"""指标的公共契约:依赖提取、可执行约束和质量结论共用这里的字段。""" from __future__ import annotations import math from typing import Any, Literal from pydantic import BaseModel, ConfigDict, Field, model_validator class RatioContract(BaseModel): model_config = ConfigDict(extra="forbid", allow_inf_nan=False) numerator: str = Field(min_length=1) denominator: str = Field(min_length=1) scale: float = Field(default=1, gt=0) subset: bool = True class MetricContract(BaseModel): """显式约束查询结果和指标值;自然语言定义仍需人工审阅。""" model_config = ConfigDict(extra="forbid", allow_inf_nan=False) shape: Literal["scalar", "table"] = "scalar" grain: list[str] = Field(default_factory=list) non_null: list[str] = Field(default_factory=list) unit: str | None = None minimum: float | None = None maximum: float | None = None ratio: RatioContract | None = None @model_validator(mode="after") def valid(self) -> "MetricContract": if self.minimum is not None and self.maximum is not None and self.minimum > self.maximum: raise ValueError("minimum 不能大于 maximum") if any(not c.strip() for c in self.grain + self.non_null): raise ValueError("粒度和非空列名不能为空") if len(set(self.grain)) != len(self.grain): raise ValueError("粒度列不能重复") return self def metric_expressions(metric: Any) -> list[str] | None: """只读取当前类型的有效输入;隐藏的旧编辑字段不产生依赖。""" if not isinstance(metric, dict): return None kind = metric.get("kind") or "formula" if kind == "external": return [] if kind == "formula": return [metric["expression"]] if isinstance(metric.get("expression"), str) else None if kind == "rule": inputs = (metric.get("rule") or {}).get("inputs") if isinstance(metric.get("rule"), dict) else None return list(inputs.values()) if isinstance(inputs, dict) and all(isinstance(v, str) for v in inputs.values()) else None if kind == "assessment": dims = (metric.get("assessment") or {}).get("dimensions") if isinstance(metric.get("assessment"), dict) else None if not isinstance(dims, list) or any(not isinstance(d, dict) or not isinstance(d.get("input"), str) for d in dims): return None return [d["input"] for d in dims] return None def contract_findings(definition: dict[str, Any], metric: dict[str, Any], snapshots: list[dict[str, Any]]) -> list[str]: """检查明确声明的约束,不把自然语言字段解释为已经执行的过滤条件。""" raw = definition.get("contract") if raw is None: return [] contract = MetricContract.model_validate(raw) problems: list[str] = [] if contract.unit is not None and contract.unit != str(definition.get("unit") or ""): problems.append("指标单位与约束声明不一致") value = metric.get("value") if contract.minimum is not None or contract.maximum is not None: if isinstance(value, bool) or not isinstance(value, (int, float)) or not math.isfinite(value): problems.append("无法检查取值范围:结果不是有限数值") elif ((contract.minimum is not None and value < contract.minimum) or (contract.maximum is not None and value > contract.maximum)): problems.append("指标值超出声明的范围") needs_snapshot = bool("shape" in contract.model_fields_set or contract.grain or contract.non_null or contract.ratio) if needs_snapshot and not snapshots: problems.append("无法检查结果形状:没有可验证的查询快照") for snap in snapshots: rows, columns = snap.get("rows"), snap.get("columns") if not isinstance(rows, list) or not isinstance(columns, list): problems.append("无法检查结果形状:查询快照不完整") continue if contract.shape == "scalar" and len(rows) != 1: problems.append(f"标量指标要求取数结果恰好一行,实际为 {len(rows)} 行") if snap.get("truncated"): problems.append("查询已截断,无法验证完整结果的粒度和行数") required = set(contract.grain + contract.non_null) if missing := required - set(columns): problems.append("查询缺少约束列:" + "、".join(sorted(missing))) continue if any(not isinstance(row, list) or len(row) != len(columns) for row in rows): problems.append("查询行与列定义不一致") continue if any(row[columns.index(col)] is None for row in rows for col in contract.non_null): problems.append("声明非空的列含有空值") if contract.grain: from app.core.artifact_store import canonical_json keys = [canonical_json([row[columns.index(col)] for col in contract.grain]) for row in rows] if len(keys) != len(set(keys)): problems.append("查询结果在声明的粒度上存在重复行") if contract.ratio is not None: ratio = contract.ratio candidates = [s for s in snapshots if ratio.numerator in (s.get("columns") or []) and ratio.denominator in (s.get("columns") or [])] if not candidates: problems.append("比率约束要求同一查询结果提供分子和分母列") else: for snap in candidates: columns, rows = snap["columns"], snap["rows"] if len(rows) != 1 or not isinstance(rows[0], list) or len(rows[0]) != len(columns): problems.append("比率约束要求分子和分母来自同一行汇总结果") continue numerator, denominator = (rows[0][columns.index(c)] for c in (ratio.numerator, ratio.denominator)) if any(isinstance(v, bool) or not isinstance(v, (int, float)) or not math.isfinite(v) for v in (numerator, denominator)): problems.append("分子和分母必须是有限数值") elif denominator <= 0: problems.append("比率分母必须大于零") elif ratio.subset and not 0 <= numerator <= denominator: problems.append("分子作为分母的子集,不能小于零或大于分母") else: expected = numerator / denominator * ratio.scale decimals = metric.get("decimals") if isinstance(decimals, int) and not isinstance(decimals, bool): expected = round(expected, decimals) if isinstance(value, bool) or not isinstance(value, (int, float)) or not math.isclose(value, expected, rel_tol=1e-9, abs_tol=1e-12): problems.append("指标值与声明的分子、分母和换算倍率不一致") return list(dict.fromkeys(problems))