Skip to content

stability.mdx

Latest commit

 

History

History
72 lines (55 loc) · 2.96 KB

stability.mdx

File metadata and controls

72 lines (55 loc) · 2.96 KB
Raw
title Stable Kernel and Compatibility
description Neural's supported kernel surface, experimental modules, and deprecation policy

Neural is the deterministic Python kernel beneath Vaticor. The stable base does not authorize provider calls, live orders, deployment, or model-driven side effects.

Stable surface

Import stable capabilities from neural.kernel:

from neural.kernel import NormalizedMarket, ReplayEvent, replay

The stable surface contains:

  • normalized market, quote, order, position, capability, and policy types;
  • dependency-free deterministic replay with a canonical SHA-256 digest;
  • neural doctor, neural capabilities, and neural replay demo.

Run the installed-wheel demo without credentials or network access:

neural --json replay demo

The result contains a digest, event count, market count, and final snapshot. The same fixture must produce the same result regardless of input order.

Capability matrix

neural --json capabilities is the machine-readable source for the current matrix.

Capability Status Install path Compatibility
Kernel normalization Stable base wheel Public neural.kernel contract
Kernel replay Stable base wheel Deterministic output and digest
CLI diagnostics Stable base wheel JSON envelope retained
Kalshi auth and collection Experimental neural-sdk[trading] May change before 1.0
Paper and venue adapters Experimental neural-sdk[trading] Paper-first; no live authority
Strategy and backtesting Experimental neural-sdk[analysis] May change before 1.0
Sentiment Deprecated neural-sdk[sentiment] Compatibility only
Generic deployment Deprecated neural-sdk[deployment] Move runtime ownership to Vaticor
FIX experiments Deprecated neural-sdk[fix] Compatibility only

Existing imports remain available during the 0.4.x compatibility window. New code should use neural.kernel for stable contracts.

Compatibility policy

  • Stable symbols retain compatible behavior within a minor release line.
  • Experimental symbols can change before 1.0; changes require release notes.
  • Deprecated symbols remain through at least one later minor release and name their replacement before removal.
  • Removal never silently enables a provider, credential, deployment, or live execution path.
  • Model output remains untrusted. Schema or replay success cannot authorize an external side effect.
  • Exact wheel compatibility requires the Neural commit, wheel digest, Python version, and validator result. Local source-path imports are insufficient.

Legacy import path

Existing neural.exchanges, neural.trading, neural.analysis, neural.data_collection, and neural.deployment imports are preserved for compatibility. Install the matching extra before using those surfaces. Moving to neural.kernel removes provider and optional-stack coupling from portable contract and replay code.