Databook
activeopen governmentAn award-winning transparency tool that combines over 50 official NYC open datasets — agencies, people, jobs, capital projects, schools, districts and procurement — into one searchable picture of how city government works.
Fixed a bug in the NYCHA vendor crosswalk generator by clearing stale, machine-derived rows before each run. This ensures the database and parquet files accurately reflect the latest matching logic instead of retaining outdated fuzzy matches.
This update fixes false-positive vendor matches in the NYCHA fuzzy-link tier. It removes the token-set shortcut, requires matched firms to share at least one non-generic token, and raises the auto-link threshold from 0.94 to 0.96. These changes eliminate known false positives and route borderline matches to human review.
Updated the text normalization function to strip apostrophes instead of replacing them with spaces. This ensures names like "ADAM'S" normalize to "ADAMS" rather than "ADAM S", allowing exact matching to succeed rather than relying on fuzzy matching.
Fixed a bug where NYCHA vendor searches returned empty results due to a DuckDB reserved word conflict with the alias "current". The alias was renamed to "cur", and the search query's exception handler was updated to log warnings instead of failing silently.
This update enhances the NYCHA vendor crosswalk script by introducing confidence tiers for matching. It adds manual curation seeds, exact matching, and token-blocked fuzzy matching with a strict 0.94 score threshold to safely auto-link minor name variations while holding lower-confidence matches for human review.
This pull request integrates NYCHA vendors into the global search and navbar typeahead, linking matched vendors to their City profiles and unmatched ones to NYCHA-native profiles. It also adds a discoverability alert on the City vendors index page that links users to the NYCHA vendor directory.