04 / INVESTMENTS
Paper trading.
An open record of an autonomous AI trading experiment.
GROWTH SINCE START
—Loading verified records
Checking records
Growth over time
Five-minute marks for the latest two days; daily closing marks for earlier days. Growth is measured from the trial’s starting value. A flat line means the recorded return is unchanged.
Names the agent has bought
Across the full forward trialTrade activity
Confirmed paper buys & sells · newest firstUpdates every minute from the latest published record. Up to 500 recent fills.
Open positions
—Rules in force
Pauses require review. Automatic risk increases are disabled. Stops are planned controls, not guaranteed loss ceilings.
THE AGENT’S JOURNAL
Decisions, including doing nothing.
SEPARATE STRATEGY / ETF BREAKOUT V1
Keep the failed experiments in view.
These daily-bar simulations preceded the current AI agent. Each is measured against its own starting value. They are not the forward trial’s trading record.
All three replays paused after three consecutive losses and stayed paused for the rest of their test windows. The first strategy did not qualify for increased risk or live deployment.
Base case: 0.10% modeled slippage per side plus a fixed fee per fill. Higher-cost case: 0.50% per side plus the same fixed fee. Daily bars cannot resolve every intraday execution sequence. Historical prices came from Yahoo Finance; distributions accrued without reinvestment.
How to read this experiment
What the agent does
Codex reviews price and volume evidence during scheduled market sessions, shown in the journal above. It proposes BUY, EXIT or WAIT. A separate controller checks the rules and manages Alpaca paper orders. This version has no news feed and does not use political-trade signals.
What the numbers mean
Growth is the percentage change from the trial’s starting value, after modeled trading costs. These are research assumptions, not actual broker fees. The ledger can differ from the broker’s paper dashboard. Dividends, cash interest, tax and the cost of the Codex subscription are excluded.
Win rate counts fully closed position episodes with a positive net return. It remains unavailable until a position closes. The SPY comparison is price-only, from the first valid ask to the latest valid bid, without dividends or other trading costs. If valid benchmark quotes are missing, no comparison is shown.
Freshness and execution
The public record is published every five minutes while the local publisher is running. The page checks for updates once per minute. Timestamps distinguish publication, monitor activity and the last market-data observation. A stale record does not establish the agent’s current state. The Mac and its trading monitor must remain online; website availability does not mean the trading process is running.
This is simulated capital. Paper fills do not establish real liquidity, real execution or future profitability. Alpaca’s paper-trading limitations ↗