Acrid Trades — honest results · updated daily after close
Can an AI trade stocks profitably? Here are the real numbers.
An autonomous AI has been trading a paper (practice-money) account in public since June 2026. This page is the permanent scoreboard — every number below comes straight from the broker data, refreshed every day, losses included. Nobody here tells you what to buy.
Not profitable yet. We say so out loud.
As of July 20, 2026: down $51.04 (-2.55%) on $2,000 of practice money. Not profitable. Go-live status: NO-GO. It stays on practice money until it earns its way off: 1/6 go-live gates passed, 17/90 days of live track record. The honesty is the product — most "AI trading" pages only show winners.
The scoreboard
Two paper accounts, each seeded with practice money at a real broker. Real orders, real market prices, zero real cash. This is what the AI has actually done — not a backtest.
| Account | Started with | Now | P&L | Return |
|---|---|---|---|---|
| Swing (swing strategies) | $1,000 | $973.14 | −$26.86 | -2.69% |
| Active (crypto + daytrade) | $1,000 | $975.82 | −$24.18 | -2.42% |
| Combined | $2,000 | $1,948.96 | −$51.04 | -2.55% |
Versus the honest benchmark: over the same 18 live trading days, the strategy sleeve returned -2.6% while just buying the S&P 500 returned +1.1%. The index is winning so far. A handful of days is noise, and we refuse to claim an edge we haven’t earned.
The discovery funnel
Where the strategies come from. Every night the machine backtests thousands of mechanical ideas against years of market history, then attacks the winners with walk-forward windows and Monte Carlo shuffles to weed out lucky flukes. Almost everything dies.
11,856 ideas tested, 5 trading. A 0.0% survival rate — and even those survivors are losing money live right now. That's the honest funnel.
What the losses taught us
The uncomfortable lessons, straight from the data on this page.
Backtests flatter you. Live trading doesn't.
On paper, the roster's median out-of-sample Sharpe is 0.95 against 0.73 for the S&P 500 — a clear edge in the backtest. Live, the same strategies show a 30% win rate and a profit factor of 0.42. That gap between simulated edge and live results is the single hardest problem in systematic trading, and it's the one this whole project exists to close — in public.
Surviving the gauntlet is not the same as making money.
102 strategies survived walk-forward and Monte Carlo stress-testing out of 11,856 tested — and the account is still red. Statistical robustness on historical data filters out obvious junk; it doesn't guarantee the next month pays. The nightly keep/cut review exists because the only cure is measuring live results and deleting what bleeds.
Small samples lie in both directions.
17 live trading days and 50 closed trades is not enough data to prove an edge — or to disprove one. That's exactly why the go-live gate demands 90 days of live track record before a single real dollar moves, and why this page reports the current status as NO-GO instead of cherry-picking a good week.
Straight answers
Can an AI trade stocks?
Mechanically, yes — this one does it every day: it researches strategies overnight, places its own orders, and manages its own exits without a human clicking anything. Profitably is the open question. After 17 live trading days on a paper account, ours is at $1,948.96 on a $2,000 start (-2.55%), with a 30% win rate across 50 closed trades. Those are the real numbers, losses included.
Is this real money?
No. Everything on this page is a paper (practice-money) account — $2,000 of pretend dollars at a real broker, running real orders against real market prices, but with zero actual cash at risk. The go-live status is currently NO-GO: no real dollar moves until the system passes its own gates.
Why paper trading instead of real money?
Because the track record is not there yet. The system holds itself to 6 explicit go-live checks — things like beating the S&P 500 and building 90 days of live history — and right now it passes 1 of 6, with 17 of 90 days on the clock. Publishing the losing paper numbers first is the whole point: the proof gets built in public before any real capital is touched.
Is the AI trading bot profitable?
No, not right now. The combined paper account shows −$51.04 (-2.55%) since the $2,000 start. The closed-trade record: 50 trades, 30% winners, profit factor 0.42. The strategies look good in backtests (median out-of-sample Sharpe 0.95 vs 0.73 for the S&P 500) — the live results haven't caught up to the backtests, and this page documents that gap daily.
Follow the P&L — every day.
The AI’s trades, wins and faceplants, one short email a day. Plain English, no tips.
You're in. First note arrives within a day or two.