风控模型开发验证报告

00 模型概要信息

项目	内容
模型名称	exp_20261010-160711-f53e
开发完成日期	2026-10-10
模型开发人员	（人工填写）
模型开发人员邮件地址	（人工填写）
模型验证人员	（人工填写）
模型验证人员邮件地址	（人工填写）
模型应用环节	preloan_approval
模型应用频率	（人工填写）
模型开发背景及业务用途描述	Local MVP full pipeline run.
模型适用产品/客户描述	（人工填写）
模型重开发原因（如有）	—
模型拟监控频率	（人工填写）
模型路线	LR + WOE 评分卡
目标变量	flag_30
实验 ID	ITE-mv247nv7-56aa

待补充：模型开发/验证人员、应用频率、产品描述等字段需人工补填。

01 模型设计

建模数据样本窗口

项目	内容
切分方式	外部 data_flag 划分（TRAIN 内 6/4 随机开发/测试，VALI 跨时间，PSI 时点）
开发样本(Train)窗口	1970-01-01 ~ 1970-01-01
预留验证样本窗口	1970-01-01 ~ 1970-01-01
跨时间验证样本窗口	1970-01-01 ~ 1970-01-01
PSI时点样本窗口	1970-01-01 ~ 1970-01-01

模型好坏定义

项目	内容
目标变量	flag_30
坏样本(BAD=1)占比	22.06%
好样本(BAD=0)占比	—
不确定样本	本流程要求目标变量已预计算为 0/1 二值，不确定样本请在上游合并口径

样本排除规则

序号	排除规则	说明
1	credit_no	主键/时间/口径字段，不入模
2	apply_dt	主键/时间/口径字段，不入模
3	flag_30 为空	排除 0 行

滚动率分析

待补充：滚动率分析（M0→M6+ 状态迁移矩阵）需要逾期状态口径数据，当前流程目标变量为上游已计算的二值标签，暂未生成。

待补充：表现窗口及坏账率分析、模型细分定义需结合业务口径人工补充。

02 数据流

开发时点 — 样本排除

类别	样本数
全部样本	30000
排除：目标变量为空	0
排除后样本	30000

注：按照01模型设计的排除规则统计；当前流程的排除项为目标变量缺失。

总体样本概况

数据集	样本量	好样本	坏样本	坏样本率	占比
全部样本	30000	23383	6617	22.06%	100.00%
开发样本	10801	8453	2348	21.74%	36.00%
测试样本	7199	5635	1564	21.73%	24.00%
训练+测试样本	18000	14088	3912	21.73%	60.00%
跨时间验证样本	6000	4624	1376	22.93%	20.00%
PSI时点样本	6000	4671	1329	22.15%	20.00%

开发/预留验证样本交叉统计

类别	总样本数	开发样本数	开发占比	预留验证样本数	预留验证占比
好	23383	8453	36.15%	5635	24.10%
坏	6617	2348	35.48%	1564	23.64%
合计	30000	10801	36.00%	7199	24.00%

细致变量分析（预分箱口径，开发样本）

变量	IV	缺失率	取值集中度	最大相关系数
risk_score	0.1434	0.00%	0.91%	0.9986
risk_score_v2	0.1390	0.00%	0.94%	0.9986
bh_a001	0.1382	0.00%	0.15%	0.9298
ScoreCust2_v2	0.1321	0.00%	1.14%	0.9952
ScoreCust2	0.1277	0.00%	1.13%	0.9952
bh_a009	0.1271	0.00%	1.73%	0.8070
cust_seg2	0.0848	0.00%	30.86%	0.6662
avg_tnr	0.0841	0.00%	1.92%	0.6947
query_cnt_3m	0.0732	0.00%	16.53%	0.5822
income_grade	0.0632	0.00%	31.12%	0.5458
repay_amt_6m_grade	0.0607	0.00%	27.69%	0.5342
rare_field	0.0330	96.32%	0.76%	0.0812
tenure_m	0.0217	0.00%	5.50%	0.3328
card_usage_ratio	0.0214	0.00%	1.69%	0.3613
bank_997_cnt_3m	0.0205	0.00%	34.62%	0.2757
edu_level	0.0004	0.00%	42.13%	—
region_code	0.0004	0.00%	20.63%	—
flag_special	0.0000	0.00%	95.41%	0.0380

注：共 18 个候选变量，按 IV 降序展示前 18 个；集中度 = 最大频次取值样本数 ÷ 非空样本数；相关性为皮尔逊（仅数值变量），完整数据见 variable_detail.json。

03 单变量分析（细分箱）

risk_score_v2 — IV 0.1426，8 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 495.5]	1406	1353	53	3.77%	0.2336
2	(495.5, 527.5]	1655	1500	155	9.37%	0.1797
3	(527.5, 563.5]	2581	2230	351	13.60%	0.1621
4	(563.5, 584.5]	1585	1243	342	21.58%	0.0022
5	(584.5, 602.5]	1181	853	328	27.77%	-0.0625
6	(602.5, 615.5]	667	442	225	33.73%	-0.0759
7	(615.5, 643.5]	986	551	435	44.12%	-0.1927
8	(643.5, inf]	740	281	459	62.03%	-0.2656

risk_score — IV 0.1272，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 492.5]	1279	1232	47	3.67%	0.2165
2	(492.5, 519.5]	1276	1182	94	7.37%	0.1696
3	(519.5, 535.5]	1032	914	118	11.43%	0.1000
4	(535.5, 546.5]	801	694	107	13.36%	0.0648
5	(546.5, 564.5]	1350	1144	206	15.26%	0.0779
6	(564.5, 587.5]	1704	1320	384	22.54%	-0.0112
7	(587.5, 600.5]	870	626	244	28.05%	-0.0507
8	(600.5, 616.5]	797	531	266	33.38%	-0.0859
9	(616.5, 643.5]	955	529	426	44.61%	-0.1917
10	(643.5, inf]	737	281	456	61.87%	-0.2638

ScoreCust2_v2 — IV 0.1262，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 507.5]	1649	1574	75	4.55%	0.2546
2	(507.5, 522.5]	1083	979	104	9.60%	0.1235
3	(522.5, 536.5]	1285	1138	147	11.44%	0.1205
4	(536.5, 544.5]	790	677	113	14.30%	0.0567
5	(544.5, 553.5]	959	776	183	19.08%	0.0237
6	(553.5, 565.5]	1232	965	267	21.67%	0.0007
7	(565.5, 583.5]	1611	1171	440	27.31%	-0.0737
8	(583.5, 593.5]	655	432	223	34.05%	-0.0766
9	(593.5, 606.5]	640	370	270	42.19%	-0.1231
10	(606.5, inf]	897	371	526	58.64%	-0.2861

ScoreCust2 — IV 0.1191，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 494.5]	957	923	34	3.55%	0.1690
2	(494.5, 508.5]	749	703	46	6.14%	0.1154
3	(508.5, 528.5]	1534	1387	147	9.58%	0.1658
4	(528.5, 545.5]	1664	1438	226	13.58%	0.1168
5	(545.5, 562.5]	1785	1421	364	20.39%	0.0198
6	(562.5, 572.5]	972	735	237	24.38%	-0.0235
7	(572.5, 579.5]	636	457	179	28.14%	-0.0392
8	(579.5, 592.5]	912	609	303	33.22%	-0.0950
9	(592.5, 604.5]	615	364	251	40.81%	-0.1111
10	(604.5, inf]	977	416	561	57.42%	-0.2967

bh_a001 — IV 0.1128，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 223.6500015258789]	1032	994	38	3.68%	0.1794
2	(223.6500015258789, 254.3499984741211]	1349	1266	83	6.15%	0.1937
3	(254.3499984741211, 280.0500030517578]	1560	1381	179	11.47%	0.1408
4	(280.0500030517578, 290.84999084472656]	786	672	114	14.50%	0.0549
5	(290.84999084472656, 317.0500030517578]	1927	1560	367	19.05%	0.0421
6	(317.0500030517578, 332.75]	1088	818	270	24.82%	-0.0301
7	(332.75, 354.0500030517578]	1197	825	372	31.08%	-0.0969
8	(354.0500030517578, 367.75]	542	334	208	38.38%	-0.0871
9	(367.75, 394.25]	756	392	364	48.15%	-0.1814
10	(394.25, inf]	564	211	353	62.59%	-0.2142

avg_tnr — IV 0.1073，7 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 2.149999976158142]	926	474	452	48.81%	-0.2194
2	(2.149999976158142, 3.350000023841858]	961	609	352	36.63%	-0.1276
3	(3.350000023841858, 4.450000047683716]	1258	903	355	28.22%	-0.0706
4	(4.450000047683716, 4.8500001430511475]	550	422	128	23.27%	-0.0083
5	(4.8500001430511475, 5.950000047683716]	1583	1247	336	21.23%	0.0069
6	(5.950000047683716, 7.3500001430511475]	2097	1722	375	17.88%	0.0646
7	(7.3500001430511475, inf]	3426	3076	350	10.22%	0.2859

cust_seg2 — IV 0.1060，4 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	2056	1867	189	9.19%	0.2166
2	(3.5, 4.5]	3333	2822	511	15.33%	0.1501
3	(4.5, 5.5]	3330	2505	825	24.77%	-0.0668
4	(5.5, inf]	2082	1259	823	39.53%	-0.2705

bh_a009 — IV 0.0991，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 6.3500001430511475]	902	433	469	52.00%	-0.2386
2	(6.3500001430511475, 7.3500001430511475]	875	548	327	37.37%	-0.1238
3	(7.3500001430511475, 8.050000190734863]	773	509	264	34.15%	-0.0891
4	(8.050000190734863, 8.849999904632568]	1062	753	309	29.10%	-0.0697
5	(8.849999904632568, 9.25]	578	430	148	25.61%	-0.0218
6	(9.25, 10.050000190734863]	1235	982	253	20.49%	0.0138
7	(10.050000190734863, 11.449999809265137]	2187	1873	314	14.36%	0.1298
8	(11.449999809265137, 12.75]	1501	1362	139	9.26%	0.1674
9	(12.75, 13.550000190734863]	664	591	73	10.99%	0.0706
10	(13.550000190734863, inf]	1024	972	52	5.08%	0.1637

rare_field — IV 0.0827，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, -1.437000036239624]	23	21	2	8.70%	0.0847
2	(-1.437000036239624, -0.8540000021457672]	60	43	17	28.33%	-0.0735
3	(-0.8540000021457672, -0.443000003695488]	50	43	7	14.00%	0.1024
4	(-0.443000003695488, -0.25699999928474426]	25	15	10	40.00%	-0.1074
5	(-0.25699999928474426, -0.05299999937415123]	30	26	4	13.33%	0.0713
6	(-0.05299999937415123, 0.10999999940395355]	24	14	10	41.67%	-0.1134
7	(0.10999999940395355, 0.3699999898672104]	34	24	10	29.41%	-0.0554
8	(0.3699999898672104, 1.0099999904632568]	89	80	9	10.11%	0.2372
9	(1.0099999904632568, 1.324500024318695]	26	19	7	26.92%	-0.0279
10	(1.324500024318695, inf]	36	22	14	38.89%	-0.1369

income_grade — IV 0.0632，5 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	1211	743	468	38.65%	-0.1736
2	(3.5, 4.5]	2510	1766	744	29.64%	-0.1417
3	(4.5, 5.5]	3361	2689	672	19.99%	0.0398
4	(5.5, 6.5]	2503	2150	353	14.10%	0.1484
5	(6.5, inf]	1216	1105	111	9.13%	0.1419

repay_amt_6m_grade — IV 0.0607，5 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	1647	1047	600	36.43%	-0.1915
2	(3.5, 4.5]	2268	1634	634	27.95%	-0.1049
3	(4.5, 5.5]	2991	2384	607	20.29%	0.0305
4	(5.5, 6.5]	2286	1944	342	14.96%	0.1227
5	(6.5, inf]	1609	1444	165	10.25%	0.1624

query_cnt_3m — IV 0.0483，9 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 0.5]	550	288	262	47.64%	-0.1355
2	(0.5, 1.5]	1123	724	399	35.53%	-0.1345
3	(1.5, 2.5]	1627	1151	476	29.26%	-0.0995
4	(2.5, 3.5]	1785	1384	401	22.46%	-0.0106
5	(3.5, 4.5]	1665	1365	300	18.02%	0.0523
6	(4.5, 5.5]	1347	1127	220	16.33%	0.0646
7	(5.5, 6.5]	1026	897	129	12.57%	0.0882
8	(6.5, 8.5]	1098	970	128	11.66%	0.1031
9	(8.5, inf]	580	547	33	5.69%	0.0940

bank_997_cnt_3m — IV 0.0246，4 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 0.5]	3447	2899	548	15.90%	0.1392
2	(0.5, 1.5]	3739	2928	811	21.69%	0.0012
3	(1.5, 2.5]	2224	1655	569	25.58%	-0.0648
4	(2.5, inf]	1391	971	420	30.19%	-0.0990

card_usage_ratio — IV 0.0183，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 0.107999999076128]	922	854	68	7.38%	0.1277
2	(0.107999999076128, 0.1578500047326088]	616	511	105	17.05%	0.0285
3	(0.1578500047326088, 0.3025500029325485]	2324	1848	476	20.48%	0.0224
4	(0.3025500029325485, 0.3381499946117401]	649	545	104	16.02%	0.0364
5	(0.3381499946117401, 0.4825499951839447]	2550	1996	554	21.73%	0.0003
6	(0.4825499951839447, 0.5283499956130981]	687	497	190	27.66%	-0.0388
7	(0.5283499956130981, 0.6137499809265137]	1203	941	262	21.78%	-0.0004
8	(0.6137499809265137, 0.6709499955177307]	594	405	189	31.82%	-0.0578
9	(0.6709499955177307, 0.7716499865055084]	715	520	195	27.27%	-0.0376
10	(0.7716499865055084, inf]	541	336	205	37.89%	-0.0844

tenure_m — IV 0.0170，10 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 2.5]	696	414	282	40.52%	-0.1218
2	(2.5, 10.5]	1179	891	288	24.43%	-0.0281
3	(10.5, 23.5]	2292	1813	479	20.90%	0.0148
4	(23.5, 30.5]	1310	997	313	23.89%	-0.0245
5	(30.5, 36.5]	1080	844	236	21.85%	-0.0011
6	(36.5, 39.5]	554	448	106	19.13%	0.0143
7	(39.5, 46.5]	1268	974	294	23.19%	-0.0161
8	(46.5, 53.5]	1211	970	241	19.90%	0.0199
9	(53.5, 58.5]	647	568	79	12.21%	0.0610
10	(58.5, inf]	564	534	30	5.32%	0.0937

edu_level — IV 0.0004，4 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	大专	3260	2574	686	21.04%	0.0155
2	本科	4550	3555	995	21.87%	-0.0035
3	硕士及以上	1405	1081	324	23.06%	-0.0160
4	高中及以下	1586	1243	343	21.63%	0.0015

region_code — IV 0.0004，5 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	其他	2119	1657	462	21.80%	-0.0011
2	华东	2156	1685	471	21.85%	-0.0018
3	华北	2119	1682	437	20.62%	0.0186
4	华南	2179	1701	478	21.94%	-0.0033
5	西南	2228	1728	500	22.44%	-0.0120

flag_special — IV 0.0000，1 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, inf]	10801	8453	2348	21.74%	0.0000

data_flag — IV 0.0000，1 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	TRAIN	10801	8453	2348	21.74%	0.0000

注：统计全部候选变量的自动分箱明细（初分组），按 IV 降序展示前 19 个变量，共 19 个；完整数据见 feature_report.json。

04 单变量分析（粗分箱）

risk_score_v2 — 8 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 495.5]	1406	1353	53	3.77%	0.2336
2	(495.5, 527.5]	1655	1500	155	9.37%	0.1797
3	(527.5, 563.5]	2581	2230	351	13.60%	0.1621
4	(563.5, 584.5]	1585	1243	342	21.58%	0.0022
5	(584.5, 602.5]	1181	853	328	27.77%	-0.0625
6	(602.5, 615.5]	667	442	225	33.73%	-0.0759
7	(615.5, 643.5]	986	551	435	44.12%	-0.1927
8	(643.5, inf]	740	281	459	62.03%	-0.2656

avg_tnr — 7 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 2.149999976158142]	926	474	452	48.81%	-0.2194
2	(2.149999976158142, 3.350000023841858]	961	609	352	36.63%	-0.1276
3	(3.350000023841858, 4.450000047683716]	1258	903	355	28.22%	-0.0706
4	(4.450000047683716, 4.8500001430511475]	550	422	128	23.27%	-0.0083
5	(4.8500001430511475, 5.950000047683716]	1583	1247	336	21.23%	0.0069
6	(5.950000047683716, 7.3500001430511475]	2097	1722	375	17.88%	0.0646
7	(7.3500001430511475, inf]	3426	3076	350	10.22%	0.2859

cust_seg2 — 4 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	2056	1867	189	9.19%	0.2166
2	(3.5, 4.5]	3333	2822	511	15.33%	0.1501
3	(4.5, 5.5]	3330	2505	825	24.77%	-0.0668
4	(5.5, inf]	2082	1259	823	39.53%	-0.2705

income_grade — 5 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	1211	743	468	38.65%	-0.1736
2	(3.5, 4.5]	2510	1766	744	29.64%	-0.1417
3	(4.5, 5.5]	3361	2689	672	19.99%	0.0398
4	(5.5, 6.5]	2503	2150	353	14.10%	0.1484
5	(6.5, inf]	1216	1105	111	9.13%	0.1419

repay_amt_6m_grade — 5 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 3.5]	1647	1047	600	36.43%	-0.1915
2	(3.5, 4.5]	2268	1634	634	27.95%	-0.1049
3	(4.5, 5.5]	2991	2384	607	20.29%	0.0305
4	(5.5, 6.5]	2286	1944	342	14.96%	0.1227
5	(6.5, inf]	1609	1444	165	10.25%	0.1624

query_cnt_3m — 9 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 0.5]	550	288	262	47.64%	-0.1355
2	(0.5, 1.5]	1123	724	399	35.53%	-0.1345
3	(1.5, 2.5]	1627	1151	476	29.26%	-0.0995
4	(2.5, 3.5]	1785	1384	401	22.46%	-0.0106
5	(3.5, 4.5]	1665	1365	300	18.02%	0.0523
6	(4.5, 5.5]	1347	1127	220	16.33%	0.0646
7	(5.5, 6.5]	1026	897	129	12.57%	0.0882
8	(6.5, 8.5]	1098	970	128	11.66%	0.1031
9	(8.5, inf]	580	547	33	5.69%	0.0940

bank_997_cnt_3m — 4 箱

分组号	分栏	Total	Good	Bad	Bad Rate	WOE
1	(-inf, 0.5]	3447	2899	548	15.90%	0.1392
2	(0.5, 1.5]	3739	2928	811	21.69%	0.0012
3	(1.5, 2.5]	2224	1655	569	25.58%	-0.0648
4	(2.5, inf]	1391	971	420	30.19%	-0.0990

注：统计入模变量经合并调整后的最终分箱明细（开发样本口径）。当前流程细/粗分箱由同一自动分箱算法直接产出。

05 变量筛选

项目	内容
筛选优先级	黑名单 > 白名单 > 自动规则
IV 下限	0.02
缺失率上限	0.8
集中度上限	0.9
相关性上限(皮尔逊)	0.7
PSI 上限	0.25
重要性 Top-K	80
黑名单	无
白名单	无

候选变量 19，成功分箱 19，入模 7

变量	白名单	入模	IV	缺失率	集中度	IV门槛	相关性	PSI	剔除原因
risk_score	—	N	0.1272	PASS	PASS	PASS	FAIL	0.0037	与 risk_score_v2 相关>0.7
ScoreCust2	—	N	0.1191	PASS	PASS	PASS	FAIL	0.0026	与 risk_score_v2 相关>0.7
cust_seg2	—	Y	0.1060	PASS	PASS	PASS	PASS	0.0005
income_grade	—	Y	0.0632	PASS	PASS	PASS	PASS	0.0009
bh_a001	—	N	0.1128	PASS	PASS	PASS	FAIL	0.0027	与 risk_score_v2 相关>0.7
bh_a009	—	N	0.0991	PASS	PASS	PASS	FAIL	0.0012	与 risk_score_v2 相关>0.7
bank_997_cnt_3m	—	Y	0.0246	PASS	PASS	PASS	PASS	0.0010
avg_tnr	—	Y	0.1073	PASS	PASS	PASS	PASS	0.0014
repay_amt_6m_grade	—	Y	0.0607	PASS	PASS	PASS	PASS	0.0007
tenure_m	—	N	0.0170	PASS	PASS	FAIL	—	—	IV<0.02
query_cnt_3m	—	Y	0.0483	PASS	PASS	PASS	PASS	0.0014
card_usage_ratio	—	N	0.0183	PASS	PASS	FAIL	—	—	IV<0.02
rare_field	—	N	0.0827	FAIL	PASS	—	—	—	缺失率>0.8
flag_special	—	N	0.0000	PASS	FAIL	—	—	—	集中度>0.9
edu_level	—	N	0.0004	PASS	PASS	FAIL	—	—	IV<0.02
region_code	—	N	0.0004	PASS	PASS	FAIL	—	—	IV<0.02
risk_score_v2	—	Y	0.1426	PASS	PASS	PASS	PASS	0.0031
ScoreCust2_v2	—	N	0.1262	PASS	PASS	PASS	FAIL	0.0020	与 risk_score_v2 相关>0.7
data_flag	—	N	0.0000	PASS	FAIL	—	—	—	集中度>0.9

06 回归结果

硬门槛：P≥0.1 / VIF>4.0 迭代剔除，共 0 个变量被剔除（1 轮）。

Parameter	Label	Estimate(无惩罚)	Standard Error	Wald Chi-Square	Pr > ChiSq	Standardized Estimate
Intercept	Intercept	-1.2148	0.0273	1986.43	<0.0001	—
WOE_risk_score_v2	risk_score_v2	-4.1925	0.2577	264.72	<0.0001	-0.6612
WOE_avg_tnr	avg_tnr	-0.8821	0.1979	19.86	<0.0001	-0.1493
WOE_cust_seg2	cust_seg2	-0.3785	0.1743	4.71	0.0300	-0.0668
WOE_income_grade	income_grade	-0.6664	0.2248	8.79	0.0030	-0.0838
WOE_repay_amt_6m_grade	repay_amt_6m_grade	-0.5144	0.2294	5.03	0.0249	-0.0635
WOE_query_cnt_3m	query_cnt_3m	-0.7941	0.3372	5.54	0.0185	-0.0690
WOE_bank_997_cnt_3m	bank_997_cnt_3m	-0.7880	0.2959	7.09	0.0077	-0.0703

注：Estimate 列为无惩罚（精确推断）拟合；生产系数保持 L2 正则（model_results.json / score.py 口径），两套估计并列展示。标准误基于无惩罚似然观测信息矩阵。

07 VIF

VIF

Parameter	VIF
risk_score_v2	2.81974
avg_tnr	1.66222
cust_seg2	1.62665
income_grade	1.31737
repay_amt_6m_grade	1.33112
query_cnt_3m	1.45011
bank_997_cnt_3m	1.06696

	risk_score_v2	avg_tnr	cust_seg2	income_grade	repay_amt_6m_grade	query_cnt_3m	bank_997_cnt_3m
risk_score_v2	1.000	0.622	0.610	0.480	0.488	0.548	0.243
avg_tnr	0.622	1.000	0.434	0.344	0.353	0.382	0.174
cust_seg2	0.610	0.434	1.000	0.333	0.352	0.385	0.176
income_grade	0.480	0.344	0.333	1.000	0.277	0.310	0.152
repay_amt_6m_grade	0.488	0.353	0.352	0.277	1.000	0.305	0.137
query_cnt_3m	0.548	0.382	0.385	0.310	0.305	1.000	0.155
bank_997_cnt_3m	0.243	0.174	0.176	0.152	0.137	0.155	1.000

08 最终评分卡

标准分600

标准ODD50

PDO20

变量	分栏	WOE	Coefficient	Scorecard Points
risk_score_v2	(-inf, 495.5]	0.2336	-4.0165	102
	(495.5, 527.5]	0.1797	-4.0165	95
	(527.5, 563.5]	0.1621	-4.0165	93
	(563.5, 584.5]	0.0022	-4.0165	75
	(584.5, 602.5]	-0.0625	-4.0165	67
	(602.5, 615.5]	-0.0759	-4.0165	66
	(615.5, 643.5]	-0.1927	-4.0165	52
	(643.5, inf]	-0.2656	-4.0165	44
avg_tnr	(-inf, 2.149999976158142]	-0.2194	-0.9134	69
	(2.149999976158142, 3.350000023841858]	-0.1276	-0.9134	71
	(3.350000023841858, 4.450000047683716]	-0.0706	-0.9134	73
	(4.450000047683716, 4.8500001430511475]	-0.0083	-0.9134	74
	(4.8500001430511475, 5.950000047683716]	0.0069	-0.9134	75
	(5.950000047683716, 7.3500001430511475]	0.0646	-0.9134	76
	(7.3500001430511475, inf]	0.2859	-0.9134	82
cust_seg2	(-inf, 3.5]	0.2166	-0.4366	77
	(3.5, 4.5]	0.1501	-0.4366	76
	(4.5, 5.5]	-0.0668	-0.4366	74
	(5.5, inf]	-0.2705	-0.4366	71
income_grade	(-inf, 3.5]	-0.1736	-0.7042	71
	(3.5, 4.5]	-0.1417	-0.7042	72
	(4.5, 5.5]	0.0398	-0.7042	75
	(5.5, 6.5]	0.1484	-0.7042	78
	(6.5, inf]	0.1419	-0.7042	77
repay_amt_6m_grade	(-inf, 3.5]	-0.1915	-0.5547	72
	(3.5, 4.5]	-0.1049	-0.5547	73
	(4.5, 5.5]	0.0305	-0.5547	75
	(5.5, 6.5]	0.1227	-0.5547	77
	(6.5, inf]	0.1624	-0.5547	77
query_cnt_3m	(-inf, 0.5]	-0.1355	-0.8021	71
	(0.5, 1.5]	-0.1345	-0.8021	71
	(1.5, 2.5]	-0.0995	-0.8021	72
	(2.5, 3.5]	-0.0106	-0.8021	74
	(3.5, 4.5]	0.0523	-0.8021	76
	(4.5, 5.5]	0.0646	-0.8021	76
	(5.5, 6.5]	0.0882	-0.8021	77
	(6.5, 8.5]	0.1031	-0.8021	77
	(8.5, inf]	0.0940	-0.8021	77
bank_997_cnt_3m	(-inf, 0.5]	0.1392	-0.7590	78
	(0.5, 1.5]	0.0012	-0.7590	75
	(1.5, 2.5]	-0.0648	-0.7590	73
	(2.5, inf]	-0.0990	-0.7590	72

09 模型表现

模型区分度评估（按样本）

模型	样本	KS	AUC	GINI
LR + WOE 评分卡	开发样本	0.3745	0.7516	0.5032
LR + WOE 评分卡	测试样本	0.3677	0.7458	0.4916
LR + WOE 评分卡	训练+测试样本	0.3711	0.7493	0.4986
LR + WOE 评分卡	跨时间验证样本	0.3433	0.7295	0.4590
LR + WOE 评分卡	PSI时点样本	0.3762	0.7449	0.4898

LR + WOE 评分卡 — 4 样本评估表（score_cut 粒度）

开发样本（KS 0.3745，AUC 0.7516）

cuts	cnt	bad	good	bad_rate	population_pct	cum_bad_rate	cum_good_rate
(0.0526, 0.0691]	1081	42	1039	3.89%	10.01%	1.79%	12.29%
(0.0691, 0.0858]	1080	74	1006	6.85%	10.00%	4.94%	24.19%
(0.0858, 0.0996]	1080	106	974	9.81%	10.00%	9.45%	35.72%
(0.0996, 0.1159]	1080	129	951	11.94%	10.00%	14.95%	46.97%
(0.1159, 0.1467]	1080	172	908	15.93%	10.00%	22.27%	57.71%
(0.1467, 0.2214]	1080	201	879	18.61%	10.00%	30.83%	68.11%
(0.2214, 0.2760]	1080	267	813	24.72%	10.00%	42.21%	77.72%
(0.2760, 0.3403]	1080	303	777	28.06%	10.00%	55.11%	86.92%
(0.3404, 0.4827]	1080	435	645	40.28%	10.00%	73.64%	94.55%
(0.4828, 0.6418]	1080	619	461	57.31%	10.00%	100.00%	100.00%

测试样本（KS 0.3677，AUC 0.7458）

cuts	cnt	bad	good	bad_rate	population_pct	cum_bad_rate	cum_good_rate
(0.0526, 0.0686]	720	32	688	4.44%	10.00%	2.05%	12.21%
(0.0687, 0.0852]	720	49	671	6.81%	10.00%	5.18%	24.12%
(0.0852, 0.0993]	721	74	647	10.26%	10.02%	9.91%	35.60%
(0.0993, 0.1158]	719	85	634	11.82%	9.99%	15.35%	46.85%
(0.1158, 0.1418]	720	110	610	15.28%	10.00%	22.38%	57.68%
(0.1418, 0.2195]	719	140	579	19.47%	9.99%	31.33%	67.95%
(0.2196, 0.2778]	720	184	536	25.56%	10.00%	43.09%	77.46%
(0.2779, 0.3383]	720	215	505	29.86%	10.00%	56.84%	86.42%
(0.3383, 0.4742]	720	269	451	37.36%	10.00%	74.04%	94.43%
(0.4742, 0.6418]	720	406	314	56.39%	10.00%	100.00%	100.00%

训练+测试样本（KS 0.3711，AUC 0.7493）

cuts	cnt	bad	good	bad_rate	population_pct	cum_bad_rate	cum_good_rate
(0.0526, 0.0690]	1800	73	1727	4.06%	10.00%	1.87%	12.26%
(0.0690, 0.0855]	1800	124	1676	6.89%	10.00%	5.04%	24.16%
(0.0856, 0.0994]	1802	179	1623	9.93%	10.01%	9.61%	35.68%
(0.0994, 0.1159]	1798	215	1583	11.96%	9.99%	15.11%	46.91%
(0.1159, 0.1444]	1800	283	1517	15.72%	10.00%	22.34%	57.68%
(0.1444, 0.2207]	1800	339	1461	18.83%	10.00%	31.01%	68.05%
(0.2207, 0.2766]	1800	453	1347	25.17%	10.00%	42.59%	77.61%
(0.2766, 0.3394]	1800	517	1283	28.72%	10.00%	55.80%	86.72%
(0.3394, 0.4784]	1800	703	1097	39.06%	10.00%	73.77%	94.51%
(0.4786, 0.6418]	1800	1026	774	57.00%	10.00%	100.00%	100.00%

跨时间验证样本（KS 0.3433，AUC 0.7295）

cuts	cnt	bad	good	bad_rate	population_pct	cum_bad_rate	cum_good_rate
(0.0526, 0.0702]	600	18	582	3.00%	10.00%	1.31%	12.59%
(0.0702, 0.0849]	600	66	534	11.00%	10.00%	6.10%	24.13%
(0.0850, 0.0997]	600	66	534	11.00%	10.00%	10.90%	35.68%
(0.0998, 0.1165]	600	82	518	13.67%	10.00%	16.86%	46.89%
(0.1166, 0.1454]	600	103	497	17.17%	10.00%	24.35%	57.63%
(0.1459, 0.2248]	600	129	471	21.50%	10.00%	33.72%	67.82%
(0.2249, 0.2729]	600	165	435	27.50%	10.00%	45.71%	77.23%
(0.2729, 0.3393]	600	178	422	29.67%	10.00%	58.65%	86.35%
(0.3396, 0.4778]	601	250	351	41.60%	10.02%	76.82%	93.94%
(0.4782, 0.6418]	599	319	280	53.26%	9.98%	100.00%	100.00%

注：score_cut 表基于模型输出 PD 等频分 10 档（分数升序），每表附 KS/AUC；训练+测试 = TRAIN 全体。

10 KS和GINI

LR + WOE 评分卡

样本	KS	GINI	AUC
开发样本	0.3745	0.5032	0.7516
测试样本	0.3677	0.4916	0.7458
训练+测试样本	0.3711	0.4986	0.7493
跨时间验证样本	0.3433	0.4590	0.7295
PSI时点样本	0.3762	0.4898	0.7449

开发样本

测试样本

训练+测试样本

跨时间验证样本

注：GINI = 2×AUC−1。KS 图 X 轴为分数（升序），红虚线为最大间隔位置；ROC 图例标注 AUC。

11 PSI

模型稳定性评估

模型	训练集与测试集	训练集与跨时间验证	训练集与oot验证集	训练集与psi验证集
LR + WOE 评分卡	0.0023	0.0042	—	0.0027

分数段 PSI 明细（开发样本 vs 跨时间验证样本）

分数段	开发样本占比	跨时间验证样本占比	PSI 贡献	累计 PSI
(-inf,0.0691]	10.00%	9.47%	0.0003	0.0003
(0.0691,0.0858]	10.00%	10.97%	0.0009	0.0012
(0.0858,0.0996]	10.01%	9.45%	0.0003	0.0015
(0.0996,0.1159]	9.99%	9.83%	0.0000	0.0015
(0.1159,0.1467]	10.00%	10.35%	0.0001	0.0017
(0.1467,0.2214]	10.01%	9.20%	0.0007	0.0023
(0.2214,0.276]	10.00%	11.25%	0.0015	0.0038
(0.276,0.3403]	9.99%	9.55%	0.0002	0.0040
(0.3403,0.4827]	10.00%	10.25%	0.0001	0.0041
(0.4827,inf]	10.01%	9.68%	0.0001	0.0042

分数段 PSI 明细（开发样本 vs PSI时点样本）

分数段	开发样本占比	PSI时点占比	PSI 贡献	累计 PSI
(-inf,0.0691]	10.00%	9.92%	0.0000	0.0000
(0.0691,0.0858]	10.00%	10.48%	0.0002	0.0002
(0.0858,0.0996]	10.01%	9.33%	0.0005	0.0007
(0.0996,0.1159]	9.99%	10.38%	0.0002	0.0009
(0.1159,0.1467]	10.00%	10.88%	0.0007	0.0016
(0.1467,0.2214]	10.01%	9.82%	0.0000	0.0016
(0.2214,0.276]	10.00%	9.53%	0.0002	0.0019
(0.276,0.3403]	9.99%	9.18%	0.0007	0.0025
(0.3403,0.4827]	10.00%	10.33%	0.0001	0.0027
(0.4827,inf]	10.01%	10.13%	0.0000	0.0027

模型样本量分布评估

评分段	开发样本样本量	开发样本占比	测试样本样本量	测试样本占比	训练+测试样本样本量	训练+测试样本占比	跨时间验证样本样本量	跨时间验证样本占比	PSI时点样本样本量	PSI时点样本占比
(-inf, inf]	1080	10.00%	737	10.24%	1817	10.09%	568	9.47%	595	9.92%
(0.0691, 0.0858]	1080	10.00%	728	10.11%	1808	10.04%	658	10.97%	629	10.48%
(0.0858, 0.0996]	1081	10.01%	715	9.93%	1796	9.98%	567	9.45%	560	9.33%
(0.0996, 0.1159]	1079	9.99%	709	9.85%	1788	9.93%	590	9.83%	623	10.38%
(0.1159, 0.1467]	1080	10.00%	768	10.67%	1848	10.27%	621	10.35%	653	10.88%
(0.1467, 0.2214]	1081	10.01%	682	9.47%	1763	9.79%	552	9.20%	589	9.82%
(0.2214, 0.276]	1080	10.00%	680	9.45%	1760	9.78%	675	11.25%	572	9.53%
(0.276, 0.3403]	1079	9.99%	763	10.60%	1842	10.23%	573	9.55%	551	9.18%
(0.3403, 0.4827]	1080	10.00%	751	10.43%	1831	10.17%	615	10.25%	620	10.33%
(0.4827, inf]	1081	10.01%	666	9.25%	1747	9.71%	581	9.68%	608	10.13%
sum	10801	100.00%	7199	100.00%	18000	100.00%	6000	100.00%	6000	100.00%

变量 PSI（开发样本 vs PSI时点样本，按入模变量分箱）

特征项	PSI	状态
risk_score_v2	0.0031	稳定
avg_tnr	0.0014	稳定
query_cnt_3m	0.0014	稳定
bank_997_cnt_3m	0.0010	稳定
income_grade	0.0009	稳定
repay_amt_6m_grade	0.0007	稳定
cust_seg2	0.0005	稳定

risk_score_v2 — PSI 0.0031

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 495.5]	13.02%	12.23%	-0.78%	0.0005
(495.5, 527.5]	15.32%	16.52%	1.19%	0.0009
(527.5, 563.5]	23.90%	24.53%	0.64%	0.0002
(563.5, 584.5]	14.67%	14.03%	-0.64%	0.0003
(584.5, 602.5]	10.93%	10.13%	-0.80%	0.0006
(602.5, 615.5]	6.18%	5.78%	-0.39%	0.0003
(615.5, 643.5]	9.13%	9.52%	0.39%	0.0002
(643.5, inf]	6.85%	7.25%	0.40%	0.0002

avg_tnr — PSI 0.0014

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 2.149999976158142]	8.57%	8.88%	0.31%	0.0001
(2.149999976158142, 3.350000023841858]	8.90%	8.87%	-0.03%	0.0000
(3.350000023841858, 4.450000047683716]	11.65%	11.57%	-0.08%	0.0000
(4.450000047683716, 4.8500001430511475]	5.09%	5.18%	0.09%	0.0000
(4.8500001430511475, 5.950000047683716]	14.66%	14.48%	-0.17%	0.0000
(5.950000047683716, 7.3500001430511475]	19.41%	18.15%	-1.26%	0.0009
(7.3500001430511475, inf]	31.72%	32.87%	1.15%	0.0004

cust_seg2 — PSI 0.0005

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 3.5]	19.04%	19.07%	0.03%	0.0000
(3.5, 4.5]	30.86%	30.50%	-0.36%	0.0000
(4.5, 5.5]	30.83%	31.77%	0.94%	0.0003
(5.5, inf]	19.28%	18.67%	-0.61%	0.0002

income_grade — PSI 0.0009

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 3.5]	11.21%	10.70%	-0.51%	0.0002
(3.5, 4.5]	23.24%	23.15%	-0.09%	0.0000
(4.5, 5.5]	31.12%	31.07%	-0.05%	0.0000
(5.5, 6.5]	23.17%	24.25%	1.08%	0.0005
(6.5, inf]	11.26%	10.83%	-0.42%	0.0002

repay_amt_6m_grade — PSI 0.0007

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 3.5]	15.25%	14.83%	-0.42%	0.0001
(3.5, 4.5]	21.00%	21.75%	0.75%	0.0003
(4.5, 5.5]	27.69%	26.92%	-0.78%	0.0002
(5.5, 6.5]	21.16%	21.43%	0.27%	0.0000
(6.5, inf]	14.90%	15.07%	0.17%	0.0000

query_cnt_3m — PSI 0.0014

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 0.5]	5.09%	5.28%	0.19%	0.0001
(0.5, 1.5]	10.40%	10.45%	0.05%	0.0000
(1.5, 2.5]	15.06%	14.95%	-0.11%	0.0000
(2.5, 3.5]	16.53%	16.27%	-0.26%	0.0000
(3.5, 4.5]	15.42%	15.28%	-0.13%	0.0000
(4.5, 5.5]	12.47%	13.08%	0.61%	0.0003
(5.5, 6.5]	9.50%	9.20%	-0.30%	0.0001
(6.5, 8.5]	10.17%	10.67%	0.50%	0.0002
(8.5, inf]	5.37%	4.82%	-0.55%	0.0006

bank_997_cnt_3m — PSI 0.0010

分栏	开发基准比例	PSI时点样本比例	变更	贡献
(-inf, 0.5]	31.91%	32.48%	0.57%	0.0001
(0.5, 1.5]	34.62%	35.27%	0.65%	0.0001
(1.5, 2.5]	20.59%	20.35%	-0.24%	0.0000
(2.5, inf]	12.88%	11.90%	-0.98%	0.0008

注：PSI ≤ 0.1 视为稳定，0.1~0.25 需关注，> 0.25 不稳定。

12 单变量分析

注：统计所有入模变量的单变量分析（开发样本口径）。预留验证/总体/OOT 口径的分组明细待扩展。

risk_score_v2

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 495.5]	1406	13.02%	1353	53	3.77%	25.53
2	(495.5, 527.5]	1655	28.34%	1500	155	9.37%	9.68
3	(527.5, 563.5]	2581	52.24%	2230	351	13.60%	6.35
4	(563.5, 584.5]	1585	66.91%	1243	342	21.58%	3.63
5	(584.5, 602.5]	1181	77.84%	853	328	27.77%	2.60
6	(602.5, 615.5]	667	84.02%	442	225	33.73%	1.96
7	(615.5, 643.5]	986	93.15%	551	435	44.12%	1.27
8	(643.5, inf]	740	100.00%	281	459	62.03%	0.61
合计		10801		8453	2348	21.74%	3.60

avg_tnr

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 2.149999976158142]	926	8.57%	474	452	48.81%	1.05
2	(2.149999976158142, 3.350000023841858]	961	17.47%	609	352	36.63%	1.73
3	(3.350000023841858, 4.450000047683716]	1258	29.12%	903	355	28.22%	2.54
4	(4.450000047683716, 4.8500001430511475]	550	34.21%	422	128	23.27%	3.30
5	(4.8500001430511475, 5.950000047683716]	1583	48.87%	1247	336	21.23%	3.71
6	(5.950000047683716, 7.3500001430511475]	2097	68.28%	1722	375	17.88%	4.59
7	(7.3500001430511475, inf]	3426	100.00%	3076	350	10.22%	8.79
合计		10801		8453	2348	21.74%	3.60

cust_seg2

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 3.5]	2056	19.04%	1867	189	9.19%	9.88
2	(3.5, 4.5]	3333	49.89%	2822	511	15.33%	5.52
3	(4.5, 5.5]	3330	80.72%	2505	825	24.77%	3.04
4	(5.5, inf]	2082	100.00%	1259	823	39.53%	1.53
合计		10801		8453	2348	21.74%	3.60

income_grade

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 3.5]	1211	11.21%	743	468	38.65%	1.59
2	(3.5, 4.5]	2510	34.45%	1766	744	29.64%	2.37
3	(4.5, 5.5]	3361	65.57%	2689	672	19.99%	4.00
4	(5.5, 6.5]	2503	88.74%	2150	353	14.10%	6.09
5	(6.5, inf]	1216	100.00%	1105	111	9.13%	9.95
合计		10801		8453	2348	21.74%	3.60

repay_amt_6m_grade

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 3.5]	1647	15.25%	1047	600	36.43%	1.75
2	(3.5, 4.5]	2268	36.25%	1634	634	27.95%	2.58
3	(4.5, 5.5]	2991	63.94%	2384	607	20.29%	3.93
4	(5.5, 6.5]	2286	85.10%	1944	342	14.96%	5.68
5	(6.5, inf]	1609	100.00%	1444	165	10.25%	8.75
合计		10801		8453	2348	21.74%	3.60

query_cnt_3m

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 0.5]	550	5.09%	288	262	47.64%	1.10
2	(0.5, 1.5]	1123	15.49%	724	399	35.53%	1.81
3	(1.5, 2.5]	1627	30.55%	1151	476	29.26%	2.42
4	(2.5, 3.5]	1785	47.08%	1384	401	22.46%	3.45
5	(3.5, 4.5]	1665	62.49%	1365	300	18.02%	4.55
6	(4.5, 5.5]	1347	74.97%	1127	220	16.33%	5.12
7	(5.5, 6.5]	1026	84.46%	897	129	12.57%	6.95
8	(6.5, 8.5]	1098	94.63%	970	128	11.66%	7.58
9	(8.5, inf]	580	100.00%	547	33	5.69%	16.58
合计		10801		8453	2348	21.74%	3.60

bank_997_cnt_3m

组号	Value	Tot. Records	累计占比	#Goods	#Bads	Bad Rate	G/B Odds
1	(-inf, 0.5]	3447	31.91%	2899	548	15.90%	5.29
2	(0.5, 1.5]	3739	66.53%	2928	811	21.69%	3.61
3	(1.5, 2.5]	2224	87.12%	1655	569	25.58%	2.91
4	(2.5, inf]	1391	100.00%	971	420	30.19%	2.31
合计		10801		8453	2348	21.74%	3.60

13 压力测试

注：验证压力情境下模型的有效性和区分能力稳定性：a.坏浓度更高样本 b.时间外样本 c.特定特征客群 d.极端时间样本。当前流程自动生成情景1（跨时间样本）与情景2（跨时间样本剔除一半好样本，批注16）。

测试情景1：跨时间验证样本

变量贡献度

变量	贡献度
risk_score_v2	0.6612
avg_tnr	0.1493
income_grade	0.0838
bank_997_cnt_3m	0.0703
query_cnt_3m	0.0690
cust_seg2	0.0668
repay_amt_6m_grade	0.0635

区分度验证 — LR + WOE 评分卡 @ 跨时间样本

KS = 0.3433，GINI = 0.4590

组号	分数区间(PD)	样本数	累计占比	Good	Bad	累计Good占比	累计Bad占比	分组坏率	累计坏率	分组Odds	累计Odds
1	0.0526 ~ 0.0702	600	10.00%	582	18	12.59%	1.31%	3.00%	3.00%	32.33	32.33
2	0.0702 ~ 0.0849	600	20.00%	534	66	11.55%	6.10%	11.00%	7.00%	8.09	13.29
3	0.0850 ~ 0.0997	600	30.00%	534	66	11.55%	10.90%	11.00%	8.33%	8.09	11.00
4	0.0998 ~ 0.1165	600	40.00%	518	82	11.20%	16.86%	13.67%	9.67%	6.32	9.34
5	0.1166 ~ 0.1454	600	50.00%	497	103	10.75%	24.35%	17.17%	11.17%	4.83	7.96
6	0.1459 ~ 0.2248	600	60.00%	471	129	10.19%	33.72%	21.50%	12.89%	3.65	6.76
7	0.2249 ~ 0.2729	600	70.00%	435	165	9.41%	45.71%	27.50%	14.98%	2.64	5.68
8	0.2729 ~ 0.3393	600	80.00%	422	178	9.13%	58.65%	29.67%	16.81%	2.37	4.95
9	0.3396 ~ 0.4774	599	89.98%	351	248	7.59%	76.67%	41.40%	19.54%	1.42	4.12
10	0.4778 ~ 0.6418	601	100.00%	280	321	6.06%	100.00%	53.41%	22.93%	0.87	3.36
TOTALS		6000		4624	1376			22.93%		—	3.36

测试情景2：跨时间样本剔除一半好样本

指标	数值
样本量	6000 → 3688（剔除好样本 2312）
KS	0.3356
AUC	0.7261
GINI	0.4522

注：情景2 模拟好样本分布右移的压力情形（固定 seed=42，可复现）；剔除后 KS/AUC 下降幅度反映模型对好样本结构的敏感度。