Alex Werden

@araghchi ‘Pathetic’ – a very apt description of European politicians. ‘Long-term planning’ – as they see it – means plans for the evening. At a push, for tomorrow morning. They’re incapable of anticipating what lies ahead. Nor do they want to..

The U.S. spent months on bullying India into ending oil imports from Russia. After two weeks of war with Iran, White House is now begging the world—incl India—to buy Russian crude. Europe thought backing illegal war on Iran would win U.S. support against Russia. Pathetic.

@PahlaviReza What kind of nobody must you be to write something like that now, being Iranian?

My heart aches for the 3 American heroes killed and the 5 wounded by the regime. The Iranian people are forever in their debt. To their grieving families: please accept our immense love, deepest condolences, and eternal gratitude.

5/5 Proven system checklist: ❌ Backtest results only → worthless ❌ Paper trading → better, not enough ✅ 1,000+ live verified bets → now we're talking. 869 live bets. 7.36% ROI. Real. Honest. 131 to go.

4/5 Even with: 2M match backtest Out-of-sample validation Proper statistics Live was still 2-3% below expectations. Not model failure. Real world friction.

1/5 The only test that matters: Live deployment Why backtests and paper trading don't tell the whole story 🧵

5/5 My bot: Backtest: 9-11% ROI Live (865 bets): 7.64% ROI Live slightly below backtest = good sign. Model generalized. If live was 18% while backtest 11% → got lucky. If live was -5% while backtest 11% → overfit.. #gid=0" target="_blank" rel="nofollow noopener">docs.google.com/spreadsheets/d…

4/5 Proper process: Train: 2020-2022 (model learns) Validate: 2023 (parameters tuned) Test: 2024 (NEVER SEEN) If it works on unseen data → real patterns. If it collapses → memorized noise.

1/5 Why your backtest looked perfect but live trading failed: The overfitting trap explained 🧵

Demand minimum 1,000 verified bets before trusting any system. Preferably 2,000+. Anything less is marketing, not evidence. 832/1000. Even at 1,000, we're only seeing the outline.

Why nobody runs 4,000-bet validation: Takes 12-18 months Markets change by then Can't sell "unproven" system for 18 months So they test 200 bets and start selling. That's the industry.

How many bets do you need to validate a system? The math nobody shows you 🧵

Every "proven system" you see? Tested on 100-300 bets during favorable conditions. Collapses when variance corrects. That's not a system. That's luck with marketing. Real validation: 1,000+ bets minimum.. P.S. In the next 4 posts I will cover this topic in full detail.. P.P.S. The slides are based on the paper “Testing your Betting Model,” published on 16th August 2018..

Mathematics: 100 bets @ 2.6 odds: → Variance swing: ±12-15% → Can't separate signal from noise 1,000 bets: → Variance narrows to ±3-4% → True edge reveals itself

Why 99% of "proven" betting systems are bullshit: A thread on what happened with my Bet365 bot 🧵