Methodology
graphGURU.ai turns raw market data into a structured, repeatable read of the chart. This page explains exactly how that happens — what the engine measures, what the two numbers on screen actually mean, and where the AI review and paper-trading fit in. No black boxes, no promises.
What the engine does
At its core is a deterministic, causal analysis engine. "Deterministic" means the same candles always produce the same result — there is no randomness. "Causal" means every calculation uses only data available at that moment in time; the engine never peeks at future candles.
It runs across 5 coins (BTC, ETH, SOL, BNB, XAUT) and 9 timeframes (5m, 15m, 1h, 4h, 1d, 3d, 1w, 15d, 30d). For each coin-and-timeframe combination it computes:
- Classic indicators — RSI, StochRSI, MACD, EMA, Bollinger Bands, ATR, ADX and OBV.
- Chart structure and patterns — pivots, support/resistance levels, double top/bottom, divergences, breakout-pullbacks, wedges and Donchian breakouts.
These are scored together into a single read: a direction (UP, DOWN or SIDEWAYS), an expected move %, a price target, a horizon (how far ahead the read applies), and a stop-loss level. The stop is placed at the opposite-side support/resistance level, but never tighter than 1.5× the ATR (average true range), so normal market noise is less likely to wick it out.
"Confidence" vs "Signal strength" — two different numbers
The app shows two figures, and they mean different things. Confusing them is the most common mistake.
Confidence is calibrated. We ran the engine over roughly six years of history and measured how often each kind of call was actually right, then mapped the displayed number onto that real hit-rate. So a Confidence of 53 means "in the backtest, calls like this landed about 53% of the time." This is why the scale sits around 49–54 and nothing scores 90%. Earlier, uncalibrated systems that flashed "70–92% confidence" were, when checked honestly, only right about 52–59% of the time. We would rather show you an accurate 53 than a flattering, meaningless 88.
Signal strength (roughly 20–92) is different: it measures how strongly the individual indicators and patterns agree with each other. It is a raw agreement score, not a probability. A high signal strength means many things line up in the same direction — which is useful context, but it does not mean the move is 90% likely to happen.
The AI review and paper-trading
Once an hour, a large language model reads the engine's output across all 9 timeframes for a coin and writes a structured review: a snapshot, probabilities, key levels, short- vs long-term view, reversal watch, buying/selling pressure, and conditional trade scenarios (for example, "if price closes above X, the setup favours Y"). The AI does not fetch its own data — it only interprets what the deterministic engine already computed.
To keep everyone honest, each trade scenario the review produces is automatically paper-traded — simulated with fake money against real subsequent price. The Backtest tab shows how those simulated plans actually played out (target hit, invalidated, timed out, and so on). This is how you can judge whether a given style of setup is worth anything, rather than taking the write-up at face value.
Data source
All analysis is built from Binance candle data — and specifically from closed candles only. The engine never analyses a candle that is still forming, because a half-formed candle can reverse before it closes and would produce unstable, misleading reads. The live price ticker you see is a separate, real-time feed for convenience; the analysis itself waits for candles to close.