AI-WORKSPACE

A unified backtest framework and agent orchestration layer for the trading research vault. Replaces the scattered, copy-pasted Strategies/*.py scripts with a single reusable library, a runner CLI, and a workflow orchestrator that mirrors the prompts in agents.md.

Quick Start

Run every registered strategy and print the ranking table:

Run a single strategy by name:

Run the end-to-end agent workflow (research, design, backtest, validate, deploy, monitor, learn) for a given idea:

Design Principles

• Next-open execution by default. execution="close" is explicit,

never the default. See Strategies/Volume-Scaled IBS.md for the data-quality reason.

• Named cost models. Strategies pull costs from backtest.costs

(etf_0.1pct, vix_etn_40, per_share_0.01). No magic numbers in strategy code.

• Registry-based strategies. Every strategy lives in

backtest/strategies/ and self-registers via @register("name"). Adding a new strategy = one decorator + one function.

• Agents without external LLM calls. The orchestrator prints

populated prompts and invokes the local backtest runner where applicable; it does not hit any external API.

• Obsidian vault stays intact. Only Python scripts are reorganised.

All .md strategy notes, Concepts/, and Reflections/ remain.

Strategy Notes

For the rationale behind each strategy and the backtest results, see Strategies/README.md and the per-strategy .md notes under Strategies/.