qiaoqichaoandClaude Code
d6bc8485f5
覆盖 AC-15/16/34/37/41（BR-15/16/40、E-11/24）。
后端：
- contracts/improvement.py：ImprovementCandidate + ContrastReport/ContrastMetrics。
零分母不得给比率或覆盖率（样本不足不是 0%）；非独立评测不得给改进/退化结论；
已发布必须引用既有不可变版本；没有对照证据不得进入待审核或已发布。
- services/improvements.py：
· evaluate_contrast 在服务端判定泄漏——预测锁定前读到标签即整次评测作废（E-11），
保留集来源进开发集的样本作废并单列，指标按分子/可观察分母/覆盖率/未观察分开报；
· transition 状态机：draft→evaluating→ready→in_review→approved→released，
自批拒绝、退化与结论不明拒绝发布、发布幂等且只引用既有版本、撤回只影响新运行。
- api/prescreen.py：候选列表、提出与状态机三个端点；评测请求体提交样本，
泄漏与指标一律由服务端计算，不接受客户端交来的"成品报告"。
- contracts/prescreen.py：补 E-11/E-24 错误码与 409 映射。
前端一致性修正：
- src/lib/prescreen-run.ts 的呈现判定改用后端既有词汇
（empty/list/grouped/radar/faceted，与 services/review_plan.presentation 一致），
避免前后端各一套判定导致界面画出后端判定为不该画雷达的图。
演示交付：
- scripts/xingbang_prescreen_demo.py：按 08 §7 八步演示，只驱动已实现入口，
未实现的步骤（批量清单、历史提炼、卡片治理）如实标注原因；
工程演示与模型效果分列，本地专项/Jev/真实概率一律标未验证或未接通。
- backend/test_demo_script.py：冒烟断言八步状态、未实现原因与分列口径。
验证：backend 全量 unittest 519 项 OK；前端 npm run build 通过。
Co-Authored-By: Claude Code <noreply@anthropic.com>
219 lines
13 KiB
Python
219 lines
13 KiB
Python
"""兴邦预审 P0 模拟闭环演示脚本（T11）。
|
||
|
||
依据：docs/v2/08-data-pilot-and-demo.md §7 的八步演示顺序。
|
||
|
||
**诚实性要求**（08 §7 末、06 §6）：脚本只驱动本仓库已经实现并验证过的入口；
|
||
没有实现的步骤打印"未实现"及其原因，不伪造屏幕状态；模型效果与工程演示
|
||
分列，模拟标签不得包装成真实业务验证。
|
||
|
||
用法：backend/venv/Scripts/python.exe scripts/xingbang_prescreen_demo.py
|
||
全程在临时目录内使用临时数据库，不触碰真实 xingbang_data。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import sys
|
||
import tempfile
|
||
from pathlib import Path
|
||
|
||
ROOT = Path(__file__).resolve().parent.parent
|
||
sys.path.insert(0, str(ROOT / "backend"))
|
||
|
||
from contracts.prescreen import ErrorCode # noqa: E402
|
||
from services import (assistant_store as store, improvements, judgment_adapters, # noqa: E402
|
||
judgment_provider as jp, prescreen, prescreen_policy,
|
||
prescreen_sandbox, prescreen_store, review_plan, trajectory)
|
||
|
||
DONE = "已执行"
|
||
NOT_DONE = "未实现"
|
||
MODEL_UNVERIFIED = "未验证（无权重与授权）"
|
||
|
||
FINDINGS = [
|
||
{"code": "corporate.cashflow", "result": "unknown", "stage": 1, "evidence_ids": ["EV-1"]},
|
||
{"code": "corporate.title", "result": "unknown", "stage": 1, "evidence_ids": ["EV-2"]},
|
||
{"code": "corporate.deal", "result": "checked", "stage": 2, "evidence_ids": ["EV-3"]},
|
||
]
|
||
DIMENSIONS = [
|
||
{"id": "DC-CASH", "component_refs": [{"id": "corporate.cashflow"}]},
|
||
{"id": "DC-TITLE", "component_refs": [{"id": "corporate.title"}]},
|
||
{"id": "DC-DEAL", "component_refs": [{"id": "corporate.deal"}]},
|
||
]
|
||
PLAN = {
|
||
"plan": {"id": "corporate.default", "version": "1.0.0"},
|
||
"skills": [{"id": "corporate.cashflow", "version": "1.0.0"}],
|
||
"rules": [], "template": {"id": "template.ai-pre-review", "version": "1.0.0"},
|
||
"content_hash": "demo-hash", "dimensions": DIMENSIONS,
|
||
}
|
||
STEPS = [
|
||
(1, "3,000 条候选清单、生成版本与多结果分布", NOT_DONE,
|
||
"T04 的服务已在单测中验证，但本脚本不重复驱动批量生成；清单需真实数据授权"),
|
||
(2, "选开发集案例，展开问询/补件/会议事件，提炼 Skill 与 advisory Rule", NOT_DONE,
|
||
"历史提炼走 T05 的治理入口，需要真实历史材料；此处不伪造提炼结果"),
|
||
(3, "自然语言改适用范围 + 人工新增卡片，统一审核发布", NOT_DONE,
|
||
"卡片治理入口属 T05；演示脚本不代替审核动作"),
|
||
(4, "数据卡 + Skills + Rules 组方案，维度随方案变化，Tools 只在依据中展示", DONE, None),
|
||
(5, "缺证 → 主审核查 → 问题与补件 → 演示概率来源", DONE, None),
|
||
(6, "勾选动作比较情景、撤销恢复基线", DONE, None),
|
||
(7, "回放完整 trajectory，对齐模拟会议记录，查看材料版本差异", DONE, None),
|
||
(8, "从漏检提出改进候选，隔离样本对照后人工审核发布，旧版仍可回放", DONE, None),
|
||
]
|
||
|
||
|
||
def run_demo() -> dict:
|
||
facts: dict = {"steps": [], "sandbox": {}, "trajectory": {}, "improvement": {}}
|
||
with tempfile.TemporaryDirectory(ignore_cleanup_errors=True) as tmp:
|
||
db_path = Path(tmp) / "demo.sqlite3"
|
||
store.initialize(db_path)
|
||
|
||
# 步骤 4：方案维度（去重与刻度判定）
|
||
dimensions = review_plan.resolve_dimensions([
|
||
{"id": "DC-CASH", "label": "回款核验", "scale_id": "evidence-stage",
|
||
"scale_version": "1.0.0", "order": 1, "component_refs": [{"id": "corporate.cashflow", "version": "1.0.0"}],
|
||
"levels": [{"value": i, "label": name, "criterion": f"证据阶段：{name}"} for i, name in
|
||
enumerate(("未提供", "仅有说明", "资料待核", "交叉核验", "人工复核"))]},
|
||
{"id": "DC-TITLE", "label": "权属核验", "scale_id": "evidence-stage",
|
||
"scale_version": "1.0.0", "order": 2, "component_refs": [{"id": "corporate.title", "version": "1.0.0"}],
|
||
"levels": [{"value": i, "label": name, "criterion": f"证据阶段：{name}"} for i, name in
|
||
enumerate(("未提供", "仅有说明", "资料待核", "交叉核验", "人工复核"))]},
|
||
{"id": "DC-DEAL", "label": "交易要素", "scale_id": "evidence-stage",
|
||
"scale_version": "1.0.0", "order": 3, "component_refs": [{"id": "corporate.deal", "version": "1.0.0"}],
|
||
"levels": [{"value": i, "label": name, "criterion": f"证据阶段：{name}"} for i, name in
|
||
enumerate(("未提供", "仅有说明", "资料待核", "交叉核验", "人工复核"))]},
|
||
])
|
||
presentation = review_plan.presentation(dimensions)
|
||
facts["presentation"] = presentation["layout"]
|
||
|
||
# 步骤 5：预审执行（专项关闭，无校准概率）
|
||
config = jp.resolve_config(False)
|
||
decision = prescreen_policy.decide(FINDINGS)
|
||
decision["findings"] = FINDINGS
|
||
decision["missing_inputs"] = ["近六个月银行流水", "设备权属材料"]
|
||
screening = prescreen.assemble(
|
||
{"id": "run-demo", "project_id": "DEMO-026", "revision": 1,
|
||
"created_at": store.now(), "input_snapshot_id": "snap-demo"},
|
||
PLAN, decision, config)
|
||
with store.transaction(db_path) as db:
|
||
prescreen_store.save(db, screening)
|
||
facts["recommendation"] = screening["recommendation"]
|
||
facts["probability_is_null"] = screening["probability"] is None
|
||
|
||
# 步骤 6：沙盘（演示值只在演示域、组合查表）
|
||
with store.transaction(db_path) as db:
|
||
stored = prescreen_store.load(db, "DEMO-026", "run-demo")
|
||
baseline = prescreen_sandbox.simulate(stored, [])
|
||
both = prescreen_sandbox.simulate(stored, ["assume_bank_flow", "assume_title"])
|
||
facts["sandbox"] = {
|
||
"baseline": baseline["probability_view"]["demo_estimate"]["value"],
|
||
"combined": both["probability_view"]["demo_estimate"]["value"],
|
||
"label": baseline["probability_view"]["demo_estimate"]["label"],
|
||
"unchanged_dimensions": [item["id"] for item in both["dimensions"]
|
||
if not item["assumed"]],
|
||
}
|
||
|
||
# 步骤 7：轨迹回放与模拟会议结果
|
||
with store.transaction(db_path) as db:
|
||
store.put(db, "run", {"id": "run-demo", "project_id": "DEMO-026",
|
||
"status": "completed", "input_snapshot_id": "snap-demo",
|
||
"created_at": store.now(), "steps": [
|
||
{"run_id": "run-demo", "step_id": "corporate.cashflow",
|
||
"skill": {"id": "corporate.cashflow", "version": "1.0.0"},
|
||
"status": "completed", "input_snapshot_id": "snap-demo",
|
||
"started_at": store.now(), "outputs": {"note": "已核查"},
|
||
"findings": [{"code": "cashflow", "result": "unknown",
|
||
"evidence_ids": ["EV-1"]}],
|
||
"errors": [], "missing_inputs": []},
|
||
{"run_id": "run-demo", "step_id": "corporate.title",
|
||
"skill": {"id": "corporate.title", "version": "1.0.0"},
|
||
"status": "failed", "input_snapshot_id": "snap-demo",
|
||
"started_at": store.now(), "outputs": {},
|
||
"findings": [], "missing_inputs": [],
|
||
"errors": [{"code": "SKILL_TIMEOUT",
|
||
"message": "演示中的失败步骤"}]}]})
|
||
replay = trajectory.replay(db, "DEMO-026", "run-demo")
|
||
trajectory.record(db, "DEMO-026",
|
||
{"raw_code": "附条件通过", "meeting_id": "M-DEMO",
|
||
"occurred_at": store.now()}, "demo-reviewer",
|
||
known_snapshots=["snap-demo"])
|
||
outcomes = trajectory.listing(db, "DEMO-026")
|
||
facts["trajectory"] = {
|
||
"steps": [item["status"] for item in replay["steps"]],
|
||
"failed_preserved": any(item["status"] == "failed" for item in replay["steps"]),
|
||
"unmatched": len(outcomes["unmatched"]),
|
||
"observed_denominator": outcomes["counts_toward_denominator"]["observed"],
|
||
}
|
||
|
||
# 步骤 8：改进候选（合成样本、独立评测、可回放旧版）
|
||
with store.transaction(db_path) as db:
|
||
candidate = improvements.propose(db, "DEMO-026", {
|
||
"target": "knowledge", "target_ref": {"id": "card-demo", "version": "1.0.0"},
|
||
"proposed_revision_ref": {"id": "card-demo", "version": "1.1.0"},
|
||
"change_summary": "把回款核验经验扩到设备直租",
|
||
"source_run_ids": ["run-demo"]}, "demo-manager", "demo")
|
||
evaluation = {"baseline_ref": {"id": "current", "version": "1.0.0"},
|
||
"candidate_ref": {"id": "prescreen.v2", "version": "0.2.0"},
|
||
"dataset_ref": "demo/synthetic-manifest.json",
|
||
"split_ref": "demo-split.v1", "independent": True,
|
||
"samples": [
|
||
{"id": "S-1", "split": "dev", "observed": True,
|
||
"baseline_pass": True, "candidate_pass": True,
|
||
"predicted_at": "2026-09-29T01:00:00+00:00",
|
||
"label_revealed_at": "2026-09-29T03:00:00+00:00"},
|
||
{"id": "S-2", "split": "dev", "observed": True,
|
||
"baseline_pass": False, "candidate_pass": True,
|
||
"predicted_at": "2026-09-29T01:00:00+00:00",
|
||
"label_revealed_at": "2026-09-29T03:00:00+00:00"}]}
|
||
evaluated = improvements.transition(db, candidate["id"], "evaluate", "demo-reviewer",
|
||
evaluation=evaluation)
|
||
improvements.transition(db, evaluated["id"], "submit", "demo-reviewer")
|
||
approved = improvements.transition(db, evaluated["id"], "approve", "demo-reviewer")
|
||
released = improvements.transition(
|
||
db, approved["id"], "release", "demo-reviewer",
|
||
publisher=lambda body: {"id": "card-demo", "version": "1.1.0"})
|
||
facts["improvement"] = {"verdict": evaluated["contrast"]["verdict"],
|
||
"status": released["status"],
|
||
"released_ref": released["released_ref"]}
|
||
|
||
# 专项：离线 fixture 可跑，live 未接通
|
||
facts["judgment_offline"] = judgment_adapters.load_fixture(
|
||
"question_bank_extraction.json")["rubric_version"]
|
||
facts["demo_table_version"] = prescreen_sandbox.DEMO_TABLE_VERSION
|
||
facts["budget_code"] = ErrorCode.JUDGMENT_CONTEXT_EXCEEDED.value
|
||
return facts
|
||
|
||
|
||
def main() -> int:
|
||
facts = run_demo()
|
||
print("兴邦预审 P0 模拟闭环演示（合成数据 · 不连接真实数据与模型）")
|
||
print("=" * 78)
|
||
for index, title, status, reason in STEPS:
|
||
print(f"{index}. [{status}] {title}")
|
||
if reason:
|
||
print(f" 原因：{reason}")
|
||
print("-" * 78)
|
||
print("工程演示（本仓库已实现并验证的入口）")
|
||
print(f" · 维度呈现：{facts['presentation']}")
|
||
print(f" · 主审意见：{facts['recommendation']} · 真实概率为空：{facts['probability_is_null']}")
|
||
sandbox = facts["sandbox"]
|
||
print(f" · 沙盘：基线 {sandbox['baseline']:.0%} → 组合 {sandbox['combined']:.0%}"
|
||
f"（{sandbox['label']}）")
|
||
print(f" · 未变化的维度仍列出：{sandbox['unchanged_dimensions']}")
|
||
trace = facts["trajectory"]
|
||
print(f" · 轨迹步骤状态：{trace['steps']} · 失败步骤保留：{trace['failed_preserved']}")
|
||
print(f" · 会议结果：unmatched {trace['unmatched']} 条 · 已观察分母 {trace['observed_denominator']}")
|
||
imp = facts["improvement"]
|
||
print(f" · 改进对照：{imp['verdict']} → {imp['status']}（{imp['released_ref']}）")
|
||
print("-" * 78)
|
||
print("模型效果（与上面的工程演示分开计）")
|
||
print(f" · 本地专项：{MODEL_UNVERIFIED}；当前只有离线 fixture（{facts['judgment_offline']}）")
|
||
print(f" · Jev：未接通（无端点与凭据）")
|
||
print(f" · 真实审批概率：未校准，恒为空；界面上的百分数来自演示表 "
|
||
f"{facts['demo_table_version']}，不是模型输出")
|
||
print(f" · 超预算时的拒绝码（{facts['budget_code']}）证明预算守卫生效，与概率无关")
|
||
print("=" * 78)
|
||
print("以上均为合成数据与演示值，不能作为真实业务效果或审批依据。")
|
||
return 0
|
||
|
||
|
||
if __name__ == "__main__":
|
||
raise SystemExit(main())
|