Rather than relying on one fixed trend signal, this indicator combines an adaptive neural network, continuous model validation, and an independent crisis detector. The result is a simple visual interface backed by a fully causal machine learning process.
structure
At its core is a compact 6-5-1 neural network trained directly on the chart.
Analyze the following six normalized features:
- Short-term and medium-term trend structure
- RSI momentum
- Deviations from linear regression
- Directional price efficiency
- relative volatility
- Candle pressure is regulated by relative volume
The network learns sequentially from completed market results. Each confirmed daily bar allows you to train only based on information from previous bars for which results are known. Future data is never used in current forecasts.
Training uses nonlinear neurons, RMS-scaled gradient updates, error clipping, and regularization. This is an adaptive online model, not a set of fixed coefficients labeled as machine learning.
Self-auditing machine learning
The neural network is continuously compared to an independent structural propensity model.
As mature predictions in the network provide useful additional information, their impact increases. If recent errors become worse than the structural baseline, their impact is automatically mitigated.
This live validation mechanism prevents the indicator from trusting its machine learning component unconditionally.
Crisis detection
A separate stress engine monitors:
- 10 days of rapid decline
- Drawdown from 63-day high
- Abnormal expansion of ATR
- long-term price structure
This layer can trigger defensive states independently of the neural model, allowing the indicator to react to sudden deterioration in the market.
How to read
long — green
Models, trend structures, and confirmation rules support a constructive market environment.
See — Amber
The market remains structurally long, but the evidence of risk or exit is increasing.
Cash — red
Persistent weakness or confirmed critical stress causes the environment to become defensive. This indicator does not take short positions.
Colored neuraxes and surrounding halos display active states without covering the chart with labels. Transition pulses identify confirmed changes, and the dashboard displays bullish probabilities, neural risks, and current machine learning audits.
What’s the difference?
This script integrates four different functions:
1. Online neural network learning
2. Live error-based model validation
3. Independent downside stress detection
4. Verified state machine designed to limit excessive switching
These components are not combined into a simple index vote. Each plays a separate role in learning, validating, protecting, or stabilizing the state.
setting
The ML response controls the adaptation speed and signal stability.
- First reacts faster.
- We recommend something balanced as a starting point.
- Smoothness prioritizes stability.
ML selectivity controls how much evidence is required before LONG or CASH is confirmed.
This indicator is designed exclusively for standard daily charts.
Built-in comparison function
The dashboard includes a lagged long/cash comparison with buy-and-hold. Apply selected migration costs and openly display periods of poor model performance.
This comparison is a diagnostic tool and not a complete strategy backtest. It does not include all possible spreads, slippage, taxes, financing or execution constraints.
Important limitations
Bullish probability is an internally normalized score, not a statistically adjusted probability of profit. This model is slow to react, can generate false transitions in sideways markets, and cannot eliminate gap risk.
The unfolding daily bar may change before it ends. The confirmed historical state does not use future data, look-ahead, or longer time frame security calls.
This indicator provides market conditions and does not provide financial advice or guarantees of performance. Online learning does not mean that you will perform better in the future.
