On October 10, 2025, USDe traded at $0.65 on Binance. On Curve, where most of its liquidity actually sat, the peg moved about 0.3%. Bybit, Uniswap and the rest barely flinched.

Same asset. Same hour. If your collateral happened to live on Binance, your margin system marked it down 35% and you could have been liquidated by a price that didn't exist anywhere else in the market.

Your backtest assumes there's one market

There isn't. Every exchange runs its own order book, its own fee schedule, its own funding calculation and its own rules for which price triggers your stop.

Your strategy's code is identical everywhere. The environment it runs in is not.

I think most traders know this in the abstract. Very few price it in. The exchange gets treated like plumbing, a place where orders go, when it's really a set of parameters your strategy never got to choose.

Start with fees, because they look like nothing. Base-tier taker fees on perpetuals are 0.055% on Bybit and about 0.05% on Binance. A system that makes 200 round trips a year, paying taker on both sides, spends 22% of its traded notional on fees on one venue and 20% on the other.

Two points of notional doesn't sound like much. Run it at 3x and it's six points of equity every year, before a single trade goes wrong. That's the whole return of some strategies.

Funding and wicks are where it gets real

Funding rates look standardised. Binance, Bybit and OKX all use a 0.01% interest component per eight-hour interval. But the other half of the formula is a premium index measured on each venue's own book, so when one exchange is crowded with longs, its rate can pull sharply away from the others.

A trend system that holds positions for weeks pays or collects that rate three times a day. Hold through a crowded month on the wrong venue and the difference shows up in your P&L as a slow leak you can't trace to any trade.

Then there are wicks. Your stop fires on that exchange's last price or mark price, not on some global Bitcoin price. A thin book can print a spike that never happens elsewhere, and your system will act on it with perfect discipline.

USDe was the extreme version. The everyday version is a stop that gets hit on one venue and not on another, and a trade log that looks nothing like the backtest.

Why people misread the gap

Here's the mistake I see most. Someone backtests on Binance data, because it has the longest clean history, and then trades on a different exchange. Six months later the results have drifted and they decide the strategy stopped working.

Maybe it did. But often they're comparing two different strategies without realising it. Different fees, different funding, different fills.

The reverse happens too. A trader copies someone's published results from one venue, runs the same rules somewhere else, underperforms, and blames themselves.

They didn't break a rule. They changed an input and nobody told them.

It's the same logic as comparing your live bot to a backtest that ignored slippage. The gap is real, but it's not evidence about your edge until you've removed the parts that have nothing to do with your edge.

What to do about it

Treat the exchange as a strategy parameter, the same way you'd treat a lookback period. Before you go live, rerun the backtest on candles from the exact venue you'll trade, with your actual fee tier and that venue's funding history applied to every hour you'd be in a position.

If the result moves a little, good. That's normal friction. If it moves a lot, you've learned that your edge is thinner than it looked, and it's far cheaper to learn that in a spreadsheet than in a live account.

One more habit. When live results start to diverge from the backtest, check the venue before you touch the rules. More often than not, in my experience, the explanation is fees, funding or fills, not a dead strategy.

This is why our own published numbers state the exchange they came from. The v33 results are Bybit data with Bybit fees, and I'd expect them to look different somewhere else. The full backtest and methodology are at v33systematic.com if you want to see how that's laid out.