Downloads & Data Dictionary

Everything this site shows is downloadable. Each table ships as Parquet (typed, compact) and CSV (opens anywhere), generated in the same export that feeds these pages — the files and the site always match.

The tables

CSVs larger than 25 MB ship gzip-compressed (.csv.gz) — most tools open them directly; otherwise unzip first. Parquet is the recommended format for analysis.

The data dictionary itself: Parquet ↓ · CSV ↓

Data dictionary

Every column of every table, generated from the same schema documentation the pipeline tests against.

Using this data responsibly

If you analyze or republish these files, carry these rules with you — they are part of the data's meaning:

• Scope: every number is bounded to the corpus above. Do not present results as regional or national benchmarks.
• One price type at a time: never rank gross_charge (list), discounted_cash (cash), and negotiated_dollar (negotiated) against each other.
• Respect the floor: rows or contexts marked insufficient_denominator have fewer than 3 hospitals and deliberately carry no market statistics. Recomputing statistics on them defeats the methodology.
• Scores ≠ compliance: usability scores describe published-file usability, not legal compliance or care quality.
• Prices ≠ bills: standard charges are not quotes and not out-of-pocket costs.

Citing this data

Hospital Price Lens, "Nashville metro" public data export, exported 2026-07-17 (build e15531a).

The build identifier pins your citation to the exact pipeline version that produced the numbers. Methodology for citation: /methodology.

What is deliberately not here

The public surface is the documented presentation tables above. Raw hospital files, intermediate pipeline layers, and the atomic observation-level fact table are not published in v1 — the aggregates above carry their documented comparison semantics with them, which is exactly what raw extracts would lose.