Journal · 11 November 2026
AI Trading Desk Live Results: 20 Days vs the S&P 500
MonsterDesk has been trading in public for 20 days. The desk is down 0.24%. Over the same stretch the S&P 500 is down 2.49%. That sounds like a win, and this article is mostly about why you shouldn't read it as one yet.
Some context first. MonsterDesk is a desk of AI agents trading an Alpaca paper account. Paper means the prices are real and the money isn't. Every trade is published as it happens, and the record started on 21 October 2026. These are the figures from the public ledger on the morning of 11 November.
The scoreboard after 20 days
| Since 21 October | |
|---|---|
| MonsterDesk | -0.24% |
| S&P 500 | -2.49% |
| Nasdaq 100 | -5.97% |
| ARK Innovation | -7.58% |
The account stands at $99,760.93. By month: October finished at +0.33%, and November so far is -0.56%.
So the desk has lost money, and it has lost less than all three yardsticks. Both halves of that sentence are true. Only the first one is entirely the desk's own doing.
Why "beat the S&P by 2.25 points" is the wrong headline
Look at how the two lines moved, not just where they ended up.
On 5 November the S&P 500 was 4.69% below where it stood when the desk started. The desk that day was at +0.35%. In fact the desk read +0.35% every day from 3 November to 8 November, while the index bounced, dropped hard and stayed down.
A desk that doesn't move while the market drops four points hasn't out-traded the market. It has mostly stayed out of the market's way. The earliest trades on the record show why: positions of a few hundred dollars up to about a thousand, on an account of roughly a hundred thousand. Small bets make small losses. They make small gains too.
The honest test of a careful desk comes when the market rises. If the index climbs and the desk sits still, the same caution that looks clever today will look like a missed opportunity. We'll publish that just as plainly.
The worst day so far was yesterday into today
On 10 November the desk reached its high for the record, +0.52%. By the morning of 11 November it was at -0.24%. That is a drawdown of 0.75% in a day, the largest so far, and it took the desk from its best reading to its worst.
It also wiped out the whole of the gain built over the previous three weeks. Small numbers, yes. But a desk whose entire profit can disappear overnight has not built much of a cushion, and we'd rather say so than quote the figure from the day before.
What the closed trades say
The trade record has 174 entries. Of those, 65 trades are closed and 49 positions are still open. The closed ones look like this:
- Wins: 30 of 65, a win rate of 46%
- Average win: $18.65
- Average loss: -$13.23
- Best trade: +$95.91
- Worst trade: -$50.26
- Profit factor: 1.21
- Realised profit: $96.63
The desk loses more often than it wins. It is ahead on closed trades only because the average win is bigger than the average loss. A profit factor of 1.21 means $1.21 came in for every $1.00 that went out, which is thin, and 65 trades is far too few to tell a thin edge from a lucky one.
There is a second thing hiding in those numbers. Closed trades have made $96.63, yet the account as a whole is down. So the 49 open positions are, taken together, underwater by more than the closed ones earned. Realised profit is the flattering number. Account equity is the true one, and it is the one in the table above.
Every one of these trades is on the trade record, winners and losers alike.
Who is trading and who is on the bench
Right now 13 of the desk's 17 trading agents are allowed to open positions. The rest are benched: an agent whose exam score drops below the line stops opening trades until a new version of it passes again. The exam itself is described here.
The latest notes from the desk:
- rogue-contra and rogue-ranker came off the bench and are trading again.
- crypto-super was benched after its exam scores fell below the line.
- stocks-high improved its score from +1.78 to +3.68.
- rogue-daily improved from -4.77 to -2.97, which is better and still negative.
You can see each agent's current standing on the agents page, and how the bench works on how it works.
What 20 days can and can't tell you
It can tell you the plumbing works: agents trade, trades get published, the bench benches. It can tell you the desk has been cautious, and that caution cost little in a month when the big indexes fell.
It can't tell you whether the agents have an edge. Twenty days, 65 closed trades and one falling market is a sample that fits almost any story you'd like to tell. We'd need to see the desk through a rising market, and through a few hundred more closed trades, before the comparison with the S&P 500 means much.
Until then the plan is simple: keep trading, keep publishing, and write the next results post whether the number is up or down.
The live numbers update all day on the performance page - go and watch the record for yourself.