Trust Layer
The trust layer evaluates the historical accuracy of FYOS predictions, creating accountability for displayed metrics.
Purpose
The trust layer answers: "How reliable are FYOS predictions?"
By comparing historical predictions against realized outcomes, we:
- Grade reliability per opportunity
- Identify systematic prediction errors
- Adjust confidence in current predictions
Components
1. Realized Outcome Evaluator
Captures actual funding outcomes at the end of holding periods:
- Actual funding received vs. predicted
- Actual fees paid vs. assumed
- Actual capacity effects vs. modeled
2. Prediction Error Profiler
Measures systematic errors:
Tracks error distributions per:
- Opportunity
- Time horizon
- Market condition
3. Reliability Grading Engine
Assigns reliability grades based on historical accuracy:
| Grade | Error Range | Interpretation |
|---|---|---|
| A | < 10% | Highly reliable predictions |
| B | 10-20% | Good reliability |
| C | 20-35% | Moderate reliability |
| D | 35-50% | Low reliability |
| F | > 50% | Unreliable predictions |
Edge Reliability Score
The Edge Reliability Score (0-100) synthesizes:
- Historical prediction accuracy
- Sample size (more history = more confidence)
- Recent vs. historical performance
- Consistency across market conditions
Confidence Levels
Based on reliability, we assign confidence:
| Reliability Score | Confidence Level |
|---|---|
| > 80 | High confidence |
| 60-80 | Medium confidence |
| 40-60 | Low confidence |
| < 40 | Insufficient data / unreliable |
Trust-Adjusted Metrics
In some views, metrics are adjusted by reliability:
This provides a more conservative estimate that accounts for prediction uncertainty.
Display in UI
Opportunity Detail
- Reliability Score — 0-100 grade
- Prediction Error — Historical error percentage
- Sample Size — Number of evaluated predictions
- Confidence Badge — High/Medium/Low/Insufficient
Screener
- Reliability can be used as a filter criterion
- Low reliability opportunities can be flagged or hidden
Building Trust Over Time
The trust layer improves with:
- More predictions — Larger sample sizes increase confidence
- Model updates — Systematic errors inform model improvements
- Transparency — Published error metrics build user trust
Limitations
The trust layer cannot account for:
- Future market regime changes
- Black swan events
- Exchange-specific anomalies not in historical data
Users should treat reliability scores as relative guidance, not absolute guarantees.