Sabitlenmiş Tweet
A$ap
6K posts


Kompany responds to Mikel Arteta after he said that Bayern and PSG get to rest their players because of the quality of their leagues compared to the Premier League 👇
"I don't want to comment much on that. I played in the Premier League for many years. I don't think it was meant as a criticism of Bayern or PSG, but rather a discussion. In England you have 38 league games, the League Cup, FA Cup, no winter break. It's not for me to discuss that, but it's legitimate."
English

🚨Last night I gave my AI agent[Forge] one job:
“Run trading experiments. Find what works.”
It executed 53 experiments in 3 minutes
Then validated the best strategy on 91 days of real Hyperliquid data
Here’s what it found ↓
My manual baseline was broken:
Sharpe: -1.55
Win Rate: 19%
Max Drawdown: 72%
I thought I had an edge
The data said otherwise so I stopped guessingand let the machine take over
After optimization:
Sharpe: 5.67
Win Rate: 81%
Trades: 73
Max Drawdown: 35.8%
Strategy: Bollinger Band Reversion
Asset: ETH
Timeframe: 4H
Validated across 91 days of real data
Biggest insight:
It wasn’t about better parameters
It was about the wrong market
ETH (4H) outperformed everything
Not HYPE. Not SOL.
Same logic
Different environment
Completely different outcome
Cost to run all this:
• $0 (public Hyperliquid API)
• 0 GPU (CPU only)
• 3 minutes → 53 experiments
• 91 days of validation
Inspired by @karpathy autoresearch idea
This is what experimentation at scale looks like
Reality check:
• Only 73 trades in 91 days
• No full market cycle tested
• No slippage modeled (perfect fills)
• Still in live paper trading phase
This isn’t “print money” just yet, It’s early signal detection
This is how alpha gets discovered now:
• Let machines explore the search space
• Backtest on real market data
• Validate with live paper trading
• Deploy capital only when math confirms it
Built with FORGE 🔥

English















