On time to the gate, not to the sky: at O'Hare in May, a third of punctual departures still taxied 30-plus minutes before takeoff

The U.S. Department of Transportation's on-time database for May 2026 — the most recent month posted (file in the "1987–present" reporting-carrier table) — records 611,735 scheduled domestic flights operated by 13 reporting carriers across 347 origin airports, of which 5,655 were cancelled (0.924%) and 1,734 diverted (0.283%). DOT measures a flight's departure punctuality at the gate: "departure delay" is the actual gate-pushback time minus the scheduled gate time. What happens next — the taxi from gate to runway and the wait in the takeoff queue — is recorded separately, as "taxi-out" time, and does not enter the on-time departure figure. In May, the national mean taxi-out was 18.46 minutes and the median 16.0. Of the 368,670 flights that pushed back from the gate on time or early (departure delay ≤ 0), 33,147 — 8.99% — went on to record a taxi-out of 30 minutes or more before wheels-up. Applying DOT's own "not delayed" cutoff instead (off the gate fewer than 15 minutes late), 471,659 flights qualify and 9.18% taxied 30 minutes or more. The concentration is by airport. Among origins with 3,000-plus operated departures, the share of on-time-or-early gate departures that then taxied 30 minutes or more was: Chicago O'Hare (ORD) 33.3%, with a median taxi-out of 25 minutes for those on-time departures; New York JFK 31.0%; New York LaGuardia (LGA) 23.5%; San Francisco (SFO) 23.4%. At the other end: St. Louis (STL) 1.46%, median taxi-out 11 minutes; Houston Hobby (HOU) 1.19%; Chicago Midway (MDW) 2.03%. Atlanta (ATL), the third-busiest origin in the file, ran 4.49%, median 15.

The following is analysis, not fact. The number a traveler is sold — "on-time departure" — is a gate event. It certifies that the aircraft door closed and the plane left the stand on schedule. It says nothing about whether the plane then joined a 20-deep conga line for the runway. In May 2026, at O'Hare, a flight that pushed back on time or early was, one time in three, still on the ground 30 or more minutes later. Its passengers experienced a punctual departure and a half-hour of tarmac. The gap between airports is not a gap in volume. O'Hare was the single busiest origin in the file, with 33,393 operated departures; Atlanta and Denver ran close behind, at 27,210 and 27,328 — yet an on-time O'Hare pushback was more than seven times as likely to eat 30-plus tarmac minutes as an on-time Atlanta one (33.3% vs 4.49%), and 23 times as likely as an on-time St. Louis one (1.46%). The median on-time O'Hare departure spent 25 minutes taxiing; the median on-time St. Louis departure, 11 — a 14-minute difference between two flights that both, on the government's books, "left on time". What separates them is runway geometry and traffic density at the surface, not the punctuality of the crews. May matters to the reading: it is a low-cancellation, low-de-icing month (0.924% cancelled, against the 1.55% of the prior December this desk examined), so the tarmac time here is overwhelmingly congestion and airport layout, not snow being sprayed off wings. And the padding works: only 3.69% of on-time-gate departures went on to arrive 15 or more minutes late. Airlines budget the taxi into scheduled block time, so a long crawl to the runway rarely shows up as a late arrival. That is precisely why it is invisible — the delay is real minutes lived on the aircraft, absorbed by the schedule before it can dent any published statistic.

The strongest counter is that taxi-out is not "delay" and calling it one is a category error. DOT does not treat taxi time as a departure delay because, at a congested hub, 20-plus minutes of taxi is the normal, expected cost of operating there, already written into the timetable. A skeptic can argue the honest denominator is arrivals: if 96.3% of on-time-gate departures still land within 15 minutes of schedule, the tarmac time is a cost the airline pre-paid on the traveler's behalf, not a failure. Three more cautions. The 30-minute threshold is this desk's cut; a 20- or 45-minute line would move every share. Taxi-out conflates distance with congestion — O'Hare's and JFK's runways sit far from their gates, so part of the 25-minute median is geography, not a queue, while compact St. Louis is short on both counts. And one month is one month: May 2026 is a specific traffic and weather draw, not a verdict on any airport's year. The defensible claim is narrow — that the on-time departure statistic, by construction, stops measuring at the gate, and that in May 2026 the unmeasured minutes after the gate were concentrated at a handful of airports.

This comparison is possible only because U.S. law publishes flight-level taxi-out times that anyone can re-pull. Elsewhere the same operational reality is tracked but not exposed the same way: EUROCONTROL reports additional taxi-out time as a monthly airport-level average — a modeled "excess" over an unimpeded reference taxi — rather than a per-flight record, and most national authorities publish nothing comparable at all. The metric a European traveler can see is the airport's average congestion; the metric a U.S. traveler can see is their own flight's number, once someone reads all 611,735 of them at once.

A traveler whose flight leaves the gate at the scheduled minute counts it a punctual departure, and the airline counts it the same way — because the clock both are watching stops at pushback. Reading all 611,735 May flights at once shows what that clock cannot: nearly 9% of the country's on-time-or-early departures were still on the ground 30 minutes later, and at O'Hare it was one in three. The tarmac time never becomes a late arrival, because the schedule already swallowed it — which is exactly why it goes uncounted. The most honest number about a departure is not whether the door closed on time; it is how long after that the wheels left the ground, and that number only exists once someone reads the runway, not just the gate.