1);
+    fixture.detectChanges();
+
+    const texte = fixture.nativeElement.textContent as string;
+    expect(texte).toContain('Actualisé à');
+    expect(texte).toContain('7 sites suivis');
+    expect(texte).toContain('Marge confortable');
+    expect(fixture.nativeElement.querySelector('.progress-bar__fill--success')).not.toBeNull();
+  });
+
+  it('présente les prévisions en tableau avec un lien vers chaque site', () => {
+    const { fixture } = setup({
+      predictions: predictionsMock([
+        {
+          site_id: 'SITE001',
+          site_name: 'Usine Nantes',
+          prediction: {
+            target_at: '2026-09-18T10:00:00Z',
+            target_metric: 'consumption_kwh',
+            period_minutes: 60,
+            predicted_value: 118.4,
+            status: 'available',
+            failure_reason: null,
+            model_reference: 'lightgbm-v1',
+            created_at: '2026-09-18T09:00:00Z',
+          },
+        },
+        { site_id: 'SITE002', site_name: 'Bureau Lille', prediction: null },
+      ]),
+    });
+
+    fixture.detectChanges();
+
+    const table = fixture.nativeElement.querySelector('table.ev-table');
+    expect(table).not.toBeNull();
+    expect(table.textContent).toContain('118.4 kWh');
+    expect(table.textContent).toContain('Pas encore de prévision');
+    expect(fixture.nativeElement.querySelector('a[href="/sites/SITE001"]')).not.toBeNull();
+  });
+
+  it('colore la charge moyenne selon les seuils 70 % et 90 %', () => {
+    const { fixture } = setup();
     const dashboard = fixture.componentInstance;
 
-    expect(dashboard.badgeToneForSeverity('low')).toBe('success');
+    expect(dashboard.loadTone(69.9)).toBe('success');
-    expect(dashboard.badgeToneForSeverity('medium')).toBe('warning');
+    expect(dashboard.loadTone(70)).toBe('warning');
-    expect(dashboard.badgeToneForSeverity('high')).toBe('danger');
+    expect(dashboard.loadTone(89.9)).toBe('warning');
-    expect(dashboard.badgeToneForSeverity('critical')).toBe('critical');
+    expect(dashboard.loadTone(90)).toBe('danger');
-    expect(dashboard.badgeToneForSeverity('high')).not.toBe(
+    expect(dashboard.loadHint(95)).toBe('Proche de la capacité du parc');
-      dashboard.badgeToneForSeverity('critical'),
-    );
   });
 
   it('distingue le ton des statuts de prévision', () => {
-    const statsMock = { getSummary: vi.fn().mockReturnValue(of({ total_sites: 7, sites: [] })) };
+    const { fixture } = setup();
-    const alertsMock = { getAlerts: vi.fn().mockReturnValue(of([])) };
-
-    TestBed.configureTestingModule({
-      imports: [Dashboard],
-      providers: [
-        { provide: StatsService, useValue: statsMock },
-        { provide: AlertsService, useValue: alertsMock },
-        { provide: PredictionsService, useValue: predictionsMock() },
-        provideRouter([]),
-      ],
-    });
-
-    const fixture = TestBed.createComponent(Dashboard);
     const dashboard = fixture.componentInstance;
 
     expect(dashboard.badgeToneForPredictionStatus('available')).toBe('success');

apps/frontend/src/app/features/dashboard/dashboard.ts

@@ -6,11 +6,10 @@ import { Router, RouterLink } from '@angular/router';
 import { StatsService } from '../../core/services/stats.service';
 import { ConsumptionGauge } from '../../shared/components/consumption-gauge/consumption-gauge';
 import { SiteLoadChart } from '../../shared/components/site-load-chart/site-load-chart';
-import { AlertsService } from '../../core/services/alerts.service';
+import { AlertFeed } from '../../shared/components/alert-feed/alert-feed';
 import { PredictionsService } from '../../core/services/predictions.service';
 import { AuthService } from '../../core/services/auth.service';
 import { StatsSummary } from '../../shared/models/stats.model';
-import { Alert, AlertSeverity } from '../../shared/models/alert.model';
 import { PredictionStatus, SitePredictionSummary } from '../../shared/models/prediction.model';
 import { Card } from '../../shared/components/ui/card/card';
 import { Alert as EvAlert } from '../../shared/components/ui/alert/alert';
@@ -22,22 +21,19 @@ const REFRESH_INTERVAL_MS = 10000;
 const UNAVAILABLE_MESSAGE =
   'Données indisponibles, les valeurs affichées datent du dernier relevé.';
 
-const TON_PAR_SEVERITE: Record<AlertSeverity, BadgeTone> = {
+// `error` n'a pas encore de précédent côté API mais figure dans `ck_prediction_status` :
-  low: 'success',
+// mieux vaut un ton défini que `undefined` le jour où ce statut apparaît.
-  medium: 'warning',
-  high: 'danger',
-  critical: 'critical',
-};
-
-// `error` n'a pas de précédent dans les fixtures ou l'API à ce jour, mais figure dans le
-// domaine du schéma backend (`ck_prediction_status`) : mieux vaut une couleur définie que
-// tomber sur `undefined` si ce statut apparaît un jour.
 const TON_PAR_STATUT_PREDICTION: Record<PredictionStatus, BadgeTone> = {
   available: 'success',
   insufficient_data: 'warning',
   error: 'danger',
 };
 
+const SEUIL_CHARGE_SOUTENUE = 70;
+const SEUIL_CHARGE_CRITIQUE = 90;
+
+export type LoadTone = 'success' | 'warning' | 'danger';
+
 @Component({
   selector: 'app-dashboard',
   standalone: true,
@@ -47,6 +43,7 @@ const TON_PAR_STATUT_PREDICTION: Record<PredictionStatus, BadgeTone> = {
     RouterLink,
     ConsumptionGauge,
     SiteLoadChart,
+    AlertFeed,
     Card,
     EvAlert,
     Badge,
@@ -58,32 +55,20 @@ const TON_PAR_STATUT_PREDICTION: Record<PredictionStatus, BadgeTone> = {
 })
 export class Dashboard implements OnInit {
   private statsService = inject(StatsService);
-  private alertsService = inject(AlertsService);
   public auth = inject(AuthService);
   private predictionsService = inject(PredictionsService);
   private router = inject(Router);
   private destroyRef = inject(DestroyRef);
 
   stats = signal<StatsSummary | null>(null);
-  alerts = signal<Alert[]>([]);
   predictions = signal<SitePredictionSummary[]>([]);
 
-  // Un signal par flux, pas un seul `error` partagé : sinon le tick suivant de `timer` (stats)
+  // Piège : un signal d'erreur par flux, sinon le tick suivant de `timer` (stats) efface en
-  // efface silencieusement un message d'échec des prévisions ou des alertes après 10s au plus,
+  // silence l'échec des prévisions après 10 s au plus, sans retry ni indication à l'utilisateur.
-  // sans retry ni indication pour l'utilisateur que la section correspondante est restée vide.
   statsError = signal<string | null>(null);
-  alertsError = signal<string | null>(null);
   predictionsError = signal<string | null>(null);
 
   ngOnInit(): void {
-    this.alertsService
-      .getAlerts()
-      .pipe(catchError(() => this.reportUnavailable(this.alertsError)))
-      .subscribe((alerts) => {
-        this.alertsError.set(null);
-        this.alerts.set(alerts);
-      });
-
     // Les prévisions viennent d'un scoring hors ligne, pas d'un calcul à la demande : un seul
     // chargement au démarrage suffit, pas besoin du rafraîchissement périodique de `stats`.
     this.predictionsService
@@ -99,7 +84,9 @@ export class Dashboard implements OnInit {
     timer(0, REFRESH_INTERVAL_MS)
       .pipe(
         switchMap(() =>
-          this.statsService.getSummary().pipe(catchError(() => this.reportUnavailable(this.statsError))),
+          this.statsService
+            .getSummary()
+            .pipe(catchError(() => this.reportUnavailable(this.statsError))),
         ),
         takeUntilDestroyed(this.destroyRef),
       )
@@ -109,14 +96,28 @@ export class Dashboard implements OnInit {
       });
   }
 
-  badgeToneForSeverity(severity: AlertSeverity): BadgeTone {
-    return TON_PAR_SEVERITE[severity];
-  }
-
   badgeToneForPredictionStatus(status: PredictionStatus): BadgeTone {
     return TON_PAR_STATUT_PREDICTION[status];
   }
 
+  loadTone(percent: number): LoadTone {
+    if (percent >= SEUIL_CHARGE_CRITIQUE) {
+      return 'danger';
+    }
+    return percent >= SEUIL_CHARGE_SOUTENUE ? 'warning' : 'success';
+  }
+
+  loadHint(percent: number): string {
+    switch (this.loadTone(percent)) {
+      case 'danger':
+        return 'Proche de la capacité du parc';
+      case 'warning':
+        return 'Charge soutenue';
+      default:
+        return 'Marge confortable';
+    }
+  }
+
   onLogout(): void {
     this.auth.logout().subscribe({
       next: () => this.router.navigate(['/login']),

apps/frontend/src/app/features/recommendations/recommendations.html

@@ -0,0 +1,58 @@
+<div class="recommendations">
+  <nav class="ev-breadcrumb">
+    <a routerLink="/dashboard">Tableau de bord</a>
+  </nav>
+
+  <header class="recommendations__header">
+    <a routerLink="/dashboard" class="ev-brand-link">
+      <ev-brand class="recommendations__logo" />
+    </a>
+    <div>
+      <h1>Recommandations</h1>
+      <p class="recommendations__subtitle">
+        Actions proposées par le moteur de règles à partir des alertes
+      </p>
+    </div>
+  </header>
+
+  <div class="recommendations__toolbar">
+    <label class="recommendations__filter">
+      <span class="form-label">Site</span>
+      <select class="form-select" data-testid="site-filter" (change)="onSiteChange($event)">
+        <option value="" [selected]="!siteFilter()">Tous les sites</option>
+        @for (site of sites(); track site.site_id) {
+          <option [value]="site.site_id" [selected]="site.site_id === siteFilter()">
+            {{ site.site_name }}
+          </option>
+        }
+      </select>
+    </label>
+    @if (isAdmin()) {
+      <ev-button
+        variant="secondary"
+        [fullWidth]="false"
+        [disabled]="generating()"
+        data-testid="generate"
+        (click)="onGenerate()"
+      >
+        {{ generating() ? 'Génération en cours…' : 'Générer les recommandations' }}
+      </ev-button>
+    }
+  </div>
+
+  @if (generationReport(); as report) {
+    <ev-alert severity="success" class="recommendations__banner">{{ bilan(report) }}.</ev-alert>
+  }
+  @if (generationError(); as message) {
+    <ev-alert severity="danger" class="recommendations__banner">{{ message }}</ev-alert>
+  }
+
+  @if (alertId(); as id) {
+    <p class="recommendations__focus">
+      Alerte n° {{ id }} ·
+      <a routerLink="/recommendations" class="ev-link">Toutes les recommandations</a>
+    </p>
+  }
+
+  <app-recommendation-list [siteId]="siteFilter()" [alertId]="alertId()" [sites]="sites()" />
+</div>

apps/frontend/src/app/features/recommendations/recommendations.scss

@@ -0,0 +1,59 @@
+:host {
+  display: block;
+  color: var(--color-text);
+  padding: 2.5rem 2rem;
+  max-width: 1100px;
+  margin: 0 auto;
+}
+
+.recommendations__header {
+  display: flex;
+  align-items: center;
+  gap: 0.85rem;
+  margin-bottom: 2rem;
+
+  h1 {
+    margin: 0;
+    font-size: 1.75rem;
+    font-weight: 700;
+  }
+}
+
+.recommendations__logo {
+  font-size: 1.3rem;
+}
+
+.recommendations__subtitle {
+  margin: 0.25rem 0 0;
+  color: var(--color-text-muted);
+}
+
+.recommendations__toolbar {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: flex-end;
+  justify-content: space-between;
+  gap: var(--space-3);
+  margin-bottom: var(--space-4);
+}
+
+.recommendations__filter {
+  display: flex;
+  flex-direction: column;
+  min-width: 14rem;
+
+  .form-label {
+    margin-top: 0;
+  }
+}
+
+.recommendations__banner {
+  display: block;
+  margin-bottom: var(--space-3);
+}
+
+.recommendations__focus {
+  margin: 0 0 var(--space-3);
+  font-size: 0.9rem;
+  color: var(--color-text-muted);
+}

apps/frontend/src/app/features/recommendations/recommendations.spec.ts

@@ -0,0 +1,169 @@
+import { TestBed } from '@angular/core/testing';
+import { By } from '@angular/platform-browser';
+import { ActivatedRoute, convertToParamMap, provideRouter } from '@angular/router';
+import { vi } from 'vitest';
+import { BehaviorSubject, of, throwError } from 'rxjs';
+import { RecommendationsView, parseAlertId } from './recommendations';
+import { RecommendationList } from '../../shared/components/recommendation-list/recommendation-list';
+import { SitesService } from '../../core/services/sites.service';
+import { AlertsService } from '../../core/services/alerts.service';
+import { RecommendationsService } from '../../core/services/recommendations.service';
+import { AuthService } from '../../core/services/auth.service';
+
+const SITES = [
+  {
+    site_id: 'SITE001',
+    site_name: 'Usine Nantes',
+    site_type: 'industriel',
+    location: 'Nantes',
+    capacity_kw: 500,
+    status: 'actif',
+  },
+];
+
+const BILAN = { alerts_examined: 2, recommendations_created: 3, already_present: 1 };
+
+function setup(options: { query?: Record<string, string>; role?: string } = {}) {
+  const query = options.query ?? {};
+  const queryParamMap = new BehaviorSubject(convertToParamMap(query));
+  const generate = vi.fn().mockReturnValue(of(BILAN));
+  const getRecommendations = vi.fn().mockReturnValue(of([]));
+  const getAlerts = vi.fn().mockReturnValue(of([]));
+  TestBed.configureTestingModule({
+    imports: [RecommendationsView],
+    providers: [
+      provideRouter([]),
+      {
+        provide: ActivatedRoute,
+        useValue: { queryParamMap, snapshot: { queryParamMap: convertToParamMap(query) } },
+      },
+      { provide: SitesService, useValue: { getSites: vi.fn().mockReturnValue(of(SITES)) } },
+      { provide: AlertsService, useValue: { getAlerts } },
+      { provide: RecommendationsService, useValue: { getRecommendations, generate } },
+      {
+        provide: AuthService,
+        useValue: { principal: vi.fn().mockReturnValue({ role: options.role ?? 'lecteur' }) },
+      },
+    ],
+  });
+  const fixture = TestBed.createComponent(RecommendationsView);
+  fixture.detectChanges();
+  fixture.detectChanges();
+  return { fixture, queryParamMap, generate, getRecommendations, getAlerts };
+}
+
+function listeEnfant(fixture: ReturnType<typeof setup>['fixture']): RecommendationList {
+  return fixture.debugElement.query(By.directive(RecommendationList)).componentInstance;
+}
+
+describe('parseAlertId', () => {
+  it("n'accepte qu'un entier strictement positif", () => {
+    expect(parseAlertId('12')).toBe(12);
+    expect(parseAlertId('0')).toBeNull();
+    expect(parseAlertId('-3')).toBeNull();
+    expect(parseAlertId('abc')).toBeNull();
+    expect(parseAlertId('12abc')).toBeNull();
+    expect(parseAlertId(null)).toBeNull();
+  });
+});
+
+describe('RecommendationsView', () => {
+  it("cible l'alerte donnée par ?alert= et la transmet à la liste", () => {
+    const { fixture } = setup({ query: { alert: '12' } });
+
+    expect(fixture.componentInstance.alertId()).toBe(12);
+    expect(listeEnfant(fixture).alertId()).toBe(12);
+    expect(fixture.nativeElement.textContent).toContain('Alerte n° 12');
+    expect(fixture.nativeElement.querySelector('a[href="/recommendations"]')).not.toBeNull();
+  });
+
+  it('ignore un paramètre alert invalide', () => {
+    const { fixture } = setup({ query: { alert: 'abc' } });
+
+    expect(fixture.componentInstance.alertId()).toBeNull();
+    expect(fixture.nativeElement.textContent).not.toContain('Alerte n°');
+  });
+
+  it('applique le site donné par ?site= au filtre et à la liste', () => {
+    const { fixture, getAlerts } = setup({ query: { site: 'SITE001' } });
+
+    expect(getAlerts).toHaveBeenCalledWith({ site_id: 'SITE001' });
+    const option = fixture.nativeElement.querySelector(
+      'option[value="SITE001"]',
+    ) as HTMLOptionElement;
+    expect(option.selected).toBe(true);
+  });
+
+  it('relance la liste sur le site choisi dans le filtre', () => {
+    const { fixture, getAlerts } = setup();
+    const select = fixture.nativeElement.querySelector(
+      '[data-testid="site-filter"]',
+    ) as HTMLSelectElement;
+
+    select.value = 'SITE001';
+    select.dispatchEvent(new Event('change'));
+    fixture.detectChanges();
+    fixture.detectChanges();
+
+    expect(getAlerts).toHaveBeenLastCalledWith({ site_id: 'SITE001' });
+    expect(listeEnfant(fixture).siteId()).toBe('SITE001');
+  });
+
+  it('cache le bouton de génération aux lecteurs', () => {
+    const { fixture } = setup({ role: 'lecteur' });
+
+    expect(fixture.nativeElement.querySelector('[data-testid="generate"]')).toBeNull();
+  });
+
+  it('permet à un admin de générer pour le site filtré, affiche le bilan et recharge la liste', () => {
+    const { fixture, generate, getRecommendations } = setup({
+      role: 'admin',
+      query: { site: 'SITE001' },
+    });
+
+    fixture.nativeElement.querySelector('[data-testid="generate"]').click();
+    fixture.detectChanges();
+    fixture.detectChanges();
+
+    expect(generate).toHaveBeenCalledWith('SITE001');
+    expect(fixture.nativeElement.textContent).toContain(
+      '3 recommandations créées, 1 déjà présente, 2 alertes examinées.',
+    );
+    expect(getRecommendations).toHaveBeenCalledTimes(2);
+    expect(fixture.componentInstance.generating()).toBe(false);
+  });
+
+  it('génère pour tout le parc quand aucun site n’est filtré', () => {
+    const { fixture, generate } = setup({ role: 'admin' });
+
+    fixture.componentInstance.onGenerate();
+
+    expect(generate).toHaveBeenCalledWith(undefined);
+  });
+
+  it("signale l'échec de la génération sans casser la page", () => {
+    const { fixture, generate } = setup({ role: 'admin' });
+    generate.mockReturnValue(throwError(() => new Error('403')));
+
+    fixture.componentInstance.onGenerate();
+    fixture.detectChanges();
+
+    expect(fixture.componentInstance.generationError()).not.toBeNull();
+    expect(fixture.nativeElement.textContent).toContain(
+      'La génération des recommandations a échoué',
+    );
+    expect(fixture.componentInstance.generating()).toBe(false);
+  });
+
+  it('accorde le bilan au singulier', () => {
+    const { fixture } = setup();
+
+    expect(
+      fixture.componentInstance.bilan({
+        alerts_examined: 1,
+        recommendations_created: 1,
+        already_present: 0,
+      }),
+    ).toBe('1 recommandation créée, 0 déjà présente, 1 alerte examinée');
+  });
+});

apps/frontend/src/app/features/recommendations/recommendations.ts

@@ -0,0 +1,85 @@
+import { Component, computed, inject, signal, viewChild } from '@angular/core';
+import { toSignal } from '@angular/core/rxjs-interop';
+import { ActivatedRoute, RouterLink } from '@angular/router';
+import { catchError, map, of } from 'rxjs';
+import { SitesService } from '../../core/services/sites.service';
+import { RecommendationsService } from '../../core/services/recommendations.service';
+import { AuthService } from '../../core/services/auth.service';
+import { Site } from '../../shared/models/site.model';
+import { RecommendationGenerationReport } from '../../shared/models/recommendation.model';
+import { RecommendationList } from '../../shared/components/recommendation-list/recommendation-list';
+import { Alert as EvAlert } from '../../shared/components/ui/alert/alert';
+import { Brand } from '../../shared/components/ui/brand/brand';
+import { Button } from '../../shared/components/ui/button/button';
+
+const GENERATION_FAILED_MESSAGE =
+  'La génération des recommandations a échoué, réessayez plus tard.';
+
+export function parseAlertId(raw: string | null): number | null {
+  return raw !== null && /^[1-9]\d*$/.test(raw) ? Number(raw) : null;
+}
+
+function pluriel(nombre: number, singulier: string, plurielForme: string): string {
+  return `${nombre} ${nombre > 1 ? plurielForme : singulier}`;
+}
+
+@Component({
+  selector: 'app-recommendations',
+  standalone: true,
+  imports: [RouterLink, RecommendationList, EvAlert, Brand, Button],
+  templateUrl: './recommendations.html',
+  styleUrl: './recommendations.scss',
+})
+export class RecommendationsView {
+  private route = inject(ActivatedRoute);
+  private sitesService = inject(SitesService);
+  private recommendationsService = inject(RecommendationsService);
+  private auth = inject(AuthService);
+
+  alertId = toSignal(
+    this.route.queryParamMap.pipe(map((params) => parseAlertId(params.get('alert')))),
+    { initialValue: null },
+  );
+  siteFilter = signal<string | null>(this.route.snapshot.queryParamMap.get('site'));
+  sites = toSignal(this.sitesService.getSites().pipe(catchError(() => of([] as Site[]))), {
+    initialValue: [] as Site[],
+  });
+
+  list = viewChild.required(RecommendationList);
+
+  isAdmin = computed(() => this.auth.principal()?.role === 'admin');
+  generating = signal(false);
+  generationReport = signal<RecommendationGenerationReport | null>(null);
+  generationError = signal<string | null>(null);
+
+  onSiteChange(event: Event): void {
+    this.siteFilter.set((event.target as HTMLSelectElement).value || null);
+  }
+
+  onGenerate(): void {
+    if (this.generating()) {
+      return;
+    }
+    this.generating.set(true);
+    this.generationError.set(null);
+    this.recommendationsService.generate(this.siteFilter() ?? undefined).subscribe({
+      next: (report) => {
+        this.generating.set(false);
+        this.generationReport.set(report);
+        this.list().reload();
+      },
+      error: () => {
+        this.generating.set(false);
+        this.generationError.set(GENERATION_FAILED_MESSAGE);
+      },
+    });
+  }
+
+  bilan(report: RecommendationGenerationReport): string {
+    return [
+      pluriel(report.recommendations_created, 'recommandation créée', 'recommandations créées'),
+      pluriel(report.already_present, 'déjà présente', 'déjà présentes'),
+      pluriel(report.alerts_examined, 'alerte examinée', 'alertes examinées'),
+    ].join(', ');
+  }
+}

apps/frontend/src/app/features/sites/site-detail/site-detail.html

@@ -81,5 +81,16 @@
     }
   }
 
+  <section class="recommendations-section">
+    <h2>Recommandations</h2>
+    <app-recommendation-list [siteId]="siteId() ?? null" [sites]="siteAsList()" />
+    <a
+      routerLink="/recommendations"
+      [queryParams]="{ site: siteId() }"
+      class="ev-link recommendations-section__link"
+      >Voir dans la vue recommandations</a
+    >
+  </section>
+
   <a routerLink="/sites" class="ev-link">Retour aux sites</a>
 </div>

apps/frontend/src/app/features/sites/site-detail/site-detail.scss

@@ -117,3 +117,12 @@ h2 {
 .chart-section {
   margin-bottom: 2rem;
 }
+
+.recommendations-section {
+  margin: 2.5rem 0 1.5rem;
+}
+
+.recommendations-section__link {
+  display: inline-block;
+  margin-top: 1rem;
+}

apps/frontend/src/app/features/sites/site-detail/site-detail.spec.ts

@@ -5,6 +5,8 @@ import { BehaviorSubject, of, throwError } from 'rxjs';
 import { SiteDetail } from './site-detail';
 import { SitesService } from '../../../core/services/sites.service';
 import { ReadingsService } from '../../../core/services/readings.service';
+import { AlertsService } from '../../../core/services/alerts.service';
+import { RecommendationsService } from '../../../core/services/recommendations.service';
 
 const SITE = {
   site_id: 'SITE001',
@@ -69,6 +71,7 @@ function setup(
   readingsMock: Partial<ReadingsService>,
 ) {
   const paramMap = new BehaviorSubject(convertToParamMap({ siteId }));
+  const getAlerts = vi.fn().mockReturnValue(of([]));
   TestBed.configureTestingModule({
     imports: [SiteDetail],
     providers: [
@@ -76,9 +79,14 @@ function setup(
       { provide: ActivatedRoute, useValue: { paramMap } },
       { provide: SitesService, useValue: sitesMock },
       { provide: ReadingsService, useValue: readingsMock },
+      { provide: AlertsService, useValue: { getAlerts } },
+      {
+        provide: RecommendationsService,
+        useValue: { getRecommendations: vi.fn().mockReturnValue(of([])) },
+      },
     ],
   });
-  return { fixture: TestBed.createComponent(SiteDetail), paramMap };
+  return { fixture: TestBed.createComponent(SiteDetail), paramMap, getAlerts };
 }
 
 describe('SiteDetail', () => {
@@ -261,6 +269,27 @@ describe('SiteDetail', () => {
     );
   });
 
+  it('demande les recommandations du site consulté à travers ses alertes', () => {
+    const { fixture, getAlerts } = setup(
+      'SITE001',
+      {
+        getSite: vi.fn().mockReturnValue(of(SITE)),
+        getCurrent: vi.fn().mockReturnValue(of(CURRENT_COMPLET)),
+      },
+      { getHistory: vi.fn().mockReturnValue(of([])) },
+    );
+
+    fixture.detectChanges();
+    fixture.detectChanges();
+
+    expect(getAlerts).toHaveBeenCalledWith({ site_id: 'SITE001' });
+    expect(fixture.nativeElement.querySelector('app-recommendation-list')).not.toBeNull();
+    expect(fixture.nativeElement.textContent).toContain('Recommandations');
+    expect(
+      fixture.nativeElement.querySelector('a[href="/recommendations?site=SITE001"]'),
+    ).not.toBeNull();
+  });
+
   it("annonce l'absence de mesure sans interroger l'historique quand timestamp est null", () => {
     const getHistory = vi.fn().mockReturnValue(of([]));
     const { fixture } = setup(

apps/frontend/src/app/features/sites/site-detail/site-detail.ts

@@ -13,6 +13,7 @@ import { Badge, BadgeTone } from '../../../shared/components/ui/badge/badge';
 import { Brand } from '../../../shared/components/ui/brand/brand';
 import { ConsumptionGauge } from '../../../shared/components/consumption-gauge/consumption-gauge';
 import { ReadingHistoryChart } from '../../../shared/components/reading-history-chart/reading-history-chart';
+import { RecommendationList } from '../../../shared/components/recommendation-list/recommendation-list';
 
 const UNAVAILABLE_MESSAGE = 'Détail du site indisponible, réessayez plus tard.';
 const NO_MEASUREMENT_MESSAGE = 'Aucune mesure remontée pour ce site.';
@@ -96,7 +97,16 @@ export interface MetricView {
 @Component({
   selector: 'app-site-detail',
   standalone: true,
-  imports: [RouterLink, Card, Alert, Badge, Brand, ConsumptionGauge, ReadingHistoryChart],
+  imports: [
+    RouterLink,
+    Card,
+    Alert,
+    Badge,
+    Brand,
+    ConsumptionGauge,
+    ReadingHistoryChart,
+    RecommendationList,
+  ],
   templateUrl: './site-detail.html',
   styleUrl: './site-detail.scss',
 })
@@ -117,6 +127,11 @@ export class SiteDetail {
 
   hasMeasurement = computed(() => this.current()?.timestamp != null);
 
+  siteAsList = computed<Site[]>(() => {
+    const site = this.site();
+    return site ? [site] : [];
+  });
+
   consumptionKw = computed(() => this.current()?.consumption_kw ?? null);
 
   consumptionLabel = computed(() => {

apps/frontend/src/app/features/sites/site-list/site-list.html

@@ -17,8 +17,8 @@
     <ev-alert severity="danger" class="banner-error">{{ message }}</ev-alert>
   }
 
-  <ev-card class="table-card">
+  <ev-card class="ev-table-card">
-    <table class="sites-table">
+    <table class="ev-table">
       <thead>
         <tr>
           <th>Nom</th>
@@ -36,7 +36,9 @@
             <td>{{ site.site_type }}</td>
             <td>{{ site.location || '-' }}</td>
             <td>{{ site.capacity_kw ?? '-' }}</td>
-            <td><ev-badge [tone]="badgeToneForStatus(site.status)">{{ site.status ?? '-' }}</ev-badge></td>
+            <td>
+              <ev-badge [tone]="badgeToneForStatus(site.status)">{{ site.status ?? '-' }}</ev-badge>
+            </td>
             <td><a [routerLink]="['/sites', site.site_id]" class="ev-link">Détail</a></td>
           </tr>
         }

apps/frontend/src/app/features/sites/site-list/site-list.scss

@@ -32,32 +32,3 @@
   display: block;
   margin: 0 0 1.5rem;
 }
-
-.table-card {
-  padding: 0;
-  overflow: hidden;
-}
-
-.sites-table {
-  width: 100%;
-  border-collapse: collapse;
-
-  th,
-  td {
-    padding: 0.85rem 1.25rem;
-    text-align: left;
-    border-bottom: 1px solid var(--color-border-light);
-  }
-
-  th {
-    font-size: 0.8rem;
-    color: var(--color-text-muted);
-    text-transform: uppercase;
-    letter-spacing: 0.02em;
-    font-weight: 600;
-  }
-
-  tr:last-child td {
-    border-bottom: none;
-  }
-}

apps/frontend/src/app/shared/components/alert-feed/alert-feed.html

@@ -0,0 +1,95 @@
+<section class="alert-feed">
+  <header class="alert-feed__header">
+    <div>
+      <h2 class="alert-feed__title">Alertes actives</h2>
+      @if (!loading() || alerts().length > 0) {
+        <p class="alert-feed__count">
+          {{ alerts().length }} {{ alerts().length > 1 ? 'alertes' : 'alerte' }}
+        </p>
+      }
+    </div>
+    <div class="alert-feed__filters">
+      @if (!siteId()) {
+        <label class="alert-feed__filter">
+          <span class="form-label">Site</span>
+          <select class="form-select" data-testid="site-filter" (change)="onSiteChange($event)">
+            <option value="">Tous les sites</option>
+            @for (site of sites(); track site.site_id) {
+              <option [value]="site.site_id">{{ site.site_name }}</option>
+            }
+          </select>
+        </label>
+      }
+      <label class="alert-feed__filter">
+        <span class="form-label">Sévérité</span>
+        <select
+          class="form-select"
+          data-testid="severity-filter"
+          (change)="onSeverityChange($event)"
+        >
+          <option value="">Toutes</option>
+          @for (severite of severites; track severite) {
+            <option [value]="severite">{{ severityLabel(severite) }}</option>
+          }
+        </select>
+      </label>
+    </div>
+  </header>
+
+  @if (error(); as message) {
+    <ev-alert severity="danger" class="alert-feed__banner">{{ message }}</ev-alert>
+  }
+
+  @if (loading() && alerts().length === 0 && !error()) {
+    <p class="alert-feed__state" aria-live="polite">Chargement des alertes…</p>
+  } @else if (alerts().length === 0 && !error()) {
+    <ev-alert severity="success" class="alert-feed__banner"
+      >Aucune alerte pour ces critères.</ev-alert
+    >
+  }
+
+  <ul class="alert-feed__list" [attr.aria-busy]="loading()">
+    @for (alert of visibleAlerts(); track alert.alert_id) {
+      <li class="alert-feed__item" [class]="'alert-feed__item--' + alert.severity">
+        <span class="alert-feed__icon">
+          <ev-icon [name]="alert.type" [label]="typeLabel(alert.type)" />
+        </span>
+        <div class="alert-feed__body">
+          <div class="alert-feed__meta">
+            <ev-badge [tone]="toneFor(alert.severity)">{{
+              severityLabel(alert.severity)
+            }}</ev-badge>
+            <span class="alert-feed__type">{{ typeLabel(alert.type) }}</span>
+            <span class="alert-feed__site">{{ siteName(alert.site_id) }}</span>
+            <time class="alert-feed__time" [attr.datetime]="alert.timestamp">
+              {{ alert.timestamp | date: 'dd/MM/yyyy HH:mm' }}
+            </time>
+          </div>
+          <p class="alert-feed__message">{{ alert.message }}</p>
+          @if (alert.value !== null) {
+            <p class="alert-feed__values">
+              <strong>{{ alert.value | number: '1.0-1' }} {{ unitFor(alert.metric) }}</strong>
+              @if (alert.threshold !== null) {
+                <span
+                  >seuil {{ alert.threshold | number: '1.0-1' }} {{ unitFor(alert.metric) }}</span
+                >
+              }
+            </p>
+          }
+        </div>
+      </li>
+    }
+  </ul>
+
+  @if (hiddenCount() > 0) {
+    <ev-button
+      variant="secondary"
+      [fullWidth]="false"
+      class="alert-feed__more"
+      data-testid="show-more"
+      (click)="showMore()"
+    >
+      Afficher plus ({{ hiddenCount() }} restantes)
+    </ev-button>
+  }
+</section>

apps/frontend/src/app/shared/components/alert-feed/alert-feed.scss

@@ -0,0 +1,154 @@
+:host {
+  display: block;
+}
+
+.alert-feed__header {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: flex-end;
+  justify-content: space-between;
+  gap: var(--space-3);
+  margin-bottom: var(--space-3);
+}
+
+.alert-feed__title {
+  margin: 0;
+  font-size: var(--font-size-lg);
+  font-weight: 600;
+}
+
+.alert-feed__count {
+  margin: 0.15rem 0 0;
+  font-size: var(--font-size-sm);
+  color: var(--color-text-muted);
+}
+
+.alert-feed__filters {
+  display: flex;
+  flex-wrap: wrap;
+  gap: var(--space-2);
+}
+
+.alert-feed__filter {
+  display: flex;
+  flex-direction: column;
+  min-width: 10rem;
+
+  .form-label {
+    margin-top: 0;
+  }
+}
+
+.alert-feed__banner {
+  display: block;
+  margin-bottom: var(--space-3);
+}
+
+.alert-feed__state {
+  margin: 0 0 var(--space-3);
+  font-size: var(--font-size-sm);
+  color: var(--color-text-muted);
+}
+
+.alert-feed__list {
+  list-style: none;
+  margin: 0;
+  padding: 0;
+  display: flex;
+  flex-direction: column;
+  gap: var(--space-2);
+}
+
+.alert-feed__item {
+  display: flex;
+  gap: var(--space-3);
+  padding: var(--space-2) var(--space-3);
+  border-radius: var(--radius-md);
+  background: var(--color-surface);
+  border: 1px solid var(--color-border-light);
+  border-left: 4px solid var(--color-border);
+}
+
+.alert-feed__item--medium {
+  border-left-color: var(--color-warning);
+  background: var(--color-warning-bg);
+
+  .alert-feed__icon {
+    color: var(--color-warning-text);
+  }
+}
+
+.alert-feed__item--high {
+  border-left-color: var(--color-danger);
+  background: var(--color-danger-bg);
+
+  .alert-feed__icon {
+    color: var(--color-danger);
+  }
+}
+
+.alert-feed__item--critical {
+  border-left-color: var(--color-critical);
+  background: var(--color-danger-bg);
+
+  .alert-feed__icon {
+    color: var(--color-critical);
+  }
+}
+
+.alert-feed__icon {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  flex-shrink: 0;
+  width: 2.25rem;
+  height: 2.25rem;
+  border-radius: 50%;
+  background: var(--color-surface);
+  color: var(--color-text-muted);
+  font-size: var(--font-size-lg);
+  box-shadow: var(--shadow-card);
+}
+
+.alert-feed__body {
+  display: flex;
+  flex-direction: column;
+  gap: 0.25rem;
+  min-width: 0;
+}
+
+.alert-feed__meta {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: var(--space-2);
+  font-size: var(--font-size-xs);
+  color: var(--color-text-muted);
+}
+
+.alert-feed__type {
+  font-weight: 600;
+  color: var(--color-text);
+}
+
+.alert-feed__message {
+  margin: 0;
+  font-size: var(--font-size-sm);
+}
+
+.alert-feed__values {
+  display: flex;
+  gap: var(--space-2);
+  margin: 0;
+  font-size: var(--font-size-xs);
+  color: var(--color-text-muted);
+
+  strong {
+    color: var(--color-text);
+  }
+}
+
+.alert-feed__more {
+  display: inline-block;
+  margin-top: var(--space-3);
+}

apps/frontend/src/app/shared/components/alert-feed/alert-feed.spec.ts

@@ -0,0 +1,218 @@
+import { ComponentFixture, TestBed } from '@angular/core/testing';
+import { vi } from 'vitest';
+import { of, throwError } from 'rxjs';
+import { AlertFeed } from './alert-feed';
+import { AlertsService } from '../../../core/services/alerts.service';
+import { SitesService } from '../../../core/services/sites.service';
+import { Alert } from '../../models/alert.model';
+import { Site } from '../../models/site.model';
+
+const SITES: Site[] = [
+  {
+    site_id: 'SITE001',
+    site_name: 'Usine Nantes',
+    site_type: 'industriel',
+    location: 'Nantes',
+    capacity_kw: 500,
+    status: 'actif',
+  },
+];
+
+function alerte(surcharges: Partial<Alert> = {}): Alert {
+  return {
+    alert_id: 1,
+    site_id: 'SITE001',
+    timestamp: '2026-09-15T11:12:00Z',
+    type: 'spike',
+    severity: 'critical',
+    message: 'Variation brutale entre deux lectures consécutives',
+    value: 812.5,
+    threshold: 400,
+    metric: 'consumption_kw',
+    prediction_id: null,
+    ...surcharges,
+  };
+}
+
+function setup(
+  alertsMock: { getAlerts: ReturnType<typeof vi.fn> },
+  sitesMock: { getSites: ReturnType<typeof vi.fn> } = {
+    getSites: vi.fn().mockReturnValue(of(SITES)),
+  },
+) {
+  TestBed.configureTestingModule({
+    imports: [AlertFeed],
+    providers: [
+      { provide: AlertsService, useValue: alertsMock },
+      { provide: SitesService, useValue: sitesMock },
+    ],
+  });
+  return TestBed.createComponent(AlertFeed);
+}
+
+function premierChargement(fixture: ComponentFixture<AlertFeed>) {
+  fixture.detectChanges();
+  vi.advanceTimersByTime(1);
+  fixture.detectChanges();
+}
+
+function texte(fixture: ComponentFixture<AlertFeed>): string {
+  return (fixture.nativeElement as HTMLElement).textContent ?? '';
+}
+
+function choisir(fixture: ComponentFixture<AlertFeed>, testId: string, value: string) {
+  const select = fixture.nativeElement.querySelector(
+    `[data-testid="${testId}"]`,
+  ) as HTMLSelectElement;
+  select.value = value;
+  select.dispatchEvent(new Event('change'));
+  fixture.detectChanges();
+  vi.advanceTimersByTime(1);
+  fixture.detectChanges();
+}
+
+describe('AlertFeed', () => {
+  beforeEach(() => vi.useFakeTimers());
+  afterEach(() => vi.useRealTimers());
+
+  it('charge les alertes au démarrage sans filtre et les affiche avec leur contexte', () => {
+    const getAlerts = vi.fn().mockReturnValue(of([alerte()]));
+    const fixture = setup({ getAlerts });
+
+    premierChargement(fixture);
+
+    expect(getAlerts).toHaveBeenCalledTimes(1);
+    expect(getAlerts.mock.calls[0][0]).toEqual({});
+    const contenu = texte(fixture);
+    expect(contenu).toContain('Usine Nantes');
+    expect(contenu).toContain('Critique');
+    expect(contenu).toContain('Pic de consommation');
+    expect(contenu).toContain('15/09/2026');
+    expect(contenu).toContain('812.5 kW');
+    expect(contenu).toContain('seuil 400 kW');
+    expect(fixture.nativeElement.querySelector('ev-icon svg')).not.toBeNull();
+    expect(fixture.nativeElement.querySelector('.alert-feed__item--critical')).not.toBeNull();
+  });
+
+  it('annonce le chargement avant la première réponse', () => {
+    const fixture = setup({ getAlerts: vi.fn().mockReturnValue(of([])) });
+
+    fixture.detectChanges();
+
+    expect(fixture.componentInstance.loading()).toBe(true);
+    expect(texte(fixture)).toContain('Chargement des alertes');
+  });
+
+  it("annonce l'absence d'alerte pour les critères choisis", () => {
+    const fixture = setup({ getAlerts: vi.fn().mockReturnValue(of([])) });
+
+    premierChargement(fixture);
+
+    expect(texte(fixture)).toContain('Aucune alerte pour ces critères.');
+    expect(fixture.nativeElement.querySelectorAll('li').length).toBe(0);
+  });
+
+  it('relance la requête avec la sévérité choisie et revient à la première page', () => {
+    const getAlerts = vi.fn().mockReturnValue(of([alerte()]));
+    const fixture = setup({ getAlerts });
+    premierChargement(fixture);
+    fixture.componentInstance.showMore();
+
+    choisir(fixture, 'severity-filter', 'high');
+
+    expect(getAlerts).toHaveBeenCalledTimes(2);
+    expect(getAlerts.mock.calls[1][0]).toEqual({ severity: 'high' });
+    expect(fixture.componentInstance.visibleCount()).toBe(10);
+  });
+
+  it('relance la requête avec le site choisi dans le filtre', () => {
+    const getAlerts = vi.fn().mockReturnValue(of([]));
+    const fixture = setup({ getAlerts });
+    premierChargement(fixture);
+
+    choisir(fixture, 'site-filter', 'SITE001');
+
+    expect(getAlerts.mock.calls[1][0]).toEqual({ site_id: 'SITE001' });
+  });
+
+  it('masque le filtre site et force site_id quand le parent fixe le site', () => {
+    const getAlerts = vi.fn().mockReturnValue(of([]));
+    const fixture = setup({ getAlerts });
+    fixture.componentRef.setInput('siteId', 'SITE001');
+
+    premierChargement(fixture);
+
+    expect(getAlerts.mock.calls[0][0]).toEqual({ site_id: 'SITE001' });
+    expect(fixture.nativeElement.querySelector('[data-testid="site-filter"]')).toBeNull();
+    expect(fixture.nativeElement.querySelector('[data-testid="severity-filter"]')).not.toBeNull();
+  });
+
+  it("signale l'indisponibilité en gardant la liste, puis repart au rafraîchissement suivant", () => {
+    const getAlerts = vi
+      .fn()
+      .mockReturnValueOnce(of([alerte()]))
+      .mockReturnValueOnce(throwError(() => new Error('API injoignable')))
+      .mockReturnValue(of([alerte(), alerte({ alert_id: 2 })]));
+    const fixture = setup({ getAlerts });
+    premierChargement(fixture);
+
+    vi.advanceTimersByTime(60_000);
+    fixture.detectChanges();
+
+    expect(getAlerts).toHaveBeenCalledTimes(2);
+    expect(fixture.componentInstance.error()).not.toBeNull();
+    expect(fixture.componentInstance.alerts().length).toBe(1);
+    expect(texte(fixture)).toContain('Alertes indisponibles');
+
+    vi.advanceTimersByTime(60_000);
+    fixture.detectChanges();
+
+    expect(getAlerts).toHaveBeenCalledTimes(3);
+    expect(fixture.componentInstance.error()).toBeNull();
+    expect(fixture.componentInstance.alerts().length).toBe(2);
+  });
+
+  it('pagine côté client par dix et dévoile le reste à la demande', () => {
+    const alertes = Array.from({ length: 25 }, (_, i) => alerte({ alert_id: i + 1 }));
+    const fixture = setup({ getAlerts: vi.fn().mockReturnValue(of(alertes)) });
+    premierChargement(fixture);
+
+    expect(fixture.nativeElement.querySelectorAll('li').length).toBe(10);
+    expect(texte(fixture)).toContain('Afficher plus (15 restantes)');
+
+    fixture.nativeElement.querySelector('[data-testid="show-more"]').click();
+    fixture.detectChanges();
+    expect(fixture.nativeElement.querySelectorAll('li').length).toBe(20);
+
+    fixture.nativeElement.querySelector('[data-testid="show-more"]').click();
+    fixture.detectChanges();
+    expect(fixture.nativeElement.querySelectorAll('li').length).toBe(25);
+    expect(fixture.nativeElement.querySelector('[data-testid="show-more"]')).toBeNull();
+  });
+
+  it("replie sur l'identifiant quand le site est inconnu ou que la liste des sites échoue", () => {
+    const fixture = setup(
+      { getAlerts: vi.fn().mockReturnValue(of([alerte({ site_id: 'SITE999' })])) },
+      { getSites: vi.fn().mockReturnValue(throwError(() => new Error('nope'))) },
+    );
+
+    premierChargement(fixture);
+
+    expect(texte(fixture)).toContain('SITE999');
+  });
+
+  it("n'affiche pas de mesure pour une alerte sans valeur", () => {
+    const fixture = setup({
+      getAlerts: vi
+        .fn()
+        .mockReturnValue(
+          of([alerte({ type: 'outage', value: null, threshold: null, metric: null })]),
+        ),
+    });
+
+    premierChargement(fixture);
+
+    expect(fixture.nativeElement.querySelector('.alert-feed__values')).toBeNull();
+    expect(texte(fixture)).toContain('Coupure');
+  });
+});

apps/frontend/src/app/shared/components/alert-feed/alert-feed.ts

@@ -0,0 +1,130 @@
+import { Component, DestroyRef, computed, inject, input, signal } from '@angular/core';
+import { takeUntilDestroyed, toObservable, toSignal } from '@angular/core/rxjs-interop';
+import { DatePipe, DecimalPipe } from '@angular/common';
+import { catchError, EMPTY, Observable, of, switchMap, tap, timer } from 'rxjs';
+import { AlertFilters, AlertsService } from '../../../core/services/alerts.service';
+import { SitesService } from '../../../core/services/sites.service';
+import { Alert, AlertMetric, AlertSeverity, AlertType } from '../../models/alert.model';
+import { Site } from '../../models/site.model';
+import {
+  LIBELLE_PAR_SEVERITE,
+  LIBELLE_PAR_TYPE,
+  SEVERITES,
+  TON_PAR_SEVERITE,
+  UNITE_PAR_METRIQUE,
+} from '../../models/alert-presentation';
+import { Badge, BadgeTone } from '../ui/badge/badge';
+import { Button } from '../ui/button/button';
+import { Alert as EvAlert } from '../ui/alert/alert';
+import { Icon } from '../ui/icon/icon';
+
+// Le DAG de détection tourne toutes les heures : une minute suffit largement pour suivre le flux.
+const REFRESH_INTERVAL_MS = 60_000;
+const PAGE_SIZE = 10;
+const UNAVAILABLE_MESSAGE = 'Alertes indisponibles, la liste affichée date du dernier chargement.';
+
+@Component({
+  selector: 'app-alert-feed',
+  standalone: true,
+  imports: [DatePipe, DecimalPipe, Badge, Button, EvAlert, Icon],
+  templateUrl: './alert-feed.html',
+  styleUrl: './alert-feed.scss',
+})
+export class AlertFeed {
+  private alertsService = inject(AlertsService);
+  private sitesService = inject(SitesService);
+  private destroyRef = inject(DestroyRef);
+
+  siteId = input<string | null>(null);
+
+  readonly severites = SEVERITES;
+  severity = signal<AlertSeverity | null>(null);
+  siteFilter = signal<string | null>(null);
+
+  alerts = signal<Alert[]>([]);
+  loading = signal(true);
+  error = signal<string | null>(null);
+  visibleCount = signal(PAGE_SIZE);
+
+  sites = toSignal(this.sitesService.getSites().pipe(catchError(() => of([] as Site[]))), {
+    initialValue: [] as Site[],
+  });
+
+  private filters = computed<AlertFilters>(() => ({
+    site_id: this.siteId() ?? this.siteFilter() ?? undefined,
+    severity: this.severity() ?? undefined,
+  }));
+
+  private siteNameById = computed(
+    () => new Map(this.sites().map((site) => [site.site_id, site.site_name])),
+  );
+
+  visibleAlerts = computed(() => this.alerts().slice(0, this.visibleCount()));
+  hiddenCount = computed(() => Math.max(this.alerts().length - this.visibleCount(), 0));
+
+  constructor() {
+    toObservable(this.filters)
+      .pipe(
+        tap(() => {
+          this.loading.set(true);
+          this.visibleCount.set(PAGE_SIZE);
+        }),
+        // Piège : catchError sur l'observable interne ; sur le flux externe il terminerait le
+        // timer et le rafraîchissement ne repartirait jamais.
+        switchMap((filters) =>
+          timer(0, REFRESH_INTERVAL_MS).pipe(
+            switchMap(() =>
+              this.alertsService
+                .getAlerts(filters)
+                .pipe(catchError(() => this.reportUnavailable())),
+            ),
+          ),
+        ),
+        takeUntilDestroyed(this.destroyRef),
+      )
+      .subscribe((alerts) => {
+        this.loading.set(false);
+        this.error.set(null);
+        this.alerts.set(alerts);
+      });
+  }
+
+  onSiteChange(event: Event): void {
+    this.siteFilter.set((event.target as HTMLSelectElement).value || null);
+  }
+
+  onSeverityChange(event: Event): void {
+    const value = (event.target as HTMLSelectElement).value;
+    this.severity.set(value ? (value as AlertSeverity) : null);
+  }
+
+  showMore(): void {
+    this.visibleCount.update((count) => count + PAGE_SIZE);
+  }
+
+  toneFor(severity: AlertSeverity): BadgeTone {
+    return TON_PAR_SEVERITE[severity];
+  }
+
+  severityLabel(severity: AlertSeverity): string {
+    return LIBELLE_PAR_SEVERITE[severity];
+  }
+
+  typeLabel(type: AlertType): string {
+    return LIBELLE_PAR_TYPE[type];
+  }
+
+  siteName(siteId: string): string {
+    return this.siteNameById().get(siteId) ?? siteId;
+  }
+
+  unitFor(metric: AlertMetric | null): string {
+    return metric ? UNITE_PAR_METRIQUE[metric] : '';
+  }
+
+  private reportUnavailable(): Observable<never> {
+    this.loading.set(false);
+    this.error.set(UNAVAILABLE_MESSAGE);
+    return EMPTY;
+  }
+}

apps/frontend/src/app/shared/components/recommendation-list/recommendation-list.html

@@ -0,0 +1,48 @@
+@if (error(); as message) {
+  <ev-alert severity="danger" class="reco-list__banner">{{ message }}</ev-alert>
+} @else if (loading() && !hasData()) {
+  <p class="reco-list__state" aria-live="polite">Chargement des recommandations…</p>
+} @else if (visibleGroups().length === 0) {
+  <ev-alert severity="success" class="reco-list__banner">{{ emptyMessage() }}</ev-alert>
+}
+
+<div class="reco-list" [attr.aria-busy]="loading()">
+  @for (group of visibleGroups(); track group.alert.alert_id) {
+    <ev-card
+      class="reco-group"
+      [class.reco-group--focus]="group.alert.alert_id === alertId()"
+      [id]="'alerte-' + group.alert.alert_id"
+    >
+      <header class="reco-group__alert">
+        <div class="reco-group__meta">
+          <ev-badge [tone]="toneFor(group.alert.severity)">{{
+            severityLabel(group.alert.severity)
+          }}</ev-badge>
+          <span class="reco-group__type">{{ typeLabel(group.alert.type) }}</span>
+          @if (!siteId()) {
+            <a [routerLink]="['/sites', group.alert.site_id]" class="ev-link">{{
+              group.siteName
+            }}</a>
+          }
+          <time [attr.datetime]="group.alert.timestamp">{{
+            group.alert.timestamp | date: 'dd/MM/yyyy HH:mm'
+          }}</time>
+        </div>
+        <p class="reco-group__message">{{ group.alert.message }}</p>
+      </header>
+      <ol class="reco-group__items">
+        @for (reco of group.recommendations; track reco.recommendation_id) {
+          <li class="reco">
+            <div class="reco__head">
+              <strong class="reco__action">{{ reco.action }}</strong>
+              <ev-badge [tone]="ruleTone(reco.rule_reference)">{{
+                ruleLabel(reco.rule_reference)
+              }}</ev-badge>
+            </div>
+            <p class="reco__explanation">{{ reco.explanation }}</p>
+          </li>
+        }
+      </ol>
+    </ev-card>
+  }
+</div>

apps/frontend/src/app/shared/components/recommendation-list/recommendation-list.scss

@@ -0,0 +1,94 @@
+:host {
+  display: block;
+}
+
+.reco-list__banner {
+  display: block;
+  margin-bottom: var(--space-3);
+}
+
+.reco-list__state {
+  margin: 0 0 var(--space-3);
+  font-size: 0.9rem;
+  color: var(--color-text-muted);
+}
+
+.reco-list {
+  display: flex;
+  flex-direction: column;
+  gap: var(--space-3);
+}
+
+.reco-group {
+  padding: var(--space-4);
+  gap: var(--space-3);
+}
+
+.reco-group--focus {
+  border-color: var(--color-primary);
+  box-shadow: 0 0 0 3px var(--color-primary-light);
+}
+
+.reco-group__alert {
+  display: flex;
+  flex-direction: column;
+  gap: var(--space-1);
+  padding-bottom: var(--space-3);
+  border-bottom: 1px solid var(--color-border-light);
+}
+
+.reco-group__meta {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: var(--space-2);
+  font-size: 0.8rem;
+  color: var(--color-text-muted);
+}
+
+.reco-group__type {
+  font-weight: 600;
+  color: var(--color-text);
+}
+
+.reco-group__message {
+  margin: 0;
+  font-size: 0.9rem;
+}
+
+.reco-group__items {
+  list-style: none;
+  margin: 0;
+  padding: 0;
+  display: flex;
+  flex-direction: column;
+  gap: var(--space-2);
+}
+
+.reco {
+  display: flex;
+  flex-direction: column;
+  gap: 0.25rem;
+  padding: var(--space-2) var(--space-3);
+  border-radius: var(--radius-sm);
+  background: var(--color-bg);
+  border-left: 3px solid var(--color-primary);
+}
+
+.reco__head {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  justify-content: space-between;
+  gap: var(--space-2);
+}
+
+.reco__action {
+  font-size: 0.95rem;
+}
+
+.reco__explanation {
+  margin: 0;
+  font-size: 0.85rem;
+  color: var(--color-text-muted);
+}

apps/frontend/src/app/shared/components/recommendation-list/recommendation-list.spec.ts

@@ -0,0 +1,244 @@
+import { ComponentFixture, TestBed } from '@angular/core/testing';
+import { provideRouter } from '@angular/router';
+import { vi } from 'vitest';
+import { NEVER, of, throwError } from 'rxjs';
+import { RecommendationList, joinByAlert } from './recommendation-list';
+import { AlertsService } from '../../../core/services/alerts.service';
+import { RecommendationsService } from '../../../core/services/recommendations.service';
+import { Alert } from '../../models/alert.model';
+import { Recommendation } from '../../models/recommendation.model';
+import { Site } from '../../models/site.model';
+
+const SITES: Site[] = [
+  {
+    site_id: 'SITE001',
+    site_name: 'Usine Nantes',
+    site_type: 'industriel',
+    location: 'Nantes',
+    capacity_kw: 500,
+    status: 'actif',
+  },
+  {
+    site_id: 'SITE002',
+    site_name: 'Bureau Lille',
+    site_type: 'bureau',
+    location: 'Lille',
+    capacity_kw: 80,
+    status: 'actif',
+  },
+];
+
+function alerte(surcharges: Partial<Alert>): Alert {
+  return {
+    alert_id: 1,
+    site_id: 'SITE001',
+    timestamp: '2026-09-15T09:00:00Z',
+    type: 'threshold',
+    severity: 'high',
+    message: 'Puissance appelée au-dessus de la capacité du site',
+    value: 812.5,
+    threshold: 720,
+    metric: 'consumption_kw',
+    prediction_id: null,
+    ...surcharges,
+  };
+}
+
+function reco(surcharges: Partial<Recommendation>): Recommendation {
+  return {
+    recommendation_id: 1,
+    alert_id: 1,
+    action: 'Ramener la puissance appelée sous le seuil contractuel',
+    explanation: 'Seuil de consommation dépassé sur le site SITE001.',
+    rule_reference: 'threshold-reduction-v1',
+    created_at: '2026-09-15T09:05:00Z',
+    ...surcharges,
+  };
+}
+
+const ALERTES: Alert[] = [
+  alerte({ alert_id: 1, site_id: 'SITE001', timestamp: '2026-09-15T09:00:00Z' }),
+  alerte({
+    alert_id: 2,
+    site_id: 'SITE002',
+    timestamp: '2026-09-15T11:00:00Z',
+    severity: 'critical',
+    type: 'spike',
+    message: 'Variation brutale entre deux lectures consécutives',
+  }),
+  alerte({ alert_id: 3, site_id: 'SITE001', timestamp: '2026-09-15T10:00:00Z', severity: 'low' }),
+];
+
+const RECOMMANDATIONS: Recommendation[] = [
+  reco({
+    recommendation_id: 3,
+    alert_id: 2,
+    action: "Escalader à l'astreinte sous une heure",
+    rule_reference: 'escalade-astreinte-v1',
+  }),
+  reco({ recommendation_id: 1, alert_id: 1 }),
+  reco({
+    recommendation_id: 2,
+    alert_id: 2,
+    action: 'Délester les équipements non prioritaires sur le créneau du pic',
+    rule_reference: 'spike-delestage-v1',
+  }),
+  reco({ recommendation_id: 4, alert_id: 99, rule_reference: 'orpheline-v1' }),
+];
+
+function setup(
+  alertsMock: { getAlerts: ReturnType<typeof vi.fn> },
+  recosMock: { getRecommendations: ReturnType<typeof vi.fn> },
+  inputs: Record<string, unknown> = {},
+) {
+  TestBed.configureTestingModule({
+    imports: [RecommendationList],
+    providers: [
+      provideRouter([]),
+      { provide: AlertsService, useValue: alertsMock },
+      { provide: RecommendationsService, useValue: recosMock },
+    ],
+  });
+  const fixture = TestBed.createComponent(RecommendationList);
+  for (const [nom, valeur] of Object.entries(inputs)) {
+    fixture.componentRef.setInput(nom, valeur);
+  }
+  return fixture;
+}
+
+function rendre(fixture: ComponentFixture<RecommendationList>) {
+  fixture.detectChanges();
+  fixture.detectChanges();
+}
+
+function texte(fixture: ComponentFixture<RecommendationList>): string {
+  return (fixture.nativeElement as HTMLElement).textContent ?? '';
+}
+
+const recosOk = () => ({ getRecommendations: vi.fn().mockReturnValue(of(RECOMMANDATIONS)) });
+
+describe('joinByAlert', () => {
+  it('groupe par alerte, du plus récent au plus ancien, recommandations par identifiant', () => {
+    const groupes = joinByAlert(ALERTES, RECOMMANDATIONS, new Map([['SITE001', 'Usine Nantes']]));
+
+    expect(groupes.map((g) => g.alert.alert_id)).toEqual([2, 1]);
+    expect(groupes[0].recommendations.map((r) => r.recommendation_id)).toEqual([2, 3]);
+    expect(groupes[1].siteName).toBe('Usine Nantes');
+    expect(groupes[0].siteName).toBe('SITE002');
+  });
+
+  it('ignore les alertes sans recommandation et les recommandations orphelines', () => {
+    const groupes = joinByAlert(ALERTES, RECOMMANDATIONS, new Map());
+
+    expect(groupes.some((g) => g.alert.alert_id === 3)).toBe(false);
+    expect(groupes.flatMap((g) => g.recommendations).some((r) => r.alert_id === 99)).toBe(false);
+  });
+});
+
+describe('RecommendationList', () => {
+  it('charge alertes et recommandations puis affiche les groupes avec leur contexte', () => {
+    const getAlerts = vi.fn().mockReturnValue(of(ALERTES));
+    const fixture = setup({ getAlerts }, recosOk(), { sites: SITES });
+
+    rendre(fixture);
+
+    expect(getAlerts).toHaveBeenCalledWith({});
+    expect(fixture.nativeElement.querySelectorAll('.reco-group').length).toBe(2);
+    const contenu = texte(fixture);
+    expect(contenu).toContain('Usine Nantes');
+    expect(contenu).toContain('Bureau Lille');
+    expect(contenu).toContain('Critique');
+    expect(contenu).toContain('Pic de consommation');
+    expect(contenu).toContain('Escalade astreinte');
+    expect(contenu).toContain('Délester les équipements');
+    expect(contenu).toContain('15/09/2026');
+    expect(fixture.nativeElement.querySelector('a[href="/sites/SITE002"]')).not.toBeNull();
+    expect(fixture.componentInstance.total()).toBe(3);
+    expect(fixture.componentInstance.error()).toBeNull();
+  });
+
+  it('filtre les alertes du site côté API et masque le lien vers le site', () => {
+    const getAlerts = vi.fn().mockReturnValue(of(ALERTES.filter((a) => a.site_id === 'SITE001')));
+    const fixture = setup({ getAlerts }, recosOk(), { siteId: 'SITE001', sites: SITES });
+
+    rendre(fixture);
+
+    expect(getAlerts).toHaveBeenCalledWith({ site_id: 'SITE001' });
+    expect(fixture.nativeElement.querySelectorAll('.reco-group').length).toBe(1);
+    expect(fixture.nativeElement.querySelector('a[href^="/sites/"]')).toBeNull();
+  });
+
+  it("ne garde que le groupe de l'alerte ciblée et le met en évidence", () => {
+    const fixture = setup({ getAlerts: vi.fn().mockReturnValue(of(ALERTES)) }, recosOk(), {
+      alertId: 2,
+    });
+
+    rendre(fixture);
+
+    const groupes = fixture.nativeElement.querySelectorAll('.reco-group');
+    expect(groupes.length).toBe(1);
+    expect(groupes[0].classList.contains('reco-group--focus')).toBe(true);
+    expect(groupes[0].id).toBe('alerte-2');
+  });
+
+  it("annonce l'absence de recommandation pour une alerte inconnue", () => {
+    const fixture = setup({ getAlerts: vi.fn().mockReturnValue(of(ALERTES)) }, recosOk(), {
+      alertId: 123,
+    });
+
+    rendre(fixture);
+
+    expect(texte(fixture)).toContain('Aucune recommandation pour cette alerte.');
+  });
+
+  it("annonce l'absence de recommandation pour le site consulté", () => {
+    const fixture = setup(
+      { getAlerts: vi.fn().mockReturnValue(of([])) },
+      { getRecommendations: vi.fn().mockReturnValue(of([])) },
+      { siteId: 'SITE001' },
+    );
+
+    rendre(fixture);
+
+    expect(texte(fixture)).toContain('Aucune recommandation pour ce site.');
+  });
+
+  it("signale l'indisponibilité et n'affiche aucun groupe si un des deux appels échoue", () => {
+    const fixture = setup(
+      { getAlerts: vi.fn().mockReturnValue(of(ALERTES)) },
+      { getRecommendations: vi.fn().mockReturnValue(throwError(() => new Error('nope'))) },
+    );
+
+    rendre(fixture);
+
+    expect(fixture.componentInstance.error()).not.toBeNull();
+    expect(fixture.componentInstance.groups()).toEqual([]);
+    expect(texte(fixture)).toContain('Recommandations indisponibles');
+    expect(fixture.nativeElement.querySelectorAll('.reco-group').length).toBe(0);
+  });
+
+  it('annonce le chargement tant que la réponse ne vient pas', () => {
+    const fixture = setup(
+      { getAlerts: vi.fn().mockReturnValue(NEVER) },
+      { getRecommendations: vi.fn().mockReturnValue(NEVER) },
+    );
+
+    rendre(fixture);
+
+    expect(fixture.componentInstance.loading()).toBe(true);
+    expect(texte(fixture)).toContain('Chargement des recommandations');
+  });
+
+  it('recharge les deux flux à la demande', () => {
+    const getAlerts = vi.fn().mockReturnValue(of(ALERTES));
+    const recos = recosOk();
+    const fixture = setup({ getAlerts }, recos);
+    rendre(fixture);
+
+    fixture.componentInstance.reload();
+    rendre(fixture);
+
+    expect(getAlerts).toHaveBeenCalledTimes(2);
+    expect(recos.getRecommendations).toHaveBeenCalledTimes(2);
+  });
+});

apps/frontend/src/app/shared/components/recommendation-list/recommendation-list.ts

@@ -0,0 +1,163 @@
+import { Component, DestroyRef, computed, inject, input, signal } from '@angular/core';
+import { takeUntilDestroyed, toObservable } from '@angular/core/rxjs-interop';
+import { DatePipe } from '@angular/common';
+import { RouterLink } from '@angular/router';
+import { catchError, EMPTY, forkJoin, Observable, switchMap, tap } from 'rxjs';
+import { AlertsService } from '../../../core/services/alerts.service';
+import { RecommendationsService } from '../../../core/services/recommendations.service';
+import { Alert, AlertSeverity, AlertType } from '../../models/alert.model';
+import { Recommendation } from '../../models/recommendation.model';
+import { Site } from '../../models/site.model';
+import {
+  LIBELLE_PAR_SEVERITE,
+  LIBELLE_PAR_TYPE,
+  TON_PAR_SEVERITE,
+} from '../../models/alert-presentation';
+import { libelleRegle, tonRegle } from '../../models/recommendation-presentation';
+import { Card } from '../ui/card/card';
+import { Badge, BadgeTone } from '../ui/badge/badge';
+import { Alert as EvAlert } from '../ui/alert/alert';
+
+const UNAVAILABLE_MESSAGE = 'Recommandations indisponibles, réessayez plus tard.';
+
+export interface RecommendedAlertView {
+  alert: Alert;
+  siteName: string;
+  recommendations: Recommendation[];
+}
+
+interface Chargement {
+  alerts: Alert[];
+  recommendations: Recommendation[];
+}
+
+// Pourquoi : une recommandation ne porte que alert_id, jamais site_id, et /recommendations n'a
+// aucun filtre ; la jointure se fait ici, en O(alertes), acceptable à la taille du jeu de données.
+export function joinByAlert(
+  alerts: Alert[],
+  recommendations: Recommendation[],
+  siteNames: Map<string, string>,
+): RecommendedAlertView[] {
+  const parAlerte = new Map<number, Recommendation[]>();
+  for (const recommandation of recommendations) {
+    const liste = parAlerte.get(recommandation.alert_id) ?? [];
+    liste.push(recommandation);
+    parAlerte.set(recommandation.alert_id, liste);
+  }
+  return alerts
+    .filter((alert) => parAlerte.has(alert.alert_id))
+    .map((alert) => ({
+      alert,
+      siteName: siteNames.get(alert.site_id) ?? alert.site_id,
+      recommendations: [...(parAlerte.get(alert.alert_id) ?? [])].sort(
+        (a, b) => a.recommendation_id - b.recommendation_id,
+      ),
+    }))
+    .sort((a, b) => Date.parse(b.alert.timestamp) - Date.parse(a.alert.timestamp));
+}
+
+@Component({
+  selector: 'app-recommendation-list',
+  standalone: true,
+  imports: [DatePipe, RouterLink, Card, Badge, EvAlert],
+  templateUrl: './recommendation-list.html',
+  styleUrl: './recommendation-list.scss',
+})
+export class RecommendationList {
+  private alertsService = inject(AlertsService);
+  private recommendationsService = inject(RecommendationsService);
+  private destroyRef = inject(DestroyRef);
+
+  siteId = input<string | null>(null);
+  alertId = input<number | null>(null);
+  sites = input<Site[]>([]);
+
+  private data = signal<Chargement | null>(null);
+  private reloadTick = signal(0);
+  loading = signal(true);
+  error = signal<string | null>(null);
+
+  private trigger = computed(() => ({ siteId: this.siteId(), tick: this.reloadTick() }));
+
+  private siteNameById = computed(
+    () => new Map(this.sites().map((site) => [site.site_id, site.site_name])),
+  );
+
+  hasData = computed(() => this.data() !== null);
+
+  groups = computed<RecommendedAlertView[]>(() => {
+    const data = this.data();
+    return data ? joinByAlert(data.alerts, data.recommendations, this.siteNameById()) : [];
+  });
+
+  visibleGroups = computed(() => {
+    const alertId = this.alertId();
+    const groups = this.groups();
+    return alertId === null ? groups : groups.filter((group) => group.alert.alert_id === alertId);
+  });
+
+  total = computed(() =>
+    this.visibleGroups().reduce((somme, group) => somme + group.recommendations.length, 0),
+  );
+
+  emptyMessage = computed(() => {
+    if (this.alertId() !== null) {
+      return 'Aucune recommandation pour cette alerte.';
+    }
+    return this.siteId()
+      ? 'Aucune recommandation pour ce site.'
+      : 'Aucune recommandation pour le moment.';
+  });
+
+  constructor() {
+    toObservable(this.trigger)
+      .pipe(
+        tap(() => this.loading.set(true)),
+        switchMap(({ siteId }) =>
+          forkJoin({
+            alerts: this.alertsService.getAlerts(siteId ? { site_id: siteId } : {}),
+            recommendations: this.recommendationsService.getRecommendations(),
+          }).pipe(catchError(() => this.reportUnavailable())),
+        ),
+        takeUntilDestroyed(this.destroyRef),
+      )
+      .subscribe((data) => {
+        this.loading.set(false);
+        this.error.set(null);
+        this.data.set(data);
+      });
+  }
+
+  reload(): void {
+    this.reloadTick.update((tick) => tick + 1);
+  }
+
+  toneFor(severity: AlertSeverity): BadgeTone {
+    return TON_PAR_SEVERITE[severity];
+  }
+
+  severityLabel(severity: AlertSeverity): string {
+    return LIBELLE_PAR_SEVERITE[severity];
+  }
+
+  typeLabel(type: AlertType): string {
+    return LIBELLE_PAR_TYPE[type];
+  }
+
+  ruleLabel(reference: string): string {
+    return libelleRegle(reference);
+  }
+
+  ruleTone(reference: string): BadgeTone {
+    return tonRegle(reference);
+  }
+
+  // Piège : vider les données avec l'erreur ; une demi-jointure (alertes sans recommandations,
+  // ou l'inverse) afficherait des groupes faux plutôt que rien.
+  private reportUnavailable(): Observable<never> {
+    this.loading.set(false);
+    this.error.set(UNAVAILABLE_MESSAGE);
+    this.data.set(null);
+    return EMPTY;
+  }
+}

apps/frontend/src/app/shared/components/ui/icon/icon.html

@@ -0,0 +1,37 @@
+<svg
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="currentColor"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+  focusable="false"
+  [attr.role]="label() ? 'img' : null"
+  [attr.aria-label]="label()"
+  [attr.aria-hidden]="label() ? null : 'true'"
+>
+  @switch (name()) {
+    @case ('spike') {
+      <polyline points="3 17 8 12 11 15 21 5" />
+      <polyline points="15 5 21 5 21 11" />
+    }
+    @case ('threshold') {
+      <line x1="3" y1="10" x2="21" y2="10" stroke-dasharray="3 3" />
+      <polyline points="4 19 9 13 13 15 20 6" />
+    }
+    @case ('anomaly') {
+      <polyline points="3 12 7 12 9.5 6 12.5 18 15 12 21 12" />
+    }
+    @case ('outage') {
+      <path d="M7.5 6.5a7 7 0 1 0 9 0" />
+      <line x1="12" y1="3" x2="12" y2="11" />
+    }
+    @case ('sensor') {
+      <circle cx="12" cy="13" r="2" />
+      <path d="M8.5 9.5a5 5 0 0 0 0 7" />
+      <path d="M15.5 9.5a5 5 0 0 1 0 7" />
+      <path d="M5.5 6.5a9 9 0 0 0 0 13" />
+      <path d="M18.5 6.5a9 9 0 0 1 0 13" />
+    }
+  }
+</svg>

apps/frontend/src/app/shared/components/ui/icon/icon.scss

@@ -0,0 +1,12 @@
+:host {
+  display: inline-flex;
+  flex-shrink: 0;
+  width: 1em;
+  height: 1em;
+  vertical-align: -0.125em;
+}
+
+svg {
+  width: 100%;
+  height: 100%;
+}

apps/frontend/src/app/shared/components/ui/icon/icon.spec.ts

@@ -0,0 +1,49 @@
+import { TestBed } from '@angular/core/testing';
+import { Icon, IconName } from './icon';
+
+const NOMS: IconName[] = ['spike', 'threshold', 'anomaly', 'outage', 'sensor'];
+
+function rendre(name: IconName, label: string | null = null) {
+  const fixture = TestBed.createComponent(Icon);
+  fixture.componentRef.setInput('name', name);
+  fixture.componentRef.setInput('label', label);
+  fixture.detectChanges();
+  return fixture.nativeElement as HTMLElement;
+}
+
+describe('Icon', () => {
+  beforeEach(() => {
+    TestBed.configureTestingModule({ imports: [Icon] });
+  });
+
+  it('dessine un tracé distinct pour chacun des cinq types', () => {
+    const traces = NOMS.map((name) => rendre(name).querySelector('svg')?.innerHTML.trim());
+
+    for (const trace of traces) {
+      expect(trace).toBeTruthy();
+    }
+    expect(new Set(traces).size).toBe(NOMS.length);
+  });
+
+  it('reste décoratif sans libellé', () => {
+    const svg = rendre('spike').querySelector('svg');
+
+    expect(svg?.getAttribute('aria-hidden')).toBe('true');
+    expect(svg?.hasAttribute('role')).toBe(false);
+  });
+
+  it('expose un rôle image et un libellé accessible quand on lui en donne un', () => {
+    const svg = rendre('outage', 'Coupure').querySelector('svg');
+
+    expect(svg?.getAttribute('role')).toBe('img');
+    expect(svg?.getAttribute('aria-label')).toBe('Coupure');
+    expect(svg?.hasAttribute('aria-hidden')).toBe(false);
+  });
+
+  it('hérite de la couleur du parent via currentColor', () => {
+    const svg = rendre('sensor').querySelector('svg');
+
+    expect(svg?.getAttribute('stroke')).toBe('currentColor');
+    expect(svg?.getAttribute('fill')).toBe('none');
+  });
+});

apps/frontend/src/app/shared/components/ui/icon/icon.ts

@@ -0,0 +1,14 @@
+import { Component, input } from '@angular/core';
+
+export type IconName = 'spike' | 'threshold' | 'anomaly' | 'outage' | 'sensor';
+
+@Component({
+  selector: 'ev-icon',
+  standalone: true,
+  templateUrl: './icon.html',
+  styleUrl: './icon.scss',
+})
+export class Icon {
+  name = input.required<IconName>();
+  label = input<string | null>(null);
+}

apps/frontend/src/app/shared/models/alert-presentation.spec.ts

@@ -0,0 +1,37 @@
+import {
+  LIBELLE_PAR_SEVERITE,
+  LIBELLE_PAR_TYPE,
+  SEVERITES,
+  TON_PAR_SEVERITE,
+  TYPES_ALERTE,
+  UNITE_PAR_METRIQUE,
+} from './alert-presentation';
+
+describe('alert-presentation', () => {
+  it('distingue le ton des sévérités high et critical', () => {
+    expect(TON_PAR_SEVERITE.high).toBe('danger');
+    expect(TON_PAR_SEVERITE.critical).toBe('critical');
+    expect(TON_PAR_SEVERITE.high).not.toBe(TON_PAR_SEVERITE.critical);
+  });
+
+  it("n'affiche pas une alerte faible avec le ton de succès", () => {
+    expect(TON_PAR_SEVERITE.low).toBe('neutral');
+    expect(TON_PAR_SEVERITE.medium).toBe('warning');
+  });
+
+  it('donne un libellé français à chaque sévérité et à chaque type', () => {
+    for (const severite of SEVERITES) {
+      expect(LIBELLE_PAR_SEVERITE[severite]).toBeTruthy();
+    }
+    for (const type of TYPES_ALERTE) {
+      expect(LIBELLE_PAR_TYPE[type]).toBeTruthy();
+    }
+    expect(SEVERITES.length).toBe(4);
+    expect(TYPES_ALERTE.length).toBe(5);
+  });
+
+  it('associe une unité à chaque métrique du contrat', () => {
+    expect(UNITE_PAR_METRIQUE.consumption_kw).toBe('kW');
+    expect(UNITE_PAR_METRIQUE.consumption_kwh).toBe('kWh');
+  });
+});

apps/frontend/src/app/shared/models/alert-presentation.ts

@@ -0,0 +1,41 @@
+import { BadgeTone } from '../components/ui/badge/badge';
+import { AlertMetric, AlertSeverity, AlertType } from './alert.model';
+
+// Pourquoi : `low` en neutre plutôt qu'en vert, une alerte faible reste une alerte ; le vert se
+// lisait comme « tout va bien » à côté des rouges.
+export const TON_PAR_SEVERITE: Record<AlertSeverity, BadgeTone> = {
+  low: 'neutral',
+  medium: 'warning',
+  high: 'danger',
+  critical: 'critical',
+};
+
+export const LIBELLE_PAR_SEVERITE: Record<AlertSeverity, string> = {
+  low: 'Faible',
+  medium: 'Moyenne',
+  high: 'Élevée',
+  critical: 'Critique',
+};
+
+export const LIBELLE_PAR_TYPE: Record<AlertType, string> = {
+  spike: 'Pic de consommation',
+  threshold: 'Seuil dépassé',
+  anomaly: 'Anomalie',
+  outage: 'Coupure',
+  sensor: 'Capteur',
+};
+
+export const UNITE_PAR_METRIQUE: Record<AlertMetric, string> = {
+  consumption_kw: 'kW',
+  consumption_kwh: 'kWh',
+};
+
+export const SEVERITES: readonly AlertSeverity[] = ['low', 'medium', 'high', 'critical'];
+
+export const TYPES_ALERTE: readonly AlertType[] = [
+  'spike',
+  'threshold',
+  'anomaly',
+  'outage',
+  'sensor',
+];

apps/frontend/src/app/shared/models/alert.model.ts

@@ -1,13 +1,16 @@
 export type AlertSeverity = 'low' | 'medium' | 'high' | 'critical';
 export type AlertType = 'spike' | 'threshold' | 'anomaly' | 'outage' | 'sensor';
+export type AlertMetric = 'consumption_kw' | 'consumption_kwh';
 
 export interface Alert {
-  alert_id: string;
+  alert_id: number;
-  timestamp: string;
   site_id: string;
-  severity: AlertSeverity;
+  timestamp: string;
   type: AlertType;
+  severity: AlertSeverity;
   message: string;
-  value: number;
+  value: number | null;
-  threshold: number;
+  threshold: number | null;
+  metric: AlertMetric | null;
+  prediction_id: number | null;
 }

apps/frontend/src/app/shared/models/recommendation-presentation.spec.ts

@@ -0,0 +1,23 @@
+import { libelleRegle, tonRegle } from './recommendation-presentation';
+
+describe('recommendation-presentation', () => {
+  it('traduit les sept règles connues du moteur', () => {
+    expect(libelleRegle('spike-delestage-v1')).toBe('Délestage');
+    expect(libelleRegle('threshold-reduction-v1')).toBe('Réduction de puissance');
+    expect(libelleRegle('outage-secours-v1')).toBe('Alimentation de secours');
+    expect(libelleRegle('sensor-maintenance-v1')).toBe('Maintenance capteur');
+    expect(libelleRegle('anomaly-verification-v1')).toBe('Vérification');
+    expect(libelleRegle('escalade-astreinte-v1')).toBe('Escalade astreinte');
+    expect(libelleRegle('contrat-puissance-v1')).toBe('Contrat de puissance');
+  });
+
+  it('affiche telle quelle une référence de règle inconnue', () => {
+    expect(libelleRegle('spike-delestage-v2')).toBe('spike-delestage-v2');
+  });
+
+  it("réserve le ton critique à l'escalade vers l'astreinte", () => {
+    expect(tonRegle('escalade-astreinte-v1')).toBe('critical');
+    expect(tonRegle('spike-delestage-v1')).toBe('neutral');
+    expect(tonRegle('inconnue-v9')).toBe('neutral');
+  });
+});

apps/frontend/src/app/shared/models/recommendation-presentation.ts

@@ -0,0 +1,23 @@
+import { BadgeTone } from '../components/ui/badge/badge';
+
+// Contrainte : une règle dont le sens change reçoit un suffixe -v2 côté backend (ADR 0006) ;
+// une référence inconnue s'affiche donc telle quelle plutôt que de casser la vue.
+const LIBELLE_PAR_REGLE: Record<string, string> = {
+  'spike-delestage-v1': 'Délestage',
+  'threshold-reduction-v1': 'Réduction de puissance',
+  'outage-secours-v1': 'Alimentation de secours',
+  'sensor-maintenance-v1': 'Maintenance capteur',
+  'anomaly-verification-v1': 'Vérification',
+  'escalade-astreinte-v1': 'Escalade astreinte',
+  'contrat-puissance-v1': 'Contrat de puissance',
+};
+
+const REGLE_ESCALADE = 'escalade-astreinte-v1';
+
+export function libelleRegle(reference: string): string {
+  return LIBELLE_PAR_REGLE[reference] ?? reference;
+}
+
+export function tonRegle(reference: string): BadgeTone {
+  return reference === REGLE_ESCALADE ? 'critical' : 'neutral';
+}

apps/frontend/src/app/shared/models/recommendation.model.ts

@@ -0,0 +1,14 @@
+export interface Recommendation {
+  recommendation_id: number;
+  alert_id: number;
+  action: string;
+  explanation: string;
+  rule_reference: string;
+  created_at: string;
+}
+
+export interface RecommendationGenerationReport {
+  alerts_examined: number;
+  recommendations_created: number;
+  already_present: number;
+}

apps/frontend/src/styles.scss

@@ -2,6 +2,7 @@
 @use 'styles/forms';
 @use 'styles/auth-page';
 @use 'styles/links';
+@use 'styles/tables';
 
 body {
   margin: 0;

apps/frontend/src/styles/_forms.scss

@@ -29,3 +29,18 @@
   color: var(--color-disabled);
   margin-top: 0.25rem;
 }
+
+// Piège : le chevron est un SVG en data URI, où aucun token CSS n'est lisible ; sa couleur
+// reprend en dur la valeur de --color-text-muted.
+.form-select {
+  @extend .form-input;
+  padding-right: 2.25rem;
+  color: var(--color-text);
+  background-color: var(--color-surface);
+  background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 20 20' fill='none' stroke='%236b7280' stroke-width='1.5' stroke-linecap='round' stroke-linejoin='round'%3E%3Cpath d='M6 8l4 4 4-4'/%3E%3C/svg%3E");
+  background-repeat: no-repeat;
+  background-position: right 0.6rem center;
+  background-size: 1rem;
+  appearance: none;
+  cursor: pointer;
+}

apps/frontend/src/styles/_tables.scss

@@ -0,0 +1,43 @@
+// Piège : `ev-card` pose son padding sur `:host`, compilé en `[_nghost-…]` (même spécificité
+// qu'une classe) et injecté après la feuille globale ; il faut le sélecteur d'élément pour gagner.
+ev-card.ev-table-card {
+  padding: 0;
+  overflow: hidden;
+}
+
+.ev-table {
+  width: 100%;
+  border-collapse: collapse;
+
+  th,
+  td {
+    padding: 0.85rem 1.25rem;
+    text-align: left;
+    border-bottom: 1px solid var(--color-border-light);
+  }
+
+  th {
+    font-size: var(--font-size-xs);
+    font-weight: 600;
+    color: var(--color-text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+  }
+
+  tbody tr:last-child td {
+    border-bottom: none;
+  }
+
+  tbody tr:hover td {
+    background: var(--color-bg);
+  }
+}
+
+.ev-table__number {
+  font-weight: 600;
+  font-variant-numeric: tabular-nums;
+}
+
+.ev-table__muted {
+  color: var(--color-text-muted);
+}

apps/frontend/src/styles/_tokens.scss

@@ -29,10 +29,17 @@
 
   // Typo, rayons, ombre
   --font-family: 'Segoe UI', system-ui, sans-serif;
+  --font-size-xs: 0.75rem;
+  --font-size-sm: 0.85rem;
+  --font-size-md: 1rem;
+  --font-size-lg: 1.25rem;
+  --font-size-xl: 1.75rem;
+  --font-size-2xl: 2.25rem;
   --radius-sm: 8px;
   --radius-md: 12px;
   --radius-pill: 999px;
   --shadow-card: 0 1px 3px rgba(0, 0, 0, 0.06);
+  --shadow-card-hover: 0 6px 16px rgba(0, 0, 0, 0.08);
 
   // Espacements
   --space-1: 0.35rem;

apps/frontend/tsconfig.json

@@ -3,6 +3,7 @@
 {
   "compileOnSave": false,
   "compilerOptions": {
+    "strict": true,
     "noImplicitOverride": true,
     "noPropertyAccessFromIndexSignature": true,
     "noImplicitReturns": true,

db/init/120-airflow-database.sql

@@ -1,4 +1,4 @@
--- Base de metadonnees Airflow (webserver + scheduler, LocalExecutor). Separee de la base
+-- Base de metadonnees Airflow (api-server + scheduler + dag-processor, LocalExecutor). Separee de la base
 -- applicative : les tables internes d'Airflow (dag_run, task_instance, ...) n'ont rien a faire
 -- dans le schema metier. Meme conteneur Postgres que `enervision`/`enervision_test` plutot qu'un
 -- service dedie, pour ne pas ajouter un conteneur de plus (issue #115).

docker-compose.prod.yml

@@ -5,6 +5,8 @@
 # moyen de dépublier 8000 et 3000 : sans lui, l'API resterait joignable en clair à côté du proxy.
 # Piège : pas de `:?` sur `PUBLIC_HOST`. Compose interpole tout le fichier, y compris pour
 # `stop` et `logs` : la garde vit dans `make stack-up`, qui la compare au certificat servi.
+# Pourquoi : ports du proxy et origine publique en variables, pour que deux environnements
+# cohabitent sur la même machine, chacun dans son projet Compose (ADR 0009).
 
 name: enervision
 
@@ -12,12 +14,15 @@ services:
   db:
     ports: !override
       - "127.0.0.1:${POSTGRES_PORT:-5433}:5432"
+    environment:
+      TS_TUNE_MEMORY: ${TS_TUNE_MEMORY:-2GB}
+      TS_TUNE_NUM_CPUS: ${TS_TUNE_NUM_CPUS:-2}
 
   mailpit:
     ports: !override
       - "127.0.0.1:${MAILPIT_UI_PORT:-8025}:8025"
 
-  airflow-webserver:
+  airflow-apiserver:
     ports: !override
       - "127.0.0.1:${AIRFLOW_PORT:-8080}:8080"
 
@@ -37,22 +42,22 @@ services:
       APP_ENV: prod
       APP_DEBUG: "false"
       APP_TRUST_PROXY_HEADERS: "true"
-      APP_CORS_ORIGINS: https://${PUBLIC_HOST:-enervision.local}
+      APP_CORS_ORIGINS: ${PUBLIC_ORIGIN:-https://${PUBLIC_HOST:-enervision.local}}
-      APP_FRONTEND_RESET_PASSWORD_URL: https://${PUBLIC_HOST:-enervision.local}/reset-password
+      APP_FRONTEND_RESET_PASSWORD_URL: ${PUBLIC_ORIGIN:-https://${PUBLIC_HOST:-enervision.local}}/reset-password
 
   frontend:
     ports: !reset null
 
   proxy:
-    image: nginx:1.28-alpine
+    image: nginx:1.31-alpine
     depends_on:
       backend:
         condition: service_healthy
       frontend:
         condition: service_started
     ports:
-      - "80:80"
+      - "${PROXY_HTTP_PORT:-80}:80"
-      - "443:443"
+      - "${PROXY_HTTPS_PORT:-443}:443"
     volumes:
       - ./infra/proxy/nginx.conf:/etc/nginx/nginx.conf:ro
       - ./infra/proxy/conf.d:/etc/nginx/conf.d:ro

docker-compose.yml

@@ -5,9 +5,9 @@
 
 name: enervision
 
-# Piege : LocalExecutor fait tourner les taches comme sous-processus du scheduler, jamais du
+# Piege : LocalExecutor fait tourner les taches comme sous-processus du scheduler, jamais de
-# webserver. `airflow_ml_state` (modele entraine, magasin MLflow) n'a donc besoin d'etre monte
+# l'api-server. `airflow_ml_state` (modele entraine, magasin MLflow) n'a donc besoin d'etre monte
-# que sur `airflow-scheduler` en pratique, mais reste partage avec le webserver pour que ce
+# que sur `airflow-scheduler` en pratique, mais reste partage avec l'api-server pour que ce
 # dernier puisse au besoin l'inspecter sans en devenir dependant.
 x-airflow-common: &airflow-common
   build:
@@ -19,17 +19,31 @@ x-airflow-common: &airflow-common
     # Piege : pas de `:?` sur les secrets Airflow. Compose interpole le fichier entier avant de
     # filtrer les services : une variable requise manquante casserait aussi `make db-up`,
     # `make dev`... pour quiconque n'a pas encore complete son `.env`. Le refus est porte par
-    # `airflow-init` (ci-dessous), dont `webserver` et `scheduler` dependent.
+    # `airflow-init` (ci-dessous), dont `api-server`, `dag-processor` et `scheduler` dependent.
     AIRFLOW__CORE__FERNET_KEY: ${AIRFLOW_FERNET_KEY:-}
-    AIRFLOW__WEBSERVER__SECRET_KEY: ${AIRFLOW_WEBSERVER_SECRET_KEY:-}
+    AIRFLOW__API__SECRET_KEY: ${AIRFLOW_API_SECRET_KEY:-}
+    # Signe les jetons entre scheduler, tâches et api-server. Conteneurs distincts : un secret
+    # généré au démarrage ne serait pas partagé, il doit venir du .env.
+    AIRFLOW__API_AUTH__JWT_SECRET: ${AIRFLOW_JWT_SECRET:-}
+    AIRFLOW__CORE__EXECUTION_API_SERVER_URL: http://airflow-apiserver:8080/execution/
+    # FabAuthManager plutôt que le SimpleAuthManager par défaut d'Airflow 3 : seul le provider
+    # FAB sait créer le compte admin que `airflow-init` pose via `_AIRFLOW_WWW_USER_*`.
+    AIRFLOW__CORE__AUTH_MANAGER: airflow.providers.fab.auth_manager.fab_auth_manager.FabAuthManager
     AIRFLOW__DATABASE__SQL_ALCHEMY_CONN: postgresql+psycopg2://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:5432/airflow
     # Role `enervision_ml` dedie pas encore provisionne (dette assumee, cf. ADR 0003) :
     # memes identifiants que le backend en attendant.
     ML_DATABASE_URL: postgresql+psycopg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:5432/${POSTGRES_DB}
     MLFLOW_TRACKING_URI: sqlite:////opt/ml/state/mlflow.db
+    # Le DAG `alertes` lance le backend en sous-processus : il lit `DATABASE_URL`, en
+    # dialecte asyncpg, là où le pipeline ML lit `ML_DATABASE_URL`.
+    DATABASE_URL: postgresql+asyncpg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:5432/${POSTGRES_DB}
+    # Clé distincte de celle de l'API : la détection ne signe ni ne vérifie aucun jeton, et
+    # Airflow permet d'exécuter du code depuis son interface (cf. ADR 0008).
+    APP_SECRET_KEY: ${AIRFLOW_APP_SECRET_KEY:-}
   volumes:
     - ./etl/airflow/dags:/opt/airflow/dags
     - ./etl/airflow/plugins:/opt/airflow/plugins
+    - ./data/raw:/opt/data/raw:ro
     - airflow_logs:/opt/airflow/logs
     - airflow_ml_state:/opt/ml/state
   restart: unless-stopped
@@ -104,11 +118,11 @@ services:
   # Conteneur unique, jamais redemarre. La migration et la creation du premier compte sont
   # portees par l'entrypoint de l'image (`_AIRFLOW_DB_MIGRATE`, `_AIRFLOW_WWW_USER_*`), qui porte
   # aussi leur code de sortie : une migration ratee (ex. base `airflow` absente sur un volume
-  # `pgdata` deja peuple) fait echouer ce service, et `webserver`/`scheduler`, qui attendent son
+  # `pgdata` deja peuple) fait echouer ce service, et api-server, dag-processor et scheduler,
-  # succes, ne demarrent pas sur une base non migree. Le mot de passe passe par l'environnement,
+  # qui attendent son succes, ne demarrent pas sur une base non migree. Le mot de passe passe
-  # jamais par `argv` (ni `ps`, ni `docker compose config`).
+  # par l'environnement, jamais par `argv` (ni `ps`, ni `docker compose config`).
   # Sans mot de passe, l'entrypoint refuse lui-meme de creer le compte ; la commande ci-dessous
-  # refuse en plus les deux cles de chiffrement vides.
+  # refuse en plus les cles et secrets vides.
   airflow-init:
     <<: *airflow-common
     restart: "no"
@@ -128,12 +142,14 @@ services:
       - |
         set -euo pipefail
         : "$${AIRFLOW__CORE__FERNET_KEY:?AIRFLOW_FERNET_KEY manquant dans .env}"
-        : "$${AIRFLOW__WEBSERVER__SECRET_KEY:?AIRFLOW_WEBSERVER_SECRET_KEY manquant dans .env}"
+        : "$${AIRFLOW__API__SECRET_KEY:?AIRFLOW_API_SECRET_KEY manquant dans .env}"
+        : "$${AIRFLOW__API_AUTH__JWT_SECRET:?AIRFLOW_JWT_SECRET manquant dans .env}"
+        : "$${APP_SECRET_KEY:?AIRFLOW_APP_SECRET_KEY manquant dans .env}"
         exec airflow version
 
-  airflow-webserver:
+  airflow-apiserver:
     <<: *airflow-common
-    command: webserver
+    command: api-server
     ports:
       - "${AIRFLOW_PORT:-8080}:8080"
     depends_on:
@@ -142,7 +158,7 @@ services:
       airflow-init:
         condition: service_completed_successfully
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://localhost:8080/health"]
+      test: ["CMD", "curl", "--fail", "http://localhost:8080/api/v2/monitor/health"]
       interval: 30s
       timeout: 10s
       retries: 5
@@ -157,6 +173,16 @@ services:
       airflow-init:
         condition: service_completed_successfully
 
+  # Obligatoire depuis Airflow 3 : le scheduler ne parse plus les fichiers de dags/ lui-même.
+  airflow-dag-processor:
+    <<: *airflow-common
+    command: dag-processor
+    depends_on:
+      db:
+        condition: service_healthy
+      airflow-init:
+        condition: service_completed_successfully
+
 volumes:
   pgdata:
   airflow_logs:

docs/ML-START.md

@@ -0,0 +1,197 @@
+# ML-START : accès aux données, scoring, frontière API et ML
+
+Document de référence du module `ml/`, cité par le code (`enervision_ml/config.py`, `data.py`,
+`train.py`, `score.py`, `features.py`), par l'[ADR 0005](adr/0005-modele-prediction-lightgbm.md)
+et par les vues d'architecture. Il répond à trois questions, et à elles seules :
+
+1. **comment le pipeline accède aux données**, et pourquoi pas par l'API ;
+2. **ce que fait un run de scoring**, étape par étape ;
+3. **où passe la frontière entre l'API et le ML**, et pourquoi elle est là.
+
+Le mode d'emploi (installation, commandes, options) est dans [`ml/README.md`](../ml/README.md).
+Le choix du modèle est dans l'ADR 0005. Ce document ne les répète pas.
+
+---
+
+## 1. Mécanisme d'accès aux données
+
+### Deux sources, un seul schéma de sortie
+
+`enervision_ml.data` expose trois chargeurs qui produisent **exactement les mêmes neuf colonnes**
+(`site_id`, `timestamp`, `consumption_kwh`, `temperature_celsius`, `humidity_percent`,
+`solar_irradiance_wm2`, `is_working_hours`, `site_type`, `capacity_kw`) :
+
+| Fonction | Source | Usage |
+|---|---|---|
+| `load_from_csv(path)` | `ml/data/all_sites_combined.csv` | Chemin de démarrage, tant que la base n'est pas peuplée |
+| `load_from_database(connection)` | `reading` joint à `site`, **historique complet** | Entraînement |
+| `load_recent_from_database(connection, since=…)` | `reading` joint à `site`, **borné par `since`** | Scoring |
+
+L'égalité des schémas n'est pas un confort : c'est ce qui permet de valider tout le pipeline sur
+CSV, sans base joignable, et d'obtenir le même comportement une fois la base peuplée. Une
+divergence entre les deux chemins ne se verrait pas au chargement, elle se verrait en production
+sous forme de prédictions silencieusement fausses.
+
+### Connexion directe à PostgreSQL, pas l'API
+
+Le pipeline lit `reading` et `site` **en SQL direct**, jamais par `GET /api/v1/readings`. Trois
+raisons, à défendre telles quelles :
+
+- **Volume.** L'entraînement lit l'historique complet d'une hypertable TimescaleDB. Le faire
+  passer par une API REST paginée, sérialisée en JSON et contrôlée route par route, c'est payer
+  trois fois pour un `SELECT`.
+- **Couplage.** Le pipeline n'est pas un client de l'application, c'est un consommateur du
+  schéma. Passer par l'API le rendrait dépendant du contrat HTTP, de l'authentification et de la
+  disponibilité du service, pour lire des données dont il connaît déjà la forme.
+- **Droits.** Un rôle de lecture sur deux tables est une surface plus petite qu'un compte
+  applicatif porteur d'un rôle métier.
+
+### `ML_DATABASE_URL`, et pourquoi ce n'est pas `DATABASE_URL`
+
+La chaîne de connexion est lue dans **`ML_DATABASE_URL`**, jamais dans `DATABASE_URL`. Ce n'est
+pas une préférence de nommage : `DATABASE_URL` est celle du backend applicatif, **propriétaire du
+schéma**, avec les droits d'écriture complets. Réutiliser cette variable par défaut ferait tourner
+l'entraînement et le scoring avec ces droits, **en silence**. `enervision_ml.config.database_url()`
+lève donc plutôt que de retomber sur une valeur par défaut.
+
+**Dette assumée, à dire à l'oral et non à masquer** : le rôle PostgreSQL dédié `enervision_ml`,
+restreint en lecture sur `reading` et `site`, **n'est pas provisionné**. En développement,
+`ML_DATABASE_URL` pointe sur la même base que le backend. La cible est un rôle séparé, cohérente
+avec le principe de moindre privilège posé par l'[ADR 0003](adr/0003-autorisation-rbac-a-trois-roles.md).
+
+### Le seul endroit qui construit les features
+
+`enervision_ml.features.build_features` est **l'unique** constructeur de features, à
+l'entraînement comme au scoring. Le piège que cela évite : si les deux divergent, même d'une
+fenêtre de moyenne glissante, le modèle reçoit en service des features qui ne ressemblent plus à
+ce qu'il a appris, et ses prédictions se dégradent **sans qu'aucune erreur ne se déclenche**.
+Ne jamais réécrire cette logique ailleurs : importer le module.
+
+Conséquence sur la validation : la coupure entraînement / validation est **chronologique**, jamais
+un tirage aléatoire de lignes. Un tirage aléatoire laisserait des lignes de validation voir des
+lignes d'entraînement à travers leurs lags et leurs moyennes glissantes, une fuite qui masquerait
+un surapprentissage.
+
+---
+
+## 2. Les étapes d'un run de scoring
+
+`python -m enervision_ml.score` calcule, pour chaque site ou pour un seul avec `--site-id`, la
+consommation prévue de **l'heure suivant sa dernière lecture connue**, et écrit une ligne dans
+`prediction`.
+
+| # | Étape | Point de vigilance |
+|---|---|---|
+| 1 | Charger une **fenêtre récente** de `reading` joint à `site` : 21 jours par défaut | Une marge au-dessus des 168 h qu'exige le lag hebdomadaire. Un `SELECT` non borné sur l'hypertable serait la même erreur que celle corrigée sur `GET /readings` |
+| 2 | Ajouter **une ligne future par site**, l'heure suivante, et calculer ses features par `build_features` | La même fonction qu'à l'entraînement, cf. section 1 |
+| 3 | Si le **lag de 168 h est absent** (moins d'une semaine d'historique) : écrire `status = "insufficient_data"` | **LightGBM n'est jamais appelé.** Un modèle interrogé sans son lag principal rendrait un nombre, et ce nombre serait faux sans le dire |
+| 4 | Sinon : `booster.predict(...)`, puis écrire `status = "available"` et la valeur prévue | |
+
+### Ce que le run écrit, et ce qu'il n'écrase pas
+
+La table `prediction` **n'a pas de contrainte d'unicité sur `(site_id, target_at)`** : chaque run
+insère une ligne de plus au lieu d'écraser la précédente. C'est délibéré, et c'est ce qui rend
+possible la comparaison prévision contre réalisé. La surveillance de dérive s'en sert : elle
+retient, pour chaque `(site_id, target_at)`, la ligne du run le plus récent, celle-là même que
+sert `GET /api/v1/predictions`. Voir l'[ADR 0011](adr/0011-surveillance-de-derive-dans-le-backend.md).
+
+Trois contraintes de cohérence sont portées par la base et non par le code applicatif :
+`status = 'available'` exige une `predicted_value` et interdit un `failure_reason` ;
+`insufficient_data` et `error` exigent l'inverse ; `target_metric` est bornée à
+`consumption_kwh` ou `consumption_kw`, et la forme énergie impose une `period_minutes`. Elles
+sont vérifiées depuis le code qui écrit par `ml/tests/test_score_integration.py`, sur une vraie
+base : un double ne prouverait rien d'une contrainte SQL.
+
+**`--now` borne la fenêtre des deux côtés.** `load_recent_from_database` exige un `until` autant
+qu'un `since`, et le scoring lui passe l'instant de référence. Sans cette borne haute,
+`build_scoring_frame` repartait de la dernière lecture de toute la table quelle que soit la valeur
+demandée : `target_at` valait toujours « fin du jeu + 1 h », et l'âge de la dernière lecture
+devenait négatif sans franchir le seuil de péremption. Rejouer le scoring sur des instants passés
+produit désormais des prévisions dont le réalisé existe déjà, ce dont la surveillance de dérive a
+besoin pour se démontrer sur un jeu figé.
+
+### `model_reference` est un hachage, pas un nom de fichier
+
+`train.py` réécrit **toujours le même chemin** (`models/lightgbm-consumption.txt`) à chaque
+entraînement. Le nom de fichier ne distinguerait donc pas deux versions du modèle. `prediction`
+porte pour cela le **SHA-256 tronqué du fichier modèle**. C'est ce qui permet, devant une
+prédiction douteuse, de savoir quel modèle l'a produite.
+
+### Mode CSV : rien n'est écrit en base
+
+En `--csv`, le run ne touche pas la base. L'heure future calculée depuis la fin du CSV n'existe
+dans aucune base réelle : ce serait inscrire une prévision pour un instant déjà passé. Le mode
+sert à valider le pipeline sans base joignable.
+
+### Limite assumée
+
+La feature `is_working_hours` de la ligne future est **recopiée** depuis la dernière lecture
+réelle, pas recalculée : il n'existe aucune règle d'heures ouvrables dans ce dépôt, elle vit dans
+le générateur du jeu de données d'origine. L'approximation n'est fausse qu'aux heures de bascule,
+sur une feature parmi une dizaine, pour une prévision à un seul pas.
+
+---
+
+## 3. La frontière entre l'API et le ML
+
+```mermaid
+flowchart LR
+    subgraph ml["ml/ · projet Python indépendant"]
+        train["enervision_ml.train<br/>LightGBM + MLflow"]
+        score["enervision_ml.score<br/>prévision à un pas"]
+    end
+    subgraph db["PostgreSQL + TimescaleDB"]
+        reading[("reading, site")]
+        prediction[("prediction")]
+    end
+    subgraph api["apps/backend · FastAPI"]
+        route["GET /api/v1/predictions"]
+    end
+
+    reading -- "SQL direct, ML_DATABASE_URL" --> train
+    reading -- "fenêtre récente" --> score
+    train -- "models/*.txt + run MLflow" --> score
+    score -- "INSERT" --> prediction
+    prediction -- "lecture seule" --> route
+    prediction -- "prévu" --> derive["app.monitoring.drift<br/>écart prévu / réalisé"]
+    reading -- "réalisé" --> derive
+    derive -- "INSERT" --> rapport[("drift_report")]
+    rapport -- "lecture seule" --> monitoring["GET /api/v1/monitoring/drift"]
+```
+
+**La règle, en une phrase : FastAPI ne fait jamais tourner LightGBM.**
+`GET /api/v1/predictions` lit la dernière prévision par site dans `prediction`, jamais un recalcul
+à la volée. Ce qui en découle, et qui est l'argument à tenir devant le jury :
+
+- **La latence de l'API ne dépend pas du modèle.** Une route de lecture indexée
+  (`ix_prediction_site_target`) répond en temps constant, qu'un run de scoring dure une seconde
+  ou une minute.
+- **Le service de production n'embarque ni LightGBM ni MLflow.** `ml/` est un projet Python
+  séparé, avec son propre `uv.lock`. Le backend n'a aucune raison de porter ces dépendances, ni
+  leur surface de vulnérabilités, pour un script lancé hors du chemin de requête.
+- **Une panne du pipeline dégrade, elle n'interrompt pas.** Si le scoring ne tourne plus, l'API
+  continue de servir la dernière prévision connue, avec son `created_at` et son
+  `model_reference`, au lieu de rendre une erreur.
+- **Le contrat est la table, pas un appel.** Ce qui traverse la frontière, ce sont des lignes de
+  `prediction` et leurs contraintes de cohérence, vérifiables en SQL.
+
+Le corollaire est qu'il n'y a **aucune prévision à la demande** : la fraîcheur d'une prévision est
+celle du dernier run de scoring. Ce run est ordonnancé par Airflow, DAG `ml_score` en `@hourly`
+(issue #115) ; seuls le mode `--csv` et un lancement local restent manuels, tout comme
+l'entraînement, dont le DAG `ml_train` n'a pas de planification.
+
+La surveillance de dérive traverse cette frontière **dans le sens de la table vers le backend**,
+sans la percer : elle relit `prediction` et `reading` en SQL, ne charge aucun modèle, et n'appelle
+pas MLflow. Son calcul, son seuil et son refus de comparer à la métrique d'entraînement sont dans
+l'[ADR 0011](adr/0011-surveillance-de-derive-dans-le-backend.md).
+
+---
+
+## Voir aussi
+
+- [`ml/README.md`](../ml/README.md) : installation, commandes, options, où écrire les tests
+- [ADR 0005](adr/0005-modele-prediction-lightgbm.md) : pourquoi LightGBM, et les 6 candidats écartés
+- [ADR 0006](adr/0006-moteur-de-regles-dans-le-backend.md) : ce qui consomme les prédictions
+- [`architecture/20-backend.md`](architecture/20-backend.md) : le contrat de `GET /predictions`
+- [`architecture/40-data.md`](architecture/40-data.md) : le modèle de données
+- [ADR 0011](adr/0011-surveillance-de-derive-dans-le-backend.md) : la surveillance de dérive