How it works, and what happened when we tested it
Every tool in this category tells you what it shows. Almost none tell you whether it works. We backtested our own levels against a control, and this page publishes the result — including the part that does not flatter us.
What the engine computes
Ampeld takes standard OHLCV candles and derives four things: stop-loss concentration bands from average true range and position decay, order blocks and fair value gaps from candle structure, modelled cumulative volume delta reconstructed from price and volume, and high-timeframe levels from daily and weekly ranges.
Everything is computed from price and volume. There is no broker data, no order book, and no exchange liquidation feed behind any of it.
What we do not have
Real liquidation data exists on crypto derivatives venues, which publish forced-exit prints. It does not exist for spot forex, gold or equities in any form: those markets have no central clearing, no consolidated open interest, and stop orders sit inside individual brokers where nobody can see them. The one public dataset that ever showed real retail stop placement — OANDA's order book endpoint — was withdrawn in 2024.
So for forex, metals, indices and equities, every “liquidation level” on the market is a model. Ours included. We would rather say that than imply otherwise.
How we tested it
For each level the engine produced historically, we asked whether price traded into it within the next 24 hours. On its own that number is unreadable — price wanders, so some level at that distance gets touched regardless. Every level was therefore paired with a placebo: a level the same distance away, mirrored to the other side of price. The placebo holds distance constant, so what remains is whatever the level construction itself contributes.
Levels closer than 0.25% to price were excluded, because those get touched almost always and forecast nothing. Samples were spaced out so one setup is not counted repeatedly. The structure detector only ever saw candles that existed at that moment.
What we found
| Layer | Levels tested | Reached | Placebo | Difference |
|---|---|---|---|---|
| Magnet bands | 8,248 | 52.9% | 53.3% | −0.4 pp |
| Order blocks | 18,282 | 14.7% | 15.6% | −0.9 pp |
| Fair value gaps | 6,695 | 15.6% | 17.5% | −1.9 pp |
We then asked the better question — given price reaches a level, does it react there? Among 4,244 touched bands, price reversed 59.8% of the time. At the placebo: 59.7%.
And we tested whether other well-known level constructions do better. They do not. Note the last row: the ATR band is a control that should come out flat, and it does, which is how we know the harness itself is sound.
| Construction | Levels tested | Difference vs placebo |
|---|---|---|
| Previous day high / low | 10,741 | −1.7 pp |
| Swing high / low, 20 bars | 11,021 | −0.5 pp |
| Round numbers | 6,151 | −0.6 pp |
| ATR band (control) | 10,560 | +0.0 pp |
Finally we tested positioning data — funding rates across twelve crypto pairs over three years, ranked cross-sectionally so market-wide moves cancel out. Also flat.
What that means
Our levels do not predict direction. Across roughly 33,000 measured levels, they are indistinguishable from an arbitrary price the same distance away. If you are looking for a tool that tells you where price is going, this is not one, and we have not found one.
What the levels do describe is where the model places retail stop concentration, computed the same way on every instrument, updated continuously. That is a picture of market structure, not a forecast.
What the terminal is for, then
- Seeing where the crowd's stops plausibly sit before you decide where to put your own — the levels are most useful as places to avoid, not places to target.
- One consistent method across crypto, forex, metals, indices and equities. Every other tool in this category is crypto-only, because that is where real liquidation data exists.
- Reading structure — order blocks, gaps, volume profile, session levels — without switching platforms.
Check it yourself
Every figure here comes from a backtest you can reproduce in principle: take any level the terminal draws, note the price and time, and look at what happened next. The comparison that matters is not “did price reach it” but “did it reach it more often than a point the same distance the other way”.
If you find something in the data that we have not, we would genuinely like to hear it — support@ampeld.com.
Ampeld is an analytics and data-visualisation tool. It does not provide investment advice, trading signals, or access to broker order books. All levels are modelled estimates. Trading leveraged instruments carries a high risk of loss. Terms.