™Backtest
Before the models went live we tested them on nineteen months of market data they had never seen. This page says how, and shows every result.
How it was done
- The data Binance trade data for BTC/USDT and ETH/USDT from 17 August 2017 to 31 July 2026, built into waves at the five degrees the models predict.
- The split The models were trained, and their confidence levels set, only on data before 1 January 2025. Everything from 1 January 2025 to 31 July 2026 was kept back for testing and played no part in either.
- As it happens live Each test prediction used only what was known at that moment: the waves already finished, and as much of the wave in progress as had formed so far.
- What counts as right A prediction is a price range. It is right when the wave ends inside that range. The live track record also counts a wave that goes through the range and beyond; here that does not count.
- Every wave counts once The figures use the prediction made when the previous wave was confirmed as ended. Live, a prediction is made again as the wave in progress grows; those repeats are not counted here.
Confidence levels
Each prediction carries a confidence level, and each level has a target that was set before testing: a High prediction should be right at least 85% of the time, Medium 75%, Low 60%. A level is only used at a degree where older data showed it could be met, so some degrees have no High predictions at all.
Results: the wave in progress
The first prediction is for the wave in progress: the price range it is expected to end in.
| Series | Waves | High target 85% | Medium target 75% | Low target 60% | No range |
|---|---|---|---|---|---|
| BTCUSDT degree 3 | 98,672 | 85.3%
of 19,644 | 81.3%
of 63,393 | 88.9%
of 15,394 | 0.2% |
| BTCUSDT degree 4 | 32,054 | None | 82.7%
of 27,871 | 66.3%
of 4,061 | 0.4% |
| BTCUSDT degree 5 | 10,592 | None | 84.3%
of 7,423 | 66.5%
of 3,127 | 0.4% |
| BTCUSDT degree 6 | 3,436 | None | 81.8%
of 1,951 | 75.4%
of 1,454 | 0.9% |
| BTCUSDT degree 7 | 1,025 | None | 76.2%
of 1,021 | None | 0.4% |
| ETHUSDT degree 3 | 166,195 | 90.0%
of 49,606 | 85.4%
of 43,989 | 71.4%
of 71,046 | 0.9% |
| ETHUSDT degree 4 | 53,765 | 86.0%
of 5,175 | 85.7%
of 14,049 | 74.8%
of 34,203 | 0.6% |
| ETHUSDT degree 5 | 17,542 | None | 87.2%
of 6,969 | 74.7%
of 10,369 | 1.2% |
| ETHUSDT degree 6 | 5,684 | None | 79.7%
of 2,107 | 74.7%
of 3,442 | 2.4% |
| ETHUSDT degree 7 | 1,842 | None | None | 79.0%
of 1,147 | 37.7% |
Results: the wave after
The second prediction is for the wave that follows the wave in progress.
| Series | Waves | High target 85% | Medium target 75% | Low target 60% | No range |
|---|---|---|---|---|---|
| BTCUSDT degree 3 | 98,672 | None | 81.0%
of 98,047 | 85.5%
of 145 | 0.5% |
| BTCUSDT degree 4 | 32,054 | None | 83.7%
of 26,872 | 76.6%
of 4,976 | 0.6% |
| BTCUSDT degree 5 | 10,592 | None | 84.3%
of 8,954 | 77.1%
of 1,562 | 0.7% |
| BTCUSDT degree 6 | 3,436 | None | 85.4%
of 1,291 | 76.9%
of 2,134 | 0.3% |
| BTCUSDT degree 7 | 1,025 | None | 84.7%
of 294 | 79.6%
of 705 | 2.5% |
| ETHUSDT degree 3 | 166,195 | None | 80.3%
of 164,985 | None | 0.7% |
| ETHUSDT degree 4 | 53,765 | None | 80.0%
of 43,937 | 75.4%
of 9,384 | 0.8% |
| ETHUSDT degree 5 | 17,542 | None | 82.6%
of 14,863 | 77.9%
of 2,466 | 1.2% |
| ETHUSDT degree 6 | 5,684 | None | 83.2%
of 2,815 | 78.9%
of 2,802 | 1.2% |
| ETHUSDT degree 7 | 1,842 | None | 81.1%
of 1,779 | None | 3.4% |
Each percentage is the share of that level's predictions whose wave ended inside the predicted price range, to be read against that level's target. They are not shares of one total, so they do not add up to 100%. None: no prediction at this degree was given this confidence level. No range: waves for which the models gave no price range at all.
What is different live, and is it as good as the backtest?
Live, a prediction is about the wave in progress: a wave we can be certain is forming, with a fixed start point and an end point that is not yet known. So the models only ever have part of that wave to work with. As the wave in progress (the dashed line on the chart) develops there is more to go on, and the prediction is made again with the new data. This can happen many times as a wave grows, and in the backtest the later predictions were better than the earlier ones.
The backtest was run the same way, on partly formed waves, re-predictions included. What it could not reproduce is the live market itself. The figures above are for predictions made once the previous wave had been confirmed as ended; live, a first prediction is often made sooner than that, on a wave end called early, and some of those calls turn out wrong. And the market since the test period is new to the models.
So expect live results to fall short of the backtest. We do not yet have enough live data to say by how much; the track record is where that will show.
What this does not tell you
- When We predict where a wave will end, not when. How long a wave takes depends on the pace of the market, and that is not something we claim to predict. The width of the prediction boxes shown on the chart is a rough guess, there simply to make the chart easier to use. DO NOT RELY ON THE TIME OF THE PREDICTION BOX, IT'S A GUESS.
- How it is performing in reality A backtest is a test on past data. How the predictions have performed on the live market is on the track record, which began on 19 September 2026.
- The higher degrees rest on far fewer waves BTCUSDT degree 7 had 1,025 waves in the test period against 98,672 at degree 3. And at ETHUSDT degree 7 the models gave no price range for the wave in progress on 37.7% of waves. Weigh each figure by the number beside it.
- The future A result on past data is not a promise about tomorrow, and nothing here is financial advice. See the terms.
How not to use the chart
- Do not assume a prediction is correct Nothing is certain. There are only probabilities.
- Do not treat a prediction as final Re-predictions happen often, particularly at the lower degrees, and the new one can be radically different.
- Do not glance at the chart, decide and walk away The predictions change with the market in real time, so be prepared for them to change.
News and sudden moves
The next scheduled news event that may move markets is shown at the top of the app, and scheduled events are marked on the chart. They are widely known calendar events, taken from a public economic calendar, and we do not guarantee them to be complete or correct. Spontaneous, sporadic or random events are unknown to the system.
Prices can shoot up or down wildly and cause huge losses to market participants. Those movements are unpredictable, and we explicitly make no effort to predict them. We make no allowance for such events and no specific predictions based on them: they are normal market risks, of which we have no knowledge and which we have no intention of forecasting.
We may switch predictions off entirely for some time before, during and after a significant news event, to avoid unintentionally misleading anyone. Today we do not: predictions carry on through news events, so treat them with extra care at those times.