NYC residential sales linked to ACRIS deeds and mortgages, 2016–2025

New York City publishes its property sales and its recorded deeds and mortgages as separate open datasets that do not join directly. Condominiums are the hardest case: the Department of Finance's parcel identifier for a unit sale is usually the building's billing lot, while ACRIS records the deed and mortgage under the unit's own lot. This dataset makes the join once, with fixed and documented rules. It covers all 729,047 one- to three-family house, condominium and co-operative sales recorded by the Department of Finance in 2016–2025. Linking is attempted for the 315,530 house and condominium sales with a positive price in the four boroughs ACRIS covers. A unique deed is found for 97.2% of those house sales and 74.0% of those condominium sales (78.9% in Manhattan, 44.7% in the Bronx); the other 38,232 are left unlinked rather than guessed. For houses and uniquely linked condominiums, the dataset records whether a mortgage was filed against the parcel in three windows around the sale, its amount and ACRIS document identifiers. A recorded mortgage near the sale date is not verified purchase finance, and financing is not measured for unlinked condominiums. Each linked deed carries flags for company buyers, same-surname transfers, shared mailing addresses and lender take-title events whose recorded amount can follow a legal rule rather than a price. No party names are included. The released rules reproduce the archived outputs of the companion papers: financing flags agree for every house and for all but 4 of 82,348 compared condominiums, and the mortgage-recording-tax statistics of The Shape of the Tax are reproduced at their published precision. This shows that the computations agree on overlapping sources, not that the links are accurate; the release reports coverage and agreement, not accuracy against verified records. Since its version 2.5, The Shape of the Tax takes its condominium loan amounts from this dataset, so that comparison is no longer independent. Files: nyc_linked_dataset_v1.0.zip holds the three linked tables, the codebook, the raw extracts with their retrieval manifests and standard-library Python scripts. It is unchanged from version 1.0. The sales and link tables can be rebuilt offline from it. The deed flags are a fixed snapshot: they were computed from ACRIS party records that are not deposited, and a new retrieval is a new computation. nyc_linked_dataset_v1.0.1_examples.zip unpacks examples/ into the same folder: a worked example of annual condominium linkage and recorded mortgages, linked-versus-unlinked condominium diagnostics and a test. README.md and CODEBOOK.md supersede the copies inside the version 1.0 archive. Version 1.0.1 adds the examples and corrects the documentation: not every eligible sale has a deed, a recorded mortgage is not verified purchase finance, the deed flags cannot be rebuilt offline, and agreement with earlier code is not accuracy. No data file changes. Companion papers: Who Gets the House? (doi:10.5281/zenodo.22421850), The Shape of the Tax (doi:10.5281/zenodo.22925301) and When a Deed Is Not a Market Sale (doi:10.5281/zenodo.22925383).