feat: Intelligence Pipeline v3 — full implementation

Multi-stage evidence-grounded inference architecture replacing the
monolithic 9B model extraction pipeline. CPU-first specialist services
handle routine extraction while the 9B vLLM model is preserved for
semantic adjudication of ambiguous cases.

Key components:
- Capability-aware inference gateway (OpenAI-compatible + Ollama)
- Endpoint registry with DB migrations and REST API
- Sentence-aware document segmenter (property tests)
- Deterministic financial parsing with offset integrity
- Symbol resolution with ambiguity detection
- Specialist service (GLiNER2, dynamic batching, K8s deployment)
- Company-specific sentiment (FinBERT, calibration)
- Retrieval-based novelty and duplicate detection
- Confidence calibration pipeline
- Deterministic routing engine (property tests)
- 9B adjudication layer with VRAM gating
- Stock-specific impact model (features, labels, baseline, trained)
- Pipeline orchestrator (state machine, queues, leases, feature flags)
- Bounded parallelism (async workers, semaphore, load shedding)
- Observability (tracing, metrics, alerts)
- Compatibility adapter (v3→v2 golden mapping tests)
- Shadow/canary promotion framework
- Active learning and fine-tuning pipeline

Test results: 1,161 tests pass, ruff lint clean.
All 282 spec tasks completed.

services/intelligence_pipeline_v3/sentiment/models.py

@@ -0,0 +1,91 @@
+"""Pydantic models for company-specific sentiment analysis."""
+
+from __future__ import annotations
+
+from pydantic import BaseModel, Field, field_validator
+
+
+class EvidenceGroup(BaseModel):
+    """A group of evidence spans associated with a specific company.
+
+    A single evidence span can appear in multiple groups when the
+    underlying text mentions multiple companies.
+    """
+
+    company_id: str = Field(description="Resolved company identifier")
+    evidence_ids: list[str] = Field(description="IDs of evidence spans in this group")
+    texts: list[str] = Field(description="Text snippets from evidence spans")
+
+    @field_validator("evidence_ids")
+    @classmethod
+    def evidence_ids_non_empty(cls, v: list[str]) -> list[str]:
+        if not v:
+            raise ValueError("evidence_ids must not be empty")
+        return v
+
+    @field_validator("texts")
+    @classmethod
+    def texts_non_empty(cls, v: list[str]) -> list[str]:
+        if not v:
+            raise ValueError("texts must not be empty")
+        return v
+
+
+class TextSentiment(BaseModel):
+    """Per-text sentiment probability distribution with evidence linkage.
+
+    Stores the raw FinBERT output for a single evidence span text,
+    enabling full provenance from probability to source evidence.
+    """
+
+    evidence_id: str = Field(description="Evidence span ID this score belongs to")
+    positive_prob: float = Field(ge=0.0, le=1.0, description="Probability of positive sentiment")
+    negative_prob: float = Field(ge=0.0, le=1.0, description="Probability of negative sentiment")
+    neutral_prob: float = Field(ge=0.0, le=1.0, description="Probability of neutral sentiment")
+
+    @property
+    def dominant_label(self) -> str:
+        """Return the label with highest probability."""
+        if self.positive_prob >= self.negative_prob and self.positive_prob >= self.neutral_prob:
+            return "positive"
+        elif self.negative_prob >= self.positive_prob and self.negative_prob >= self.neutral_prob:
+            return "negative"
+        return "neutral"
+
+
+class CompanySentimentResult(BaseModel):
+    """Sentiment classification result for a single company.
+
+    Contains full probability distribution, supporting evidence IDs,
+    per-text scores, and model/calibration versioning for lineage tracking.
+    """
+
+    company_id: str = Field(description="Resolved company identifier")
+    label: str = Field(description="Derived label: positive, negative, neutral, or mixed")
+    positive_prob: float = Field(ge=0.0, le=1.0, description="Probability of positive sentiment")
+    negative_prob: float = Field(ge=0.0, le=1.0, description="Probability of negative sentiment")
+    neutral_prob: float = Field(ge=0.0, le=1.0, description="Probability of neutral sentiment")
+    evidence_ids: list[str] = Field(description="All evidence span IDs contributing to this result")
+    is_mixed: bool = Field(default=False, description="Whether mixed sentiment was detected from disagreement")
+    per_text_scores: list[TextSentiment] = Field(
+        default_factory=list,
+        description="Full probability distributions per evidence text",
+    )
+    model_version: str = Field(description="Sentiment model name and version")
+    calibration_version: str = Field(default="uncalibrated", description="Calibration artifact version")
+
+    @field_validator("label")
+    @classmethod
+    def label_valid(cls, v: str) -> str:
+        valid_labels = {"positive", "negative", "neutral", "mixed"}
+        if v not in valid_labels:
+            raise ValueError(f"label must be one of {valid_labels}, got '{v}'")
+        return v
+
+
+class SentimentBatchResult(BaseModel):
+    """Result of sentiment classification for a batch of companies."""
+
+    results: list[CompanySentimentResult] = Field(description="Per-company sentiment results")
+    model_version: str = Field(description="Sentiment model version used for batch")
+    processing_time_ms: int = Field(ge=0, description="Total processing time in milliseconds")