The fine print, in large type

Methodology

Everything on Elevator Radar traces to a public filing. This page says exactly where the data comes from, what we do to it, and how to get a correction made.

Sources and freshness

Permit data comes from official open-data portals published by each building department. Device registries come from state and city conveyance programs. Each source is fetched on its own cadence (daily for permit feeds, weekly to monthly for registries).

What counts as an elevator permit

New York and Chicago publish elevator permits as their own filing type, so inclusion is the agency’s own classification. Markets whose portals bury elevator work inside general building permits (Philadelphia, Austin, San Francisco) go through a conservative classifier: deterministic device-term rules first, a language-model assist on ambiguous rows, and a human review queue. Precision is audited on random samples per source; a leaderboard poisoned by HVAC permits is worse than a smaller correct one.

Job-level dedupe

Filing systems record initial and subsequent filings against one job. Leaderboards count distinct jobs, never raw filings, so a project with five paperwork amendments still counts once. Device counts are parsed from the work description when the filing states them ("(2) passenger elevators"); otherwise a filing counts as one device and is marked assumed, and assumed counts never set records.

Contractor identity

City systems spell the same company a dozen ways. Every observed spelling is bound to one canonical company through, in order: a human-maintained override file, exact alias history, license numbers where the filing carries them, deterministic name normalization, and finally a conservative fuzzy match whose uncertain cases go to human review instead of the leaderboard. Companies are only credited, never penalized: an unmatched filing simply stays off the boards until a human resolves it.

The Count

The national elevator count is a separate product with stricter rules: hard counts from state and city device registries, cited agency totals where a registry is not public, and a disclosed statistical range for the states that have not released data. The live counter never interpolates; every movement is a row in the public event log.

How we model the national total

The counter is the verified floor. The national estimate is a separate, deliberately conservative model. For every jurisdiction we have hard-counted, we know the population it actually covers (state population minus the home-rule metros we exclude; New York is its five boroughs). We measure elevators-per-person across that counted population, roughly 3.4 per thousand today, and apply it to the US population. It is coverage-weighted, not a flat per-capita, so it self-corrects as we add states, and it is reported as a band rather than a false-precision single number. The next refinement joins ZIP-level Census building-stock data so dense and sparse areas are weighted directly rather than blended. The “one million” figure everyone repeats traces to a single estimate made around 2007; this is a bottom-up derivation from live public records.

Corrections

If your company is misattributed, split, merged wrongly, or you want a filing reviewed, corrections are made publicly: the change lands in the data, and material changes appear in the event log. Nothing is edited silently. Request a correction.

What this is not

• Not a license verification service; always verify licensure with the authority having jurisdiction.
• Not a complete picture of any company: permits are one lens, and some work never requires one.
• Not affiliated with any building department or agency.