Level 3 · our own model
NF-Learn
Every upstream read is persisted on our side — raw payload, distilled series, feed health. NF-Learn learns from that accumulating ledger. It is not another weather model: it learns the residual the physical families leave behind, and the fragility that precedes their failure. Where the ledger is too thin, it abstains rather than inventing a number.
Observations kept
171,819
typed, deduplicated series
Raw snapshots
21,713
every upstream payload, verbatim
Receipts issued
32,338
18,035 still open
Receipts scored
14,303
outcome became available
Head A — residual learner
One weight set per channel, shared across every entity in it. A new city inherits everything the channel has learned.
precipitationAGRICULTURE
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +72h | v75 | 24,184 | -0.11 | — | ×0.50 | 2026-09-09 12:05 |
| +6h | v75 | 51,904 | 0.01 | 97% | ×0.50 | 2026-09-09 12:05 |
| +24h | v75 | 44,334 | -0.06 | 97% | ×0.50 | 2026-09-09 12:05 |
learned weights, +72h
bias
-0.101
level_z
-0.830
lag1_delta
-0.059
lag2_delta
-0.072
mean6_delta
0.073
mean24_delta
-0.214
slope
-0.056
dispersion
-0.193
hour_sin
0.007
hour_cos
-0.007
surface_pressureAGRICULTURE
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v75 | 51,936 | -0.01 | 90% | ×1.02 | 2026-09-09 12:05 |
| +24h | v75 | 44,376 | 0.22 | 89% | ×1.13 | 2026-09-09 12:05 |
| +72h | v75 | 24,169 | 0.62 | — | ×1.12 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.013
level_z
-0.383
lag1_delta
1.358
lag2_delta
0.634
mean6_delta
-1.012
mean24_delta
0.624
slope
-2.247
dispersion
0.298
hour_sin
0.413
hour_cos
0.232
temperature_2mAGRICULTURE
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v124 | 406,490 | 0.12 | 90% | ×1.04 | 2026-09-09 12:05 |
| +24h | v99 | 371,024 | -0.13 | 90% | ×1.00 | 2026-09-09 12:05 |
| +72h | v91 | 299,171 | -0.10 | 90% | ×0.99 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.267
level_z
-0.746
lag1_delta
0.448
lag2_delta
-0.024
mean6_delta
-0.560
mean24_delta
0.110
slope
0.695
dispersion
0.533
hour_sin
0.051
hour_cos
-1.303
wind_speed_10mAGRICULTURE
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +24h | v99 | 370,977 | 0.23 | 90% | ×0.95 | 2026-09-09 12:05 |
| +72h | v91 | 299,094 | 0.25 | 90% | ×0.92 | 2026-09-09 12:05 |
| +6h | v124 | 406,427 | 0.14 | 90% | ×0.98 | 2026-09-09 12:05 |
learned weights, +24h
bias
-0.225
level_z
-1.358
lag1_delta
-0.063
lag2_delta
-0.001
mean6_delta
0.083
mean24_delta
0.656
slope
-0.070
dispersion
0.084
hour_sin
-0.178
hour_cos
0.097
pm2_5AIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v125 | 526,280 | -0.04 | 90% | ×0.94 | 2026-09-09 12:05 |
| +24h | v100 | 504,275 | -0.03 | 90% | ×1.03 | 2026-09-09 12:05 |
| +72h | v91 | 462,313 | 0.12 | 90% | ×1.03 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.273
level_z
-0.998
lag1_delta
0.067
lag2_delta
0.056
mean6_delta
0.464
mean24_delta
0.346
slope
-0.287
dispersion
-0.616
hour_sin
0.141
hour_cos
-0.623
pm25AIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 432 | 0.73 | — | ×1.00 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.014
level_z
-0.477
lag1_delta
0.004
lag2_delta
-0.034
mean6_delta
0.019
mean24_delta
-0.084
slope
0.047
dispersion
-0.068
hour_sin
-0.224
hour_cos
-0.848
precipitationAIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v75 | 31,648 | 0.04 | — | ×0.90 | 2026-09-09 12:05 |
| +24h | v75 | 24,138 | 0.35 | — | ×0.60 | 2026-09-09 12:05 |
| +72h | v36 | 6,307 | 0.32 | — | ×1.10 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.246
level_z
-1.401
lag1_delta
-0.048
lag2_delta
-0.035
mean6_delta
0.132
mean24_delta
0.166
slope
0.276
dispersion
-0.173
hour_sin
0.093
hour_cos
-0.055
surface_pressureAIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v75 | 31,636 | 0.33 | — | ×1.01 | 2026-09-09 12:05 |
| +24h | v75 | 24,136 | 0.33 | — | ×1.06 | 2026-09-09 12:05 |
| +72h | v35 | 6,286 | 0.04 | — | ×0.93 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.114
level_z
-0.890
lag1_delta
0.814
lag2_delta
0.210
mean6_delta
-0.260
mean24_delta
0.164
slope
-2.171
dispersion
0.185
hour_sin
-0.319
hour_cos
-0.123
temperature_2mAIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v125 | 526,559 | 0.35 | 90% | ×1.04 | 2026-09-09 12:05 |
| +24h | v100 | 504,537 | -0.10 | 90% | ×1.09 | 2026-09-09 12:05 |
| +72h | v91 | 462,589 | 0.10 | 90% | ×1.06 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.468
level_z
-1.081
lag1_delta
0.461
lag2_delta
0.233
mean6_delta
-0.011
mean24_delta
0.392
slope
-0.371
dispersion
0.207
hour_sin
0.288
hour_cos
0.655
wind_speed_10mAIR
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v125 | 526,415 | -0.03 | 90% | ×0.98 | 2026-09-09 12:05 |
| +24h | v100 | 504,377 | 0.15 | 90% | ×1.00 | 2026-09-09 12:05 |
| +72h | v91 | 462,441 | 0.25 | 90% | ×1.00 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.040
level_z
-1.219
lag1_delta
-0.046
lag2_delta
0.047
mean6_delta
-0.188
mean24_delta
0.534
slope
-0.081
dispersion
-0.150
hour_sin
0.402
hour_cos
0.029
ml_anomaly_scoreEARTH
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v89 | 35,595 | 0.00 | 81% | ×0.74 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.000
level_z
0.000
lag1_delta
0.000
lag2_delta
0.000
mean6_delta
0.000
mean24_delta
0.000
slope
0.000
dispersion
0.000
hour_sin
0.000
hour_cos
0.000
ml_rate_ratio_7dEARTH
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v89 | 35,591 | 0.00 | 84% | ×0.56 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.000
level_z
0.000
lag1_delta
0.000
lag2_delta
0.000
mean6_delta
0.000
mean24_delta
0.000
slope
0.000
dispersion
0.000
hour_sin
0.000
hour_cos
0.000
ml_swarm_scoreEARTH
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v89 | 35,594 | -0.01 | 95% | ×0.50 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.005
level_z
-0.905
lag1_delta
0.050
lag2_delta
0.070
mean6_delta
0.112
mean24_delta
-0.146
slope
0.004
dispersion
-0.232
hour_sin
-0.044
hour_cos
0.060
ocean_current_speedSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.43 | — | ×1.03 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.14 | — | ×0.95 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.208
level_z
-3.116
lag1_delta
-0.109
lag2_delta
-0.049
mean6_delta
1.126
mean24_delta
1.126
slope
0.243
dispersion
-0.156
hour_sin
0.165
hour_cos
-0.373
precipitationSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +72h | v75 | 21,200 | 0.27 | — | ×0.80 | 2026-09-09 12:05 |
| +6h | v75 | 48,571 | -0.28 | 94% | ×0.50 | 2026-09-09 12:05 |
| +24h | v75 | 41,019 | 0.15 | — | ×0.55 | 2026-09-09 12:05 |
learned weights, +72h
bias
-0.038
level_z
-1.230
lag1_delta
-0.161
lag2_delta
-0.084
mean6_delta
0.165
mean24_delta
0.291
slope
-0.496
dispersion
-0.272
hour_sin
-0.128
hour_cos
0.068
sea_level_mSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.77 | — | ×0.95 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.18 | — | ×1.05 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.695
level_z
-3.281
lag1_delta
-0.555
lag2_delta
-0.309
mean6_delta
1.222
mean24_delta
1.286
slope
-0.939
dispersion
-0.819
hour_sin
0.273
hour_cos
-0.049
sea_surface_temperatureSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v124 | 375,145 | -0.72 | 90% | ×0.98 | 2026-09-09 12:05 |
| +24h | v99 | 340,561 | -0.05 | 91% | ×0.82 | 2026-09-09 12:05 |
| +72h | v91 | 273,231 | -0.58 | 90% | ×0.88 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.092
level_z
-0.083
lag1_delta
-0.070
lag2_delta
-0.076
mean6_delta
-0.083
mean24_delta
-0.131
slope
0.279
dispersion
0.046
hour_sin
0.098
hour_cos
0.191
secondary_swell_directionSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +24h | v1 | 19 | 0.32 | — | ×1.05 | 2026-09-09 12:05 |
| +6h | v12 | 608 | 0.12 | — | ×1.28 | 2026-09-09 12:05 |
learned weights, +24h
bias
-0.219
level_z
-0.908
lag1_delta
0.035
lag2_delta
0.070
mean6_delta
0.145
mean24_delta
0.145
slope
0.073
dispersion
0.359
hour_sin
-0.000
hour_cos
0.219
secondary_swell_height_mSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +24h | v1 | 19 | 0.49 | — | ×0.95 | 2026-09-09 12:05 |
| +6h | v12 | 608 | -0.02 | — | ×1.00 | 2026-09-09 12:05 |
learned weights, +24h
bias
0.091
level_z
-1.343
lag1_delta
0.004
lag2_delta
-0.282
mean6_delta
0.197
mean24_delta
0.197
slope
0.177
dispersion
-0.143
hour_sin
0.000
hour_cos
-0.091
sstSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.61 | — | ×1.11 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.10 | — | ×1.05 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.496
level_z
-1.785
lag1_delta
-0.347
lag2_delta
-0.078
mean6_delta
0.731
mean24_delta
0.817
slope
0.073
dispersion
-0.613
hour_sin
0.902
hour_cos
0.304
storm_surge_mSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.51 | — | ×0.99 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.60 | — | ×1.15 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.260
level_z
-2.595
lag1_delta
-0.480
lag2_delta
0.023
mean6_delta
0.845
mean24_delta
1.045
slope
0.398
dispersion
0.246
hour_sin
-0.043
hour_cos
-0.410
surface_pressureSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +24h | v75 | 40,989 | 0.32 | — | ×0.93 | 2026-09-09 12:05 |
| +72h | v75 | 21,205 | 0.38 | — | ×1.10 | 2026-09-09 12:05 |
| +6h | v75 | 48,522 | -0.05 | 90% | ×1.00 | 2026-09-09 12:05 |
learned weights, +24h
bias
-0.565
level_z
-1.126
lag1_delta
-0.271
lag2_delta
-0.170
mean6_delta
0.308
mean24_delta
0.668
slope
-0.342
dispersion
1.302
hour_sin
0.004
hour_cos
0.146
swell_directionSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.42 | — | ×1.05 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.41 | — | ×1.05 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.135
level_z
-2.659
lag1_delta
-0.105
lag2_delta
-0.185
mean6_delta
0.973
mean24_delta
0.973
slope
0.071
dispersion
-0.247
hour_sin
-0.154
hour_cos
0.021
swell_heightSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.14 | — | ×1.07 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.33 | — | ×1.15 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.043
level_z
-1.803
lag1_delta
0.049
lag2_delta
-0.076
mean6_delta
0.660
mean24_delta
0.660
slope
0.059
dispersion
0.212
hour_sin
0.135
hour_cos
-0.092
swell_periodSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.35 | — | ×0.94 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.72 | — | ×1.05 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.189
level_z
-2.091
lag1_delta
-0.253
lag2_delta
-0.040
mean6_delta
0.958
mean24_delta
0.958
slope
-0.375
dispersion
0.175
hour_sin
-0.007
hour_cos
0.078
temperature_2mSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v91 | 374,073 | -0.13 | 90% | ×0.98 | 2026-09-09 12:05 |
| +24h | v99 | 340,547 | -0.05 | 90% | ×0.98 | 2026-09-09 12:05 |
| +72h | v91 | 273,252 | 0.28 | 90% | ×0.95 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.009
level_z
-0.460
lag1_delta
0.121
lag2_delta
-0.006
mean6_delta
0.041
mean24_delta
0.523
slope
-0.227
dispersion
-0.081
hour_sin
-0.013
hour_cos
-0.031
wave_heightSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v124 | 375,848 | 0.10 | 90% | ×0.95 | 2026-09-09 12:05 |
| +24h | v99 | 340,635 | 0.03 | 90% | ×1.04 | 2026-09-09 12:05 |
| +72h | v91 | 273,258 | 0.35 | 91% | ×0.85 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.043
level_z
-0.091
lag1_delta
0.267
lag2_delta
0.031
mean6_delta
0.598
mean24_delta
0.039
slope
-0.936
dispersion
-0.316
hour_sin
-0.030
hour_cos
0.044
wave_periodSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.06 | — | ×1.08 | 2026-09-09 12:05 |
| +24h | v1 | 19 | -0.07 | — | ×1.05 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.076
level_z
-1.811
lag1_delta
-0.101
lag2_delta
-0.186
mean6_delta
0.567
mean24_delta
0.835
slope
-0.150
dispersion
-0.133
hour_sin
-0.126
hour_cos
0.155
wind_speed_10mSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v124 | 375,257 | 0.12 | 90% | ×0.99 | 2026-09-09 12:05 |
| +24h | v99 | 340,623 | 0.29 | 90% | ×1.04 | 2026-09-09 12:05 |
| +72h | v91 | 273,280 | 0.29 | 90% | ×1.03 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.023
level_z
-0.493
lag1_delta
-0.080
lag2_delta
-0.053
mean6_delta
0.197
mean24_delta
0.377
slope
-0.090
dispersion
0.153
hour_sin
0.059
hour_cos
0.018
wind_wave_directionSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.34 | — | ×1.09 | 2026-09-09 12:05 |
| +24h | v1 | 19 | -0.32 | — | ×0.95 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.040
level_z
-2.131
lag1_delta
0.065
lag2_delta
0.062
mean6_delta
0.541
mean24_delta
0.874
slope
-0.182
dispersion
0.001
hour_sin
-0.157
hour_cos
0.110
wind_wave_heightSEA
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v12 | 608 | 0.34 | — | ×1.10 | 2026-09-09 12:05 |
| +24h | v1 | 19 | 0.05 | — | ×0.95 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.644
level_z
-2.810
lag1_delta
0.148
lag2_delta
0.093
mean6_delta
0.936
mean24_delta
0.936
slope
0.384
dispersion
-0.682
hour_sin
0.072
hour_cos
-0.257
pm2_5URBAN
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v125 | 461,743 | 0.11 | 90% | ×0.91 | 2026-09-09 12:05 |
| +24h | v99 | 432,011 | -0.04 | 90% | ×0.95 | 2026-09-09 12:05 |
| +72h | v91 | 369,306 | 0.28 | 90% | ×0.91 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.177
level_z
-0.575
lag1_delta
0.154
lag2_delta
-0.064
mean6_delta
-0.082
mean24_delta
-0.081
slope
0.635
dispersion
-0.464
hour_sin
-0.082
hour_cos
0.091
precipitationURBAN
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +24h | v75 | 35,623 | -1.41 | — | ×0.66 | 2026-09-09 12:05 |
| +72h | v75 | 16,515 | -0.07 | — | ×1.24 | 2026-09-09 12:05 |
| +6h | v75 | 43,127 | -0.73 | — | ×0.57 | 2026-09-09 12:05 |
learned weights, +24h
bias
0.178
level_z
-1.319
lag1_delta
-0.154
lag2_delta
-0.098
mean6_delta
0.148
mean24_delta
0.477
slope
-0.251
dispersion
-0.063
hour_sin
-0.342
hour_cos
-0.036
surface_pressureURBAN
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +72h | v75 | 16,519 | 0.76 | — | ×1.03 | 2026-09-09 12:05 |
| +6h | v75 | 43,162 | 0.13 | — | ×1.16 | 2026-09-09 12:05 |
| +24h | v75 | 35,665 | 0.27 | — | ×1.08 | 2026-09-09 12:05 |
learned weights, +72h
bias
-0.019
level_z
-1.551
lag1_delta
-1.286
lag2_delta
-0.585
mean6_delta
1.220
mean24_delta
-0.457
slope
2.218
dispersion
-0.301
hour_sin
-0.017
hour_cos
0.064
temperature_2mURBAN
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v91 | 460,501 | 0.09 | 90% | ×0.98 | 2026-09-09 12:05 |
| +24h | v99 | 431,962 | -0.02 | 90% | ×0.98 | 2026-09-09 12:05 |
| +72h | v91 | 369,300 | 0.05 | 90% | ×1.03 | 2026-09-09 12:05 |
learned weights, +6h
bias
0.025
level_z
-0.815
lag1_delta
0.775
lag2_delta
0.301
mean6_delta
-0.645
mean24_delta
0.444
slope
1.226
dispersion
0.047
hour_sin
0.014
hour_cos
-0.743
wind_speed_10mURBAN
| horizon | version | samples seen | skill vs persistence | empirical coverage | interval scale | updated |
|---|---|---|---|---|---|---|
| +6h | v125 | 461,776 | 0.10 | 90% | ×1.01 | 2026-09-09 12:05 |
| +24h | v99 | 432,061 | 0.11 | 90% | ×1.02 | 2026-09-09 12:05 |
| +72h | v91 | 369,312 | 0.26 | 90% | ×1.02 | 2026-09-09 12:05 |
learned weights, +6h
bias
-0.074
level_z
-1.057
lag1_delta
0.044
lag2_delta
0.033
mean6_delta
-0.091
mean24_delta
0.355
slope
-0.274
dispersion
0.309
hour_sin
0.177
hour_cos
-0.613
Head B — fragility
P(validity breach within 72h), trained on our own scored receipts. Features are observable before the breach, never after.
Brier skill
-0.14
vs base rate
Base rate
0.63
observed breach frequency
Samples seen
355
v1
bias
0.637
recent_error_z
-1.782
error_trend
0.313
coverage_deficit
4.047
staleness
0.000
dispersion
-0.610
horizon_z
-0.239
Scoreboard
NF-Learn's own receipts, scored against what actually happened.
| model | n | MAE | coverage |
|---|---|---|---|
| nf-learn:sea:wind_wave_direction | 77 | 40.12 | 62% |
| nf-learn:sea:swell_height | 77 | 0.09 | 88% |
| nf-learn:sea:ocean_current_speed | 77 | 0.52 | 40% |
| nf-learn:sea:swell_period | 77 | 0.93 | 75% |
| nf-learn:sea:wave_period | 77 | 0.53 | 38% |
| nf-learn:sea:wind_wave_height | 77 | 0.16 | 30% |
| nf-learn:sea:secondary_swell_height_m | 77 | 0.09 | 38% |
| nf-learn:sea:swell_direction | 77 | 18.86 | 62% |
| nf-learn:sea:wave_height | 77 | 0.08 | 62% |
| nf-learn:sea:sst | 77 | 0.58 | 4% |
| nf-learn:sea:storm_surge_m | 77 | 0.02 | 27% |
| nf-learn:sea:sea_level_m | 77 | 0.41 | 68% |
| nf-learn:sea:secondary_swell_direction | 76 | 43.60 | 71% |
Learning cycles
Ingest → train → issue → score. Idempotent: a repeated cycle re-reads the same window without duplicating a single row.
| started | trigger | observations | issued | scored | train samples |
|---|---|---|---|---|---|
| 2026-09-09 12:05 | scheduler | 638 | 882 | 1,248 | 181,525 |
| 2026-09-09 11:05 | scheduler | 539 | 666 | 0 | 181,145 |
| 2026-09-09 10:17 | open-meteo | 6,720 | 28,011 | 192 | 0 |
| 2026-09-09 10:05 | scheduler | 640 | 737 | 0 | 181,078 |
| 2026-09-09 09:17 | open-meteo | 6,720 | 28,011 | 612 | 0 |
| 2026-09-09 09:05 | scheduler | 641 | 737 | 0 | 180,848 |
| 2026-09-09 08:17 | open-meteo | 6,720 | 28,011 | 517 | 0 |
| 2026-09-09 08:05 | scheduler | 642 | 666 | 240 | 180,794 |
| 2026-09-09 07:17 | open-meteo | 6,720 | 28,011 | 1,488 | 0 |
| 2026-09-09 07:05 | scheduler | 643 | 666 | 0 | 180,902 |
| 2026-09-09 06:17 | open-meteo | 6,720 | 28,011 | 1,380 | 0 |
| 2026-09-09 06:05 | scheduler | 644 | 666 | 0 | 180,845 |
Ledger opened 2026-06-13 00:00. Discrete events — earthquakes above all — are stored as events and never resampled into a continuous series, so no head can learn from a curve that does not exist. Scheduler endpoint: /api/public/engine-cycle.