arithmion-core

The interface tier of Arithmion, an AI-native quantitative research and trading platform. This package contains the contracts the platform is built on: abstract base classes, typed data carriers, and the entry-point plugin registry. It contains no engine, no providers, and no agent runtime — those live in the Arithmion product, which implements these contracts.

Links — arithmion.com for the product; arithmion.com/learn for the learning center, a free bilingual (English / 中文) quant curriculum covering market structure, statistics, machine learning, and a validation track on multiple testing, PBO and the deflated Sharpe ratio — also available as markdown in quant-learning-path.

The package installs into the apexquant namespace (import apexquant.core), which is the platform's historical package name; the import path is shared with the full product so that code written against these contracts runs unchanged in both.

What is in here

Data contracts — the shapes that cross every boundary:

• bar / signals / result / trade_plan — OHLCV bars, signal proxies, prediction results, trade plans
• backtest.result / backtest.serialize — the backtest result carrier and its canonical JSON wire shape
• panel — the cross-sectional factor-panel shape and its OHLCV converters
• config_schema — the typed schema plugins use to declare their configuration surface
• model_meta — the meta.json schema describing trained-model artifacts

Plugin contracts — subclass one of these, register it via an entry point, and the platform discovers it:

• strategy / portfolio_strategy — bar-driven and rebalance-driven strategies
• predictor — per-bar inference
• datasource / streaming — bounded (batch) and unbounded (live) data
• broker — order routing
• engine — the backtest-engine protocol and factory
• metric / validation — result metrics and trustworthiness validators
• risk — risk manager, position sizer, stop policy
• training — the trainer contract and the walk-forward window partitioner
• universe / calendar — point-in-time stock pools and trading calendars

Mechanism:

• registry / plugins — the domain registry and the entry-point scanner that performs discovery

Install

pip install -e .

Requires Python ≥ 3.11. Runtime dependencies are pandas and loguru only.

A minimal strategy

from apexquant.core.bar import Bar
from apexquant.core.strategy import BaseStrategy, Signal

class Momentum(BaseStrategy):
    name = "momentum"
    freq = "daily"

    def on_bar(self, bar: Bar) -> Signal:
        if bar.close > bar.open:
            return Signal.BUY
        return Signal.HOLD

Third-party packages expose implementations through entry-point groups (apexquant.strategies, apexquant.datasources, apexquant.metrics, …); see plugins.py for the group inventory and registry.py for the domain registry the scanner populates.

Stability

Interface versioning follows the platform's roadmap: during v0.x no interface is frozen, and breaking changes are recorded in CHANGELOG.md. Contracts carry an __apexquant_api_version__ tuple where versioned.

Relationship to the main repository

The Arithmion product repository is the source of truth; this repository is a curated export of its src/apexquant/core interface tier (scripts/sync_from_apexquant.py performs the sync and enforces the scope). Issues and contributions against the contracts are welcome here.

Why apexquant (and aq)?

Arithmion grew out of ApexQuant, a graduate thesis project on applying deep time-series learning to stock forecasting and quantitative trading:

• Applying Deep Time-Series Learning to Stock Forecasting and Quant Trading — the thesis
• Apexquant — the thesis-era codebase this platform descends from

When the project became the Arithmion platform, the brand changed but the Python package kept its original apexquant name — frankly, because renaming it was not worth the churn: imports, entry-point groups (apexquant.strategies, …), and the __apexquant_api_version__ markers would all have had to move for zero functional gain. A convenient side effect is that thesis-era and current code interoperate as-is. The same applies to the aq shorthand you may see elsewhere in the platform (CSS tokens, UI prefixes). Wherever you read apexquant or aq, it means Arithmion.

License

MIT — see LICENSE.

中文说明

arithmion-core 是 Arithmion（AI 原生量化平台）的接口层：平台赖以构建的 契约——抽象基类、类型化数据载体、以及基于 entry-point 的插件注册机制。 本仓库不包含引擎实现、数据/券商 provider 或 agent 运行时；它们属于实现了 这些契约的 Arithmion 产品本体。

相关链接——产品在 arithmion.com；学习中心在 arithmion.com/learn：免费的中英双语量化课程， 覆盖市场结构、数理统计、机器学习，以及围绕多重检验、PBO 与 Deflated Sharpe 的验证专线，markdown 版在 quant-learning-path。

包安装在 apexquant 命名空间下（import apexquant.core），与完整产品共享 导入路径：针对这些契约编写的代码在两侧无需修改即可运行。

安装：pip install -e .，要求 Python ≥ 3.11，运行时依赖仅 pandas 与 loguru。

关于 apexquant 与 aq 命名：Arithmion 的前身是毕业设计项目 ApexQuant（论文、 当时的代码库）。品牌更名后 Python 包沿用了原名 apexquant——说白了是改名不值得折腾：导入路径、 entry-point 组名、__apexquant_api_version__ 标记全都得跟着动，却没有 任何功能收益。顺带的好处是论文时期与当前的代码无需迁移即可互通；平台 其它位置的 aq 简写同理。代码中的 apexquant 与 aq 均指 Arithmion。

v0.x 期间接口均不冻结，破坏性变更记录于 CHANGELOG.md。许可证为 MIT。