feat(platform): close AI expense value loop

Add tenant-safe value, telemetry, connector, commercial, and production-readiness foundations.

server/tests/test_risk_rule_golden_evaluator.py

@@ -1,9 +1,6 @@
 from __future__ import annotations
 
-from collections.abc import Generator
-from datetime import datetime
-from decimal import Decimal
-from unittest.mock import MagicMock, patch
+from unittest.mock import patch
 
 import pytest
 from sqlalchemy import create_engine
@@ -12,11 +9,8 @@ from sqlalchemy.pool import StaticPool
 
 from app.db.base import Base
 from app.models.agent_asset import AgentAsset, AgentAssetTestRun
-from app.models.employee import Employee
-from app.models.financial_record import ExpenseClaim
 from app.models.golden_case import GoldenCase
 from app.services.risk_rule_golden_evaluator import (
-    GoldenEvalReport,
     RiskRuleGoldenEvaluator,
     _aggregate,
     _run_single_case,
@@ -141,19 +135,36 @@ def test_aggregate_with_failure() -> None:
 
     results = [
         GoldenCaseResult("1", "a", True, True, "high", "high", True),
-        GoldenCaseResult("2", "b", True, False, "high", "none", False),  # FP
+        GoldenCaseResult("2", "b", True, False, "high", "none", False),  # FN
     ]
     report = _aggregate(results)
     assert report.passed_count == 1
     assert report.failed_count == 1
     assert report.accuracy == 0.5
     assert report.all_passed is False
-    assert report.precision == 0.5  # 1/(1+1)
+    assert report.precision == 1.0
+    assert report.recall == 0.5  # 1/(1+1)
+
+
+def test_aggregate_false_positive_reduces_precision_not_recall() -> None:
+    from app.services.risk_rule_golden_evaluator import GoldenCaseResult
+
+    report = _aggregate(
+        [
+            GoldenCaseResult("1", "a", True, True, "high", "high", True),
+            GoldenCaseResult("2", "b", False, True, "", "high", False),
+        ]
+    )
+
+    assert report.precision == 0.5
+    assert report.recall == 1.0
 
 
 def test_evaluate_for_rule_empty_returns_passed() -> None:
     with _build_session() as db:
-        report = RiskRuleGoldenEvaluator().evaluate_for_rule(db, _keyword_manifest(), "risk.test.keyword")
+        report = RiskRuleGoldenEvaluator().evaluate_for_rule(
+            db, _keyword_manifest(), "risk.test.keyword"
+        )
         assert report.total == 0
         assert report.all_passed is True
 
@@ -163,7 +174,9 @@ def test_evaluate_for_rule_all_pass() -> None:
         db.add(_golden_case("g1", reason="虚假发票", expected_hit=True))
         db.add(_golden_case("g2", reason="正常报销", expected_hit=False))
         db.commit()
-        report = RiskRuleGoldenEvaluator().evaluate_for_rule(db, _keyword_manifest(), "risk.test.keyword")
+        report = RiskRuleGoldenEvaluator().evaluate_for_rule(
+            db, _keyword_manifest(), "risk.test.keyword"
+        )
         assert report.total == 2
         assert report.all_passed is True
         assert report.accuracy == 1.0
@@ -172,9 +185,11 @@ def test_evaluate_for_rule_all_pass() -> None:
 def test_evaluate_for_rule_with_failure() -> None:
     with _build_session() as db:
         db.add(_golden_case("g1", reason="虚假发票", expected_hit=False))  # 期望不命中但实际命中
-        db.add(_golden_case("g2", reason="正常报销", expected_hit=True))   # 期望命中但实际不命中
+        db.add(_golden_case("g2", reason="正常报销", expected_hit=True))  # 期望命中但实际不命中
         db.commit()
-        report = RiskRuleGoldenEvaluator().evaluate_for_rule(db, _keyword_manifest(), "risk.test.keyword")
+        report = RiskRuleGoldenEvaluator().evaluate_for_rule(
+            db, _keyword_manifest(), "risk.test.keyword"
+        )
         assert report.total == 2
         assert report.all_passed is False
         assert report.failed_count == 2
@@ -206,6 +221,7 @@ def test_require_pass_passes_when_all_green() -> None:
         # 应写一条 test_type='golden' 记录
         run = db.query(AgentAssetTestRun).filter_by(asset_id="a1", test_type="golden").one()
         assert run.passed is True
+        assert run.status == "passed"
 
 
 def test_require_pass_raises_on_failure() -> None:
@@ -220,18 +236,22 @@ def test_require_pass_raises_on_failure() -> None:
             )
         run = db.query(AgentAssetTestRun).filter_by(asset_id="a2", test_type="golden").one()
         assert run.passed is False
+        assert run.status == "failed"
+        assert run.result_json["failure_reason"] == "golden_case_regression"
 
 
-def test_require_pass_empty_golden_set_passes() -> None:
+def test_require_pass_empty_golden_set_fails_closed() -> None:
     with _build_session() as db:
         asset = _asset("a3", "R3")
         db.add(asset)
         db.commit()
-        report = RiskRuleGoldenEvaluator().require_pass(
-            db, asset, "v1", _keyword_manifest(), "risk.test.keyword", actor="tester"
-        )
-        assert report.total == 0
-        assert report.all_passed is True
+        with pytest.raises(PermissionError, match="缺少必要"):
+            RiskRuleGoldenEvaluator().require_pass(
+                db, asset, "v1", _keyword_manifest(), "risk.test.keyword", actor="tester"
+            )
+        run = db.query(AgentAssetTestRun).filter_by(asset_id="a3", test_type="golden").one()
+        assert run.passed is False
+        assert run.result_json["failure_reason"] == "missing_active_golden_cases"
 
 
 def test_require_pass_respects_feature_flag(monkeypatch: pytest.MonkeyPatch) -> None:
@@ -246,17 +266,45 @@ def test_require_pass_respects_feature_flag(monkeypatch: pytest.MonkeyPatch) ->
             db, asset, "v1", _keyword_manifest(), "risk.test.keyword", actor="tester"
         )
         assert report.total == 0
+        assert report.gate_status == "skipped"
+        run = db.query(AgentAssetTestRun).filter_by(asset_id="a4", test_type="golden").one()
+        assert run.status == "skipped"
+        assert run.result_json["failure_reason"] == "gate_explicitly_disabled"
 
 
-def test_require_pass_swallows_evaluator_exception() -> None:
+def test_require_pass_records_and_blocks_precondition_error() -> None:
+    with _build_session() as db:
+        asset = _asset("a-precondition", "R-PRECONDITION")
+        db.add(asset)
+        db.commit()
+
+        with pytest.raises(PermissionError, match="前置条件失败"):
+            RiskRuleGoldenEvaluator().require_pass(
+                db,
+                asset,
+                "v1",
+                {},
+                "",
+                actor="tester",
+                precondition_error="missing_rule_document",
+            )
+
+        run = db.query(AgentAssetTestRun).filter_by(asset_id=asset.id).one()
+        assert run.status == "failed"
+        assert run.result_json["failure_reason"] == "missing_rule_document"
+
+
+def test_require_pass_fails_closed_on_evaluator_exception() -> None:
     with _build_session() as db:
         asset = _asset("a5", "R5")
         db.add(asset)
         db.commit()
         evaluator = RiskRuleGoldenEvaluator()
         with patch.object(evaluator, "evaluate_for_rule", side_effect=RuntimeError("boom")):
-            report = evaluator.require_pass(
-                db, asset, "v1", _keyword_manifest(), "risk.test.keyword", actor="tester"
-            )
-        assert report.total == 0
-        assert report.all_passed is True  # 降级放行
+            with pytest.raises(PermissionError, match="fail-closed"):
+                evaluator.require_pass(
+                    db, asset, "v1", _keyword_manifest(), "risk.test.keyword", actor="tester"
+                )
+        run = db.query(AgentAssetTestRun).filter_by(asset_id="a5", test_type="golden").one()
+        assert run.passed is False
+        assert run.result_json["failure_reason"] == "evaluation_error:RuntimeError"