AI Stock Signals Explained — How Vantage Beacon (powered by TradeVertex AI) Builds a Single Confluence Score
Vantage Beacon powered by TradeVertex AI turns decades of high-low ranges, daily opens, and time-of-day performance into a single, decisive readout: a direction (LONG or SHORT), a confidence score, and a risk-to-reward ratio. This article walks through how that readout is built, what each factor contributes, and how the engine validates itself against history before any signal reaches the scanner.
For the reader who wants the model end-to-end: range score × volatility regime × session win-rate → confidence; confidence ≥ 0.65, with a positive R:R, and a 90-day walk-back.
Why one confluence score lives above the noise
The signal engine takes three independent readings from each bar — range, volatility regime, session win-rate — and combines them into a single confidence value. The combination is weighted: 0.40 × rangeScore, 0.30 × volatilityScore, 0.30 × sessionScore. Range is the loudest of the three, regime is close behind, and session is the smallest but most directional when present. The intent of the weighting is that any one factor can lift confidence into the surfaced range, and no single factor can carry a weak signal through on its own.
A wide-range bar in a contracting-volatility regime is noise. A quiet-range bar in a high-historical-win-rate hour is noise. A bar with strong range and regime but a quiet session bucket is, at best, marginal confidence. The weighting is what makes the readout a confluence rather than a single-indicator model: only bars where the three legs line up clear the 0.65 confidence threshold. Below 0.65, no signal surfaces — regardless of how strong one factor looks in isolation. That single rule is the difference between a scanner that fires often and one that fires only when the evidence converges across the three legs the engine scores.
R:R from the swing, confidence from the factors, the top three from both
Risk-to-reward is read from the swing, not from a forecast. The engine takes the last six bars around the signal candle, walks the swing high and swing low on that lookback, and treats those as the natural stop and limit. The ratio is computed against the bar's close and rounded to one decimal place — so a 1.80 R:R surfaces as 1.8 and a 2.57 surfaces as 2.6. A signal at 1.8 R:R will not pull ahead of a 2.6 R:R signal at the same confidence; the sort key is confidence descending, then R:R descending, with the top three surfaced per session.
The factor scores — rangeScore, volatilityScore, sessionScore — are diagnostic, not inputs to the sort. They tell you which leg is carrying the signal: a 0.78 confidence at 1.6 R:R with a 0.95 rangeScore and a 0.42 sessionScore is a signal leaning on today's bar width, not on the hour-of-day history. Reading the diagnostic first is the discipline; the sort order is the consequence. Two signals can land at the same confidence and the same R:R; when they do, the factor mix on each side is what separates an entry from a skip.
The 90-day walk-back is the final confirmation
Every surfaced signal walks back 90 days and asks the same question: did the same direction, on the same hour-of-day bucket, win at the offered R:R over that prior window? The output is the historicalWinRate field, with a missing-value floor when fewer than ten prior candles are available — when the lookback is too thin, the field returns null rather than a thin-sample rate. That floor exists so a confidence never surfaces as repeatable when it is built on one or two lucky matches.
The operating rule is two lines. High live confidence plus high historical win-rate means the setup has been repeatable recently — size up into the entry, scaled to how clean the walk-back looks. High live confidence plus low historical win-rate means the live readout is leading the walk-back rather than the other way around — size down, or skip. The scanner can surface two signals at identical live confidences that sit against very different priors when the 90-day window is laid next to them. The walk-back is what turns the confluence score from a one-bar reading into a recent-history reading.
What this readout is not
The Vantage Beacon (powered by TradeVertex AI) signal is a confluence readout — it is not a price-prediction engine, a news-sentiment model, or a portfolio allocator. The scanner surfaces range, regime, session, and R:R, validated against the 90 days that just happened. It does not tell you where the next bar will land; it tells you where the setup is currently aligned across the three legs the engine scores, and what the recent walk-back has to say about it.