hfdatalibrary.comRoast Review

The hero opens with HF Data Library: High-Frequency U.S. Equity Data (1-Minute OHLCV), then stacks 1,391 tickers, 1,551,364,273 one-minute bars, 23+ years of data, and 25 academic variables into four large counters. The subhead adds “Free, research-grade, high-frequency intraday data” and promises exact ARR figure[redacted] that it is documented, version-controlled, and updated daily. It is a strong, specific proposition, but the page presents a dataset like a SaaS launch where every statistic needs to arrive above the fold wearing a little medal.

The section called “What is this?” explains the coverage, two cleaning versions, pre-computed variables, methodology, and the claim that this is a free, citable alternative to TAQ and CRSP. Then the page offers Raw and Clean cards, a nine-step cleaning pipeline, preserved gaps, outlier filtering, and a note warning that LOCF introduces biases. This is the rare feature grid that brings footnotes, which makes the roast less about vapor and more about watching academic caution get laid out in the same card rhythm as a productivity app.

Under “Multiple ways to access the data”, the library offers browser downloads, a REST API, and AI & MCP access for Claude, Cursor, Gemini, and ChatGPT. The comparison table then lines up HF Data Library against Polygon.io and other sources across academic variables, data-quality scores, REST API access, MCP, DOI, license, update cadence, and real-time availability. It is all useful, but the page still manages to make downloading a Parquet file sound like a platform ecosystem, complete with the assertion that this is the only source with its own free MCP server.

The closing stretch answers every conceivable question, repeats “free” and “no paywall” until the reader understands the bargain, and introduces One account. Every library: HF Data Library, Econ Data Library, and IP Data Library. With the University of Central Arkansas affiliation, Zenodo DOI, CC BY 4.0 license, known-issues page, citation instructions, and actual sample code, the trust signals are unusually concrete; the only thing missing is a button asking the data to schedule a demo. The documentation does the convincing, while the oversized stats, comparison table, FAQ, and AI-access banner insist on giving the conversion funnel a dissertation defense.