WeatherBlend

Multi-model forecast blending for Membury, East Devon

Rain models

Per-station P(wet ≥ 0.1 mm/h) — 3a (lean) + 3c (rich), Brier, lower better. Plus per-station rainfall amount (mm/h) — 3f distributional, CRPS, lower better. Δ vs best single NWP — negative = blend wins.

Precipitation — Chards Snowdon Hill

Phase 3c · v2026-06-15_165236_phase3c Δ -0.002 vs prev train

Rich P(wet) classifier, 101 features — rich surface (humidity/dew-depression/cloud/EA persistence) + the multi-level pressure (upper-air) block (850/700/500 hPa temps, heights, winds, RH850 across 4 exact NWPs, lead-matched by a leak-free backward ASOF). UA added in-place 2026-06-02 (−3.7%→−6.8% Brier across 24→72h). Trained 2026-06-15. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.071 precip_icon (0.136) -47.6%
+48h 0.081 precip_icon (0.135) -39.6%
+72h 0.091 precip_icon (0.164) -44.1%
+96h 0.097 precip_icon (0.186) -47.8%
+120h 0.102 precip_icon (0.226) -54.9%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +12h+24h+48h+72h+96h+120h
v2026-06-07_144138_phase3c
v2026-06-10_175508_phase3c		 304 0.053
icon: 0.023		0.011
icon: 0.005		0.001
gfs: 0.000		0.001
gfs: 0.000		0.043
ukmo: 0.067		0.002
gfs: 0.000
v2026-05-31_210952_phase3c
v2026-06-07_144138_phase3c
v2026-06-10_175508_phase3c		 600 0.002
gfs: 0.000		0.078
icon: 0.108		0.079
icon: 0.091		0.003
ecmwf: 0.000		0.115
gem: 0.182		0.172
icon: 0.200
v2026-05-31_210952_phase3c 646 —0.103
icon: 0.197		0.133
icon: 0.207		0.113
icon: 0.178		0.116
gem: 0.193		0.228
icon: 0.167
v2026-05-31_210952_phase3c 504 —0.118
icon: 0.220		0.152
icon: 0.230		0.117
icon: 0.204		0.119
gem: 0.197		0.366
icon: 0.274
v2026-05-26_102922_phase3c
v2026-05-31_210952_phase3c		 458 —0.099
icon: 0.146		0.134
jma: 0.095		0.115
ecmwf: 0.000		0.136
gfs: 0.179		0.158
jma: 0.247
v2026-05-24_142008_phase3c
v2026-05-26_102922_phase3c		 376 —0.019
mf: 0.019		0.017
ecmwf: 0.017		0.026
icon: 0.027		0.000
gfs: 0.000		0.024
icon: 0.023
v2026-05-17_174859_phase3c
v2026-05-24_142008_phase3c
v2026-05-26_102922_phase3c		 724 —0.000
icon: 0.000		0.000
icon: 0.000		0.000
gfs: 0.000		0.001
icon: 0.025		0.004
aifs: 0.000
v2026-05-17_174859_phase3c 592 —0.051
mf: 0.091		0.034
ukmo: 0.073		0.014
ukmo: 0.033		0.005
gfs: 0.075		0.030
aifs: 0.000
v2026-05-11_200152_phase3c
v2026-05-17_174859_phase3c		 568 —0.138
mf: 0.269		0.142
icon: 0.281		0.216
gfs: 0.300		0.198
ukmo: 0.273		0.197
gfs: 0.248
v2026-05-11_200152_phase3c 472 —0.098
icon: 0.180		0.104
icon: 0.182		0.007
icon: 0.025		0.004
ecmwf: 0.000		—
v2026-05-11_200152_phase3c 80 —0.201
gfs: 0.200		————

Phase 3a · v2026-06-14_152008 Δ -0.005 vs prev train

Lean P(wet) classifier, 27 features. Trained 2026-06-14. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.075 precip_icon (0.138) -45.9%
+48h 0.084 precip_icon (0.136) -38.2%
+72h 0.094 precip_icon (0.167) -43.4%
+96h 0.100 precip_icon (0.189) -46.8%
+120h 0.106 precip_icon (0.228) -53.4%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_143956 560 0.054
icon: 0.059		0.057
icon: 0.063		0.050
mf: 0.067		0.049
gem: 0.072		0.002
gfs: 0.000
v2026-05-31_210814
v2026-06-07_143956		 600 0.085
icon: 0.108		0.080
icon: 0.091		0.002
ecmwf: 0.000		0.119
gem: 0.182		0.156
icon: 0.200
v2026-05-31_210814 646 0.102
icon: 0.197		0.134
icon: 0.207		0.122
icon: 0.178		0.118
gem: 0.193		0.197
icon: 0.167
v2026-05-31_210814 504 0.117
icon: 0.220		0.150
icon: 0.230		0.126
icon: 0.204		0.115
gem: 0.197		0.304
icon: 0.274
v2026-05-26_102758
v2026-05-31_210814		 458 0.102
icon: 0.146		0.143
jma: 0.095		0.106
ecmwf: 0.000		0.133
gfs: 0.179		0.162
jma: 0.247
v2026-05-24_141845
v2026-05-26_102758		 376 0.019
mf: 0.019		0.017
ecmwf: 0.017		0.026
icon: 0.027		0.001
gfs: 0.000		0.025
icon: 0.023
v2026-05-17_174051
v2026-05-24_141845
v2026-05-26_102758		 724 0.001
icon: 0.000		0.000
icon: 0.000		0.000
gfs: 0.000		0.006
icon: 0.025		0.010
aifs: 0.000
v2026-05-17_174051 592 0.049
mf: 0.091		0.036
mf: 0.083		0.020
mf: 0.036		0.019
gfs: 0.075		0.059
aifs: 0.000
v2026-05-11_194110
v2026-05-17_174051		 640 0.128
mf: 0.269		0.141
icon: 0.281		0.222
gfs: 0.300		0.196
gfs: 0.366		0.200
gfs: 0.259
v2026-05-11_194110 544 0.099
icon: 0.171		0.124
icon: 0.203		0.018
icon: 0.038		0.005
icon: 0.000		0.004
gfs: 0.000
v2026-05-11_194110 124 0.174
aifs: 0.161		0.258
mf: 0.200		———

Precipitation — Goren

Phase 3c · v2026-06-15_165243_phase3c Δ -0.001 vs prev train

Rich P(wet) classifier, 101 features — rich surface (humidity/dew-depression/cloud/EA persistence) + the multi-level pressure (upper-air) block (850/700/500 hPa temps, heights, winds, RH850 across 4 exact NWPs, lead-matched by a leak-free backward ASOF). UA added in-place 2026-06-02 (−3.7%→−6.8% Brier across 24→72h). Trained 2026-06-15. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.081 precip_ukmo (0.152) -46.6%
+48h 0.090 precip_icon (0.143) -37.0%
+72h 0.101 precip_icon (0.167) -39.2%
+96h 0.107 precip_icon (0.196) -45.4%
+120h 0.114 precip_icon (0.235) -51.5%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +12h+24h+48h+72h+96h+120h
v2026-06-07_144213_phase3c
v2026-06-10_175544_phase3c		 304 0.115
icon: 0.125		0.051
icon: 0.053		0.012
gfs: 0.012		0.001
gfs: 0.000		0.084
ukmo: 0.088		0.023
gfs: 0.023
v2026-05-31_211026_phase3c
v2026-06-07_144213_phase3c
v2026-06-10_175544_phase3c		 600 0.075
mf: 0.000		0.090
icon: 0.124		0.089
icon: 0.114		0.084
gfs: 0.100		0.160
gem: 0.238		0.170
icon: 0.216
v2026-05-31_211026_phase3c 636 —0.142
icon: 0.201		0.156
ukmo: 0.228		0.153
ukmo: 0.208		0.189
gem: 0.286		0.214
icon: 0.204
v2026-05-31_211026_phase3c 494 —0.164
icon: 0.223		0.190
ukmo: 0.256		0.188
ukmo: 0.270		0.252
gem: 0.385		0.422
aifs: 0.250
v2026-05-26_102952_phase3c
v2026-05-31_211026_phase3c		 464 —0.142
icon: 0.194		0.144
ecmwf: 0.085		0.149
ecmwf: 0.000		0.138
gfs: 0.188		0.246
ecmwf: 0.224
v2026-05-24_142038_phase3c
v2026-05-26_102952_phase3c		 366 —0.038
mf: 0.038		0.034
ecmwf: 0.034		0.052
icon: 0.054		0.001
gfs: 0.000		0.044
icon: 0.045
v2026-05-17_174956_phase3c
v2026-05-24_142038_phase3c
v2026-05-26_102952_phase3c		 708 —0.000
icon: 0.000		0.000
icon: 0.000		0.000
gfs: 0.000		0.001
icon: 0.025		0.005
aifs: 0.000
v2026-05-17_174956_phase3c 584 —0.045
mf: 0.091		0.033
ukmo: 0.067		0.014
mf: 0.030		0.004
gfs: 0.075		0.037
aifs: 0.000
v2026-05-11_201109_phase3c
v2026-05-17_174956_phase3c		 560 —0.113
icon: 0.257		0.104
icon: 0.271		0.193
gfs: 0.222		0.217
ukmo: 0.286		0.262
gfs: 0.352
v2026-05-11_201109_phase3c 472 —0.116
icon: 0.197		0.111
gem: 0.191		0.007
icon: 0.025		0.014
ecmwf: 0.000		—
v2026-05-11_201109_phase3c 80 —0.156
icon: 0.213		————

Phase 3a · v2026-06-14_152042 Δ -0.004 vs prev train

Lean P(wet) classifier, 27 features. Trained 2026-06-14. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.084 precip_gfs (0.170) -50.7%
+48h 0.093 precip_icon (0.144) -35.6%
+72h 0.101 precip_icon (0.169) -40.1%
+96h 0.114 precip_icon (0.199) -42.7%
+120h 0.117 precip_icon (0.237) -50.4%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_144030 560 0.100
icon: 0.129		0.097
gfs: 0.131		0.095
mf: 0.108		0.079
gfs: 0.108		0.023
gfs: 0.023
v2026-05-31_210847
v2026-06-07_144030		 600 0.096
icon: 0.124		0.095
icon: 0.114		0.079
gfs: 0.100		0.163
gem: 0.238		0.164
icon: 0.216
v2026-05-31_210847 636 0.145
icon: 0.201		0.155
gfs: 0.246		0.156
mf: 0.264		0.194
gem: 0.286		0.208
icon: 0.204
v2026-05-31_210847 494 0.167
icon: 0.223		0.190
ecmwf: 0.294		0.189
mf: 0.316		0.252
gem: 0.385		0.388
aifs: 0.250
v2026-05-26_102826
v2026-05-31_210847		 464 0.150
icon: 0.194		0.145
ecmwf: 0.085		0.083
ecmwf: 0.000		0.141
gfs: 0.188		0.218
ecmwf: 0.224
v2026-05-24_141913
v2026-05-26_102826		 366 0.039
mf: 0.038		0.034
ecmwf: 0.034		0.050
icon: 0.054		0.002
gfs: 0.000		0.045
icon: 0.045
v2026-05-17_174147
v2026-05-24_141913
v2026-05-26_102826		 708 0.001
icon: 0.000		0.002
icon: 0.000		0.001
gfs: 0.000		0.007
icon: 0.025		0.007
aifs: 0.000
v2026-05-17_174147 584 0.047
mf: 0.091		0.036
mf: 0.071		0.018
mf: 0.030		0.020
gfs: 0.075		0.042
aifs: 0.000
v2026-05-11_195050
v2026-05-17_174147		 632 0.126
icon: 0.257		0.123
icon: 0.271		0.163
gfs: 0.222		0.215
gfs: 0.384		0.249
gfs: 0.350
v2026-05-11_195050 544 0.117
icon: 0.186		0.131
icon: 0.212		0.026
icon: 0.038		0.006
icon: 0.000		0.004
gfs: 0.000
v2026-05-11_195050 124 0.135
icon: 0.169		0.178
gfs: 0.200		———

Precipitation — Raymonds Hill

Phase 3c · v2026-06-15_165251_phase3c Δ -0.001 vs prev train

Rich P(wet) classifier, 101 features — rich surface (humidity/dew-depression/cloud/EA persistence) + the multi-level pressure (upper-air) block (850/700/500 hPa temps, heights, winds, RH850 across 4 exact NWPs, lead-matched by a leak-free backward ASOF). UA added in-place 2026-06-02 (−3.7%→−6.8% Brier across 24→72h). Trained 2026-06-15. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.076 precip_ukmo (0.149) -48.8%
+48h 0.085 precip_icon (0.139) -38.7%
+72h 0.095 precip_icon (0.161) -41.4%
+96h 0.103 precip_icon (0.187) -44.9%
+120h 0.109 precip_gfs (0.208) -47.3%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +12h+24h+48h+72h+96h+120h
v2026-06-07_144250_phase3c
v2026-06-10_175620_phase3c		 304 0.097
icon: 0.109		0.024
gfs: 0.024		0.001
gfs: 0.000		0.001
gfs: 0.000		0.059
ukmo: 0.067		0.003
gfs: 0.000
v2026-05-31_211102_phase3c
v2026-06-07_144250_phase3c
v2026-06-10_175620_phase3c		 600 0.008
gfs: 0.000		0.036
icon: 0.065		0.022
icon: 0.023		0.004
ecmwf: 0.000		0.120
gem: 0.188		0.138
ukmo: 0.188
v2026-05-31_211102_phase3c 634 —0.102
icon: 0.156		0.123
ukmo: 0.197		0.120
ukmo: 0.164		0.161
gem: 0.250		0.198
icon: 0.183
v2026-05-31_211102_phase3c 492 —0.115
icon: 0.171		0.145
ukmo: 0.214		0.136
mf: 0.220		0.199
gem: 0.308		0.361
aifs: 0.167
v2026-05-26_103024_phase3c
v2026-05-31_211102_phase3c		 450 —0.097
ukmo: 0.108		0.119
gfs: 0.110		0.110
ecmwf: 0.000		0.147
icon: 0.162		0.213
ecmwf: 0.234
v2026-05-24_142109_phase3c
v2026-05-26_103024_phase3c		 376 —0.019
mf: 0.019		0.017
ecmwf: 0.017		0.026
icon: 0.027		0.001
gfs: 0.000		0.022
icon: 0.023
v2026-05-17_175055_phase3c
v2026-05-24_142109_phase3c
v2026-05-26_103024_phase3c		 716 —0.000
icon: 0.000		0.000
icon: 0.000		0.000
gfs: 0.000		0.005
icon: 0.027		0.004
aifs: 0.000
v2026-05-17_175055_phase3c 574 —0.050
mf: 0.092		0.029
ukmo: 0.065		0.023
ukmo: 0.040		0.017
gfs: 0.086		0.026
aifs: 0.000
v2026-05-11_201250_phase3c
v2026-05-17_175055_phase3c		 560 —0.110
mf: 0.257		0.100
icon: 0.271		0.190
gfs: 0.222		0.222
ukmo: 0.280		0.251
gfs: 0.320
v2026-05-11_201250_phase3c 472 —0.103
icon: 0.174		0.100
icon: 0.188		0.008
icon: 0.025		0.006
ecmwf: 0.000		—
v2026-05-11_201250_phase3c 80 —0.218
gfs: 0.200		————

Phase 3a · v2026-06-14_152116 Δ -0.006 vs prev train

Lean P(wet) classifier, 27 features. Trained 2026-06-14. Metric: Test Brier.

Lead Blend Best single Δ vs best
+24h 0.079 precip_icon (0.140) -43.6%
+48h 0.087 precip_icon (0.140) -38.0%
+72h 0.096 precip_icon (0.163) -41.5%
+96h 0.104 precip_icon (0.189) -44.8%
+120h 0.112 precip_gfs (0.210) -46.7%
Verify history (11 runs)

Mon + Thu rolling Brier. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_144104 560 0.074
icon: 0.098		0.080
icon: 0.086		0.093
mf: 0.092		0.056
gem: 0.093		0.003
gfs: 0.000
v2026-05-31_210920
v2026-06-07_144104		 600 0.037
icon: 0.065		0.023
icon: 0.023		0.003
ecmwf: 0.000		0.123
gem: 0.188		0.142
icon: 0.196
v2026-05-31_210920 634 0.105
icon: 0.156		0.141
mf: 0.221		0.132
mf: 0.197		0.164
gem: 0.250		0.211
icon: 0.183
v2026-05-31_210920 492 0.121
icon: 0.171		0.169
mf: 0.269		0.146
mf: 0.220		0.198
gem: 0.308		0.372
aifs: 0.167
v2026-05-26_102854
v2026-05-31_210920		 450 0.103
icon: 0.129		0.158
gfs: 0.110		0.134
ecmwf: 0.000		0.158
icon: 0.162		0.198
ecmwf: 0.234
v2026-05-24_141940
v2026-05-26_102854		 376 0.019
mf: 0.019		0.017
ecmwf: 0.017		0.026
icon: 0.027		0.001
gfs: 0.000		0.022
icon: 0.023
v2026-05-17_174244
v2026-05-24_141940
v2026-05-26_102854		 716 0.001
icon: 0.000		0.000
icon: 0.000		0.000
gfs: 0.000		0.004
icon: 0.027		0.008
aifs: 0.000
v2026-05-17_174244 574 0.048
mf: 0.092		0.032
mf: 0.070		0.025
mf: 0.044		0.017
gfs: 0.086		0.061
aifs: 0.000
v2026-05-11_195436
v2026-05-17_174244		 632 0.101
mf: 0.257		0.116
icon: 0.271		0.150
gfs: 0.222		0.214
gfs: 0.377		0.251
gfs: 0.321
v2026-05-11_195436 544 0.107
icon: 0.165		0.109
icon: 0.203		0.018
icon: 0.035		0.005
icon: 0.000		0.002
gfs: 0.000
v2026-05-11_195436 124 0.200
aifs: 0.161		0.263
aifs: 0.200		———

Rainfall amount — Chards Snowdon Hill

Phase 3f · v2026-06-14_131629_phase3f

Two-stage distributional rainfall intensity (mm/h). Stage 1 = Phase 3a's P(wet); stage 2 = per-(station, lead) NGBoost-LogNormal fit on wet-hour rows (15-feature lean spec: 7 NWP precip + 4 spread + 4 calendar). Mixed predictive distribution F(x) = (1−π)·δ_0(x) + π·LogNormal(μ_log, σ_log)(x). CRPS is the headline skill metric — see Skill → Rain. Trained 2026-06-14. Metric: Test CRPS.

Lead Blend Best single Δ vs best
+24h NaN ngboost_lognormal (NaN val) —
+48h NaN ngboost_lognormal (NaN val) —
+72h NaN ngboost_lognormal (NaN val) —
+96h NaN ngboost_lognormal (NaN val) —
+120h NaN ngboost_lognormal (NaN val) —
Verify history (7 runs)

Mon + Thu rolling CRPS. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_130145_phase3f 126 0.059
—		0.044
—		0.044
—		0.037
—		0.000
—
v2026-05-31_125257_phase3f
v2026-06-07_130145_phase3f		 150 0.084
—		0.040
—		0.000
—		0.117
—		0.135
—
v2026-05-31_125257_phase3f 160 0.195
—		0.142
—		0.144
—		0.137
—		0.175
—
v2026-05-31_125257_phase3f 128 0.224
—		0.159
—		0.165
—		0.159
—		0.254
—
v2026-05-26_102409_phase3f
v2026-05-31_125257_phase3f		 104 0.352
—		0.200
—		0.313
—		0.219
—		0.304
—
v2026-05-26_102409_phase3f 82 0.003
—		0.003
—		0.006
—		0.000
—		—
v2026-05-26_102409_phase3f 19 0.000
—		————

Rainfall amount — Goren

Phase 3f · v2026-06-14_131629_phase3f

Two-stage distributional rainfall intensity (mm/h). Stage 1 = Phase 3a's P(wet); stage 2 = per-(station, lead) NGBoost-LogNormal fit on wet-hour rows (15-feature lean spec: 7 NWP precip + 4 spread + 4 calendar). Mixed predictive distribution F(x) = (1−π)·δ_0(x) + π·LogNormal(μ_log, σ_log)(x). CRPS is the headline skill metric — see Skill → Rain. Trained 2026-06-14. Metric: Test CRPS.

Lead Blend Best single Δ vs best
+24h NaN ngboost_lognormal (NaN val) —
+48h NaN ngboost_lognormal (NaN val) —
+72h NaN ngboost_lognormal (NaN val) —
+96h NaN ngboost_lognormal (NaN val) —
+120h NaN ngboost_lognormal (NaN val) —
Verify history (7 runs)

Mon + Thu rolling CRPS. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_130145_phase3f 126 0.080
—		0.063
—		0.074
—		0.066
—		0.003
—
v2026-05-31_125257_phase3f
v2026-06-07_130145_phase3f		 150 0.098
—		0.039
—		0.037
—		0.214
—		0.187
—
v2026-05-31_125257_phase3f 160 0.267
—		0.216
—		0.252
—		0.287
—		0.265
—
v2026-05-31_125257_phase3f 128 0.316
—		0.267
—		0.331
—		0.415
—		0.462
—
v2026-05-26_102409_phase3f
v2026-05-31_125257_phase3f		 104 0.335
—		0.144
—		0.246
—		0.200
—		0.282
—
v2026-05-26_102409_phase3f 79 0.008
—		0.011
—		0.017
—		0.000
—		—
v2026-05-26_102409_phase3f 17 0.000
—		————

Rainfall amount — Raymonds Hill

Phase 3f · v2026-06-14_131629_phase3f

Two-stage distributional rainfall intensity (mm/h). Stage 1 = Phase 3a's P(wet); stage 2 = per-(station, lead) NGBoost-LogNormal fit on wet-hour rows (15-feature lean spec: 7 NWP precip + 4 spread + 4 calendar). Mixed predictive distribution F(x) = (1−π)·δ_0(x) + π·LogNormal(μ_log, σ_log)(x). CRPS is the headline skill metric — see Skill → Rain. Trained 2026-06-14. Metric: Test CRPS.

Lead Blend Best single Δ vs best
+24h NaN ngboost_lognormal (NaN val) —
+48h NaN ngboost_lognormal (NaN val) —
+72h NaN ngboost_lognormal (NaN val) —
+96h NaN ngboost_lognormal (NaN val) —
+120h NaN ngboost_lognormal (NaN val) —
Verify history (7 runs)

Mon + Thu rolling CRPS. Per-lead cells turn red when the rolling metric breaches the lead-specific drift threshold; check the verify report for the per-cell breakdown. Version column names the trained model — a fresh champion takes ~5-9d to show.

Run (UTC) Version N +24h+48h+72h+96h+120h
v2026-06-07_130145_phase3f 126 0.071
—		0.041
—		0.057
—		0.042
—		0.000
—
v2026-05-31_125257_phase3f
v2026-06-07_130145_phase3f		 150 0.099
—		0.011
—		0.000
—		0.166
—		0.163
—
v2026-05-31_125257_phase3f 158 0.351
—		0.172
—		0.179
—		0.202
—		0.211
—
v2026-05-31_125257_phase3f 126 0.423
—		0.212
—		0.233
—		0.287
—		0.367
—
v2026-05-26_102409_phase3f
v2026-05-31_125257_phase3f		 102 0.724
—		0.218
—		0.294
—		0.453
—		0.614
—
v2026-05-26_102409_phase3f 82 0.001
—		0.002
—		0.003
—		0.000
—		—
v2026-05-26_102409_phase3f 19 0.000
—		————