kostemeloo413-ux/v7-investment-workbench

What it does

V7 is a research and paper-trading workspace for China A-shares. It brings market data, news/event analysis, strategy rules, quant research and simulated accounts into one traceable workflow. The design deliberately splits responsibilities: an LLM is used for semantic handling of news and events, while deterministic Python code owns data admission, trading constraints, order matching, fees and ledger reconciliation project README.

The system is organized around several cooperating modules: market and event collection with source and data timestamps; strategy contracts that define candidates, triggers, position sizing, exits and required evidence; a quant research chain with labels, features, ranking, calibration and planning; data admission that distinguishes usable, stale and missing data; simulated execution under trading-session, T+1, board-lot, quote and fee constraints; and ledger/replay reports system composition. The repository also describes research views, strategy signals, execution eligibility, orders and fills as separate business objects, and notes that the public build includes a Blogger Lab experiment module that is disabled by default system overview.

It is intended for people who want to study and rehearse an A-share research process with explicit data states and auditable paper results, not for connecting to live brokerage order entry.

Getting started

The verified environment is Windows with 64-bit Python 3.13. From the repository root, create a virtual environment, install the public preview dependencies, and start the public preview quick start:

py -3.13 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
.\.venv\Scripts\python.exe -m scripts.public_preview

If the Python Launcher is unavailable, substitute the full path to an installed Python 3.13 interpreter for py -3.13. The server listens on port 8020 by default; add --port 8021 if that port is occupied README lines 33–37. Once it is running, three local entry points are available:

• /demo shows the workbench with clearly labeled synthetic data and does not call a quote source or LLM.
• /replay runs historical matching, T+1 restrictions, fees and reconciliation against three historical prices and preset hypothetical orders.
• /v7 shows the workbench with missing-data and unavailable states preserved rather than filled in.

After starting the server, open the local demo or replay pages in a browser. You can also verify the replay from the command line README lines 51–54:

.\.venv\Scripts\python.exe -m scripts.public_preview --check
.\.venv\Scripts\python.exe -m scripts.public_demo --json

On macOS or Linux, python3.13 and .venv/bin/python can be used for the equivalent commands, but the README states that equal platform validation has not been provided quick start.

Data and execution constraints

The public preview binds to the local loopback address by default, accepts only GET/HEAD requests, and screens private configuration files and inherited LLM/Webhook credentials before startup. The synthetic demo and historical replay use separate interfaces; neither initializes a formal competition season or accumulated forward days quick start. The replay core uses the Python standard library, while web dependencies come from requirements.txt README lines 51–54.

Live data collection, LLM connections and a formal simulated season require separate configuration and runtime validation. The repository notes that the active strategies are disabled and that the Blogger Lab needs an explicit data root and local loopback access when enabled system composition. The workbench runs with Python, FastAPI, SQLite and plain HTML/CSS/JavaScript; quant research uses NumPy, pandas, LightGBM, scikit-learn and joblib project structure.

The default test collection covers public entry points, isolated historical replay, read-only boundaries, data adaptation and trading recovery. Some historical tests still depend on old seasons or undistributed materials, so the default suite is not the full historical test set validation. More detail is available in the known issues and the validation notes.

The project is for research and simulation only and does not connect to live order placement. News citation checks do not make inferences correct, and model scores or self-reported confidence are not validated profit probabilities. The public release states that it has no qualified forward-validated profit conclusion. It does not distribute production databases, account ledgers, holdings, runtime state, credentials, raw audio/transcripts or full news corpora, and the role names are historical interface labels that do not imply participation or endorsement by real people. The repository does not grant a unified redistribution license for the whole directory; code, public model resources and third-party dependencies need separate license review usage scope and license.

HotRepo's take

HotRepo first recorded this repository on the 2026-11-02 new-project list for Python, with 106 stars and 3 forks at that snapshot and 6 stars gained during the measured window that began at 11:45 UTC [HotRepo data]. The collected star curve over that afternoon rose from 100 to 106 across three hourly observations. That is a short single-day record, so it supports only the observation of early attention, not a trend judgment. The more substantive signal for readers is the project's own separation of public preview, historical replay and optional live stack: the reproducible part is the local demo and matching replay, while live data and LLM connections are explicitly left to separate configuration and validation.