Models Dashboard

Phase 40 • Training

Model Story

A story-driven view of each run: trust (performance), drivers (what mattered), adaptation (how it shifts in stress), and blind spots (when it fails).

1. Control Panel

Select a target and a run to load the model story.

Target

Run

2. Actions

Download artifacts or share this view.

JSON

Metrics CSV Feature Usage CSV

Reliability (Calibration)

Verification: Does "60% probability" actually mean 60% win rate?

Signal: Points should closely hug the diagonal line. Divergences indicate the model is over- or under-confident.

View detailed interpretation...

A perfectly calibrated model will have points exactly on the dashed diagonal line. If points fall below the line, the model is overconfident (e.g., predicting 80% probability but only right 60% of the time). If they fall above, it's underconfident. This helps you know if you can trust the magnitude of the model's confidence scores.

Reality Check: Equity Curve

Select a run to compute backtest performance…

Help Keep This Tool Free

If these predictions add value to your market analysis, please consider supporting the project so it can remain accessible and free for everyone. Designing, maintaining, and hosting these models on cloud infrastructure involves ongoing costs.

Support the Project

Prediction Confidence

Select a run to compute prediction uncertainty…

View detailed interpretation...

Equity Curve: Simulates the return path if you strictly followed this model's signal out-of-sample.

Confidence: Shows the raw probability outputs. Tighter bands or lines hovering closer to 50% imply greater uncertainty, whereas strong divergences towards 0% or 100% show high conviction.

Evolution of Trust

Track how allocation rotates across base models over time.

Signal: As you look across the timeline (especially during volatile periods), you will see the shaded colored bands expand and contract as the meta-model shifts capital away from failing algorithms and toward defensive ones. Represents the StackedEnsemble model.

View detailed interpretation...

This chart tracks how allocation rotates across base models over time (stacked to 100%), with the target-specific benchmark overlaid to reveal who is driving decisions in each regime. Watch how weights shift dynamically as market volatility changes.

Signal Forensics

Compare the stacked ensemble against the actual target.

Signal: Look for moments where the ensemble model diverges from the majority of base models or clearly leads the target action.

View detailed interpretation...

This chart plots the stacked ensemble probability (blue line) against the actual binary outcome of the target (black line). The underlying base models (dotted lines) are also shown. This view allows you to forensically examine specific market events and see exactly how the ensemble aggregated the base model signals to form a final conviction.

Model Stability (Rolling Log Loss)

Consistency check across rolling windows.

Signal: Spikes indicate regime failures where the model stopped understanding the market.

View detailed interpretation...

Log Loss penalizes false confidence. A model that is 99% confident but wrong is heavily penalized. Sudden spikes on this chart indicate specific periods in history (like COVID-19 or the 2008 crash) where the model's fundamental assumptions failed. A stable, flat line implies a robust model across all macro conditions.

Comparison

Compare candidate models against the Conservative (Voting) ensemble. Blue indicates the Champion. Green indicates the Voting System.

The Brain

The final feature set used by this run, grouped by type (Volatility / Momentum / Macro / Other).

Numbers represent feature importance scores (relative contribution to prediction).

Signal: The exact raw inputs the model decided were relevant for this run.

Drivers

Ranked importance across the base model layer (normalized). Shows what drove predictions most.

Signal: If the top drivers are overwhelmingly macro features, the model is betting on economic conditions. If they are momentum, it's following the trend.

Phase 50 • Signal

Signal Dashboard

See what the model predicts next and why. This view reads the latest Phase 50 artifacts and explains the signal strength using the same thresholds the system uses internally.

1. Signal Control

Pick a target and (optionally) a snapshot.

Target

Run snapshot

2. Key Metrics

A high-level overview of the most recent directional forecast.

Signal Strength

—

Select a target

Forecast

—

Last Updated

—

Model

—

Forecast JSON Explanation JSON

The Speedometer

A live read of the current signal for the selected target, expressed as probability or forecast value.

Strong Buy —

Buy / Up —

Strong Sell —

Sell / Down —

View detailed interpretation...

A visual gauge of the raw directional probability or expected return from the ensemble.

Help Keep This Tool Free

Support the Project

Ensemble Composition

Select a target to reveal how this signal was generated.

Aggressive (Champion)

Pred. Return

—

Raw market call (unscaled).

Rec. Position

—

Risk-scaled allocation.

—

Forecast Uncertainty

90% Confidence Range

—

Low: —

High: —

Aggressive: —

Conservative: —

Forecast: —

The Forecast Uncertainty range spans from Low to High (90% confidence). The inner colored band highlights the spread between the Aggressive and Conservative strategies.

View detailed interpretation...

Breaks down the aggressive (champion) and conservative predictions, showing how the ensemble balances risk and reward.

Forecast Uncertainty: The 90% confidence range. A wider band implies greater historical error or model disagreement.

Conviction: The standard deviation (σ) of individual base model predictions. A lower standard deviation implies higher conviction (models agree). A higher standard deviation means models are mixed.

Live Drivers: Shows the active weight and contribution of each base model in producing the final ensemble prediction today.

How Much to Trust This Signal
—

A risk-aware view of today’s forecast. The band summarizes uncertainty (model disagreement or historical error).

Confidence: Select a target to estimate uncertainty…

View detailed interpretation...

Visualizes the historical uncertainty bands around the prediction to gauge if the model is operating within normal confidence levels.

Forecast Uncertainty (The Cone)

Range of predictions from individual base models (Level 0).

Loading uncertainty data...

Each dot is a base model prediction; the whisker shows the full min/max range for the ensemble.

View detailed interpretation...

Displays the individual predictions of all base models. A wide spread indicates high disagreement and lower conviction.

Why the Model Thinks This

Feature contributions for this specific prediction. Positive bars push the forecast up; negative bars push it down.

Up Drivers Down Drivers

Insight: If the top bars are dominated by volatility or risk-off features, the model is reacting to regime stress. If momentum and breadth dominate, the model is leaning on trend continuation.

View detailed interpretation...

Feature importance (SHAP values) driving today's forecast. Green bars push the prediction higher, red bars push it lower.

Models Dashboard

Portfolio Models Overview

No Models Found

Help Keep This Tool Free

Model Story

1. Control Panel

2. Actions

Reliability (Calibration)

Reality Check: Equity Curve

Help Keep This Tool Free

Prediction Confidence

Evolution of Trust

Signal Forensics

Model Stability (Rolling Log Loss)

Voting Ensemble

Blind Spots

Comparison

The Brain

Feature Routing Lineage

Drivers

Signal Dashboard

1. Signal Control

2. Key Metrics

The Speedometer

Help Keep This Tool Free

Ensemble Composition

How Much to Trust This Signal
—

Forecast Uncertainty (The Cone)

Why the Model Thinks This

Models Dashboard

Portfolio Models Overview

No Models Found

Help Keep This Tool Free

Model Story

1. Control Panel

2. Actions

Reliability (Calibration)

Reality Check: Equity Curve

Help Keep This Tool Free

Prediction Confidence

Evolution of Trust

Signal Forensics

Model Stability (Rolling Log Loss)

Voting Ensemble

Blind Spots

Comparison

The Brain

Feature Routing Lineage

Drivers

Signal Dashboard

1. Signal Control

2. Key Metrics

The Speedometer

Help Keep This Tool Free

Ensemble Composition

How Much to Trust This Signal —

Forecast Uncertainty (The Cone)

Why the Model Thinks This

How Much to Trust This Signal
—