Add framework management with full i18n support

backend/app/api/routes/frameworks.py

@@ -5,9 +5,15 @@
 
 from app.api.dependencies import get_current_user
 from app.core.database import get_db
+from app.models.document import Document, DocumentDefinition
 from app.models.framework import Framework
 from app.models.user import User
-from app.schemas.framework import FrameworkResponse
+from app.schemas.framework import (
+    AddFrameworkRequest,
+    AvailableFrameworkResponse,
+    FrameworkResponse,
+)
+from app.services.seed import SEED_FRAMEWORKS, _load_knowledge
 
 router = APIRouter(prefix="/api/frameworks", tags=["frameworks"])
 
@@ -34,6 +40,57 @@ def list_frameworks(
     return [_to_response(fw) for fw in frameworks]
 
 
+@router.get("/available", response_model=list[AvailableFrameworkResponse])
+def list_available_frameworks(
+    db: Session = Depends(get_db),
+    _current_user: User = Depends(get_current_user),
+):
+    existing_names = {name for (name,) in db.query(Framework.name).all()}
+    return [
+        {"name": fw["name"], "description": fw["description"], "category": fw["category"]}
+        for fw in SEED_FRAMEWORKS
+        if fw["name"] not in existing_names and fw["name"] != "EU AI Act"
+    ]
+
+
+@router.post("", response_model=FrameworkResponse)
+def add_framework(
+    body: AddFrameworkRequest,
+    db: Session = Depends(get_db),
+    _current_user: User = Depends(get_current_user),
+):
+    seed_data = next((fw for fw in SEED_FRAMEWORKS if fw["name"] == body.name), None)
+    if not seed_data:
+        raise HTTPException(status_code=404, detail="Framework not found in catalogue")
+
+    if db.query(Framework).filter(Framework.name == body.name).first():
+        raise HTTPException(status_code=409, detail="Framework already exists")
+
+    fw = Framework(
+        name=seed_data["name"],
+        description=seed_data["description"],
+        category=seed_data["category"],
+        status=seed_data["status"],
+        content=_load_knowledge(seed_data["name"]),
+    )
+    db.add(fw)
+    db.flush()
+
+    for doc_data in seed_data["documents"]:
+        db.add(
+            DocumentDefinition(
+                framework_id=fw.id,
+                name=doc_data["name"],
+                description=doc_data["description"],
+                article=doc_data["article"],
+            )
+        )
+
+    db.commit()
+    db.refresh(fw)
+    return _to_response(fw)
+
+
 @router.get("/{framework_id}", response_model=FrameworkResponse)
 def get_framework(
     framework_id: uuid.UUID,
@@ -44,3 +101,25 @@ def get_framework(
     if not fw:
         raise HTTPException(status_code=404, detail="Framework not found")
     return _to_response(fw)
+
+
+@router.delete("/{framework_id}")
+def delete_framework(
+    framework_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    _current_user: User = Depends(get_current_user),
+):
+    fw = db.query(Framework).filter(Framework.id == framework_id).first()
+    if not fw:
+        raise HTTPException(status_code=404, detail="Framework not found")
+
+    if fw.name == "EU AI Act":
+        raise HTTPException(status_code=403, detail="This framework cannot be removed")
+
+    def_ids = [d.id for d in fw.document_definitions]
+    if def_ids:
+        db.query(Document).filter(Document.definition_id.in_(def_ids)).delete()
+
+    db.delete(fw)
+    db.commit()
+    return {"detail": "Framework deleted"}

backend/app/schemas/framework.py

@@ -4,6 +4,16 @@
 from pydantic import BaseModel, ConfigDict
 
 
+class AvailableFrameworkResponse(BaseModel):
+    name: str
+    description: str
+    category: str
+
+
+class AddFrameworkRequest(BaseModel):
+    name: str
+
+
 class FrameworkResponse(BaseModel):
     model_config = ConfigDict(from_attributes=True)
 

backend/app/services/sample_evidence_en.py

@@ -760,74 +760,4 @@
     "undp-framework-alignment-FA-15-0": "The system is registered in the EU AI database with all required fields completed, including provider and deployer information.",
     "undp-framework-alignment-FA-16-0": "A Council of Europe HUDERIA-style assessment has been conducted, addressing human rights and democracy impacts.",
     "undp-framework-alignment-FA-17-0": "The system complies with the Council of Europe AI Convention provisions on transparency, oversight, and accountability.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Energy & Carbon (Principle 1-2)
-    # =========================================================================
-    "env-energy-carbon-EC-01-0": "Training energy was measured at 850 kWh for the initial model training run using GPU cluster power monitoring.",
-    "env-energy-carbon-EC-01-1": "The training energy mix comprises 78% renewable sources (wind and solar) from the data centre's certified green tariff.",
-    "env-energy-carbon-EC-02-0": "Inference energy is monitored per server; each recognition consumes approximately 0.003 kWh including camera and edge processing.",
-    "env-energy-carbon-EC-02-1": "Energy-per-query is tracked as a KPI; current rate is 10.8 kJ per recognition event including all system components.",
-    "env-energy-carbon-EC-03-0": "Total carbon footprint is estimated at 420 kg CO2e for training and 1.2 tonnes CO2e annually for inference across all sites.",
-    "env-energy-carbon-EC-03-1": "Scope 1 emissions are zero; Scope 2 covers data centre electricity; Scope 3 includes hardware manufacturing and logistics.",
-    "env-energy-carbon-EC-04-0": "A carbon reduction target of 20% by 2027 has been set, with progress reviewed quarterly against the baseline.",
-    "env-energy-carbon-EC-04-1": "Progress is tracked via quarterly carbon accounting; current trajectory shows a 12% reduction achieved since baseline.",
-    "env-energy-carbon-EC-05-0": "85% of operational energy comes from certified renewable sources; the target is 100% renewable by 2027.",
-    "env-energy-carbon-EC-05-1": "A commitment to increase renewable usage by 5% annually is documented in the environmental management plan.",
-    "env-energy-carbon-EC-06-0": "Residual emissions are offset through verified Gold Standard carbon credits; offset certificates are retained for audit.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Hardware (Principle 3)
-    # =========================================================================
-    "env-hardware-HW-01-0": "Hardware selection prioritised NVIDIA T4 GPUs for their inference efficiency, consuming 70W versus 250W for alternatives.",
-    "env-hardware-HW-01-1": "Energy-efficient edge processing units reduce data centre load by performing initial face detection at the camera location.",
-    "env-hardware-HW-02-0": "GPU utilisation is monitored; inference workloads are batched during peak hours to maintain above 60% utilisation.",
-    "env-hardware-HW-02-1": "Off-peak workloads including model evaluation and reporting are scheduled for nighttime when renewable energy availability peaks.",
-    "env-hardware-HW-03-0": "Hardware refresh cycles are set at five years; servers are refurbished and redeployed to non-critical workloads after primary use.",
-    "env-hardware-HW-03-1": "Camera hardware is designed for a seven-year lifespan with modular lens replacement to extend operational life.",
-    "env-hardware-HW-04-0": "E-waste is managed through certified WEEE recycling partners with documented chain-of-custody and material recovery rates above 95%.",
-    "env-hardware-HW-04-1": "Hazardous materials in hardware are tracked via supplier declarations; RoHS compliance is verified for all components.",
-    "env-hardware-HW-05-0": "Supply chain sustainability is assessed during procurement; camera and server vendors provide annual sustainability reports.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Data Management (Principle 4)
-    # =========================================================================
-    "env-data-management-DM-01-0": "Data storage efficiency is maximised by storing only facial templates (512 bytes each) rather than raw images after enrolment.",
-    "env-data-management-DM-01-1": "A retention policy automatically purges templates of departed employees within 30 days, minimising storage footprint.",
-    "env-data-management-DM-02-0": "Data transfer is minimised through edge processing; only compact feature vectors are transmitted to the central server.",
-    "env-data-management-DM-02-1": "Network transfer environmental cost is negligible due to small payload sizes (under 2 KB per recognition event).",
-    "env-data-management-DM-03-0": "Data centres are selected based on PUE rating (target below 1.3) and proximity to deployment sites to minimise latency and transfer.",
-    "env-data-management-DM-03-1": "Data centres are located in northern Europe to benefit from cooler ambient temperatures and higher renewable energy availability.",
-    "env-data-management-DM-04-0": "Data centre water usage effectiveness (WUE) is below 0.5 L/kWh, using free-cooling for 80% of the year.",
-    "env-data-management-DM-04-1": "Water-efficient cooling is prioritised; adiabatic cooling is used only during peak summer temperatures.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Model Efficiency (Principle 5)
-    # =========================================================================
-    "env-model-efficiency-ME-01-0": "The model size (25M parameters) is justified by the accuracy requirements; a smaller MobileNet variant was evaluated but fell below the FAR threshold.",
-    "env-model-efficiency-ME-01-1": "Three smaller architectures were benchmarked; the selected model offers the best accuracy-to-compute ratio for the deployment scenario.",
-    "env-model-efficiency-ME-02-0": "Training efficiency uses mixed-precision (FP16) training, reducing GPU memory usage by 40% and training time by 30%.",
-    "env-model-efficiency-ME-02-1": "All training runs are documented with energy consumption, duration, and accuracy metrics to track efficiency improvements.",
-    "env-model-efficiency-ME-03-0": "Inference is optimised using TensorRT quantisation (INT8), reducing latency by 60% and energy consumption by 50%.",
-    "env-model-efficiency-ME-03-1": "The accuracy loss from INT8 quantisation is less than 0.1% FAR, validated as acceptable for the access control use case.",
-    "env-model-efficiency-ME-04-0": "Model retraining occurs biannually rather than continuously, using incremental fine-tuning on new enrolment data only.",
-    "env-model-efficiency-ME-04-1": "Incremental fine-tuning consumes 15% of the energy of full retraining while maintaining equivalent accuracy performance.",
-    "env-model-efficiency-ME-05-0": "Environmental efficiency is tracked as recognitions-per-kWh; current performance is 333 authentications per kWh.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Lifecycle (Principle 3, 6)
-    # =========================================================================
-    "env-lifecycle-LC-01-0": "A lifecycle assessment covers the system from hardware manufacture through operation to decommissioning and disposal.",
-    "env-lifecycle-LC-01-1": "Embodied emissions from hardware manufacturing account for approximately 35% of the total lifecycle carbon footprint.",
-    "env-lifecycle-LC-02-0": "The decommissioning plan specifies secure data deletion, hardware recycling, and camera removal procedures.",
-    "env-lifecycle-LC-02-1": "Data deletion follows NIST 800-88 guidelines; hardware is sent to certified recycling facilities with documented material recovery.",
-    "env-lifecycle-LC-03-0": "Vendor sustainability is evaluated during procurement; data centre providers must demonstrate PUE below 1.3 and renewable energy usage.",
-    "env-lifecycle-LC-03-1": "SLA environmental criteria include uptime efficiency, power usage targets, and annual sustainability reporting obligations.",
-    "env-lifecycle-LC-04-0": "Circular economy practices include hardware refurbishment, component reuse, and material recovery at end of life.",
-    # =========================================================================
-    # ENVIRONMENTAL IMPACT FRAMEWORK — Monitoring (Principle 6)
-    # =========================================================================
-    "env-monitoring-EM-01-0": "Environmental KPIs include energy-per-recognition, total carbon emissions, hardware utilisation rate, and renewable energy percentage.",
-    "env-monitoring-EM-01-1": "KPIs are reviewed quarterly with annual targets set during the environmental management planning cycle.",
-    "env-monitoring-EM-02-0": "An annual environmental report discloses the system's energy consumption, carbon footprint, and sustainability initiatives.",
-    "env-monitoring-EM-02-1": "Reporting follows the GHG Protocol for carbon accounting and references ISO 14001 for environmental management practices.",
-    "env-monitoring-EM-03-0": "Continuous improvement identifies opportunities such as model compression, hardware upgrades, and cooling optimisation each year.",
-    "env-monitoring-EM-03-1": "Lessons learned from each improvement initiative are documented and shared across sites to accelerate sustainability gains.",
-    "env-monitoring-EM-04-0": "Environmental regulatory compliance is monitored; the system meets EU Energy Efficiency Directive requirements for ICT equipment.",
-    "env-monitoring-EM-05-0": "Sustainability targets are integrated into the project's KPI framework and reported alongside operational and compliance metrics.",
 }