version	0.9.31

AutoQuant V2

AutoQuant turns quantitative research into a versioned, testable, Agent-operable engineering workflow.

It is an AI-native quantitative workbench, not only a backtest library, strategy generator, or integration backend. A coding Agent can enter the filesystem, understand the current question and evidence, take one bounded action, evaluate through fixed contracts, resume after interruption, and leave durable work for another Agent or human reviewer.

long-lived Workspace
└── Project
    ├── research question and content-locked inputs
    ├── fixed Studies and bounded Research Sessions
    ├── factors, portfolios, ML/RL policies, and simulations
    └── immutable Runs, Reports, Reviews, and Dossiers

One Workspace may hold many self-contained Projects. A Project is one evolving body of research; a Study locks one evaluation question; a Research Session is a bounded editable investigation; a Run is an immutable measurement.

Current release: 0.9.31

Current status states what this checkout can honestly do. Release history indexes bounded outcomes, and versioning and release owns version increments, checkout behavior, audits, tags, compatibility, and host pins.

Standalone or an OpenAlice desk

AutoQuant has one product shape in both environments:

standalone clone                    OpenAlice Trading Harness
└── AutoQuant Workspace             └── AutoQuant Workspace desk
    └── Quant Agent                     └── Quant coworker
        └── Projects                        └── Projects

Standalone, a human or Agent operates the workbench directly. Inside OpenAlice, the unchanged repository becomes a specialized Workspace desk that can receive a quantitative assignment from another coworker and return an evidence-bound report. AutoQuant owns research and historical simulation; brokers, authenticated accounts, approvals, and live order submission remain outside it.

Quick start

AutoQuant requires Python 3.11 and uv.

git clone git@github.com:TraderAlice/Auto-Quant-V2.git
cd Auto-Quant-V2
uv sync

uv run aq --version
uv run aq project list .
uv run aq validate .
uv run aq orient . --json
uv run aq studio serve .

The repository root is already a Workspace. Its checked-in projects/sample-research-desk is a deterministic teaching Project with inspectable historical evidence, not a real assignment.

Start genuinely new work as a sibling Project:

uv run aq project templates --json
uv run aq project create . research-desk \
  --name "Research Desk" \
  --description "Bound and investigate the delegated question" \
  --template blank \
  --json
uv run aq orient . --project research-desk --json

Before quantitative work, the Agent rewrites the new Project's research.md in English and asks about every caller-owned ambiguity that could materially change the answer. Template choice, data acquisition, strict intake, governed iteration, evidence publication, and Studio review are covered by the operator guide.

What lives where

• Workspace: Project discovery and the standardized AutoQuant Harness.
• Project: the question, source authority, strategies/models, Studies, Runs, Sessions, and durable deliverables for one evolving body of research.
• Factor: causal predictive evidence over a caller-bound outcome.
• Portfolio: target-weight construction, risk, cost, and implementation evidence from qualified return research.
• Governed RL: bounded adaptive value beyond fixed factor sleeves; never a shortcut around the same portfolio and evidence contracts.
• Fixed research routes: descriptive Book Risk, price Event, path Stress, and strategic Allocation questions that do not pretend to need editable models.

Ordinary factor code receives one long-form pandas panel rather than a proprietary DSL. CSV works in the base environment; Parquet and Feather use uv sync --extra columnar.

Documentation map

Need	Canonical document
Operate a Workspace end to end	Operator guide
See current tested capability and limits	Status
Discover commands and JSON contracts	CLI
Understand Workspace and Project files	Project format
Understand Core ownership and boundaries	Architecture
Change versions or publish a release	Versioning and release
Scan published outcomes	Changelog
Contribute framework changes	AGENTS.md and PLANS.md

Repository structure

Auto-Quant-V2/
├── autoquant-workspace.json  # checked-in root Workspace
├── autoquant/                # Core, CLI, templates, Skills, and Studio
├── projects/                 # sample plus ordinary research Projects
├── docs/                     # operator references and design invariants
├── plans/                    # bounded engineering execution records
├── scripts/                  # repository checks
├── tests/                    # deterministic bounded verification
├── AGENTS.md                 # contributor and Agent routing guide
└── PLANS.md                  # active/completed plan index

Auto-Quant Classic and its flat Freqtrade arena are retired from the current tree; Git history remains their archive.

Development

Read AGENTS.md, PLANS.md, and the active plan before non-trivial changes. Do not launch an unbounded autonomous loop or use a long multi-year backtest as routine validation.

uv run python scripts/check_doc_links.py
uv run python -m unittest discover -s tests -v
uv build

License

MIT.