Methodology v1.0 · Current
1. Raw error metrics
R is the reference, F the single-value forecast, and A the actual outcome.
- Absolute error = |F − A|
- Percentage error = |F − A| / |A| × 100
- Signed error = (F − A) / |A| × 100
Positive signed error means too high; negative means too low. If A is zero, percentage error and single-value score are not determinable, rather than inventing a denominator.
2. Price-target base score
| Percentage error | Base score |
|---|---|
| 0% | 10.0 |
| >0–2% | 9.5 |
| >2–5% | 9.0 |
| >5–10% | 8.0 |
| >10–15% | 7.0 |
| >15–20% | 6.0 |
| >20–30% | 5.0 |
| >30–40% | 4.0 |
| >40–50% | 3.0 |
| >50–75% | 2.0 |
| >75–100% | 1.0 |
| >100% | 1.0 |
3. Direction accuracy
Actual change within ±0.5% of the reference is neutral; otherwise above is bullish and below is bearish. We compare this with the submitted direction. Matching direction adds 0.5; incorrect direction applies -0.5; not determinable adds zero. A bullish forecast with an unchanged outcome is incorrect. A zero reference makes direction not determinable. Final score = base plus adjustment, clamped to 1.0–10.0 and rounded to one decimal.
4. Range forecasts
A range is not treated as a point forecast. Inside the range, base score = 10 − 2 × distance from center / half-width: 10 at center and 8 at either boundary. A zero-width range at its center receives 10. Outside, error is the distance to the nearest boundary; percentage error divides that distance by |A|. The threshold table applies, capped at 8 so moving just outside an edge never improves the base score. The same direction adjustment applies. Signed error is measured from the nearest boundary, or zero inside.
5. Target hit requires period evidence
A bullish point target requires an observed period high reaching it; a bearish point target requires a period low. A witnessed crossing establishes a hit. A miss requires complete period evidence; partial evidence without a crossing remains not determinable. Target-date closing value alone does not establish a hit. Neutral and range forecasts have no directional point-target hit. Period high/low evidence is manually supplied; no automatic market-data feed is used.