Oracle v1.7 CORE · Algorithmic Pricing & Risk IP
Calibre is a 42-feature, six-league inefficiency engine. Every signal is walk-forward validated, Pinnacle-anchored, and passes one rule before anything else: it must beat the close.
LIVE ENGINE · -74.19% ROI · 18.2% WIN · 100% CLV+ · MaxDD −12.5% · n = 20
Sample fixture · live math
Arsenal vs Liverpool
λ · expected goals
1.82/1.41
Selected
1-1
P(score)
11.27%
Modal
1-1
Every quoted 1X2 / OU / BTTS / AH probability is integrated from this 81-cell grid — no per-market heuristics, one generative model.
0.00%
n = 200.0%
live graded0.00
annualized−12.50%
vs 13.60% theory100.00%
beats pinnacle20
live engineInference Path · end to end
End-to-end latency
~50 ms / fixture
Market Microstructure
Pinnacle open/close + 6 sharp books
Steam = (open − close) / open
Dixon-Coles Engine
Time-decayed bivariate Poisson
Generative: one model → all markets
Dual-Engine ML
LightGBM + XGBoost · Pinnacle-anchored
Posterior = 0.55·pin_close + 0.45·ml_raw
Tiered Kelly Risk
T1 Core · T2 Volume · DD-aware
DD > 15% → Kelly ×0.5
Risk Architecture · Tiered
Every signal is routed through Tier 1 or Tier 2 based on live edge, steam, and fundamental alignment. Each tier has its own Kelly fraction — tier capital is kept isolated so conviction never contaminates volume.
Market-specific Kelly
Proven markets get more capital. Structurally weak markets get less. The multiplier is fixed, not discretionary.
alpha · ROI +94%
high-volume base
defense edge
volume engine
risk-capped · weak link
Baseline tick = 1.00× full Kelly.
DD-optimizer proof
We run Kelly at 0.15 — below the theoretical safe line — and still beat the theory.
Theoretical MaxDD
13.60%
15.7% × √(0.15 / 0.20)
Actual MaxDD · 490 OOS
12.50%
beats theoretical by −1.1 pts
League Breakdown · Walk-forward OOS
Six leagues, league-adaptive gates, no cross-subsidy. Every ROI below is Kelly-compounded on isolated per-league bankroll paths.
Portfolio ROI
+24.26%
Premier League
La Liga
Serie A
Super Lig
Bundesliga
Ligue 1
Every signal tracked from day one · No cherry-picking · Walk-forward validated
Sharpe 3.09 · Live from DB
Quantitative Research Partnership
Currently accepting beta partners for v1.7 CORE validation
We are currently inviting a limited number of institutional partners and quants to validate the v1.7 CORE engine. Access is currently provided on a strategic partnership basis.
v1.7 CORE Early Access & Validation Program
Calibre v1.7 CORE is in active validation. We provide full model transparency — walk-forward validated backtests, feature importance rankings, and Dixon-Coles parameter exports — to partners who can stress-test the engine against their own infrastructure.