01
COUNTERFACTUAL HISTORIES
HURRICANE ANDREW 1992 · 60,500 COUNTERFACTUAL SEASON-YEARS · IBTrACS × NFIP × ALPHAEARTHDOWNLOAD DATASET · XLSX · 28 SHEETS↓INITIALISING GL…
SIMULATED PEAK 1-MIN WIND, 0.05° FIELD346496130 kt
Simulate Hurricane Andrew (1992) arriving on a slightly different track. What does a catalogue anchored on the one observed history miss?
INITIALISING GL…
INITIALISING GL…
INITIALISING GL…
WIND AT SIX EXPOSED COUNTIES85 kt Miami-Dade
0kt40kt80kt120kt
-1800+180
MIAMI-DADE FLBROWARD FLPALM BEACH FLCOLLIER FLLEE FLORLEANS LA
EXPOSURE VS LANDFALL SHIFT2.65× @ +140 km
$0B$10B$20B
-1800+180
wind-weighted exposure $BTIV ≥ 64 kt (÷30)
POPULATION UNDER THE EYEWALL AT EACH INSTANT2.00M at t2
0M2M4M6M
-1800+180
t1 · 08-23 18:00Zt2 · 08-24 09:00Zt3 · 08-26 09:00Z
CUMULATIVE POPULATION HIT BY EACH INSTANT2.22M by t3
0M4M8M12M
-1800+180
t1 · 08-23 18:00Zt2 · 08-24 09:00Zt3 · 08-26 09:00Z
PEOPLE AND HOUSING EXPOSED2.80× at +100 km
-1800+180
population (M)housing units (M)owner-occupied (M)aged 65+ (M)under 18 (M)
GENERATING CAPACITY EXPOSED3.28× at +120 km
-1800+180
total capacity (GW)gas (GW)nuclear (GW)solar (GW)
LABOUR, INCOME, LAND STATE3.28× at +100 km
-1800+180
unemployed (×10³)degree holders (×10³)tracts in footprintAlphaEarth PC1 sum
EXCEEDANCE CURVE1-in-234
$0B$25B$50B
1:101:1001:1k1:10k
counterfactualobserved worst season
02
CHOKEPOINT TO MAIN STREET
HORMUZ -93.1% FROM 2026-03-02 · PORTWATCH × FRED × EIA × 2,002,356 NFIP CLAIMSDOWNLOAD DATASET · XLSX · 26 SHEETS↓INITIALISING GL…
ALPHAEARTH 2022→2025 SURFACE CHANGE, LAND ONLYnonemidhigh
If the Strait of Hormuz stays closed in 2027, simulate the impact on a small business in Florida.
INITIALISING GL…
INITIALISING GL…
INITIALISING GL…
HORMUZ DAILY TRANSITS AND CRUDE-93.1%
04080120
Hormuz transits/dayWTI $/bbl
WHOLE NETWORK VS HORMUZ ALONEALL-28 91
050100150
all 28 chokepoints, transits vs 2019–23 = 100Strait of Hormuz alone100
HOW MANY CHOKEPOINTS ARE OFF NORMAL9 DOWN · 3 UP
051015
chokepoints ≥20% below their own normalchokepoints ≥20% above
SIX MAJOR CHOKEPOINTS INDEXED-91.7%
050100150200
Strait of HormuzBab el-Mandeb StraitCape of Good HopeSuez CanalMalacca StraitPanama Canal
AS-OF COEFFICIENTS, RE-ESTIMATED EACH MONTHΣβ 0.44 @ 2026-04
-1.00.01.0
crude → pump β (expanding window)+1.96 seoil → claim severity βzero
STANDING ANOMALY COUNTER, ALL MACRO SERIES10 / 32 @ 2026-04
0102030
macro series outside their own trailing 95% bandseries scored
MACRO CASCADE, z VS PRE-EVENT8 BEYOND 95%
-202612
#1#5#9#12
z-score vs pre-event window, Hormuz 2026same series, Red Sea 202395% band−95% band
AS-OF SCENARIO ANSWER AT THIS MONTHfuel +24.0%
0%25%50%75%
implied fuel-cost change, as-of β95% upper95% lower
03
INTERNATIONAL EXPOSURE DATA
40.1% NAME-JOIN → 52.9% RESOLVED · 3,130 COUNTIES OF HIDDEN TRUTH · ALPHAEARTH × GLOPOP-S × NFIPDOWNLOAD DATASET · XLSX · 23 SHEETS↓INITIALISING GL…
SUBNATIONAL HDI AS A SUBMISSION CARRIES IT — ONE NUMBER PER COUNTRY0.358mid0.940
LOCO R² 0.524 · 934 UNITS / 107 COUNTRIES
INITIALISING GL…
Bring this international submission up to the resolution of our US book: resolve the region names, then reconstruct the sub-national exposure split we were never given.
INITIALISING GL…
INITIALISING GL…
INITIALISING GL…
FRAME RESOLUTION, LOG SCALE85,529
10^310^410^5
TRACTSCOUNTIESADMIN-1GDL
units available per frame (log₁₀)
WHAT ONE COUNTRY NUMBER HAS TO STAND FOR8.6% LOST
widest unit inside the countrynarrowest unitthe one number the submission carries
COVERAGE, 262 ECONOMIES98.3%
0%50%100%
0%25%50%75%100%
World Bank indicator coverage, economies sortedUSA
MATCH RATE VS TOKEN THRESHOLD52.9%
0%25%50%75%
0.10.30.50.70.9
matched share vs token thresholdexact string join
LIFT · TRUE INSURED VALUE CAPTURED$2.0T
0%25%50%75%100%
0%25%50%75%100%
oracle (perfect ranking)AlphaEarth per capita + areaAlphaEarth + populationpopulation splitAlphaEarth alone
RECONSTRUCTION ERROR BY METHOD47.1%
0255075100
m1m3m5m7m8
top-county error %share MAE (pp) ×20R² (log TIV) ×100
TRANSFER · OBSERVED VS PREDICTEDR² 0.52
0204060
observed (sorted)predicted, leave-one-country-out
R3 · IF THE k LARGEST UNITS ARE ALREADY NAMED47% → 2%
0%25%50%
k=0k=3k=5k=8k=12k=20
top-county error %top-5 accumulation error %