quant.config.json

quant.config.json is the control plane for a quant strategy.

It should contain environment and runtime declarations, not arbitrary narrative or loose notes.

Typical shape

{
  "pair": "BTCUSDT",
  "timeframe": "1h",
  "artifact_target": "python_bundle",
  "parameters": {
    "lookback": 5
  },
  "training_requirements": [
    "numpy",
    "pandas",
    "scikit-learn"
  ],
  "inference_requirements": [
    "numpy"
  ]
}

Core fields

pair

Primary market symbol for the strategy.

Examples:

• BTCUSDT
• EURUSD
• XAUUSD

timeframe

Primary candle timeframe.

Examples:

• 1m
• 5m
• 15m
• 1h
• 4h
• 1d

artifact_target

Declares the intended artifact class.

Current serious targets are:

• python_bundle
• joblib
• onnx

parameters

Free-form strategy parameters used by train() and predict().

Examples:

• lookback
• thresholds
• feature toggles
• session filters

Requirements fields

training_requirements

Packages declared for the training environment.

Important behavior:

• default training packages are always installed
• approved extras are installed after the defaults
• this is a real mechanism, not just metadata

inference_requirements

Packages declared for live runtime compatibility.

Important behavior:

• this is validated against supported runtime packages
• unsupported requirements should block deployment
• this is a compatibility contract, not a promise that workers install arbitrary packages at runtime

Why requirements live in config

Package declarations belong in config, not in the Python file.

That keeps:

• strategy.py as real Python
• environment declarations explicit
• training and inference concerns auditable

Best practice

• keep the config small and explicit
• use parameters for strategy knobs
• use requirements only when actually needed
• keep artifact_target aligned with the model/runtime plan

If quant.config.json becomes vague or overloaded, the deployment story gets vague too.