Cookie preferences
We use essential cookies for session continuity and optional analytics cookies to improve the beta experience. You can accept or reject non-essential cookies. Learn more in our Privacy Policy.
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:
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; deprecated legacy branches stay explicitly labeled, while decay, capacity, and coverage limits show where caution still matters.
Not all metrics carry equal weight. Use this guide to interpret validation evidence correctly.
Strongest validated evidence. Use for primary decision context.
Useful context but not primary truth. Interpret alongside primary metrics.
Limited validation depth. Treat as exploratory signal, not proven truth.
Not decision-grade evidence. Provided for transparency, not as proof.
How closely predictions matched realized direction and magnitude within the intended prediction contract. Good alignment does NOT imply economic viability.
Predicted vs realized divergence exceeds 0.15%, sign flip rate is 12.46%.
Predicted return reflects gross model estimate before execution costs. Realized return reflects net outcomes after fees, decay, and execution reality. This gap is expected and is what the viability layer explains.
Strong alignment signal from realized outcomes.
Reliability reflects alignment with realized outcomes, not profitability.
Prediction error dispersion is currently high within the validated sample.
| Exchange | Median Error | Samples | Dir. Acc.Exp. |
|---|---|---|---|
| BINANCEUnder investigation | 5.47% | 614,970 | 89.12% |
| BYBITUnder investigation | 8.01% | 576,639 | 86.64% |
| OKXUnder investigation | 2.43% | 288,958 | 85.95% |
Direction metrics are experimental diagnostics. Median error is the primary alignment signal.
Prediction Alignment asks whether FYOS tracked realized direction and magnitude correctly. Economic Viability asks whether those aligned signals remained tradable after fees, decay, mirage, and execution reality.
A model can predict direction accurately while the economic opportunity still fails to survive execution costs. These are intentionally different truths - alignment measures forecast quality, viability measures real-world survivability.
Whether opportunities remained economically viable after fees, decay, mirage adjustments, and execution realities. This is separate from prediction accuracy.
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.
This is the expected gap between headline yield and net-facing return. Large values indicate high gross/net divergence, not model failure.
Cases where decay was worse than predicted
Cases where model was conservative
Based on 98634 validated observations
Primary viability signal
Post-decay sign accuracy — experimental diagnostic only, not a viability signal
Based on 308684 validated observations
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
What remains weakly validated, incomplete, or limited in scope. These caveats should inform how much weight to give other metrics.
Historical layer exists, but evidence is censorship-heavy and limited.
Censorship-heavy observation window
Deviation of 11.90h hours marks this as the weakest validated layer.
Modeled reference, not audited execution truth.
Used as a soft constraint, not an audited execution limit.
Active venue points observed
Final verdict
This page reports model accountability—alignment, drift, and validation gaps—so trust is earned empirically.
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.