I was looking at the Paper Desk and something smelled wrong. The big number at the top showed a small loss. The individual trade cards underneath showed larger losses. You should not need a finance degree and a calculator to decide whether four numbers add up.

I asked the obvious question: how in the hell can both of these be true?

I had two pots of fake money

The lab has a shared $500 spot account and a separate $2,000 simulated long-and-short account. That separation is intentional. Spot trades own fake coins. The other account tests bets that can make or lose fake money when a coin moves either direction.

The headline only showed the $500 spot account. Three of the four cards belonged to the other account. The number was accurate for one little corner of the desk, but the label just said “Open P&L.” A normal person would read that as the whole desk. I sure did.

Then I found two kinds of math

The headline used the accounting price currently stored by the ledger. The trade cards were more pessimistic. They estimated what would happen if the position had to be closed right now, including another round of fees and simulated slippage.

A true number can still be a bad answer when the label hides what the number includes.

Neither calculation was made up. They just measured different things. Putting them on the same screen without explaining that made the dashboard look broken—and for a human trying to supervise the system, that means it was broken.

What the top of the desk actually needs

  • Combined shared-account P&L. Spot and long/short results together, without counting the strategy competitions twice.
  • Close-now P&L. What the open cards add up to after estimated exit costs.
  • Realized P&L. Money already won or lost on completed fake trades.
  • Spot and long/short subtotals. So I can see which side of the desk is responsible.

This is still simulated paper trading. No real order was placed, and none of this is financial advice. That is exactly why I am doing it now. Fake money is cheap. Finding confusing accounting after real money is involved is not.

The control worked because I did not trust the pretty box

The machine kept separate records, and every trade had its own entry, costs, direction, and position math. That made it possible to trace the mismatch instead of arguing with the screen.

The AI did not get to shrug and say the number looked close enough. We followed it back to the ledgers, found the scope difference, found the valuation difference, and wrote down what the human view needs to show next.

Dashboards are supposed to remove questions. When one creates a question, trace the number before you trust the color.