feat(stage10): implement vision integration

src/nsct/storage/models.py

@@ -471,4 +471,134 @@ class EvidenceScoreRelationModel(Base):
     __table_args__ = (
         Index("ix_evidence_score_relations_score_id", "score_id"),
         Index("ix_evidence_score_relations_related_claim_id", "related_claim_id"),
-    )
+    )
+
+
+# ---------------------------------------------------------------------------
+# Stage 10 — Vision Evidence Extraction
+# ---------------------------------------------------------------------------
+
+
+class VisionCaptureType(str, enum.Enum):
+    """Klassifikation der visuellen Erfassung (Stage 10)."""
+
+    RAW_IMAGE = "raw_image"
+    DIAGRAM = "diagram"
+    CHART = "chart"
+    SCREENSHOT = "screenshot"
+    INFOGRAPHIC = "infographic"
+    PDF_LAYOUT = "pdf_layout"
+
+
+class VisionEvidenceLevel(str, enum.Enum):
+    """Evidenz-Level einer visuellen Evidenz (Stage 10)."""
+
+    HIGH = "high"
+    MEDIUM = "medium"
+    LOW = "low"
+    UNCERTAIN = "uncertain"
+
+
+class VisionEntityType(str, enum.Enum):
+    """Kategorie einer erkannten Entity aus einem Bild (Stage 10)."""
+
+    DATE = "date"
+    PERSON = "person"
+    ORGANIZATION = "organization"
+    LOCATION = "location"
+    NUMBER = "number"
+    STATISTIC = "statistic"
+    GRAPH_ELEMENT = "graph_element"
+
+
+class VisionEvidenceModel(Base):
+    """Einzelne visuell extrahierte Evidenz (Stage 10).
+
+    Felder:
+        uuid: Primärschlüssel (UUID)
+        research_run_id: Research-Run-Zuordnung
+        source_id: Quelle, von der das Bild stammt
+        capture_type: Art der visuellen Erfassung
+        extracted_text: Vom Vision-Modell extrahierter Text
+        image_data_url: Data-URL oder base64-codiertes Bild (optional)
+        entities: JSON mit erkannten Entities
+        confidence: Confidence 0.0–1.0
+        confidence_label: Confidence als Label (HIGH/MEDIUM/LOW/UNCERTAIN)
+        evidence_level: Evidenz-Level (HIGH/MEDIUM/LOW/UNCERTAIN)
+        created_at / updated_at: Zeitstempel
+    """
+
+    __tablename__ = "vision_evidence"
+
+    id = Column(String(36), primary_key=True, default=lambda: str(uuid4()))
+    research_run_id = Column(String(36), nullable=False)
+    source_id = Column(String(36), ForeignKey("sources.id"), nullable=False)
+
+    capture_type = Column(Enum(VisionCaptureType), nullable=False)
+
+    extracted_text = Column(Text, nullable=False)
+    image_data_url = Column(Text, nullable=True)
+    entities = Column(JSON, nullable=False, default=dict)
+
+    confidence = Column(Float, nullable=False, default=0.5)
+    confidence_label = Column(String(16), nullable=False, default="medium")
+    evidence_level = Column(Enum(VisionEvidenceLevel), nullable=False, default=VisionEvidenceLevel.MEDIUM)
+
+    created_at = Column(DateTime, nullable=False, default=datetime.utcnow)
+    updated_at = Column(DateTime, nullable=False, default=datetime.utcnow)
+
+    # Relationships
+    source = relationship("SourceModel")
+
+    __table_args__ = (
+        Index("ix_vision_evidence_research_run_id", "research_run_id"),
+        Index("ix_vision_evidence_capture_type", "capture_type"),
+        Index("ix_vision_evidence_source_id", "source_id"),
+    )
+
+
+class VisionEntityModel(Base):
+    """Erkannte Entity aus einem visuellen Bild (Stage 10).
+
+    Felder:
+        uuid: Primärschlüssel (UUID)
+        evidence_id: FK zur VisionEvidence
+        entity_type: Kategorie der Entity
+        entity_value: Der erkannte Wert
+        context: Kontextbeschreibung
+        confidence: Confidence 0.0–1.0
+        created_at: Zeitstempel
+    """
+
+    __tablename__ = "vision_entities"
+
+    id = Column(String(36), primary_key=True, default=lambda: str(uuid4()))
+    evidence_id = Column(
+        String(36),
+        ForeignKey("vision_evidence.id"),
+        nullable=False,
+    )
+
+    entity_type = Column(Enum(VisionEntityType), nullable=False)
+    entity_value = Column(Text, nullable=False)
+    context = Column(Text, nullable=True)
+    confidence = Column(Float, nullable=False, default=0.5)
+
+    created_at = Column(DateTime, nullable=False, default=datetime.utcnow)
+
+    # Relationships
+    evidence = relationship("VisionEvidenceModel", back_populates="entities")
+
+    __table_args__ = (
+        Index("ix_vision_entities_evidence_id", "evidence_id"),
+        Index("ix_vision_entities_entity_type", "entity_type"),
+    )
+
+
+# Back-populate the backref for VisionEntityModel
+VisionEvidenceModel.entities = relationship(
+    "VisionEntityModel",
+    back_populates="evidence",
+    cascade="all, delete-orphan",
+    foreign_keys="VisionEntityModel.evidence_id",
+)