How a signal is actually built

Price data comes from Twelve Data as OHLC candles per symbol and timeframe. We request 100 closed candles per timeframe so that even a 200-period reference has a meaningful history, and we work with closed candles only — a forming candle would silently repaint every indicator.

• ▸Coverage: FX majors and crosses, metals, indices, crypto, and US and EU equities
• ▸Requests are cached and throttled (7 requests/minute) to stay inside provider limits
• ▸If a timeframe returns fewer than 30 candles, the symbol is skipped instead of scored on thin data
• ▸No broker connection: we never see your account, orders or positions

Each of the three timeframes is reduced to one indicator snapshot with fixed, published parameters. Nothing here is optimised per symbol — curve-fitted parameters are the fastest way to build a backtest that dies in live markets.

• ▸EMA 20 / 50 / 200 on closes
• ▸RSI 14, MACD 12/26/9, Stochastic 14/3
• ▸Bollinger Bands 20 / 2σ, ATR 14
• ▸Trend classification: up when close > EMA20 > EMA50 > EMA200, down when the stack is fully inverted, otherwise sideways
• ▸Volatility classification from ATR as a percentage of price: below 0.3% low, above 1.2% high, otherwise medium

On the entry timeframe we detect candle formations and horizontal structure geometrically — no image recognition, no subjective drawing. A swing high or low needs two lower highs (or higher lows) on each side to count.

• ▸Candles: Doji, Hammer, Shooting Star, Bullish/Bearish Engulfing, Morning/Evening Star
• ▸Levels: up to three support and three resistance zones from confirmed swing points over the last 50 candles
• ▸Streak counter: consecutive green or red entry candles, as an exhaustion warning
• ▸Pro tier adds order blocks, break of structure, change of character and harmonic geometry

A long score and a short score are accumulated independently from the snapshots. Both are additive and bounded at 100, so no single condition can carry a setup on its own. The full weight table is published below.

• ▸Higher timeframe trend dominates the weighting — a counter-trend setup can never reach a high score
• ▸Overbought/oversold RSI extremes actively subtract from the aligned direction
• ▸Low volatility is penalised: correct direction plus no movement is still a losing trade after costs
• ▸Correlation and news windows are surfaced as separate warnings, not blended into the score

Only after the numbers exist does the language layer run. It receives the snapshots, patterns, levels and score — never a raw chart image alone — and produces a structured read: scenario, invalidation, risk framing, plus an explicit counter-argument and an uncertainty statement when the evidence is thin.

• ▸A transparency log shows which inputs the analysis used, step by step
• ▸Every analysis must state what would invalidate it
• ▸Low-conviction cases are labelled as such instead of being dressed up
• ▸The narrative can never overwrite the score — if they disagree, you see both