The Model
A machine learning system designed to identify statistical edges in crypto markets, validated using out-of-sample historical data.
What the System Does
Data-Driven
The model learns directly from raw market data — price, volume, and market structure — without relying on hand-crafted rules or subjective interpretation.
Path-Based Validation
Signals are validated using real trading logic: did the target get hit before the stop? This mirrors how actual trades play out, not just whether price went up or down.
Statistically Validated
Every model must pass rigorous out-of-sample testing with significance testing before going live. If the edge isn't statistically real, it doesn't ship.
Signal Specifications
Timeframe
1-Hour Candles
Asset
BTC/USDT
Frequency
Every Hour
Signal Validity
8 Hours
Trading Framework
Every signal falls into one of three categories. The model explicitly outputs NO TRADE when it doesn't see an edge.
The model identifies a statistical edge for price moving up. Entry, stop-loss, and take-profit levels are provided.
The model identifies a statistical edge for price moving down. Entry, stop-loss, and take-profit levels are provided.
No clear directional edge detected. The model stays flat rather than forcing a trade — knowing when not to trade is as important as knowing when to.
Asymmetric Risk-to-Reward (~2.5–3:1)
Targets are set wider than stop-losses, so winning trades are significantly larger than losing ones. This means the system can be profitable even with a win rate well below 50% — the math favors disciplined, asymmetric setups over high win rates.
Each directional signal has a predefined stop-loss and take-profit and remains valid for 8 hours unless either level is hit first.
Market Regime Awareness
The model classifies the current market into one of four regimes and adapts its behavior accordingly.
Trending
Sustained directional moves with follow-through momentum
Mean Reverting
Range-bound conditions where price oscillates around a level
Breakout
High-volatility expansion phases with sharp directional moves
Low Vol Chop
Quiet, directionless markets with low signal-to-noise ratio
Continuous Improvement
Every signal feeds back into a supervised learning loop that monitors performance and detects when the model's edge may be degrading.
Signal Logged
Every prediction is automatically recorded with its entry, stop-loss, target, and confidence at the time of generation.
Outcome Resolved
After the signal window expires, actual market data is checked: did the target hit before the stop, or vice versa?
Performance Monitored
Rolling win rates, cumulative PnL, drawdowns, and drift detection are tracked continuously to ensure the edge persists.
Available Models
Risk Disclaimer
CoinInsight.ai is a decision-support tool, not a decision-maker. Statistical edges are not guarantees — all models can and will have losing periods. Past performance is not indicative of future results. Never trade with money you cannot afford to lose.
Read full disclaimer →