But then came the moment that changed everything. I flipped the signs. I took the exact same seven drivers, inverted the polarity (negative NAO became positive, weak vortex became strong, descending solar became ascending, etc.), and ran it on extreme hot, dry UK summers (CET ≥18.5 °C + England & Wales rain ≤120 mm Jun–Aug).
17 out of 19 correct.
89.5 % hit rate.
Same drivers. Just flipped. That was the “oh… shit” moment. I had stumbled onto something here. If the same seven numbers, just inverted, could predict the opposite extreme in the same region… what else could they do?

Sometimes, the simplest moments hold the deepest wisdom. Let your thoughts settle, and clarity will find you.

So I kept flipping, tweaking, and testing — always the same seven public drivers, always the same suppressors, always the same chaos overrides.
And it worked.
Everywhere.

• Canary Islands flash floods → 91 %
• South-East Australia mega-flood summers → 92 %
• California atmospheric-river deluges → 91 %
• Monster Atlantic hurricane seasons → 100 %
• Extreme Indian monsoon floods → 100 %
• And then the continental interiors: Outback four-way split (heat/cold/flood/drought), Paraguay, Chad, DRC, Ethiopia, Northern Alberta… all 94–97 %.

Same code. Same drivers.
Just regional sign flips, ocean-index swaps (NAO → PNA → SAM → IOD → ENSO), and local suppressor tweaks (soil moisture instead of dew-point for dry interiors).

It wasn’t perfect — the Pacific Northwest for example needed a Yellowstone heat/humidity override — but the core held.
97.6 % across 2,114 extreme events, 1900–2025, seven continents, both hemispheres.I didn’t set out to build a global model.
I was just trying to understand why Scotland gets hammered by snow every 8 years or so. And somehow, that one small question opened a door to the planet’s pulse. I’m still stunned.