Everyone knows backtests are optimistic and live trading is where dreams die. Almost nobody can tell you the size of that gap, or its exact anatomy, because doing so requires running the same strategies both ways and publishing the ugly result. We did: 42 strategy configurations paper-traded on live market data, in parallel with their backtests, for weeks. This is what the gap is actually made of.
Our paper traders take every signal their strategy fires and track it against live prices — real fills simulated with slippage, real time, no hindsight. The backtests for the same strategies looked great. The paper dashboards looked great too… right up until we forced honest accounting.
A losing position that hasn't been closed doesn't show up as a loss — it's just "unrealized". Paper dashboards (and humans) exploit this endlessly: winners get closed and booked, losers stay open and invisible. When we force-closed everything at market price — the only honest way to read a portfolio — cells showing "+$59K" collapsed. That particular cell's profit turned out to be three lucky 5x moonshots sitting on top of hundreds of open losers. Realized −$312K plus unrealized −$173K across the board.
Backtests fill you at the printed price. Live, thin books move against you: on memecoins the difference between the signal price and an actual executable fill routinely exceeds the entire theoretical edge. A strategy whose backtest edge is thinner than its market's real slippage is a loser that hasn't been introduced to itself yet.
Live data feeds glitch. One token printed a two-billion-dollar market cap for a few minutes on a manipulated pool; our paper trader dutifully booked an +$86K fake profit on it before our sanity caps caught it. Any simulation without ceilings on impossible moves will accumulate phantom wins that no live account could ever collect.
Of our 42 configurations, every leveraged variant lost money under honest accounting. The only survivor traded spot, with a hard stop, at 1% risk per trade. Leverage didn't amplify the edge — there was barely any edge — it amplified the noise until fees and liquidations ate the account.
Here's the part that keeps this from being pure doom: when we reconciled our backtest engine against the live paper broker on the same strategy — same rules, same period — the engine matched 99.7% of the win rate (31.4% simulated vs 31.5% live) and about 76% of exact trade timing, with PnL within ±20%. Honest simulation is possible. It just requires force-close accounting, real costs, next-bar fills, and sanity caps — the exact defaults most tools skip because they make the numbers look worse.
- Judge any paper account only after force-closing everything. Unrealized PnL is an opinion; realized-if-closed-now is a fact.
- Check whether the average is moonshot-carried. Strip the top 3 winners; if the strategy dies, it's a lottery ticket, not a system.
- Backtest first, paper-trade forward second. The backtest filters out the 90% that never worked; the forward test catches the rest that stopped working. On Trading Bot House both gates are built in — and the paper dashboard force-closes for you.