A backtest describes what a strategy would have done under assumptions. A live record shows what it did. The gap between them is the sum of everything the backtest could not model: real spreads and slippage, real fills, real swap, the broker's feed, and the degree to which the strategy was fitted to its test data. That gap, tracked over time, is the best available test of whether a strategy is what its backtest claims.
What size of gap to expect
Live results below the backtest by a modest margin is the normal, healthy case. A strategy on daily bars with wide stops, tested with realistic costs, might run live at 70 to 90 per cent of its backtested return with similar drawdowns. A strategy on short timeframes with tight stops can run at half or less, because costs and fills dominate. A live result above the backtest is possible in a favourable period and should not be read as the strategy being better than tested.
What a widening gap means
If live results track the backtest's expected range for a year and then fall steadily below it, one of three things has happened: the market's regime has changed to one the strategy depends on less; costs at the broker have risen; or the edge, if there was one, is decaying. Each has a different remedy, and the first step is to identify which by looking at the trades, the spreads and the year-by-year behaviour in the backtest.
If live results fall below the backtest from the start and stay there, the backtest was fitted, and the strategy was never what it appeared.
Using the comparison
Treat the backtest as a forecast with a range, and plot the live equity against it: the backtest's expected return path with a band around it derived from the backtest's own variability. Live inside the band is consistent. Live below the band for a sustained period is a signal. This is the check that catches a failing strategy months before the drawdown alone would, because it compares the live result with what the strategy should have done in the same period, not with zero.
A strategy's backtest implies about 1.2 per cent a month with monthly results ranging from minus 4 to plus 6. After twelve months live it has returned 10 per cent, with monthly results ranging from minus 5 to plus 5, and its worst drawdown was within the backtest's range. It is running at roughly 70 per cent of the backtest, with the shortfall explained by a spread half a pip wider than the test assumed. Consistent. A second strategy's backtest implied 3 per cent a month; live it has returned 4 per cent in twelve months, below the band for nine of them. Something the backtest assumed is not true, and the likeliest thing is that its parameters were chosen on the tested period.
A vendor who publishes a backtest with its assumptions, a live record from a sealed start date, and the comparison between them, has given you the one piece of evidence that separates a fitted strategy from a real one. A vendor who publishes the backtest alone has given you a forecast with no test of it. Where both exist, read the gap, expect it to be modest and stable, and treat a widening one as the earliest warning the record can give.