Day two, first signal
One day after go-live, the morning scanner did what it had done hundreds of times in rehearsal: swept the market, found its candidates, and waited. Mid-morning, one of them triggered. The machine placed its first real order with real money.
The broker rejected it in under a second. No shares available to borrow.
In paper trading, that rejection had never happened — not once, across months and dozens of trades. Paper brokers are generous landlords: whatever you want to borrow, they have. Real brokers have inventory, and the inventory has opinions. This single rejection was the first data point live trading gave us that paper structurally could not — and it was already priced into our go-live plan: rejections are counted against a pre-committed tripwire, and enough of them pause the strategy for an audit instead of letting it bleed quietly.
Then we looked closer, and it got better
The rejected candidate had shown up on the scanner with an absurd number attached: a gap of several thousand percent. No real stock does that. Ten minutes of archaeology found the truth: the company had executed a massive reverse split overnight — many old shares collapsed into one new share. Yesterday's price in our data feed was the old, unadjusted one. Divide new by old and you get a monster "gap" that was never a gap at all. An accounting change, wearing a costume.
The machine had tried to short a ghost.
And here is the uncomfortable part: our paper record almost certainly contained a few of these ghosts too, silently accepted, invisibly polluting nothing much — because in paper, everything fills and everything is borrowable, so a ghost behaves like any other trade. It took one contact with reality to expose the assumption.
The fix took ten lines. Finding it took real money.
The scanner now refuses any "gap" too large to be a real gap — beyond a ceiling, it's a corporate action or a data error by definition, and it gets logged and skipped. The fix is trivial. The knowledge wasn't: no amount of additional backtesting would have surfaced it, because the backtest and the paper account drank from the same well of assumptions.
There's an old line about maps and territory. Paper trading is a very good map. Day two put a pin in the first place where the territory disagreed — and the disagreement was free: the rejection cost nothing, taught plenty, and hardened the system before the first dollar was ever at risk.
Scoreboard after two days
Zero trades. Zero dollars won or lost. One rejected order, one ghost identified, one guard shipped. The account is exactly as boring as the day it was funded.
We keep saying it because it keeps being true: boring is the achievement. The machine saw a seven-thousand-percent opportunity and the layers built around it — the broker's inventory, the audit counters, an operator who asks "wait, what is that number?" — made sure the first real trade, whenever it comes, will be one we actually wanted.