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 +1,406 @@
+"""Tests for inference registry seed migration helpers.
+
+Task 18: Migrate existing provider records.
+Validates:
+- Initial endpoints have correct protocol and URL
+- Initial deployments reference valid endpoints
+- Agent conversion maps ollama correctly
+- Agent conversion maps vllm correctly
+- Unknown providers raise error (don't silently convert)
+- Conflicting defaults are identified
+"""
+from __future__ import annotations
+
+import re
+from pathlib import Path
+from uuid import UUID
+
+import pytest
+
+from services.shared.inference.seed_migration import (
+    OLLAMA_DEPLOYMENT_ID,
+    OLLAMA_ENDPOINT_ID,
+    VLLM_DEPLOYMENT_ID,
+    VLLM_ENDPOINT_ID,
+    UnknownProviderError,
+    convert_agent_providers,
+    get_initial_deployments,
+    get_initial_endpoints,
+    identify_conflicting_defaults,
+)
+
+# ─── SQL migration file checks ────────────────────────────────────────────────
+
+MIGRATION_PATH = (
+    Path(__file__).resolve().parent.parent
+    / "infra"
+    / "migrations"
+    / "042_seed_inference_registry.sql"
+)
+
+
+class TestMigrationFileExists:
+    def test_file_exists(self):
+        assert MIGRATION_PATH.exists(), f"Migration file not found: {MIGRATION_PATH}"
+
+    def test_file_not_empty(self):
+        content = MIGRATION_PATH.read_text()
+        assert len(content.strip()) > 100
+
+    def test_uses_on_conflict_do_nothing(self):
+        """Migration is idempotent via ON CONFLICT DO NOTHING."""
+        content = MIGRATION_PATH.read_text()
+        # Remove comments before counting
+        sql = re.sub(r"--[^\n]*", "", content)
+        inserts = re.findall(r"INSERT INTO", sql)
+        on_conflicts = re.findall(r"ON CONFLICT", sql)
+        assert len(inserts) == len(on_conflicts), (
+            f"Expected {len(inserts)} ON CONFLICT clauses, got {len(on_conflicts)}"
+        )
+
+
+# ─── Initial endpoints ─────────────────────────────────────────────────────────
+
+
+class TestInitialEndpoints:
+    """Test that initial endpoints have correct protocol and URL."""
+
+    def test_returns_two_endpoints(self):
+        endpoints = get_initial_endpoints()
+        assert len(endpoints) == 2
+
+    def test_ollama_endpoint_protocol(self):
+        endpoints = get_initial_endpoints()
+        ollama = next(e for e in endpoints if e["name"] == "stonks-ollama")
+        assert ollama["protocol"] == "ollama_native"
+
+    def test_ollama_endpoint_url(self):
+        endpoints = get_initial_endpoints()
+        ollama = next(e for e in endpoints if e["name"] == "stonks-ollama")
+        assert ollama["base_url"] == "http://ollama.ollama-service.svc.cluster.local:11434"
+
+    def test_ollama_endpoint_id_is_uuid(self):
+        endpoints = get_initial_endpoints()
+        ollama = next(e for e in endpoints if e["name"] == "stonks-ollama")
+        assert isinstance(ollama["id"], UUID)
+
+    def test_ollama_health_path(self):
+        endpoints = get_initial_endpoints()
+        ollama = next(e for e in endpoints if e["name"] == "stonks-ollama")
+        assert ollama["health_path"] == "/api/tags"
+
+    def test_vllm_endpoint_protocol(self):
+        endpoints = get_initial_endpoints()
+        vllm = next(e for e in endpoints if e["name"] == "stonks-vllm")
+        assert vllm["protocol"] == "openai_chat"
+
+    def test_vllm_endpoint_url(self):
+        endpoints = get_initial_endpoints()
+        vllm = next(e for e in endpoints if e["name"] == "stonks-vllm")
+        assert vllm["base_url"] == "http://kube-vllm.stonks-oracle.svc.cluster.local:8000"
+
+    def test_vllm_endpoint_id_is_uuid(self):
+        endpoints = get_initial_endpoints()
+        vllm = next(e for e in endpoints if e["name"] == "stonks-vllm")
+        assert isinstance(vllm["id"], UUID)
+
+    def test_vllm_health_path(self):
+        endpoints = get_initial_endpoints()
+        vllm = next(e for e in endpoints if e["name"] == "stonks-vllm")
+        assert vllm["health_path"] == "/health"
+
+    def test_endpoints_have_unique_ids(self):
+        endpoints = get_initial_endpoints()
+        ids = [e["id"] for e in endpoints]
+        assert len(set(ids)) == len(ids)
+
+    def test_endpoints_are_enabled(self):
+        endpoints = get_initial_endpoints()
+        for ep in endpoints:
+            assert ep["enabled"] is True
+
+
+# ─── Initial deployments ───────────────────────────────────────────────────────
+
+
+class TestInitialDeployments:
+    """Test that initial deployments reference valid endpoints."""
+
+    def test_returns_two_deployments(self):
+        deployments = get_initial_deployments()
+        assert len(deployments) == 2
+
+    def test_deployments_reference_valid_endpoint_ids(self):
+        """Every deployment references an endpoint from the seed set."""
+        endpoints = get_initial_endpoints()
+        endpoint_ids = {e["id"] for e in endpoints}
+        deployments = get_initial_deployments()
+        for dep in deployments:
+            assert dep["endpoint_id"] in endpoint_ids, (
+                f"Deployment {dep['served_model_name']} references unknown endpoint {dep['endpoint_id']}"
+            )
+
+    def test_ollama_deployment_model_name(self):
+        deployments = get_initial_deployments()
+        ollama_dep = next(d for d in deployments if d["endpoint_id"] == OLLAMA_ENDPOINT_ID)
+        assert ollama_dep["served_model_name"] == "qwen3.5:9b"
+
+    def test_vllm_deployment_model_name(self):
+        deployments = get_initial_deployments()
+        vllm_dep = next(d for d in deployments if d["endpoint_id"] == VLLM_ENDPOINT_ID)
+        assert vllm_dep["served_model_name"] == "AxionML/Qwen3.5-9B-NVFP4"
+
+    def test_vllm_deployment_context_window(self):
+        deployments = get_initial_deployments()
+        vllm_dep = next(d for d in deployments if d["endpoint_id"] == VLLM_ENDPOINT_ID)
+        assert vllm_dep["context_window"] == 8192
+
+    def test_vllm_deployment_max_output_tokens(self):
+        deployments = get_initial_deployments()
+        vllm_dep = next(d for d in deployments if d["endpoint_id"] == VLLM_ENDPOINT_ID)
+        assert vllm_dep["max_output_tokens"] == 2048
+
+    def test_vllm_deployment_quantization(self):
+        deployments = get_initial_deployments()
+        vllm_dep = next(d for d in deployments if d["endpoint_id"] == VLLM_ENDPOINT_ID)
+        assert vllm_dep["quantization"] == "NVFP4"
+
+    def test_vllm_deployment_has_json_schema_capability(self):
+        deployments = get_initial_deployments()
+        vllm_dep = next(d for d in deployments if d["endpoint_id"] == VLLM_ENDPOINT_ID)
+        assert vllm_dep["capabilities"]["json_schema"] is True
+
+    def test_ollama_deployment_lacks_json_schema(self):
+        deployments = get_initial_deployments()
+        ollama_dep = next(d for d in deployments if d["endpoint_id"] == OLLAMA_ENDPOINT_ID)
+        assert ollama_dep["capabilities"]["json_schema"] is False
+
+    def test_deployments_have_unique_ids(self):
+        deployments = get_initial_deployments()
+        ids = [d["id"] for d in deployments]
+        assert len(set(ids)) == len(ids)
+
+    def test_deployments_are_enabled(self):
+        deployments = get_initial_deployments()
+        for dep in deployments:
+            assert dep["enabled"] is True
+
+
+# ─── Agent conversion: ollama ──────────────────────────────────────────────────
+
+
+class TestConvertAgentOllama:
+    """Test agent conversion maps ollama provider correctly."""
+
+    def test_ollama_agent_produces_binding(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-aaaaaaaaaaaa", "model_provider": "ollama", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert len(bindings) == 1
+
+    def test_ollama_binding_endpoint_id(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-aaaaaaaaaaaa", "model_provider": "ollama", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["endpoint_id"] == OLLAMA_ENDPOINT_ID
+
+    def test_ollama_binding_deployment_id(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-aaaaaaaaaaaa", "model_provider": "ollama", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["model_deployment_id"] == str(OLLAMA_DEPLOYMENT_ID)
+
+    def test_ollama_binding_stage_extraction(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-aaaaaaaaaaaa", "model_provider": "ollama", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["stage"] == "extraction"
+
+    def test_ollama_event_classifier_stage(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-bbbbbbbbbbbb", "model_provider": "ollama", "slug": "event-classifier"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["stage"] == "classification"
+
+    def test_ollama_thesis_rewriter_stage(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-cccccccccccc", "model_provider": "ollama", "slug": "thesis-rewriter"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["stage"] == "thesis_rewrite"
+
+    def test_ollama_binding_is_active(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-aaaaaaaaaaaa", "model_provider": "ollama", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["is_active"] is True
+
+
+# ─── Agent conversion: vllm ───────────────────────────────────────────────────
+
+
+class TestConvertAgentVllm:
+    """Test agent conversion maps vllm provider correctly."""
+
+    def test_vllm_agent_produces_binding(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-dddddddddddd", "model_provider": "vllm", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert len(bindings) == 1
+
+    def test_vllm_binding_endpoint_id(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-dddddddddddd", "model_provider": "vllm", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["endpoint_id"] == VLLM_ENDPOINT_ID
+
+    def test_vllm_binding_deployment_id(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-dddddddddddd", "model_provider": "vllm", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["model_deployment_id"] == str(VLLM_DEPLOYMENT_ID)
+
+    def test_vllm_binding_stage(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-dddddddddddd", "model_provider": "vllm", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["stage"] == "extraction"
+
+    def test_vllm_binding_is_active(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-dddddddddddd", "model_provider": "vllm", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert bindings[0]["is_active"] is True
+
+    def test_multiple_vllm_agents(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-111111111111", "model_provider": "vllm", "slug": "document-extractor"},
+            {"id": "00000000-0000-4000-8000-222222222222", "model_provider": "vllm", "slug": "event-classifier"},
+            {"id": "00000000-0000-4000-8000-333333333333", "model_provider": "vllm", "slug": "thesis-rewriter"},
+        ]
+        bindings = convert_agent_providers(agents)
+        assert len(bindings) == 3
+        stages = [b["stage"] for b in bindings]
+        assert "extraction" in stages
+        assert "classification" in stages
+        assert "thesis_rewrite" in stages
+
+
+# ─── Unknown providers raise error ────────────────────────────────────────────
+
+
+class TestUnknownProviderError:
+    """Test that unknown providers raise error (don't silently convert)."""
+
+    def test_unknown_provider_raises(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-eeeeeeeeeeee", "model_provider": "openai", "slug": "document-extractor"}
+        ]
+        with pytest.raises(UnknownProviderError) as exc_info:
+            convert_agent_providers(agents)
+        assert "openai" in str(exc_info.value)
+
+    def test_unknown_provider_includes_agent_id(self):
+        agents = [
+            {"id": "00000000-0000-4000-8000-eeeeeeeeeeee", "model_provider": "anthropic", "slug": "test"}
+        ]
+        with pytest.raises(UnknownProviderError) as exc_info:
+            convert_agent_providers(agents)
+        assert "00000000-0000-4000-8000-eeeeeeeeeeee" in str(exc_info.value)
+
+    def test_empty_provider_skipped(self):
+        """Agents with no provider set are skipped, not errored."""
+        agents = [
+            {"id": "00000000-0000-4000-8000-eeeeeeeeeeee", "model_provider": "", "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert len(bindings) == 0
+
+    def test_none_provider_skipped(self):
+        """Agents with None provider are skipped, not errored."""
+        agents = [
+            {"id": "00000000-0000-4000-8000-eeeeeeeeeeee", "model_provider": None, "slug": "document-extractor"}
+        ]
+        bindings = convert_agent_providers(agents)
+        assert len(bindings) == 0
+
+    def test_mixed_valid_and_invalid_raises_on_invalid(self):
+        """If any agent has an unknown provider, conversion fails immediately."""
+        agents = [
+            {"id": "00000000-0000-4000-8000-111111111111", "model_provider": "vllm", "slug": "document-extractor"},
+            {"id": "00000000-0000-4000-8000-222222222222", "model_provider": "unknown_thing", "slug": "test"},
+        ]
+        with pytest.raises(UnknownProviderError) as exc_info:
+            convert_agent_providers(agents)
+        assert "unknown_thing" in str(exc_info.value)
+
+
+# ─── Conflicting defaults identification ──────────────────────────────────────
+
+
+class TestConflictingDefaults:
+    """Test that conflicting defaults are identified."""
+
+    def test_returns_non_empty_list(self):
+        conflicts = identify_conflicting_defaults()
+        assert len(conflicts) > 0
+
+    def test_identifies_config_py(self):
+        conflicts = identify_conflicting_defaults()
+        config_conflicts = [c for c in conflicts if "config.py" in c]
+        assert len(config_conflicts) >= 1
+
+    def test_identifies_migrations(self):
+        conflicts = identify_conflicting_defaults()
+        migration_conflicts = [c for c in conflicts if "migrations" in c]
+        assert len(migration_conflicts) >= 1
+
+    def test_identifies_helm_values(self):
+        conflicts = identify_conflicting_defaults()
+        helm_conflicts = [c for c in conflicts if "helm" in c]
+        assert len(helm_conflicts) >= 1
+
+    def test_identifies_kube_vllm_deployment(self):
+        conflicts = identify_conflicting_defaults()
+        kube_conflicts = [c for c in conflicts if "kube-vllm" in c]
+        assert len(kube_conflicts) >= 1
+
+    def test_all_entries_are_strings(self):
+        conflicts = identify_conflicting_defaults()
+        for c in conflicts:
+            assert isinstance(c, str)
+            assert len(c) > 10  # Meaningful content
+
+
+# ─── Well-known ID consistency ─────────────────────────────────────────────────
+
+
+class TestWellKnownIds:
+    """Ensure Python constants match SQL migration UUIDs."""
+
+    def test_ollama_endpoint_id_matches_sql(self):
+        content = MIGRATION_PATH.read_text()
+        assert str(OLLAMA_ENDPOINT_ID) in content
+
+    def test_vllm_endpoint_id_matches_sql(self):
+        content = MIGRATION_PATH.read_text()
+        assert str(VLLM_ENDPOINT_ID) in content
+
+    def test_ollama_deployment_id_matches_sql(self):
+        content = MIGRATION_PATH.read_text()
+        assert str(OLLAMA_DEPLOYMENT_ID) in content
+
+    def test_vllm_deployment_id_matches_sql(self):
+        content = MIGRATION_PATH.read_text()
+        assert str(VLLM_DEPLOYMENT_ID) in content