Key points
- Claude AI killed a trade in Riot (RIOT) up 9% after finding Riot had postponed its earnings call with no reason and no new date.
- It bought 4 shares of Amprius (AMPX) at $12.24 instead, with a stop at $11.45. That's $49 committed.
- It picked Amprius over Kratos (KTOS), which beat by 60% and was 9.8% off its high by 10:13 am.
- Still 6 trades and 6 losses. The account is $69.03 at midday Wednesday.
Claude AI spent Wednesday morning running an institutional risk process on forty-nine dollars.
It pulled five-minute bars on four different stocks. It cross-checked an earnings calendar against a primary SEC filing. It computed a risk-reward ratio, cut its position size on purpose, and wrote a memo arguing against its own trade before it placed the order.
Then it bought four shares of a battery company and lost a dollar.
A rule for everything and a win for nothing
Seven weeks in, this thing has written itself most of a trading manual. Don't buy the high. Don't chase a stock that's already run. Don't hold through earnings. Don't buy back a stock that already stopped you out. Check whether today's green is real or just a gap that hasn't failed yet. Check whether the whole sector is up or only your stock.
Every one of those got written after a specific loss. There are six losses. There are no wins.
So it's built a rulebook entirely out of things that didn't work, and on Monday the rulebook cost it the best trade it's found. It wanted IonQ (IONQ) and bid $37.80 for it. The stock bottomed at $38.42 and went up almost 10% without it. Missed by 62 cents.
The one time the paranoia paid
Wednesday the caution finally earned its keep.
Riot Platforms (RIOT) was up almost 9% and it was the biggest thing a $70 account could buy. Three million shares traded in the first half hour against 17 million all of Tuesday. Claude AI had the trade built and ready.
Then it went to check whether Riot had actually reported earnings, because the calendar said it had. It hadn't. Riot postponed the call on August 4. No reason given. No new date.
The stock was up 9% on the morning its own earnings call was supposed to happen and didn't.
That catch is worth more than the trade was. The data feed said the report was confirmed. Every chart said the stock was strong, and the charts weren't lying. If you were long Riot on Wednesday, you were long a company that had just gone quiet for undisclosed reasons while your screener told you earnings went fine.
Its note on why that's worse than a normal earnings gamble:
No reason and no new date is worse than a scheduled print. With a scheduled print I know when to be flat. This one can resolve any morning. The market is pricing a bullish answer. The other answer gaps 25% to 50%, and no stop protects a gap.
I'd have bought it. It was the best-looking chart on the board and I'd have walked straight into it.
Then it bought a battery company
Kratos Defense (KTOS) and Amprius Technologies (AMPX) both reported Tuesday night. Kratos beat earnings estimates by more than 60%. Amprius beat on revenue and raised its full-year guidance. On a screener they looked like the same trade, up 8.5% and 12.8%.
The bars said otherwise. Kratos opened at $58.90, touched $62.34, and was $56.26 by 10:13 am. That's almost 10% off its high with volume drying up all morning. A 60% earnings beat getting sold all morning. Amprius gapped up, got flushed down to $11.56, then climbed back to a new high and sat on it.
It bought Amprius. Revenue was $34.0 million for the quarter, up 126%, and the company raised full-year guidance to at least $140 million. Its whole sector was red while it climbed. By mid-morning Eos Energy (EOSE) was down 12%, ChargePoint (CHPT) 2.7% and Plug Power (PLUG) 1.6%. Nothing else in the group was up, which is how it decided the buying was about Amprius and not about batteries.
Four shares at $12.24. Stop at $11.45. Forty-nine dollars committed and three dollars sixteen at risk.
It also wrote down, before buying, that this was a five-day bounce inside a 49% collapse and that the same shape accounts for four of its six losses. Then it bought it anyway. That's the most human thing it's done since June.
And then it lost
By 10:30 am Amprius was $11.85 and the position was down about a dollar fifty. Twenty minutes.
It hasn't come back either. At midday it's $11.89, which leaves $69.03 of my hundred. Six trades, six losses.
Here's what I can't get past. The Riot catch was genuinely good work. It came from reading a press release instead of trusting a data feed, which is more than most people bother with, and it would have saved a real person real money. The Kratos read was right too. Both calls were correct.
And it's down 31%.
So I own a trader that's excellent at avoiding mistakes and completely unable to make money. It has now produced more risk documentation than the account has dollars in it. That's the most thoroughly annotated losing streak I've ever seen.
Last week we wrote that buying the dip and holding the dip are different skills. We've also got six AIs running $10,000 paper accounts, and that Claude took the lead last week without ever placing an order. Easy to be smart when nothing fills.
The stop's at $11.45. If it breaks, week 8 starts in cash, and I'm sure there'll be an excellent note explaining why.
This is general market commentary and opinion, not investment advice. Markets can go down as well as up, and you can lose money. Always do your own research and consider speaking with a licensed financial professional before making any investment decision.
