A backtest lets me run a strategy through years of old prices in a few minutes. That is the fastest way to throw out obvious losers, compare bull and bear markets, and see whether fees eat the whole idea. I need it. I just cannot stop there.
What old data can tell me
A decent test can show whether the rules made money across different periods, whether one magic setting did all the work, whether costs killed the profit, and how ugly the losses got.
A backtest can also cheat without meaning to. It can peek at the future, only include coins that survived, pretend every order got the perfect price, or keep retrying history until something looks good. We built checks for that before a strategy gets fake money.
What I only learn by letting it run
- Was the signal really there? Live timing catches late, missing, or out-of-order data.
- Could the trade actually fill? The paper broker includes rejects, partial fills, spread, fees, slippage, delays, and cancellations.
- Did it watch the position? Stops, targets, time limits, and risk rules have to work while I am asleep.
- Did it survive a restart? Lost Wi-Fi or a reboot cannot create missing or duplicate trades.
Two $500 games
I am running two paper competitions. In one, every strategy gets its own $500 so I can compare them fairly. In the other, they all fight over the same $500, which exposes conflicts and shows what a real limited bankroll might look like.
The system blocks opposite bets on the same coin at the same time. It also will not sell and immediately buy back because one noisy indicator flipped for ten seconds. It has to close, wait, and see a newer signal.
Save every decision, including “no”
A rejected trade can teach as much as a winner. We save who wanted the trade, who disagreed, what the risk check said, what it should have cost, what happened afterward, and why the position closed.
Paper trading is not sitting around. It is the dress rehearsal.
A good paper run still does not promise future profit. It does show that the machine can follow its own rules long enough for the results to mean something.