Same reliability, opposite delays: in May, Envoy and Delta each ran about 4 in 5 flights on time — Envoy inherited most of its delay, Delta made most of its own

The U.S. Department of Transportation's on-time database for May 2026 — the most recent month posted (the June and July files are not yet released) — records 611,735 scheduled domestic flights by the reporting carriers: 5,655 cancelled, 1,734 diverted, and 604,346 completed gate-to-gate. DOT requires a carrier to file a cause for every arrival 15 or more minutes late, split across five categories; 129,511 of the completed flights carried such a cause. Those flights logged 8,759,842 minutes of arrival delay in total. Systemwide, the largest single category was late-aircraft — a delay a flight inherits because the inbound aircraft or crew arrived late — at 3,536,468 minutes, or 40.4%. Air-carrier causes (crew, maintenance, fueling, baggage and other airline operations) accounted for 2,894,113 minutes (33.0%), the National Aviation System (air-traffic control, congestion, non-extreme weather) 1,794,650 (20.5%), extreme weather 522,891 (6.0%), and security 11,720 (0.1%). Across the whole system, carriers inherited 1.22 minutes of late-aircraft delay for every minute they generated as air-carrier delay. That systemwide ratio hides a wide spread between airlines. Across the 12 carriers that completed at least 5,000 flights, the late-aircraft-to-air-carrier ratio ran from 0.59 at SkyWest and 0.62 at Delta — whose delays were dominated by self-generated air-carrier minutes — to 2.13 at Envoy Air and 2.24 at Frontier, whose delays were dominated by inheritance. That is a 3.8× spread in the same month, over the same airspace. The composition did not track reliability. Envoy Air landed 83.1% of its flights on time and Delta 82.2% — statistically the same — yet Envoy's delay minutes were 43.3% inherited against 20.3% self-generated (ratio 2.13), while Delta's were 28.1% inherited against 45.6% self-generated (ratio 0.62). Frontier and American were equally alike on the headline: 76.7% and 76.5% on time, delayed at 23.3% and 23.5% — essentially identical odds of a late arrival — but Frontier's delays were 52.5% inherited to 23.4% made (ratio 2.24) and American's 46.1% to 32.9% (ratio 1.40). Across all 12 carriers, the correlation between on-time rate and the inherited-to-made ratio was −0.15. The four regional feeders in the set did not cluster: SkyWest sat lowest at 0.59 and Envoy Air second-highest at 2.13, while Republic (1.37) and PSA (1.52) landed mid-pack.

The following is analysis, not fact. A carrier's headline on-time rate answers one question — how often were you late — and the industry, the booking screens and the DOT rankings are all organized around that single number. The cause decomposition answers a different one: when you were late, where did the lateness come from. May's record shows the two are nearly independent. A passenger comparing Envoy and Delta on reliability alone would see a tie; the underlying delay was made of opposite material ^mmodeled. The inherited-to-made ratio behaves like a fingerprint of network structure rather than a measure of quality ^mmodeled. A carrier that runs its aircraft on tight back-to-back turns with little schedule slack — the point-to-point pattern common to ultra-low-cost and low-cost operators — passes any early stumble straight down the day's chain of flights on the same tail, so its delay minutes pile up in the late-aircraft column: Frontier at 2.24, Southwest at 1.72 ^mmodeled. A carrier that builds more slack between turns, or absorbs disruption at a hub, tends to show delay as first-generation air-carrier minutes that stop where they start: Delta at 0.62 ^mmodeled. Part of the arithmetic is mechanical — a carrier with a small air-carrier share, like Frontier's 23.4%, will post a high ratio even with only moderate inheritance, because the denominator is small — so the ratio should be read as a description of the delay mix, not a scoreboard. What the ratio is not is a verdict on who runs a better airline. The −0.15 correlation with on-time rate means the composition explains almost none of the reliability difference between carriers. All four combinations existed in May: high-inheritance and reliable (Envoy, 2.13 and 83.1%), high-inheritance and unreliable (Southwest, 1.72 and 72.4%), low-inheritance and reliable (Delta, 0.62 and 82.2%), low-inheritance and unreliable (Allegiant, 1.05 and 72.9%). The useful read for a traveler is narrow but real: two airlines can be equally likely to run you late and be late for opposite reasons — one because of something happening at your gate, the other because of something that happened hours earlier on an aircraft you never boarded — and the on-time rate alone will never tell you which ^mmodeled.

The strongest counter is that late-aircraft is a residual, not a root cause, so the ratio measures propagation rather than blame. A delay coded late-aircraft was created somewhere upstream — by weather, congestion or an operational problem earlier in the day — and attributing it to a carrier's own operations misreads the category. A high ratio says a carrier's delays are disproportionately inherited; it does not say the carrier caused them, and it should not be read as an airline-fault index. Three more cautions. Extreme weather is undercounted in the 6.0% figure: DOT files only severe weather in the weather column and folds ordinary weather into the National Aviation System category, so the true weather-attributable share is higher than 6% and part of the NAS column is weather in disguise. The ratio also confounds with network type and route length — regional feeders, ultra-low-cost point-to-point carriers and hub legacies run structurally different operations, so the spread partly reflects business model rather than execution, and the mechanical small-denominator effect inflates ratios for carriers with low air-carrier shares. And this is one month: May 2026 carries its own weather and air-traffic pattern, small carriers such as Allegiant (10,533 completed flights) rest on thinner samples, and a different month could compress or widen the spread ^sspeculative. The durable claim is the decoupling itself — across 12 carriers in one month, delay composition and on-time rate moved almost independently — not a ranking of airlines.

This carrier-by-carrier decomposition is possible only because U.S. law compels flight-level cause reporting: DOT requires a filed cause for every late arrival, which is what lets anyone split a carrier's delay minutes into inherited versus self-made and compare across the industry. Europe's network manager, Eurocontrol, tracks the same phenomenon under the name "reactionary delay" — its term for lateness passed from one flight to the next — but publishes it largely at network-aggregate level rather than as a re-pullable, per-carrier ledger. The cascade is a feature of any hub-and-tail airline system worldwide; the ability to fingerprint each carrier's share of it is a quirk of one country's disclosure rules.

A passenger compares airlines on one number — the on-time rate — and assumes a lower one means a worse-run airline. Reading all 604,346 May flights at once shows what that number cannot: two carriers can be equally likely to run you late and be late for opposite reasons. Envoy and Delta both got about four in five flights in on time, but Envoy's delays were mostly inherited from an earlier flight and Delta's were mostly generated at the gate; Frontier and American matched each other's odds of a delay almost exactly while their delays were built of different material. Across 12 carriers the composition of delay and the frequency of delay barely moved together — a correlation of −0.15. The most revealing number in an airline's record is not how often it is late, but the mix of what its lateness is made of — and that mix is a fingerprint of how the network is flown, not a scoreboard of how well it runs.