In order of how often they turn out to be the answer.
1. Costs
Almost always this one. Compare your live results against the stressed figure rather than the modelled one — if live sits between the two, nothing is broken and the strategy is behaving as tested.
2. A different instrument than you think
Brokers quote the same name differently. Our crypto is broker-style CFD data rather than exchange spot, precisely because the two are measurably different prices, and an FX cross at your broker may carry a different rollover convention.
3. Timezone and session boundaries
A session window is evaluated in the timezone the strategy names. "11am London" is not 11:00 UTC for the seven months London spends on summer time, and a strategy that trades a session open will fire an hour off if the timezone is wrong.
4. The period, not the strategy
A backtest covers a specific stretch of history. Live trading covers a different one. A strategy can be entirely correct and still lose for months, and the drawdown figure on your results is the honest estimate of how long that can feel.
If none of the four explain it, the run is reproducible: the strategy, its version and its result are all stored, so write to us and we can look at the same numbers you are looking at.