Project

Quant Research and Paper-Trading Workbench

A personal extension of open-source InStock that connects market data, evidence, deterministic risk, paper intents, and review.

Python · Tornado · MySQL · MLflow · State Machine · Paper Trading

2026-present

Product Snapshot

Role: Personal system design and extension on an open-source base
Users: A single active trader who needs traceable research and paper execution, not autonomous recommendations or broker trading.
Stage: Local MVP in development; the current boundary ends at approved PAPER_PENDING, with fill, scan, and complete review still closed.
Focus: Data health, evidence snapshots, decision state, deterministic risk, paper trading, and review loops.
Validation: Added agent, quant, review, watchlist, strategy-lab, web, and formal test modules on top of the upstream system.
Public Proof: The local working tree modifies 25 tracked upstream files and adds dedicated services, pages, design documents, and tests; upstream code is not claimed as original.

Next Step

Portfolio PDF About the Founder Email

Open-source attribution

This is not a trading platform written from zero. It extends open-source InStock in a local branch. I claim the added system design and implementation, not the upstream history.

Product boundary

The target is a single-user decision workbench:

proposal → draft → frozen evidence → completed plan → deterministic risk → user approval → paper record → review

An LLM may organize evidence, produce counterarguments, and draft a review. It cannot approve a trade, calculate final risk, or send an order to a broker.

Local extension

The working tree changes 25 tracked upstream files and adds agent, quant, ML, review, watchlist, strategy research, backtest, web-handler, page, and test modules.

The Agent MVP currently implements DRAFT, Evidence, derived-plan confirmation, Risk, read-only Paper Intent, and explicit approval. It stops at PAPER_PENDING; transactional fills, proactive scan, and complete review remain closed.

Safety and evidence

Deterministic code checks freshness, position size, duplicate intent, plan versions, and state transitions. Formal plans and evidence snapshots use one source of truth with audit events.

This page exposes no account, holding, symbol, return, or local database information and makes no future-performance claim.

Current judgment

The most useful contribution is not another model. It is a contract that prevents the path from idea to evidence, risk, action, and review from skipping steps.