• Status: 10 November 2010 onward, so YC's 2005 to 2010 batches have no earlier readings.
• Status gap: 20 August 2019 to 29 November 2022, which no archived page covers.
• Form D filings: March 2009 onward, when Form D went electronic; earlier paper filings are absent. The amount raised and issuer details cover only filings from mid-2024 to mid-2026.
• Geography: worldwide. Form D is a US filing, so a company incorporated elsewhere appears only in status rows.
• Grain: one row per company, day, event kind and key.
• Licence: CC BY 4.0. Form D filings are US public domain. The directory and archived lists state no licence.

An archived community list covers 2010 to 2019 and YC's archived directory 2022 to 2024. From August 2024, a public mirror's git history and then YC's live directory take over; status_source names the source of each reading. The live directory states only the present, so the most recent changes are dated by the day they were first seen.

Name matching alone goes wrong in three ways:

• A different business with the same name. FUTURA is a Winter 2020 dentistry startup in Mexico City. Futura LLC, a Texas company, filed sixty Form D notices between 2011 and 2016.
• A fund named after the company. Filers such as "Stripe May 2023 a Series of Stonks SPVs LLC" buy shares in Stripe; they are not Stripe raising money.
• Two startups with one name. More than ninety names belong to more than one YC company, such as the Parse of Summer 2011, bought by Facebook, and the Parse of Fall 2025.

EDGAR's search returns the Form D notices whose issuer name contains each company's name. TypeSafe's Jev, a model built for yes-or-no decisions, reads the startup (name, batch, city, one-line description) beside the filer (name, business address, state of incorporation) and answers: is this filer the startup itself, and is it a fund or SPV holding its shares? A filing is kept only when Jev says it is the startup and not a fund, the names match, and it was filed no more than three years before the batch. filing_match_score is Jev's probability, so a reader can apply a stricter threshold.

• company_uid: one identity per company, so two that share a name never share a history.
• company_key: the name in lowercase letters and digits, corporate suffix removed.
• batch: one spelling per batch; batch_raw keeps the source's.
• status and previous_status: the status that day and on the company's previous reading. Exited is the community list's word for an exit of unstated kind.
• filing_ columns: the filing as EDGAR lists it, with exemptions, issuer name, match score and, where available, the raise and the issuer's revenue and industry answers.

An archived reading carries a company_id only when its name and batch match one directory company, and before December 2022 no slug or team_size. EDGAR returns at most 100 notices per company, so a widely shared name may miss older filings.

• Time to event models for acquisition, listing or shutdown, with right and interval censoring.
• Leak-free features, since each row is what a source said that day.
• Testing whether a raise predicts an outcome.
• Scoring language models on dated, attributed startup outcomes.
• Benchmarking entity resolution, with a model score on every filing match.