The ICT Silver Bullet Is Sold on a 70% Win Rate. We Measured 36.1%.

Most of what gets taught under the smart-money banner cannot be tested. "The market seeks liquidity" is not a claim, it is a mood. You cannot run it against data and find out you were wrong.

The Silver Bullet is different, and that is why we picked it. It names a window, it names a setup, and its teachers quote a win rate: somewhere between 70 and 90 percent, depending on who is selling it. A number like that is falsifiable. You either hit it or you do not.

So we measured it. Six markets, one minute data, 2019 to 2021, which is our in-sample window. No expert advisor, no out-of-sample data touched. If the edge is not in the raw prices there is nothing to automate.

The rules, pinned down before we ran anything

The published strategy defines three windows, not one: London at 03:00 to 04:00 New York time, the New York AM session at 10:00 to 11:00, and the afternoon at 14:00 to 15:00. Only the New York one usually gets tested, including by us the first time. Measuring all three is not fishing, it is testing the strategy as it is actually published.

Inside a window, the setup is: price sweeps a prior high or low, then leaves a fair value gap, and you enter at the midpoint of that gap with a stop beyond the sweep and a target at twice your risk. One trade per window per day per market.

We pre-registered the whole grid before running it, including the decision that all three windows get reported whether they flatter us or not, and that the headline would be the pooled result rather than the best cell.

What 2,200 setups did

The claimed win rate against the measured one, across 2,200 setups on six markets.

36.1%. Not 70, not 90. Across 2,200 filled setups the Silver Bullet wins just over a third of the time, and it loses 0.24R per trade, with a 95% confidence interval of 0.29 to 0.19 of a loss. That interval sits entirely below zero, so this is not a small sample wobbling around break-even.

Net R per trade in each of the three published windows, with 95% intervals.

And it is not one bad window dragging down two good ones. All three lose, across all six markets. The window that gets taught most is not the one that performs best, which is the sort of thing that happens when a strategy is selected by how it looks rather than by what it does.

We tested the wrong markets first, so we went back

Our first two passes ran index CFDs only: Nasdaq, S&P and Dow. That is defensible, because the Silver Bullet is most associated with NQ and ES futures.

It is also incomplete, and we only noticed when we went back to the strategy's own sources. Those sources describe EUR/USD and GBP/USD as the most widely traded instruments for this setup, producing signals in both the London and New York windows, and they name gold as a secondary market specifically in the London window.

We had measured the London window. We had never measured it on the instruments that window is supposed to suit. That is the obvious rebuttal to our verdict, so it deserved an answer rather than a footnote.

We wrote the extension down before running it: same rules, same windows, same default cell, three new markets, both fill conventions, and an explicit commitment that a positive result would rewrite the verdict. Costs on the new markets are harsher than on the old ones, because FX and gold pay commission where index CFDs do not.

Net R per trade on each market, with 95% intervals.

The FX pairs and gold lose more than twice as much as the indices. EUR/USD loses 0.37R per trade, GBP/USD 0.23R, gold 0.42R, against 0.15R pooled across the three indices. Every one of the six markets is negative with a confidence interval clear of zero, under both fill conventions.

The markets its own teachers point you toward are the markets where it performs worst.

The part that matters more than the loss

A losing backtest is not very interesting on its own. Plenty of decent ideas lose money once you charge them a spread.

What makes the Silver Bullet smart money, rather than a gap trade with a clock attached, is the liquidity sweep. Price is supposed to take out a high or a low first, and that sweep is the tell that institutions are involved. It is the reason the strategy has the name it has.

So we tested it with the sweep requirement and without it. If the sweep is doing the work, removing it should make things worse.

The strategy with the sweep requirement and without it.

It does not. The difference between the two is statistically indistinguishable from zero, on the indices and on the FX and gold extension alike.

We should be careful about how far that goes, because the two arms are not identical twins. Without a sweep there is no swept extreme to place a stop behind, so the unconditioned version has to put its stop somewhere else, and it admits roughly three times as many setups. This is not a clean isolation of the sweep on its own, and we are not going to pretend otherwise.

What it does establish is narrower and still damaging: the sweep-conditioned version does not outperform the unconditioned one. The filter the entire framework is named after does not show up in the results, and if it is doing any work at all, that work is smaller than 2,200 setups can detect.

What we got wrong along the way

We measured this once before, on 2026-08-02, using only the New York AM window. That pass produced 321 trades and we recorded two conclusions from it: a dollar expectancy, and a finding that the sweep made the Nasdaq specifically worse.

An independent statistical pass killed both. At 321 trades the pooled net figure carried a confidence interval from 6.37 dollars of loss to 4.35 of profit, which supports no conclusion at all, and the sweep comparisons came back at p values of 0.52, 0.29 and 0.80. The only thing that survived was the win rate.

So we widened to all three published windows, and later to all six markets. Each widening roughly doubled the sample without touching a single out-of-sample year. The win rate, the one finding that survived the first cull, has moved from 42.1% to 36.9% to 36.1% as the sample grew. It has never once approached 70.

The honest limits

Three years is three years, and this is an in-sample window by design. A version of this strategy that only works in some later regime is not tested here, though the sources make no such claim.

We measured index CFDs rather than the NQ and ES futures the strategy was originally built on. Futures carry tighter costs than CFDs, so a futures trader faces a friendlier cost stack than our model charges. It does not rescue the result: gross of all costs the strategy is still slightly negative, so there is no edge underneath for cheaper execution to uncover.

Our exit walk resolves the ambiguous case, where a single minute bar touches both the stop and the target, in favour of the stop. That is the conservative choice and it biases a losing strategy slightly more negative. We ran the opposite convention as a sensitivity and the pooled result barely moved.

The verdict

No edge. On 2,200 setups across six markets and all three published windows, the ICT Silver Bullet wins 36.1% of the time against a claimed 70 to 90, loses 0.24R per trade with an interval clear of zero, and is negative on every market and in every window we measured. It performs worst on the instruments its own sources recommend most. The liquidity sweep that gives the strategy its name cannot be shown to affect the outcome.

The out-of-sample years remain untouched. There was no reason to spend them.