TommyKwok/binance-top50-main

Binance Top 50 Backtesting Dataset

Institutional-grade, survivorship-bias-free crypto market data. Optimized for zero-copy research via DuckDB and the ducklake extension.

🚀 Key Features

• High-Fidelity Metadata: Standardized venues and instruments registries aligned with Databento (DBN) and Tardis.dev schemas.
• Top 50 Liquid Universe: Dynamically constructed based on trailing volume and market cap.
• Institutional Screening: Automatic exclusion of memes, leveraged tokens, and junk symbols.
• Point-in-Time Correctness: Includes daily snapshots in the gold layer to eliminate bias.
• Standardized Schemas: Aligned with Databento (DBN) and Tardis.dev conventions.

📦 Dataset Structure

This dataset follows a Natively-Managed Medallion Architecture:

• registry/: High-fidelity instrument and venue specifications (DBN and Tardis.dev aligned).
• bronze/: Raw ingested data from Binance Archive and REST API.
• silver/: Normalized, type-safe time series (natively partitioned by symbol, ts_date).
• gold/: Materialized daily stats for survivorship-bias-free reconstruction (natively partitioned by ts_date).

🛠️ How to Consume (Zero-Copy)

The recommended way to query this dataset is using DuckDB with the ducklake extension:

-- 1. Install and load the extension
INSTALL ducklake; LOAD ducklake;

-- 2. Attach the dataset directly from Hugging Face
ATTACH 'hf://datasets/TommyKwok/binance-top50-main/lake/' AS binance_lake (TYPE DUCKLAKE);

-- 3. Query the Silver layer (Optimized via Symbol/Date Pruning)
SELECT ts_event, open, close 
FROM binance_lake.silver.klines 
WHERE symbol = 'BTCUSDT' AND ts_date = '2024-05-20'
LIMIT 10;

📊 Manifest

This dataset was built with the following parameters:

• Lookback: 30 days
• Top N: 50 assets
• Trade Types: Spot, USD-M
• Data Types: klines, aggTrades, fundingRate

Built at: 2026-05-21T02:38:14Z