Calibration record
There is no win rate here and no trading results — we do not have them at all. We publish one thing: are the risk boundaries printed in every report breached at the rate they claim? An honest model at 95% confidence is breached on ~5% of days only — no more (which would understate the risk) and no less (which would overstate the caution).
Every figure below is computed by a point-in-time historical sweep, with no forward look. Where the sample is too small for a reliable judgement (minimum 20), we say so rather than guess.
EURUSD 1D
Risk-boundary breaches (VaR 95%)
5.3% target 5%
Sound calibration — the band describes the risk honestly
✓ Formal Kupiec test: calibration not rejected at 95% confidence (p=0.31)
✓ Timing (Christoffersen): breaches are dispersed — no clustering
Sample: 4985 scored readings (36 live · 4949 historical sweep) · 265 actual breaches.
Loss depth on breach days (ES 95%)
1.07× target 1.0×
Realised losses match the declared ES (ratio ~1)
Ratio = mean realised loss ÷ declared ES, over 265 breach days.
Probability cone — each horizon on its own
Each edge is a one-sided 95% bound: the healthy standard is ~5% broken per side, at every horizon.
| Horizon | Lower edge | Upper edge |
|---|---|---|
| 1 day | 5.4% ✓ | 5.1% ✓ |
| 5 day | 5.7% ⚠ | 5% ✓ |
| 10 day | 5.8% ⚠ | 4.8% ✓ |
| 20 day | 6.5% ⚠ | 4.8% ✓ |
For horizons > 1 the evaluation windows overlap, so Kupiec is indicative there rather than conclusive — the rate itself stays unbiased.
Holding rate of tested support/resistance
Supports 50.8% held (1360 of 2678 tests)
Resistances 56.7% held (1626 of 2868 tests)
XAGUSD 1D
Risk-boundary breaches (VaR 95%)
4.8% target 5%
Sound calibration — the band describes the risk honestly
✓ Formal Kupiec test: calibration not rejected at 95% confidence (p=0.96)
✓ Timing (Christoffersen): breaches are dispersed — no clustering
Sample: 21 scored readings (21 live · 0 historical sweep) · 1 actual breaches.
Loss depth on breach days (ES 95%)
Sample too small
1 breach days with a recorded ES — minimum 10.
Probability cone — each horizon on its own
Each edge is a one-sided 95% bound: the healthy standard is ~5% broken per side, at every horizon.
| Horizon | Lower edge | Upper edge |
|---|---|---|
| 1 day | 5.2% ✓ | 5.1% ✓ |
| 5 day | 5.7% ✓ | 6.2% ⚠ |
| 10 day | 6% ⚠ | 7.7% ⚠ |
| 20 day | 4.8% ✓ | 7.6% ⚠ |
For horizons > 1 the evaluation windows overlap, so Kupiec is indicative there rather than conclusive — the rate itself stays unbiased.
Holding rate of tested support/resistance
Supports 48.8% held (1014 of 2079 tests)
Resistances 48% held (983 of 2046 tests)
XAUUSD 1D
Risk-boundary breaches (VaR 95%)
5.7% target 5%
Sound calibration — the band describes the risk honestly
✓ Formal Kupiec test: calibration not rejected at 95% confidence (p=0.06)
✓ Timing (Christoffersen): breaches are dispersed — no clustering
Sample: 3935 scored readings (28 live · 3907 historical sweep) · 223 actual breaches.
Loss depth on breach days (ES 95%)
1.14× target 1.0×
Realised losses match the declared ES (ratio ~1)
Ratio = mean realised loss ÷ declared ES, over 223 breach days.
Probability cone — each horizon on its own
Each edge is a one-sided 95% bound: the healthy standard is ~5% broken per side, at every horizon.
| Horizon | Lower edge | Upper edge |
|---|---|---|
| 1 day | 5.6% ✓ | 5.1% ✓ |
| 5 day | 5.5% ✓ | 5.9% ⚠ |
| 10 day | 5.3% ✓ | 6.6% ⚠ |
| 20 day | 4.7% ✓ | 8.2% ⚠ |
For horizons > 1 the evaluation windows overlap, so Kupiec is indicative there rather than conclusive — the rate itself stays unbiased.
Holding rate of tested support/resistance
Supports 51.9% held (937 of 1805 tests)
Resistances 40.7% held (745 of 1830 tests)