Free tick-level crypto trade data samples are now available across Kaggle, HuggingFace, and Zenodo — audited, no gaps hidden, Apache Parquet format.
What's in the samples
Each sample covers a 1-week window of raw trade data for one asset (BTC, ETH, XRP, SOL, LINK, LTC, ADA, BNB, DOGE, AVAX). Every row is a single executed trade — price, quantity, timestamp, and aggressor side. No OHLCV aggregation, no smoothing, no gap-filling.
The samples come from a fully audited multi-year historical dataset (2020-2026) built and maintained by The Glitch List, a data engineering lab based in Barcelona.
Quick start with pandas
import pandas as pd
df = pd.read_parquet("sample_BTC_2025-02-19_to_2025-02-26.parquet")
print(df.shape)
print(df.head())
# Basic trade imbalance by minute
df['minute'] = pd.to_datetime(df['exchange_ts'], unit='ms').dt.floor('min')
imbalance = df.groupby('minute')['quantity'].sum()
Where to find the samples
- Kaggle: search "jalvartstudio" — 8 public datasets, one per asset (LINK/LTC/ADA bundled as a pack)
- HuggingFace: huggingface.co/jalvart — native Parquet, interactive Dataset Viewer built in, no download needed to preview
- Zenodo: DOI-backed permanent records, citable in academic work
Data quality notes
Every sample ships with a Data Quality Disclosure documenting known gaps and any timestamp normalization applied. Source raw data switched timestamp units from milliseconds to microseconds starting 2025-01-01 — this is documented and corrected in the pipeline, not silently patched.
Full historical datasets
The free samples are 1-week excerpts. Full datasets span multiple years (up to 6.5 years for BTC, 6.2 billion rows), fully audited row by row.
🎁 Launch offer — 20% off, limited to 50 uses: GLITCH20 at theglitchlist.com?utm_source=devto&utm_medium=referral&utm_campaign=free_sample
Feedback and questions welcome — this is an ongoing project and disclosure of any data quality issues found is part of how we keep the catalog honest.
Top comments (0)