Trady

Research pipeline for finding unusual Polymarket wallet activity and testing whether those signals would have been useful historically.

It ingests public market and trade data, builds wallet profiles, groups related wallets, produces heuristic signals, and can train an XGBoost score from the resulting features. A high score means "worth investigating". It does not prove that a wallet belongs to an insider.

Current status

MVP, backtest-only. Data ingestion, wallet analysis, clustering, rule-based signals, and ML scoring are implemented as local pipelines. The repository does not place orders, manage a live portfolio, or run a production trading service.

The backtest engine and terminal dashboard described in action_plan/ are design specs, not completed modules. Treat any result as research output until it has been tested on clean out-of-sample data and reviewed for leakage.

Pipeline

Polymarket Gamma / Data API / Goldsky
                 |
                 v
      Parquet market + trade data
                 |
                 v
 freshness, behaviour, funding and wallet clustering
                 |
                 v
 rule-based trade signals and position-size suggestions
                 |
                 v
      optional XGBoost scoring

The default working directory is data/processed/. Intermediate datasets are kept as Parquet files so each stage can be inspected independently.

Run

Requires Python 3.10 or newer.

git clone https://github.com/vedanth-jadhav/trady.git
cd trady
python -m venv .venv
source .venv/bin/activate
pip install -e .

Open the interactive CLI:

trady

Or run each stage directly:

trady-data --help
trady-analysis --help
trady-signals --help
trady-ml --help

A typical research run is:

trady-data run-pipeline
trady-analysis run-analysis
trady-signals run-pipeline
trady-ml train
trady-ml score
trady-ml suspects

The commands expose flags for market filters, date ranges, input paths, and output paths. Use each command's --help before a larger data pull.

Verify

pip install -e '.[dev]'
pytest

What is not built

• live order execution or exchange authentication
• a live portfolio, risk engine, or position reconciliation
• the backtest engine specified in action_plan/05_backtest_engine/
• the terminal dashboard specified in action_plan/06_terminal_dashboard/
• deployment, scheduling, monitoring, or alerting
• proof that a flagged wallet has non-public information

Repository map

• src/data/ - Polymarket clients, ingestion, validation, and Parquet storage
• src/analysis/ - wallet freshness, behaviour, funding, profiling, and clustering
• src/signals/ - signal detection, aggregation, filters, market context, and sizing
• src/ml/ - features, labels, XGBoost model, overrides, and scoring CLI
• tests/ - data, signal, and ML tests
• action_plan/ - project design notes, including modules that are not implemented yet

This is experimental trading research, not financial advice.