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.
2026-07-13 02:14:59 +00:00
parent 84634a365e
commit a72f336ad1
227 changed files with 50403 additions and 0 deletions
@@ -0,0 +1,158 @@
+"""Tests for audit/review module — Task 44."""
+
+from __future__ import annotations
+
+from uuid import uuid4
+
+from services.intelligence_pipeline_v3.audit.models import (
+    AuditRecord,
+    CorrectionEvent,
+    CorrectionType,
+    ReviewFilter,
+    ReviewStatus,
+)
+from services.intelligence_pipeline_v3.audit.store import AuditStore
+
+
+class TestAuditRecord:
+    """Task 44.1-44.2: Evidence display and specialist output tracking."""
+
+    def test_create_record(self):
+        record = AuditRecord.create(
+            document_id="doc-001",
+            run_id=uuid4(),
+            evidence_spans=[{"text": "Apple reported Q4 revenue", "start": 0, "end": 25}],
+            specialist_outputs={"sentiment": {"positive": 0.8}},
+            routing_reasons=["HIGH_CONFIDENCE"],
+            route_decision="fast_path",
+        )
+        assert record.document_id == "doc-001"
+        assert record.review_status == ReviewStatus.PENDING
+        assert len(record.evidence_spans) == 1
+
+    def test_add_correction(self):
+        record = AuditRecord.create(
+            document_id="doc-001", run_id=uuid4()
+        )
+        correction = CorrectionEvent.create(
+            record_id=record.record_id,
+            field_name="sentiment",
+            correction_type=CorrectionType.INCORRECT,
+            original_value="positive",
+            corrected_value="negative",
+            reviewer_id="reviewer-1",
+        )
+        record.add_correction(correction)
+        assert record.review_status == ReviewStatus.CORRECTED
+        assert len(record.corrections) == 1
+
+    def test_corrections_are_immutable(self):
+        correction = CorrectionEvent.create(
+            record_id=uuid4(),
+            field_name="ticker",
+            correction_type=CorrectionType.CORRECT,
+            original_value="AAPL",
+            reviewer_id="reviewer-1",
+        )
+        # Frozen dataclass — cannot modify
+        assert correction.event_id is not None
+        assert correction.timestamp is not None
+
+    def test_mark_reviewed(self):
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        record.mark_reviewed()
+        assert record.review_status == ReviewStatus.REVIEWED
+
+    def test_mark_confirmed(self):
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        record.mark_confirmed()
+        assert record.review_status == ReviewStatus.CONFIRMED
+
+
+class TestReviewFilter:
+    """Task 44.4: Filters for low confidence, unsupported claims, adjudicated."""
+
+    def test_filter_adjudicated(self):
+        f = ReviewFilter(is_adjudicated=True)
+        record_adj = AuditRecord.create(
+            document_id="doc-001",
+            run_id=uuid4(),
+            adjudicator_decision={"resolved": True},
+        )
+        record_fast = AuditRecord.create(
+            document_id="doc-002", run_id=uuid4()
+        )
+        assert f.matches(record_adj)
+        assert not f.matches(record_fast)
+
+    def test_filter_by_review_status(self):
+        f = ReviewFilter(review_status=ReviewStatus.CORRECTED)
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        assert not f.matches(record)
+        record.review_status = ReviewStatus.CORRECTED
+        assert f.matches(record)
+
+    def test_filter_unsupported_claims(self):
+        f = ReviewFilter(has_unsupported_claims=True)
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        assert not f.matches(record)
+        # Add unsupported correction
+        record.add_correction(
+            CorrectionEvent.create(
+                record_id=record.record_id,
+                field_name="fact",
+                correction_type=CorrectionType.UNSUPPORTED,
+                original_value="revenue beat",
+                reviewer_id="r1",
+            )
+        )
+        assert f.matches(record)
+
+
+class TestAuditStore:
+    """Task 44: Storage and retrieval."""
+
+    def test_store_and_retrieve(self):
+        store = AuditStore()
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        store.store(record)
+        assert store.get(record.record_id) is record
+        assert store.count() == 1
+
+    def test_get_by_document(self):
+        store = AuditStore()
+        run1 = uuid4()
+        run2 = uuid4()
+        store.store(AuditRecord.create(document_id="doc-001", run_id=run1))
+        store.store(AuditRecord.create(document_id="doc-001", run_id=run2))
+        store.store(AuditRecord.create(document_id="doc-002", run_id=uuid4()))
+        assert len(store.get_by_document("doc-001")) == 2
+
+    def test_add_correction_to_record(self):
+        store = AuditStore()
+        record = AuditRecord.create(document_id="doc-001", run_id=uuid4())
+        store.store(record)
+        correction = CorrectionEvent.create(
+            record_id=record.record_id,
+            field_name="ticker",
+            correction_type=CorrectionType.VALUE_OVERRIDE,
+            original_value="GOOG",
+            corrected_value="GOOGL",
+            reviewer_id="r1",
+        )
+        assert store.add_correction(record.record_id, correction)
+        assert store.correction_count() == 1
+
+    def test_filter(self):
+        store = AuditStore()
+        r1 = AuditRecord.create(
+            document_id="doc-001",
+            run_id=uuid4(),
+            adjudicator_decision={"x": 1},
+        )
+        r2 = AuditRecord.create(document_id="doc-002", run_id=uuid4())
+        store.store(r1)
+        store.store(r2)
+        results = store.filter(ReviewFilter(is_adjudicated=True))
+        assert len(results) == 1
+        assert results[0].document_id == "doc-001"