Detroit Permit Dataset: Sources, Coverage & Methodology

Source coverage

building — City ArcGIS source layer: 47,251 records.
trade — City ArcGIS source layer: 119,652 records.
demolition — City ArcGIS source layer: 16,264 records.
plan review — City ArcGIS source layer: 39,729 records.

222,896 source activities in total. Activity dates span January 2, 2019–September 24, 2026. Issued permits use the issued date; review activities use the task-status date, falling back to available filing dates. Source layers may be edited after this edition. The counts describe this snapshot, not all historical activity in Detroit.

Property grouping

Records are grouped using available exact parcel identifiers, address identifiers and conservatively normalized property addresses. Street suffixes and punctuation are normalized; unit distinctions are retained. Missing or ambiguous identifiers are not repaired by fuzzy matching. The output contains 89,856 deterministic property groups; 199 activity records remain ungrouped. 28,960 groups contain more than one source layer.

A property group is not a verified project, current parcel boundary or ownership history. Review and permit stages are retained rather than collapsed; related filings can otherwise be double-counted.

Activity and money fields

Activity categories use explicit source type or work-description rules. Uncertain descriptions remain Other / Unclassified. Trade labels come from trade permit types. Only positive building-layer contractor estimates populate the normalized estimate field. Department estimates, fees and actual spending are not interchangeable. No cross-permit project-cost sum is supplied.

Names and privacy boundaries

Owner, contact-business and demolition-contractor names are copied from their respective City filing fields where available. Owner-name coverage is 135,440 source rows; business/contractor-name coverage is 121,033 source rows before customer safety filtering. Names are not verified identities and do not transfer roles to unrelated filings. Personal owner names are retained in the purchased data, not ranked in the public sample.

The parties table records 256,419 name/activity relationships. BUSINESS, INDIVIDUAL and UNKNOWN are DatasetPress inferences—not City classifications. Business summaries include only business/contractor roles with an explicit company term, so they are a conservative subset, not a market-share ranking. Matching spelling does not establish legal identity.

Contact-person names and off-property contact/mailing-address columns are omitted. Detected phone, email and web-link patterns are redacted from free text; name fields containing obvious contact details are suppressed. Property addresses remain. No contact enrichment is added.

Files and spreadsheet safety

The Excel activity index provides focused navigation; the 15 CSV tables retain the full customer-facing fields. Import parcel IDs, permit IDs and ZIP codes as text. Formula-like CSV strings receive a leading apostrophe to prevent spreadsheet formula execution. Missing values remain blank rather than zero.

Quality checks and limitations

Retrieved layer totals are checked against snapshot manifests, record IDs are checked for uniqueness within layers, and property/date coverage is measured. Grouping and categorization remain rule-based approximations. Source errors, missing filings and late changes remain possible. Verify important conclusions against the linked City record.