No black box. Here’s the box.

Live prices, volume and market caps come from CoinGecko; candlestick history and chart data come from Binance. Market sentiment uses the alternative.me Fear & Greed index. Commodity quotes (gold, silver, oil) come from spot-price APIs and Yahoo Finance. Server responses are cached for 30–60 seconds, so what you see is effectively real time.

What we don’t use: we do not currently ingest on-chain data or social-media feeds. If a data source isn’t listed here, it isn’t in the signal.

For the selected timeframe, the engine computes five classic technical indicators on the candle closes: RSI(14), MACD(12, 26, 9), MA 9/21 crossover, EMA trend and Bollinger Bands(20, 2σ), plus 7-day support and resistance. Each indicator casts a bullish, bearish or neutral vote. Trend-following indicators (moving averages, MACD) carry more weight than the mean-reversion oscillators (RSI, Bollinger), which routinely fight a strong trend.

A momentum reading — the recent % move, blended from 24h and 7d changes — joins the indicators as one more weighted vote.

A coin is only rated Buy or Sell when all of these hold at once:

• Confluence — the weighted indicator votes clearly agree on a direction (mixed votes = Hold).
• Momentum doesn’t fight it — no buying a falling tape, no selling into a rally.
• Trend alignment — longs are only allowed above the long-term EMA, shorts only below it.
• Liquidity floor — thinly traded coins (low volume relative to market cap) are never rated.

Sub-hourly timeframes (5m, 15m, 30m) are never rated tradable: in our backtests they have negative expectancy, so the engine stays flat there and says so rather than inventing a signal.

The confidence % is not a vibe — it is calibrated to the engine’s real backtested win rate, then scaled down for less reliable timeframes. When a signal says ~45%, signals like it have historically won ~45% of the time. We publish the calibration table on the track record page so you can check it.

A sub-50% win rate is profitable when winners are bigger than losers — that’s what the risk/reward on each signal is for.

Risk is sized to the timeframe’s own volatility (ATR — the typical per-bar move) and anchored to market structure: the stop sits just beyond the nearest swing support or resistance, floored so it can’t hide inside the noise and capped so a distant level doesn’t blow up risk. The target reaches for the opposing level, bounded between 1.5R and 3R depending on conviction. Volatile coins get wider stops; quiet ones get tighter stops.

The numbers — direction, confidence, entry, stop, target, sentiment — come from the deterministic engine above, never from a language model. A large language model writes the prose: the plain-English rationale and market briefs that explain what the numbers mean. If the model is unavailable, a templated explanation built from the same factors is shown instead. Either way, the trade math is reproducible.

• Backtesting with no lookahead — each simulated signal only sees candles up to its decision bar; when a bar touches both stop and target, we assume the stop hit first (conservative fills).
• Live forward tracking — every signal the production engine issues is logged and followed to its real outcome. Wins and losses are published, unedited.
• Same engine everywhere — the backtest runs the exact code that generates live signals, so the published record tests what you actually get.

Technical signals are probabilities, not prophecies. Backtests can over-fit and exclude slippage, fees and funding. Crypto markets can gap through any stop. Nothing on WiseTrade is financial advice — it is research tooling to help you make your own decisions, with risk you choose and control.