Data sources

Every table and context layer on this site, with its grain, vintage, refresh cadence, the date it was last pulled, and the licence that governs it. Counts are sites carrying the layer at build 2026-10-20 out of 1902. No subscription data is used; rating-agency and servicer findings arrive secondhand through news reports and are labelled as such.

Core tableswhat a record is made of

Context layersone JSON blob per site and layer; context.csv and the API carry them

Compared with CoStarwhat a subscription database has, what the public record gives, and the gap

CoStar, Trepp and their peers are the reference for brokers (opinion: because their records are verified by research staff and licensed feeds; not checked here). The Atlas works from the public record and says where each number came from. The comparison, as far as a public-record database can judge it:

Debt and transactions: the public trailhow the capital layers are built

1. News watcher: Bing News RSS queries every six hours (mall sold, for sale, special servicing, foreclosure, auction, office park conversion); a model extracts kind, price, acres, loan amount, parties and the supporting sentence into the deal table; rows link to the site when the name and place match.
2. Wikipedia text: ownership changes and sales in the article, extracted with the sentence; corporate mergers can look like sales.
3. SEC EDGAR full-text search (efts.sec.gov): one query per US mall, the quoted mall name plus its quoted city, over forms 10-K, 10-D, 8-K, 10-Q and ABS-EE (EDGAR full-text search covers 2001 to date). Filers are classified by their SEC industry code: 6189 (asset-backed securities) is a CMBS trust, 6798 and 65xx (REITs, real-estate operators and lessors) is a public owner, anything else (a retailer listing store locations, a bank) is “other” and stays out of the tables. Stored: filer, CIK, code, filing count in the sample (first 100 hits), first and last filing date, form mix, and the search link. A second query asks for the mall name immediately followed by its city, as a loan table prints them; filers found that way are marked “strong”. Limits: still a text match (a filing that names the mall and, elsewhere, the city, matches; a hotel or a tenant named after the mall matches too), a 100-hit sample per query so dates are “in sample”, no loan numbers yet. Excluded as common-name: malls returning over 200 hits, and records that share name and city with another record (a duplicate, not a usable query).
4. Next: ABS-EE Exhibit 102 XML carries loan-level balance, maturity, DSCR, occupancy and special-servicer transfer dates for every CMBS loan since 2016; county recorder deeds and mortgages for pre-CMBS and bank debt; REIT 10-K property tables for owned malls (GLA, occupancy, sales psf).
5. SEC filings, read directly (layers10.py): the property tables in mall REITs' annual reports (10-K Item 2; Simon, Macerich, CBL, PREIT, Washington Prime, GGP and Brookfield Property REIT, Glimcher, Rouse Properties, Mills, Tanger; several vintages each so sold and closed malls still appear) give GLA, year built or acquired, occupancy or leased share, in-line sales per sq ft, anchors and the owner's share; the CMBS trusts' monthly asset-level exhibit (Form ABS-EE, Exhibit 102 XML) gives the loan balance, maturity, rate, DSCR, appraised value, occupancy, largest tenants, payment status and any special-servicer transfer. Rows are matched to a site by name and state (exact name; or the name without its generic ending plus the same city; or that name alone when only one such site exists in the state), and every value is stored with the filing URL and the verbatim table row or XML fields (origin sec-reit / sec-absee, unreviewed, sec_facts.csv). Limits: only REITs with tables EDGAR serves as HTML (2002 on); ABS-EE exists only for trusts issued after Reg AB II took effect in late 2016, so most pre-2017 loans show a trust link without numbers; a one-property loan shows the whole balance, a multi-property loan's balance covers all its properties and is marked so.

See also methods and limits, the codebook and downloads, and capital trends.