The scorecard history, as tables

Every dated check of every feed record this site covers, and every finding in each of them, as two Parquet tables with a data dictionary, a provenance file, and a license, rebuilt monthly. Licensed to one organization. The dated scorecards it is built from stay free, at their own addresses, for everyone.

The plan and its price are listed below.

See the plan and price

One-time purchase. The download link arrives on the page you land on after paying, and by email, and stays valid for 30 days. What stays free and the license are below.

GTFS Scorecard grades a transit agency's published GTFS feed for data quality and rider-information gaps, free, for any agency, at gtfsscorecard.org, and keeps every dated check it has ever published at its own address. What it did not publish, until now, is the whole corpus as one table over time: more than a hundred dated files per feed record, and no file anywhere that lists findings as rows. These tables are that: the build step is the open-source scorecard history-export command, and what is sold is the built artifact and its monthly refresh, never the facts and never the code.

What arrives

checks.parquet: one row per feed record per dated check. The grade, the overall and category scores, how many categories were measured, the expiry fields, the methodology and feed-byte identity (rubric, scoring profile, validator, reader profile, feed hash), the fetch provenance, and the registry's license record for the feed.
findings.parquet: one row per finding per check. The notice code, severity, instance count, points, its rank among the top fixes, and the reach numbers from the consequence block where the artifact has one.
DATA-DICTIONARY.md: every column, its type, and what a null means. A null is never zero and never a permissive license.
PROVENANCE.json: what was read, the snapshot date range, the artifact schema versions seen, the row counts, each table's SHA-256, and how many feed records were left out at their publisher's request.
LICENSE.md: the terms the tables are sold under, the same words as the license section on this page.

The five files come as one zip. A sweep row, where only the calendar dates were re-read that day, says so in recompute_kind, so a carried-forward score cannot be mistaken for a fresh one. Grades are comparable only within one scoring profile, validator version, rubric version, reader profile, and set of measured categories, the same rule the public site applies.

The plan and its price

What stays free

Everything that is free today. Every dated scorecard (/data/artifacts/<id>/<date>.json), index.json and its compact history, the catalog and the flat dataset, every api/v1/ file including agencies.parquet, the monthly dataset release, the query page, and the history-export command itself, which is Apache-2.0 like the rest of the pipeline. Anyone who would rather build the tables from the free files can. Nothing is subtracted from the free site to make this; it is an addition for the people who need the whole corpus as a table, and it is never the way an agency sees its own grade.

See the free open data

How it works

Pay through Stripe. GTFS Scorecard never sees your card.
Stripe sends you to a short page that confirms the payment and shows your download link. The same link is emailed to the address you paid with.
The link opens the newest monthly build and stays valid for 30 days; each click opens a fresh short-lived download, so the link itself can be kept.
If the link does not work, or anything about the order is not right, reply to the receipt Stripe emailed you, or open an issue on GitHub, and it will be reissued or refunded.

The build runs once a month in the open-source repository's own automation and is stored privately behind a link only your purchase receives. If no build exists yet when you pay, nothing is delivered and your checkout is not used; you will be told so on the page, and the tables will be sent by hand as soon as the build exists.

What a purchase does not buy

No influence over the scores. Grades, methodology, weights, and which agencies are listed are never for sale, and the tables carry the same numbers as the public site. A publisher who does not want its feed's measurements in the tables that are sold can ask to be left out, and is, the way the listing policy honors a removal request; the free site is unchanged by that request.

The license, in short

The tables are a packaged work licensed to the purchasing organization, not to the public. The full text ships in the zip as LICENSE.md and is kept in docs/history-tables.md. What it says:

One organization. Use the tables inside your own organization for any internal purpose, including work you are paid for.
No redistribution of the package. Do not publish, sell, sublicense, or share the tables outside your organization. Figures, charts, and findings drawn from them may be published freely.
The facts stay free. Every number is derived from a dated scorecard published free under CC BY 4.0, and the export command is open source. Nothing in the license restricts those free sources.
Share-alike rows. A row whose license_notice names a share-alike license is licensed to you under that license, not under the clauses above, and may be redistributed on its terms.
Publisher terms travel with the rows. Each feed's own terms, recorded in the license columns, continue to apply to what the row says about that feed.
Attribution. A published figure credits GTFS Scorecard (gtfsscorecard.org), scored on top of the MobilityData gtfs-validator, and any publisher credit the row carries.
No warranty. The grade is a data-quality signal, not a compliance determination, and the tables are provided as they are.

Who built this

Chelsea Kelly-Reif, the same person who built and maintains the free scorecard, starting from the transit systems in Davis, California. Questions before you buy: open an issue on GitHub. More about the project.