METHODOLOGY
Confidence is a signal. Not a guarantee.
This page exists because "trust the AI" isn't good enough. Here's exactly what the confidence score measures, exactly how win rate is calculated, and exactly where the model's limits are — no marketing gloss, no rounded-up numbers.
THE CONFIDENCE SCORE
What it is, what it isn't
The single most misread number in the terminal. Read both sides before you size a position off it.
What it is
- A relative conviction estimate — how strongly price action, volume and momentum agreed at the moment the signal was generated
- Computed fresh for every signal, every scan cycle
- Useful for comparing two signals against each other, not for predicting either one in isolation
What it is not
- Not a probability of profit, and not a statistically certified win chance
- Not a guarantee — a 90%+ confidence signal can still hit its stop-loss; a lower-confidence signal can still win
- Not personalised — it says nothing about your position size, account balance, or risk tolerance
HOW WIN RATE IS CALCULATED
Auto-scored on close. Live, not backtested.
The Terminal Accuracy Metric Matrix scores every signal automatically the moment it closes, based on whether price reached its take-profit or stop-loss level. There is no manual curation — losing signals are scored and counted exactly the same way winning ones are, and nothing is excluded after the fact.
Just as important: the figure you see is live-tracked, built only from signals that have actually closed since tracking began, not a backtest run against historical data. That means it starts from a small sample and grows over time.
WHY SAMPLE SIZE MATTERS
A percentage without a sample size is a marketing number
Anywhere the terminal shows a win rate or accuracy figure, it's shown next to the number of closed signals behind it. A rate built from a handful of signals carries far less statistical weight than one built from hundreds, even if the percentage looks identical — so we don't display one without the other. Early in any instrument's track record, or right after launch, treat the percentage as directional, not definitive, and weight it against the sample size shown alongside it.
This is also why you won't find a single headline accuracy percentage advertised outside the terminal. The only accurate answer to "what's the win rate" is "check the live number, next to its sample size, for the instrument and timeframe you care about" — anything else is a snapshot that's already stale.
Signal fairness
Every user on the same plan tier — Free or Premium — receives identical signals at the identical time. There is no early-access queue, no preferential delivery, and no version of a signal that goes to some accounts before others within a tier. Premium unlocks additional modules and coverage, not a faster or different copy of the same signal.
Limitations
- Accuracy scoring is based on take-profit / stop-loss levels being reached — it does not account for slippage, exchange fees, or partial fills
- Market regime changes (sudden volatility, macro shocks) can degrade signal performance faster than the live accuracy rating reflects
- Outputs come from an experimental AI model and can include predictive variance or outright errors ("hallucinations") — see the Risk Disclosure page
- A live accuracy rating describes past closed signals only and is not a forward-looking guarantee of future results
QUESTIONS
Methodology, answered
It's a relative conviction estimate — how strongly price action, volume and momentum agreed when the signal was generated. It is not a probability of profit or a guarantee of outcome.
No — it's live-tracked from signals that have actually closed since tracking began, auto-scored with no manual curation, and always shown next to its sample size.
Yes. Users on the same plan tier get identical signals at the identical time — no early or preferential access within a tier.
Fast-moving news the model can't price in ahead of time, sudden regime changes, thin liquidity, and the inherent variance of any probabilistic AI model — including occasional hallucinated or incorrect outputs.