Reality Engine — Prediction vs Execution
FYOS does not publish a success dashboard here. This surface exists to show where the model tracks reality, where it drifts, and how much trust each layer deserves.
This page separates four kinds of evidence:
- Prediction Alignment — Did predictions match realized direction/magnitude?
- Economic Viability — Did opportunities remain viable after fees, decay, downside, and execution realities?
- Bounded Planner — Portfolio-level validation on limited cohorts only
- Coverage Limits — What remains weakly validated or incomplete?
Product-wide sample
Alignment metric only
Trust layer metric
Use this surface to evaluate model trust, not opportunity attractiveness. Alignment and reliability describe how closely FYOS tracks realized outcomes; survivability, half-life, and capacity show where caution still matters.
Prediction Accuracy
How closely predictions matched realized direction and magnitude within the intended prediction contract. Good alignment does NOT imply economic viability.
Aggregated reality check
No systematic overstatement detected; mean bias is 0.06% and directional flips remain 99.89% of snapshots.
Current trust signal, not a success badge
Model behaves cautiously with mixed evidence.
Reliability reflects alignment with realized outcomes, not profitability.
Observed prediction error bands
Prediction error dispersion is currently high within the validated sample.
Accuracy by exchange
| Exchange | Median Error | Dir. Accuracy | Samples |
|---|---|---|---|
| BINANCE | 0.05% | 99.89% | 1,839 |
| BYBIT | 0.08% | 99.93% | 1,485 |
| OKX | 0.05% | 99.83% | 1,173 |
Economic Viability
Whether opportunities remained economically viable after fees, decay, mirage adjustments, and execution realities. This is separate from prediction accuracy.
Archived validation context only
Deprecated diagnostic only. Not used in primary scoring, ranking, or allocation.
Deprecated diagnostic only. Retained for archived comparison context.
Legacy observations remain visible here only to document cleanup progress, not as primary truth semantics.
Predicted vs actual reality gap
This measures gross-to-net explanatory gap, not prediction failure
Missed decay events
Conservative estimates
Based on 12520 validated observations
Post-decay return accuracy
Predictions with correct sign vs realized
Based on 12630 validated observations
Planner Validation
Portfolio-level validation on bounded cohorts only. These metrics are NOT product-wide system truth - they represent results from specific portfolio replay scenarios.
Portfolios that outperformed expectations
Sparse sample - interpret with caution
Coverage & Validation Limits
What remains weakly validated, incomplete, or limited in scope. These caveats should inform how much weight to give other metrics.
Predicted vs actual decay
Censorship-heavy observation window
Deviation of 20.19h hours marks this as the weakest validated layer.
Expected vs realized capacity
Used as a soft constraint, not an audited execution limit.
Model visibility signals
Active venue points observed
Final verdict
This page reports model accountability—alignment, drift, and validation gaps—so trust is earned empirically.
How predictions are evaluated
FYOS compares prediction snapshots against realized execution outcomes, profiles error distributions, validates realized economic drift empirically, and exposes weak areas instead of hiding them behind performance claims.