How the engine produces signals.
The core promise of the CryptoEdge engine is that everything you see is computed — not invented. This guide walks through each layer of the pipeline so you can decide which numbers to trust and how much weight to give them.
1. Inputs — live OHLCV
Every analysis on the site starts from live candles fetched from CryptoCompare and CoinGecko. We get open, high, low, close, and volume for every bar on every supported timeframe. There's no proprietary data feed; if you wanted to, you could rebuild every number on this site from public APIs.
2. Wyckoff classification
The engine identifies pivot highs and lows, fits a trading range to the recent action, classifies it as accumulation or distribution based on context (prior trend direction, volume profile, position in the larger structure), then scans the bars inside the range for textbook events: Selling Climax, Automatic Rally, Secondary Test, Spring, Sign of Strength, and so on.
The active phase (A through E) is derived from which events have printed and where the latest bar sits in the range. Full Wyckoff guide →
3. Scenario probabilities
The engine scores three mutually exclusive scenarios for the next move: bullish (range breaks up), neutral (price stays inside the range), and bearish (range breaks down). Each scenario gets points from a list of named factors — phase position, volume confirmation, momentum, the presence of a Spring or Upthrust, news bias, etc. The raw scores are softmaxed into probabilities that sum to 1.
Every factor is named and shows you its direction and weight. There's no black box. If the factor list says "Spring detected: +18 bullish, weight 0.85," that's exactly what happened.
4. Conviction alerts
On top of the scenario probabilities, a separate scoring layer asks: how confidently can we act on this? Conviction combines the scenario spread (how much the dominant scenario beats the others) with whether a key level has actually broken on volume.
Conviction score is too low for any tier. Most days look like this — that's the point.
A handful of factors are aligning but not enough to commit. Treat this as "be ready" — when watch becomes alert, the setup has matured.
Conviction is high. The structure, the scenario probabilities, and a level break are all pointing the same direction. This is the tier most actionable trade ideas come from.
The strongest tier. Conviction is high, the level break is fresh and on volume, the probability spread between the dominant scenario and the others is wide. These are rare — usually a few per coin per month.
5. Leverage tiers — bounded by liquidation math
When the engine publishes a trade idea, it includes three leverage tiers (conservative, moderate, aggressive). These are not vibes — they are computed from the distance between entry and stop, the maintenance-margin assumption (~0.5% on most major exchanges), and a 40% safety buffer for slippage and funding.
The aggressive tier is the maximum where your stop hits before you would liquidate (with that buffer). Conservative caps at 3×, moderate at 5×. None ever exceeds 20×. You can verify the math yourself in the tooltip on every tier.
6. Real track record — every idea is scored
The moment a trade idea is published, its entry, stop, and target get logged to persistent storage. Every 6 hours a cron walks the bars that have printed since publication, in order, and records which level was hit first. Wins and losses are measured in R-multiples (multiples of the risk distance). The win rate, average R, profit factor, and equity curve on the calculator page are aggregates over that real, scored sample. Nothing is backtested in hindsight.
7. Adaptive weights
As the scored sample grows, the engine recalibrates the weight of each factor based on its historical hit rate. Factors that have predicted well get more weight; factors that have underperformed get less. Recalibration runs daily. You can see which factors have been adapted and by how much in the Adaptive Factor Weights panel on the calculator.
What the engine doesn't do
- It doesn't predict the future. It identifies setups with measured historical performance and tells you the probabilistic odds.
- It doesn't tell you to trade. Every output is structured information; what you do with it is your decision.
- It doesn't guarantee anything. Past R-expectancy on a sample of N trades is the best honest estimate of future expectancy — not a promise.
- It doesn't care if you're bullish or bearish on a coin. The engine sees structure; it doesn't have feelings.