Transparency & accuracy report

Did the parsing work?

Across 3,480 hospitals we've attempted, 2,024 (58%) produced a usable price file. The rest failed — and the breakdown matters, because most failures are the hospital's file, not our parser:

• 672 — file unreachable: the hospital's link is dead (404), blocked (403), timed out, or erroring. Nothing we can parse.
• 735 — file unusable: an unrecognized layout, a ZIP with no usable CSV, or a file too large to process. These are non-standard publications, not parse mistakes.
• 47 — genuine parse errors: we fetched a recognized file but our parser failed on it. This is the bucket we own, and it's 2% of the files we could actually read.
• 2 — other / uncategorized.

Put another way: of the files we could fetch in a standard format, our parser succeeded on 98%. When a price is missing, it's almost always because the hospital hasn't published a usable file — not a parsing mistake — and you can check that yourself, because every hospital page links to its exact source file. (This attempt log covers crawls since June 2026; the coverage totals above count every hospital we currently serve, including files parsed earlier.)

How good is the data we do have?

Each parsed file gets a 0–100 quality score (completeness of codes, payers, and price types — see our methodology). Across 2,895 hospitals with a parsed file:

• Median quality score: 89/100.
• 78% score 80 or higher.
• 2,361 meet our bar to power price comparisons; files below it are kept but not surfaced as money pages.

Quality is also shown per hospital (the score on each hospital page) and per row — a "1 plan" tag flags a figure backed by a single payer, and a "shared rate" tag flags a price the hospital applies to several procedures at once (a billing tier), so you can weigh each number, not just trust the file-level score.

How fresh is it?

We re-ingest hospital files on a weekly schedule, so the data tracks what hospitals currently publish. Of the latest file we hold per hospital, 100% were ingested in the last 90 days (oldest on record: 2026-06-01). Each hospital page shows the ingestion date and links to the source file, so you can always confirm against the original.

Corrections

Every page invites a correction, and corrections go into a review-then-ingest queue rather than changing the site silently. No corrections have been submitted yet — if you spot a number that looks wrong, you'd be the first to flag it. Submit a correction →

Want the raw data?

Everything here is built on public records, and we keep it traceable. For the underlying files and bulk snapshots, see data sources & export; for how we process them, see our methodology.