// datasets

Datasets, evaluated.

What it actually takes to use a data source: license, how you get it, what format it comes in, how often it changes, what can trip you up and where it was used.

7 used · 45 more
2026-10-23
legal

InfoLeg ↗

Argentina's national norms (laws, decrees, resolutions…) and which norm modifies which, from the Ministry of Justice

license
None declared in the dataset metadata; published on the national open-data portal
access
ZIP download from datos.gob.ar, with small sample CSVs; the CKAN API (package_show) tells you when a new dump is out
format
CSV (UTF-8): one file of norms (~430k rows), one of modification links (~376k)
refresh
Re-published periodically; check last_modified in the API
watch out
Complete only from May 1997; older norms appear only when a newer one references them. A norm with several issuing bodies repeats once per body, so deduplicate by id_norma. Links only record modifications, not "issued under the powers of". Full texts are not in the CSV, only links to InfoLeg pages, and InfoLeg blocks heavy scraping. datos.gob.ar answers 403 to the default User-Agent of most HTTP clients.
used in
InfoLeg graph → · ADR-003 → · infoleg-graph.geeklogbook.com ↗
2026-10-16
health

Tuva synthetic Medicare data ↗

The synthetic claims and clinical data the Tuva Project uses for its demo: ~167k medical claims and 2.3k eligibility rows

license
No separate data license; the Tuva demo project is Apache-2.0
access
Public S3 bucket (tuva-public-resources/versioned_tuva_synthetic_data/<version>/), loaded as dbt seeds by Tuva's load_seed macro
format
One CSV per table (medical_claim, pharmacy_claim, eligibility, lab_result…), already in Tuva's Input Layer
refresh
A folder per Tuva version; pin the one you load
watch out
Each table is loaded from its own version folder, so a project can end up mixing versions. The data is clean by construction: it proves the pipeline logic, not how you handle real payer data.
used in
What 167k synthetic claims taught me ↗ · Tuva on DuckDB ↗ · DQI dashboard ↗ · data quality app (Streamlit) ↗ · tuva-data-quality repo ↗
2026-10-23
finance

Yahoo Finance via yfinance ↗

Prices, OHLCV, dividends, splits and fundamentals for stocks, ETFs, indexes and FX

license
Yahoo's terms: personal use only. yfinance is Apache-2.0, unofficial and not affiliated with Yahoo
access
Python library, no API key
format
pandas DataFrames
refresh
Daily and intraday; 1-minute bars only go back about 30 days
watch out
It reads Yahoo's undocumented endpoints, so it breaks when Yahoo changes them, and long loops over tickers get rate-limited: log the ones that fail. Check whether prices come adjusted (auto_adjust). Delisted tickers are gone, which biases backtests. Grouping and resampling are on you.
used in
Grouping with yfinance ↗ · Sharpe ratio ↗ · Invesco vs Blackstone ↗
2026-10-23
finance

SEC EDGAR ↗

US company filings (10-K, 10-Q, 8-K…), their XBRL financial data, and the ticker-to-CIK map

license
Public domain (US government)
access
JSON APIs on data.sec.gov and bulk ZIPs, no API key
format
JSON, XBRL, HTML and text filings
refresh
Filings as they are accepted; bulk files nightly
watch out
Requests without a User-Agent that names you and gives an email get 403. Fair-access limit of 10 requests per second. Companies tag the same concept with different XBRL tags, so cross-company comparisons need mapping.
used in
Downloading data from the SEC with Python ↗
2026-10-23
crypto

Whale Alert ↗

Large on-chain transfers ("whale" moves) across 14 blockchains, as alerts

license
Commercial: API keys need a paid developer plan. A free archive of the alerts posted on X and Telegram
access
REST and WebSocket APIs; the free alert archive as a download
format
JSON; bulk exports as Parquet or NDJSON (top plan only)
refresh
Real time
watch out
The API keeps only 90 days of history; full history and bulk exports need the top Enterprise plan. Alerts start at $100k. The public /whales page that a 2023 scraper read now returns 404.
used in
What is Whale Alert ↗ · Scraping Whale Alerts ↗ · Airflow + MinIO pipeline ↗ · whale-alert repo ↗
2026-10-23
crypto

CoinGecko API ↗

Prices, market caps and volumes for thousands of tokens, aggregated across exchanges

license
Free Demo plan; paid plans for more calls, history and commercial use
access
REST API. Works without a key; a free Demo key gives a steadier rate limit
format
JSON
refresh
Near real time
watch out
Keyless calls are throttled hard and return 429 quickly. Coins are addressed by CoinGecko IDs (bitcoin), not tickers (BTC). Prices are aggregated, so they won't match any single exchange.
used in
Live crypto prices in Google Sheets ↗
2026-10-23
crypto

Binance market data ↗

Tickers, order books, trades and candles from the largest crypto exchange, spot and futures

license
Free public market data, under Binance's terms
access
Public REST and WebSocket endpoints, no key; history as daily and monthly dumps on data.binance.vision ↗
format
JSON; zipped CSV for the dumps
refresh
Real time; dumps daily and monthly
watch out
Blocked from US IPs (Binance.US is a separate exchange with its own API). Rate limits are by request weight, and heavy endpoints cost more. Prices are Binance's own, not a market average.
used in
Live crypto prices in Google Sheets ↗

Not used yet. Concrete datasets on file for the roadmaps.