Measured 31 August 2026 · re-tested against three further controls and a second data source
Every trading tool draws patterns on a chart and lets you assume they mean something. We built the machinery to check ours, ran it across twenty instruments, and published the answer even though the answer was uncomfortable.
We took every pattern the chart draws — fourteen chart patterns and a dozen candle patterns — and asked one question of each: does price behave differently after this pattern than after ordinary bars that look like it?
That last clause is the whole difficulty. Comparing a pattern against all bars measures candle geometry, not the label: a Bullish Engulfing appears where volatility is already high, so it will look predictive whether or not it means anything. So every pattern is compared against a matched null — bars drawn from the same volatility band, trend position, session and day of week. Like against like.
Nothing the chart draws has a demonstrable directional edge.
One comparison out of 886 survived a Holm correction: Bearish Engulfing on ETHUSD H1, at 58.3% against the 49.3% expected of bars like it. One instrument. It replicated on nothing — five other instruments showed nothing, and BTCUSD on the same feed and the same timeframe went the other way.
Then we pulled the same window from two exchanges that were not our own data source:
| Source | Difference | p | Survives correction |
|---|---|---|---|
| Our feed, ETHUSD H1 | +14.6pp | 5.8e-6 | yes |
| Binance ETHUSDT | +5.0pp | 0.080 | no |
| Coinbase ETH-USD | +8.2pp | 0.010 | no |
The effect shrinks by two thirds on a real exchange and stops being significant. Our own bars are stitched together from a broker feed, with a median gap between one bar's open and the previous close of 0.022% against 0.000% on Binance — fatter bodies, which is exactly the geometry an engulfing detector reads. The one survivor was mostly an artefact of our own data.
These are the figures a different company would have put on a landing page:
| Pattern | Win rate | Sample |
|---|---|---|
| Inverted Hammer (before matching) | 84% | — |
| Dark Cloud Cover | 72% | 25 |
| Double Bottom | 70.4% | 27 |
| Bear Flag | 68.2% | 44 |
Not one of them survives its own confidence interval. Twenty-five samples is not evidence of anything, and 84% was measuring candle geometry rather than the pattern. Every one of them is the reason the correction exists.
Read as binary trades, the labels resolve between 48% and 51% at one, three, five and twelve-bar expiries. No label clears the 55.6% a typical binary payout needs to break even, at any expiry.
The chart used to colour those patterns by a direction — green for bullish, red for bearish. That colour is a claim, and we had just measured that we could not support it.
So we removed it. The chart still draws the shapes, because the shape is a real thing on the screen and it is useful to see where the market did something structural. It no longer tells you which way that means price is going, because we do not know, and neither does anybody selling you the opposite.
The app ships the test. This is not a study we ran once and wrote up. The same machinery lives in the Edge Lab tab and runs against your journal — it will tell you, with intervals, whether your own edge is distinguishable from luck. Frequently the honest answer is that your sample is too small to say yet, and it says that too.
Because the alternative is asking you to take a chart on faith, and because a tool that will not test itself is not a tool you should trust with an account.
Mytoolfx is a journal and a discipline tracker. What it is good at is knowing your rules better than you do at four in the afternoon on a losing day, and telling you what a trade costs you before you take it. What it will not do is tell you which way the next candle goes.