Y Combinator company status and Form D filings

• Status changes: 10 November 2010 to 12 September 2026, added daily.
• Form D filings: 17 March 2009 to 8 September 2026.
• Grain: one row per company, day, event kind, key.
• Geography: worldwide. Form D is a US filing, so a company incorporated elsewhere appears in the status rows and not the filing rows.
• Licence: CC BY 4.0. Form D filings are US federal records in the public domain. The directory and the archived lists state no licence, so their recorded basis is unknown.
• Gap: 20 August 2019 to 29 November 2022. The community list froze and the directory rendered in the browser, so no archived page holds a status then.

Three eras were reconciled onto one key. A community list read through its web archive covers 2010 to 2019. The directory read through its archive covers 2022 to 2024. The mirror's git history covers 2024 onward. Every reading carries the day it was observed and which era said so.

• company_key is the name in lowercase letters and digits, corporate suffix removed. Every era and every filing match on it.
• batch is one spelling of one batch. Ninety-five spellings name fifty-one batches across the eras, and the short form is expanded. batch_raw keeps what the source wrote.
• status and previous_status give the status that day and the one before. Active, Acquired, Public, Inactive and Missing are Y Combinator's words. Exited is the community list's own and records an exit without saying which kind.
• The filing_ columns give the filing as EDGAR lists it, every exemption claimed, the raise, the issuer's own revenue and industry answers, and its registration.

Every filing column is empty on a status reading and every status column is empty on a filing.

The raise and the issuer's registration come from the quarterly data sets, which begin at the third quarter of 2024. 202 of the 1,827 filings carry them. Of those 202, 179 issuers declined to state a revenue range, which is itself the answer the form records, and 76 gave no year of incorporation.

A status reading before December 2022 carries no company_id, slug or team_size. The archived pages that hold it never stated them.

• Training a time to event model for acquisition, listing or shutdown. Every company has a dated entry and either a dated outcome or none, so one still Active is right censored rather than missing, and the 2019 to 2022 gap is interval censoring.
• Building features that cannot leak. Each row states what a source said on one day, so a feature cut at a date uses only what was knowable then.
• Learning whether a raise predicts an outcome. A Form D is the first public trace of a private round, so the filing rows are a dated treatment and the status rows are the response.
• Scoring a language model on startup outcomes. Every claim is dated and attributed, so an answer can be checked against what was true on the day it describes.
• Benchmarking entity resolution. One company is named by three archives, by the directory and by its own SEC filings, and company_key is one published answer to reconcile them.