feat: 添加线索引擎、NLQ、场景检测、前端界面等核心功能模块

2026-06-16 08:15:15 +08:00
parent 7b1e2b10a8
commit 48340f6011
62 changed files with 6772 additions and 65 deletions
@@ -0,0 +1,100 @@
+"""全量穿透扫描编排（P1.5）。
+
+把场景检测器的结果转化为线索，记录扫描覆盖范围（证明全量性）与数据版本（可追溯）。
+当前为同步执行；后续可包装为 Celery 异步任务（接口保持不变）。
+"""
+
+from __future__ import annotations
+
+import uuid
+from dataclasses import dataclass
+
+from sqlalchemy.orm import Session
+
+from app.clues import service as clue_svc
+from app.clues.models import Clue
+from app.scenarios import churn_fraud as cf
+from app.scenarios import split_contract as sc
+
+MODEL_VERSION = "mock-llm@0.1"
+
+
+@dataclass
+class ScanResult:
+    scenario_code: str
+    scanned_count: int
+    clue: Clue | None
+
+
+def run_split_contract_scan(
+    session: Session,
+    contracts: list[sc.ContractRecord],
+    approval_threshold: float,
+    shared_controller: bool = False,
+    data_version_id: uuid.UUID | None = None,
+) -> ScanResult:
+    """场景一拆单扫描：检测→评分→（命中则）生成线索。"""
+    finding = sc.detect_threshold_edge(contracts, approval_threshold)
+    score = sc.split_risk_score(finding, shared_controller)
+    clue = None
+    if score > 0:
+        rationale = sc.build_rationale(finding, approval_threshold, shared_controller)
+        clue = clue_svc.create_clue(
+            session,
+            title="疑似政企拆单规避审批",
+            risk_domain="收入",
+            scenario_code="R8",
+            score=score,
+            rationale=rationale,
+            evidence={
+                "near_threshold_contracts": [c.contract_id for c in finding.near_threshold],
+                "edge_ratio": finding.ratio,
+                "near_threshold_amount": finding.total_amount,
+                "approval_threshold": approval_threshold,
+                "shared_controller": shared_controller,
+            },
+            subjects={"customers": sorted({c.customer_key for c in finding.near_threshold})},
+            amount_involved=finding.total_amount,
+            model_version=MODEL_VERSION,
+            data_version_id=data_version_id,
+        )
+    return ScanResult("R8", len(contracts), clue)
+
+
+def run_churn_scan(
+    session: Session,
+    retention_curve: list[cf.CohortPoint],
+    commission_paid: float,
+    active_ratio: float,
+    zero_usage_ratio: float,
+    channel_key: str,
+    data_version_id: uuid.UUID | None = None,
+) -> ScanResult:
+    """场景二养卡骗补扫描：时序断崖 + 佣金质量不匹配→线索。"""
+    finding = cf.detect_pulse_decay(retention_curve)
+    mismatch = cf.commission_quality_mismatch(commission_paid, active_ratio, zero_usage_ratio)
+    score = cf.churn_risk_score(finding, mismatch)
+    clue = None
+    if score >= 0.5:
+        rationale = cf.build_rationale(finding, mismatch)
+        clue = clue_svc.create_clue(
+            session,
+            title="疑似养卡骗补（脉冲增长+规律退订）",
+            risk_domain="成本",
+            scenario_code="R9",
+            score=score,
+            rationale=rationale,
+            evidence={
+                "cliff_month": finding.cliff_month,
+                "max_drop": finding.max_drop,
+                "commission_paid": commission_paid,
+                "active_ratio": active_ratio,
+                "zero_usage_ratio": zero_usage_ratio,
+                "mismatch": mismatch,
+            },
+            subjects={"channel": channel_key},
+            amount_involved=commission_paid,
+            model_version=MODEL_VERSION,
+            data_version_id=data_version_id,
+        )
+    return ScanResult("R9", len(retention_curve), clue)