The research system

Quantitative engine

The Python quant stack, the feature registry, and the rule that the AI never writes a raw score.

The quantitative engine is where intuition becomes mathematics. It computes the measurable evidence the agents interpret and the scoring engine combines — in Python, with a registered, versioned set of features.

The stack

NumPy
pandas
SciPy
statsmodels
scikit-learn
TA-Lib / pandas-ta

Library	Used for
NumPy	Numerical arrays, vectorized maths
pandas	Time-series, feature generation, resampling, rolling windows
SciPy	Statistical tests, optimization, signal processing
statsmodels	Regression, stationarity tests, diagnostics, hypothesis testing
scikit-learn	Classification, clustering, anomaly detection, model evaluation
TA-Lib / pandas-ta	Technical indicators — treated as features, never standalone proof

What it computes

Momentum, volatility, liquidity and microstructure features, for example:

price momentum
return acceleration
realized volatility
volume acceleration
order-book imbalance
estimated slippage
unique buyer growth
holder growth
wallet concentration
whale activity
BTC/ETH correlation
social velocity
anomaly scores

The feature registry

Every feature is registered so calculations never change invisibly. Each entry carries a unique code, description, formula, data dependencies, version, time window, expected range, missing-data policy, owner and deprecation status.

Where the formulas live

The features above are not hand-waved — each is a versioned, unit-tested formula. Two representative examples; the full catalogue is the Mathematical library, with sizing, liquidity and performance maths on the adjacent pages.

returns_v1

Simple return

R_{t} = \frac{P _{t} - P _{t - 1}}{P _{t - 1}}

The percentage change between two prices — the most basic measured evidence.

momentum_v1

Volatility-normalized momentum

N or ma l i z e d M o m e n t u m = \frac{R O C _{n}}{σ _{n}}

Momentum in units of volatility, so one threshold behaves consistently across regimes.

The scoring inputs

Signal Score — only measurable market and project data.
Risk Score — contract, liquidity, market and portfolio risk.
Execution Score — whether the trade is realistic at the expected price.
Market regime — bull, bear, high-volatility or neutral context.

The AI never writes a raw score

Each AI output follows a structured schema (evidence, confidence, sources, contrary evidence, missing information). The final score is computed by deterministic code — the model proposes interpretation, mathematics produces the number. See the scoring engine.