How it works

Deterministic allocation in six steps. Each rule is fixed, testable, and reproducible.

Data

Adj Close & Volume (Yahoo), fundamentals (Alpha Vantage demo), benchmarks (FRED), crypto caps (CoinGecko). 24 h HTTP cache; ≤ 1-day forward fill for EOD gaps.

Score (Durability-Lite)

Binary, explainable rules: ROE > 12 %, D/E < 1, Margin > 10 %, Insider > 2 %, R&D/Rev > 5 %; penalty if (D/E > 1 ∧ Insider < 1 %). Score clamped to 0–100.

Risk

Volatility σ via GARCH(1,1) on MEΩ-relative returns; CVaR<sub>95</sub> reported for tails.

Allocation

Drop last winner → keep tickers with σ ≤ median(σ) → pick highest Durability score; tie → lower σ.

Cost gate

Fee 12 bp + slippage from √impact: impact_decimal = 0.001 × √(qty / ADV10); impact_bp = 10000 × impact_decimal. Skip trade if fee_bp + impact_bp > SlipCap (35 bp default).

Numéraire & GINα

Everything is expressed in MEΩ. Performance is reported as GINα, below.

GINα_t = (1 + R_t^P) / (1 + r_t^IRS + π_t^MEΩ) − 1 − C_carry,t

r_t^IRS = global interest-rate surface (polynomial fit, CIP-constrained)
π_t^MEΩ = MEΩ-weighted global inflation
C_carry,t = funding / storage / staking net costs

GIN alpha = ((1 + R^P) / (1 + r_IRS + pi_MEO)) - 1 - carry

Use MEO for environments that cannot render Ω; sigma_63 for σ<sub>63</sub>.

All steps are deterministic. Given identical inputs and timestamps, the system produces identical outputs. No discretion, no optimization, no look-ahead bias.