Scoreboard.Json

2026-09-13T04:45:35+00:00
Generated Utc
30
Window Days

Coverage

Labelcoverage = fraction of catalog events preceded by a logged, matched forecast; NOT a skill claim. It rises as we alarm MORE, so it cannot distinguish a good forecast from a wide one -- see measured_skill.

Coverage - M5

Hit173
N187

Coverage - M55

Hit49
N50

Coverage - M6

Hit12
N12

M6 Rows

Band MaxBand MinBand Width MagDateHash PrefixHitLead DaysMagMatch Distance KmModelPlaceTightness Note
84.93.12026-09-11ac5ca35c75255ef6True20.66.51264tensor_eq_per_region126 km NNE of Teluknaga, Indonesiaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
83.34.72026-09-03713da8eaa76e8c3fTrue17.36.3854.4tensor_eq_per_region84 km SSW of Nikolski, Alaskaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.73.32026-09-02616662d965d5bfbbTrue20.46.2387.2tensor_eq_per_regionSouth Sandwich Islands regiona wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
74.52.52026-08-279cd4e28210d8d11bTrue20.16575tensor_eq_per_region145 km N of Caluula, Somaliaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.93.12026-08-2405c1221d66158b01True20.461152tensor_eq_per_region211 km NNE of Lospalos, Timor Lestea wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
8532026-08-233afc918c8a874c29True20.46593tensor_eq_per_region32 km SSW of Honchō, Japana wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
7522026-08-2204efb3e37eb76bafTrue18.56.2334.7tensor_eq_per_regionScotia Seaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.63.42026-08-20b99ca59718bac7e3True20.56.71444tensor_eq_per_region37 km NE of Tambo, Perua wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
8532026-08-16cb281bb2af36653fTrue20.46.1487.2tensor_eq_per_region55 km NNE of Port-Olry, Vanuatua wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.93.12026-08-15aba1d4d8c2306bcaTrue20.56.92191tensor_eq_per_region9 km WNW of Pematangsiantar, Indonesiaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.93.12026-08-14aba1d4d8c2306bcaTrue206.1757tensor_eq_per_region62 km N of Ende, Indonesiaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart
84.93.12026-08-14aba1d4d8c2306bcaTrue207.8757tensor_eq_per_region64 km NNW of Ende, Indonesiaa wide band or a distant centroid is COVERAGE, not precision; these two numbers are what tell them apart

Measured Skill

Age Days3.8
Area Skill Score0.4793
Area Skill Score Is TrajectoryTrue
Area Skill Score Single Point0.4939
Event Hit Rate0.8676
Generated Utc2026-09-09T10:04:37+00:00
Mean Model Bss Oos0.0528
Min Mag5
N Covered354
N Events408
N Models4
N Models Positive3
Statusok
Tau Alarm Fraction0.8799
Window Days90

Measured Skill - Labels

Area Skill ScoreMolchan Area Skill Score. 0.5 = NO SKILL. Cannot be improved by alarming more: tau -> 1 drives ASS -> 0.5. Integrated over the FULL trajectory when area_skill_score_is_trajectory is true; the single-point two-segment approximation is kept in area_skill_score_single_point and is OPTIMISTIC.
Honest ReadingASS is the operator number (where to look); BSS is the pricing number (how likely). They are different products and a monotone recalibration moves BSS while leaving ASS fixed.
Mean Model Bss OosBrier Skill Score, OUT-OF-SAMPLE from an expanding-window walk-forward. 0 = no better than quoting the base rate. n-weighted MEAN over models -- NOT a pooled refit, which destroys resolution (29% single vs 4.56% pooled).
Tau Alarm Fractionfraction of climatology-weighted seismicity held under alarm. Random alarming at this fraction catches this fraction of events.

Measured Skill - No Skill Reference

Area Skill Score0.5
Bss0

Regions

Cov 90DKeyLabelN 30DRows 30D
-italyItaly1[{"band_max": 7.5, "band_min": 2.5, "band_width_mag": 5.0, "date": "2026-08-16", "hash_prefix": "1cd5fe6594cb700e", "hit": true, "lead_days": 20.7, "mag": 4.1, "match_distance_km": 395.9, "model": "te
{"hit": 12, "min_mag": 4.5, "n": 12, "window_days": 91}greeceGreece & Aegean11[{"band_max": 7.7, "band_min": 4.4, "band_width_mag": 3.3, "date": "2026-09-12", "hash_prefix": "c5c8889bcfaaa3a8", "hit": true, "lead_days": 20.6, "mag": 4.8, "match_distance_km": 377.6, "model": "ea
-turkeyTurkey & Caucasus7[{"band_max": null, "band_min": null, "band_width_mag": null, "date": "2026-09-03", "hash_prefix": null, "hit": false, "lead_days": null, "mag": 4.1, "match_distance_km": null, "model": null, "place":
{"hit": 114, "min_mag": 4.5, "n": 128, "window_days": 91}japanJapan20[{"band_max": 7.0, "band_min": 4.8, "band_width_mag": 2.2, "date": "2026-09-13", "hash_prefix": "09a7533ed1196170", "hit": true, "lead_days": 19.3, "mag": 4.9, "match_distance_km": 107.6, "model": "te
{"hit": 66, "min_mag": 4.5, "n": 97, "window_days": 91}new_zealandNew Zealand & Kermadec20[{"band_max": 6.9, "band_min": 4.3, "band_width_mag": 2.6, "date": "2026-09-10", "hash_prefix": "74dce9f26d662fce", "hit": true, "lead_days": 20.8, "mag": 4.3, "match_distance_km": 55.1, "model": "ear
{"hit": 64, "min_mag": 4.5, "n": 71, "window_days": 91}chileChile20[{"band_max": null, "band_min": null, "band_width_mag": null, "date": "2026-09-11", "hash_prefix": null, "hit": false, "lead_days": null, "mag": 4.9, "match_distance_km": null, "model": null, "place":
{"hit": 30, "min_mag": 4.5, "n": 31, "window_days": 91}peruPeru & Ecuador19[{"band_max": 8.0, "band_min": 4.7, "band_width_mag": 3.3, "date": "2026-09-12", "hash_prefix": "1a52cbab505cf83f", "hit": true, "lead_days": 20.6, "mag": 4.5, "match_distance_km": 1497.4, "model": "t
{"hit": 51, "min_mag": 4.5, "n": 51, "window_days": 91}alaskaAlaska & Aleutians20[{"band_max": 8.0, "band_min": 3.3, "band_width_mag": 4.7, "date": "2026-09-11", "hash_prefix": "0166c7c92b8733f4", "hit": true, "lead_days": 19.9, "mag": 4.3, "match_distance_km": 1408.2, "model": "t
-californiaCalifornia1[{"band_max": 7.0, "band_min": 3.2, "band_width_mag": 3.8, "date": "2026-08-29", "hash_prefix": "066767bfe2f3eed5", "hit": true, "lead_days": 20.3, "mag": 4.7, "match_distance_km": 445.1, "model": "te
-utahUtah & Yellowstone0[]
{"hit": 103, "min_mag": 4.5, "n": 108, "window_days": 91}mexicoMexico & Central America20[{"band_max": 8.0, "band_min": 4.5, "band_width_mag": 3.5, "date": "2026-09-12", "hash_prefix": "4fce36c192c222a3", "hit": true, "lead_days": 20.6, "mag": 4.2, "match_distance_km": 361.9, "model": "te
{"hit": 20, "min_mag": 4.5, "n": 24, "window_days": 91}caribbeanCaribbean8[{"band_max": 7.0, "band_min": 3.5, "band_width_mag": 3.5, "date": "2026-09-11", "hash_prefix": "67463a06eaea7702", "hit": true, "lead_days": 20.0, "mag": 4.1, "match_distance_km": 609.4, "model": "te
{"hit": 486, "min_mag": 4.5, "n": 555, "window_days": 91}indonesiaIndonesia20[{"band_max": 8.0, "band_min": 4.9, "band_width_mag": 3.1, "date": "2026-09-12", "hash_prefix": "ac5ca35c75255ef6", "hit": true, "lead_days": 20.9, "mag": 4.8, "match_distance_km": 1443.0, "model": "t
{"hit": 143, "min_mag": 4.5, "n": 185, "window_days": 91}philippinesPhilippines & Taiwan20[{"band_max": 8.0, "band_min": 4.8, "band_width_mag": 3.2, "date": "2026-09-08", "hash_prefix": "a321137f9a8d2d65", "hit": true, "lead_days": 20.6, "mag": 4.3, "match_distance_km": 513.8, "model": "te

Skill

Labelwhen-issued rate: fraction of expired M4+ predictions whose window produced an event at or above the predicted magnitude floor, trailing 30 d
N Predictions682
When Issued Full Mag Pct65.4

Skill - Climatology Ref

Labelnaive USGS-catalog climatology, walk-forward backtest Aug 2026
Rate Pct28.7
Sourcereports/magnitude_band_calibration_backtest_v2.md

Spatial Localization

Age Days0
Frac Test Within 50Km Of Past0.956
Generated Utc2026-09-13T04:31:46.256388+00:00
Honest LabelSPATIAL localization only -- WHERE earthquakes cluster (the field-standard hazard map). This is NOT an earthquake prediction/when rate, and the ASS must NEVER be read as a percent of quakes predicted; the temporal (when) skill is measured separately and is near-zero. Reproduce: python -m earthquake.smoothed_seismicity --validate
Mean Nn Dist Km14.5
Median Nn Dist Km6.6
Metricspatial localization: Molchan Area Skill Score + median epicentral distance from a real future M5+ to the nearest past-seismicity
Molchan Ass0.9782
N Test5039
N Train66634
Notespatial_auc 0.5 / ASS 0.5 = uniform (no skill); compare to current product ASS 0.48, 1442 km
Spatial Auc Vs Uniform0.9784
Statusok

Spatial Localization - No Skill Reference

Current Product Ass0.48
Molchan Ass0.5
Spatial Auc Vs Uniform0.5

Raw JSON