How To Read Reality
The FYOS Reality surface is an evidence layer, not a promise layer.
What Reality Is
Reality shows:
- Predicted vs realized outcome comparisons
- Historical evidence of model accuracy
- Error profiles across cohorts
- Reliability grades for trust assessment
Reality answers the question:
"How has the model performed historically, and what does that suggest about confidence levels?"
What Reality Is Not
Reality is not:
- A guarantee of future accuracy
- A promise that predictions will be correct
- A replacement for careful interpretation
- A blanket endorsement of all opportunities
Reality is an evidence surface — it shows you what happened, not what will happen.
Prediction Snapshot Persistence
FYOS persists prediction snapshots:
- Captures model predictions at specific points in time
- Records the expected outcomes based on model state
- Enables later comparison against realized outcomes
This creates an audit trail for model evaluation.
Realized Outcome Evaluation
When outcomes become observable, FYOS evaluates:
- Realized return vs predicted return
- Prediction error magnitude and direction
- Error patterns across cohorts, exchanges, and time
This evidence feeds back into trust and reliability systems.
Prediction Error Profiling
Error profiling reveals:
- Systematic biases: Does the model consistently over- or under-predict?
- Cohort patterns: Which segments have better accuracy?
- Temporal patterns: Does accuracy vary over time?
Error profiles inform reliability grades and runtime haircuts.
Reliability Grading
Based on accumulated evidence, FYOS assigns reliability grades:
| Grade | Meaning |
|---|---|
A | Strong historical accuracy |
B | Good accuracy with some variance |
C | Moderate accuracy, use caution |
D | Weak accuracy, significant caution |
ungraded | Insufficient history |
Reliability grades inform:
- Trust-layer confidence
- Runtime haircut levels
- User-facing warnings
How Reality Supports Decisions
Reality helps you understand:
- Confidence calibration: How much to trust model outputs
- Segment quality: Which exchanges/assets have better track records
- Uncertainty bounds: What range of outcomes is realistic
- Red flags: Where the model has struggled historically
Limitations
Reality has important limitations:
- Past is not future: Historical evidence doesn't guarantee future accuracy
- Regime changes: Market conditions can shift
- Coverage gaps: Not all segments have equal evidence
- Sample size: Small-sample cohorts have noisy estimates
Use Reality as context, not as a crystal ball.
Reality vs Trust Layer
| Concept | Purpose |
|---|---|
| Reality | Show evidence of model accuracy |
| Trust Layer | Gate and adjust based on evidence |
Reality provides the evidence. Trust Layer uses it to inform product decisions.
Correct Interpretation
Right way:
- "Reality shows this cohort has historically been accurate within X%"
- "This is evidence, not a guarantee"
- "I should adjust my confidence based on this evidence"
Wrong way:
- "Reality proves the model is always correct"
- "This historical accuracy guarantees future accuracy"
- "I can ignore all other risk factors because Reality looks good"
Evidence-Oriented Philosophy
FYOS is intentionally designed as a pessimistic, evidence-oriented system.
Reality embodies this philosophy:
- Show what actually happened
- Don't hide model limitations
- Don't oversell accuracy
- Let users make informed decisions
Summary
- Reality is an evidence surface, not a promise surface
- It shows predicted vs realized outcomes
- It informs reliability grades and trust
- It has limitations — past is not future
- Use Reality for confidence calibration, not certainty
- Combine with other methodology components for complete interpretation