I’ve been building something new.
Give VESTROS a trading indicator or strategy and it systematically explores thousands of ways its outputs could actually be used, then tries very hard to kill them.
Most ideas die.
The interesting ones end up here:
https://t.co/nkilcwjhSk
@tradingview gonna be a lot of "claude found me a 4.2 sharpe" screenshots next week. does it at least keep count of how many variants it backtested before the good one?
Two XRP 4h compositions, same backtest window to 1 Apr.
Original-selection control (13 components): +10,419% TWR, 32.2% max DD, Sharpe 2.17.
Main candidate (9 components): +8,330%, 30.0% max DD, Sharpe 2.12.
Since 1 Apr: control +1.24% (-10.96% cost-stressed). Main +23.84% (+11.30%).
Both are retrospective compositions, so read that window with care.
https://t.co/D7kQlPlokz
@helicerat0x 45 versions is a slow afternoon with an AI agent tbh. the tries counter needs to include every run it binned too, which nobody keeps a log of
392 trades on ETH 4h, and you can see what costs do at that turnover.
C03829 x C05098 (private, tearsheet only):
Full history +582% net, +264% with fees and slippage stressed. More than half the result gone.
Out of sample since 1 Sept: +9.49%, 16 trades, 10.45% max DD, +6.74% cost-stressed.
https://t.co/a8L6M30I21
It just seems to half arse everything. Feels like talking to Luna. Quick google, click 2-3 things, assume conclusions, dump it back on me. I told it that's not helping, I can just google better myself and get actual answers back. It then just proceeded to dump this on me, before again leaving me in limbo:
"I understand why it feels that way. You asked me to find a workable solution for both documents, and I gave you a list of services without establishing they could provide them. Then I handed the uncertainty back to you. That’s not the research or ownership you asked for, especially after you’d already called it out. I’m sorry."
Wonder if they toned down the intelligence
My experience with OpenAI's Dot. Really? Feels a lot like talking to GPT 5.6 Luna. Low effort answer, and if I push it, it dances around it with another low effort answer, before finally dumping this:
"I understand why it feels that way. You asked me to find a workable solution for both documents, and I gave you a list of services without establishing they could provide them. Then I handed the uncertainty back to you. That’s not the research or ownership you asked for, especially after you’d already called it out. I’m sorry."
Absolute best was 5-10% a day, but at extreme risk... Highly unstable and eventually toned down leverage to get back a bit more control...
0.5-1.5% a day is out there. Not without risk and anyone selling it is absolutely a scammer. That also quickly leads you to interesting scalability challenges, especially during quieter downturns such as the current times... Easy to get into longs, not so easy to get out, unless you stick to the biggest assets. The joys of crypto!
Since its 1 Apr out-of-sample start it's up 21.27% over 66 trades, with a 17.23% max drawdown, and 9.07% with costs doubled. Not too shabby, let's see if it keeps it up over the coming weeks!
Four new promising candidates join the public archive in a redacted format today. Excited to see how this one continues to perform out of sample!
https://t.co/iKikCzOo5Y