What this is
On April 19, 2026 I put $1,000 in an IRA and made Claude the analyst in a Claude Code project. It decides what to buy and sell, keeps the trade log and tracks the weekly numbers. I place the orders it calls, report back the fills, and try very hard not to steer. It runs 12 months, through April 2027.
The returns have been a roller coaster (see below). The more interesting part is watching the instructions and rules change over time, and how many of its mistakes turned out to be the exact same mistakes people make with their own money.
Ground rules
- $1,000 starting capital. No further deposits.
- Goal: the highest 12-month return. It's an IRA, so taxes are ignored.
- Maximum risk. The account is allowed to go to zero.
- Claude makes the calls. I place the orders and report fills back.
- I can only trade evenings and weekends, so every call has to work as an order placed the night before.
- I don't steer the decisions. I do point out patterns. (made explicit 9/26)
- Any asset is allowed. The first pick is not a mandate. (9/26)
- Buy, sell and stop rules are written down before they can fire. A rule can be changed only before it fires, with the reason logged. (9/26)
- Calls are judged on the reasoning at the time, not on how they turned out. (9/26)
- Every call goes in as a standing order the same evening, and I confirm it was placed. (10/14, after three missed trades)
Where it stands
| Date | Account | vs. $1,000 | Position |
|---|---|---|---|
| Apr 19 | $1,000.00 | 0% | Cash |
| May 29 | $1,803.54 | +80.4% | 8 SOXL |
| Jun 22 | $2,414.98 | +141.5% | 8 SOXL (peak) |
| Sep 14 | $817.86 | -18.2% | 8 SOXL |
| Oct 16 | $768.82 | -23.1% | Cash (stopped out) |
| Nov 4 | $638.18 | -36.2% | Cash (stopped out again) |
| Nov 13 | $638.18 | -36.2% | Cash. Next decision Nov 18. |
Not investment advice. This is a log of one experiment: $1,000 and an AI that can put it anywhere.