VALIDATION DIAGNOSTICS
CIF v0.21.3
HEI diagnostics plus cross-asset replication; v0.8 scoring formulas remain unchanged.
Mode
DIAGNOSTIC
DIAGNOSTIC
NO AUTO-TUNING v0.9.2 is descriptive only. It does not change MOS weights or formulas from the HEI replay sample. Small high-coverage cohorts are explicitly labelled and cannot justify weight changes.
VALIDATION INTEGRITY v0.9.2
30-day horizon maturity audit
Mature 30d returns42full horizon reached
30d drawdowns42mature paths only
Premature drawdowns0must be 0
Mature paths missing0must be 0
A 30-day drawdown/run-up is emitted only after a fully closed daily candle reaches the complete +30d horizon. Partial future paths are NULL, never mislabeled as 30d.
FULL-MODEL FORWARD SHADOW
Production validation progress
Shadow snapshots601 per UTC day
Distinct projects1need ≥5
Historical validationCOLLECTINGmulti-horizon OOS
CalibrationCOLLECTINGordinal score validation
30d matured29target 100
90d matured0target 60
180d matured0target 30
365d matured0target 15
MOS REPLAY BASELINE
The result that triggered diagnostics
7d outcomes45
30d outcomes42
Coverage ≥709
All-point MOS→30d excess BTC-0.085
The all-point statistic mixes different historical coverage regimes. v0.9.1 therefore reports each regime separately before any model decision is considered.
COVERAGE-REGIME DIAGNOSTICS
Does the result change as MOS becomes more complete?
RegimeN 30d30d return ρExcess BTC ρExcess ETH ρBTC tertile spread
All scoreableprovisional42-0.089-0.085-0.066-4.77
Coverage ≥50%provisional42-0.089-0.085-0.066-4.77
Coverage ≥60%provisional42-0.089-0.085-0.066-4.77
Coverage ≥70%too_small5-0.500-0.500-0.500—
Full current component settoo_small5-0.500-0.500-0.500—
ρ = Spearman rank correlation. With the current HEI history, coverage ≥70 is expected to remain a small exploratory cohort; its sign is informative but not decision-grade.
DRAWDOWN DIAGNOSTICS
30-day drawdown by MOS tertile
All scoreableMature N=42 · provisional
Low MOS median DD-18.84%
High MOS median DD-35.40%
High − Low-16.56 pp
Coverage ≥60%Mature N=42 · provisional
Low MOS median DD-18.84%
High MOS median DD-35.40%
High − Low-16.56 pp
Coverage ≥70%Mature N=5 · too_small
Low MOS median DD—%
High MOS median DD—%
High − Low— pp
Full current component setMature N=5 · too_small
Low MOS median DD—%
High MOS median DD—%
High − Low— pp
COMPONENT DIAGNOSTICS — ALL SCOREABLE
Which MOS components carry forward information?
ComponentN 30dReturn ρExcess BTC ρExcess ETH ρBTC tertile spread
1D Trendprovisional42-0.218-0.154-0.105-2.01
4H / 1H Structureprovisional42-0.0120.0000.0206.31
Volume & Participationprovisional420.3770.2180.1767.57
Relative Strengthprovisional42-0.145-0.273-0.285-11.12
1M / 1W Structuretoo_small5-0.700-0.700-0.700—
Range Positioningprovisional42-0.609-0.344-0.242-19.66
COMPONENT DIAGNOSTICS — COVERAGE ≥70
Same components inside the current high-coverage regime
ComponentN 30dReturn ρExcess BTC ρExcess ETH ρReliability
1D Trend5-1.000-1.000-1.000too_small
4H / 1H Structure5-0.300-0.300-0.300too_small
Volume & Participation5-0.100-0.100-0.100too_small
Relative Strength5-0.700-0.700-0.700too_small
1M / 1W Structure5-0.700-0.700-0.700too_small
Range Positioning5-1.000-1.000-1.000too_small
ANTI-OVERFIT GUARDRAILS
What v0.10 is allowed to conclude
- No automatic MOS weight or formula changes.
- Coverage-regime diagnostics are separated before interpreting aggregate replay results.
- Component diagnostics are multiple comparisons and remain descriptive.
- At least 20 observations are required before even reviewing a weight; 50 is preferred.
- Any future formula change must be frozen first and validated on a different / forward sample.
- No AQS/Risk backfill from present-day fundamentals.
- Forward snapshots are timestamped before outcomes exist.
- MOS replay uses only candles closed at/before each checkpoint.
- Historical MOS replay cannot unlock full-model production validation.
LATEST SHADOW SNAPSHOTS
Frozen score states
2026-10-05deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK59.3 / 59.2 / 51.3
coverage73/75/73.25
2026-10-04deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK59.3 / 49.0 / 51.0
coverage73/75/73.25
2026-10-03deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.1 / 49.5 / 50.0
coverage73/75/73.25
2026-10-02deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.3 / 65.1 / 52.6
coverage73/75/73.25
2026-10-01deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK61.4 / 66.8 / 47.8
coverage73/75/73.25
2026-09-30deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK59.6 / 52.0 / 52.4
coverage73/75/73.25
2026-09-29deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK59.5 / 54.1 / 52.2
coverage73/75/73.25
2026-09-28deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.2 / 58.5 / 51.2
coverage73/75/73.25
2026-09-27deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.0 / 64.7 / 52.8
coverage73/75/73.25
2026-09-26deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.3 / 57.0 / 50.8
coverage73/75/73.25
2026-09-25deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.2 / 59.2 / 50.9
coverage73/75/73.25
2026-09-24deterministic-preview-v0.8 · coherent true
AQS / MOS / RISK60.6 / 59.7 / 50.3
coverage73/75/73.25