Methodology
Trust is a feature. Everything below is published so that any number on FlyerIntel can be checked, reproduced or disputed.
What “on time” means
FlyerIntel follows the definition used by the US Department of Transportation. A flight is delayed when it arrives at the gate 15 or more minutes after its scheduled arrival time. A flight arriving 14 minutes late is on time; a flight arriving early is on time.
Cancelled and diverted flights never arrive at their scheduled destination, so they have no arrival delay. They are excluded from on-time rates rather than counted as late, and reported separately as cancellation and diversion rates. This matters: a carrier that cancels heavily can post a respectable on-time rate, which is why FlyerIntel always shows cancellations alongside punctuality and weights them in the reliability score.
Specifically, for any group of flights:
- Completed — flights with a recorded arrival delay
- On-time rate — on-time arrivals ÷ completed flights
- Cancellation rate — cancelled flights ÷ all scheduled flights
- Severe delay rate — arrivals 60+ minutes late ÷ completed flights
- Delay when late — mean arrival delay across flights 15+ minutes late
- Median arrival delay — the middle outcome, unmoved by a few extreme days
- Departure punctuality — on-time gate departures ÷ flights that departed
Why a flight was cancelled
Every cancelled record carries one of four reason codes, filed with US DOT by the carrier that operated the flight. FlyerIntel reports them in the carrier’s own terms and does not reclassify them:
- The airline — circumstances DOT defines as within the carrier’s control: maintenance, crew, aircraft cleaning, baggage, fueling
- Extreme weather — conditions, actual or forecast, that in the carrier’s judgment prevent the flight operating
- The air traffic system — airport operations, heavy traffic volume, air traffic control, and non-extreme weather. Not a decision of the airline
- Security — terminal evacuation, a security breach, inoperative screening equipment, or screening queues over 29 minutes
Two consequences worth stating plainly. Weather appears in two of those four categories, so the extreme-weather share understates weather’s real contribution. And because the code is the airline’s own filing rather than an independent finding, carriers operating the same runways in the same conditions can report markedly different splits — the figures describe what each airline reported, not an apportionment of fault.
Reliability scores
A FlyerIntel Reliability Score is a 0–100 summary of how dependably an airport, route or airline actually performed. Higher is better. Scores are versioned (route_reliability_v1, airport_reliability_v1, airline_reliability_v1) so a change in method is visible rather than silent.
Each component is converted to a 0–100 sub-score by linear interpolation between two fixed reference points, then combined using the weights below. The reference points are fixed rather than derived from whatever is currently in the database, so a score does not drift when coverage expands.
| Component | Weight | Scores 0 at | Scores 100 at |
|---|---|---|---|
| On-time arrival rate Share of completed flights arriving less than 15 minutes after schedule. | 35% | 55% | 90% |
| Cancellation rate Share of scheduled flights that were cancelled. | 20% | 4% | 0% |
| Severe delay rate Share of completed flights arriving 60 or more minutes late. | 15% | 15% | 2% |
| Median arrival delay The typical arrival outcome, unaffected by a handful of extreme days. | 10% | 15min | -12min |
| Average delay when late How costly a delay is when one happens. | 10% | 90min | 30min |
| Worst-case arrival delay The 90th percentile arrival delay — a measure of how bad a bad day gets. | 10% | 150min | 15min |
Values outside the reference range are clamped, so a component can never contribute more than its weight or less than zero. Every published score stores its component breakdown.
Why trend is not part of the score
A score answers “how reliable is this?”. A trend answers “which way is it moving?”. Averaging the two produces a number that means neither. FlyerIntel computes trend separately — comparing the most recent 12 months against the 12 before, so seasonality cancels out — and displays it alongside the score instead of inside it.
Percentiles and cohorts
A score on its own is hard to interpret, so every score is placed in a percentile against a named cohort of comparable entities. Airports are compared against the major US airports FlyerIntel covers; routes against other routes carrying enough traffic to be comparable; airlines against US reporting carriers. The cohort and its size are shown wherever a percentile appears.
Sample-size thresholds
Statistics from small samples are noise dressed as insight. FlyerIntel does not publish a conclusion below these thresholds — the page simply omits that section.
| Conclusion | Minimum records |
|---|---|
| Airport reliability metrics | 5,000 flights |
| Route reliability metrics | 750 flights |
| Route page listed and indexed | 10,000 flights |
| Airline profile | 10,000 flights |
| Any hour, window, weekday or month bucket | 120 flights |
| Carrier comparison on a route | 200 flights |
| Carrier comparison at an airport | 500 flights |
| Flight-number analysis | 90 flights |
| Cancellation-cause split at an airport | 500 cancellations |
A “best time to fly” claim additionally requires at least 3 qualifying departure windows and a spread of at least 2 percentage points between the best and worst. Below that, timing is not a real effect.
The cancellation-cause threshold is counted in cancellations rather than flights because that is what its percentages are taken from: an airport can have a very large sample of departures and a very small sample of the thing being described. It is set higher than the bucket floor because a four-way split is a weaker statistic than an average, and because this one assigns responsibility — at 500 records the widest 95% interval on any single share is about 4 percentage points, narrow enough that the ordering of causes is not an artifact of the sample.
Time-of-day, weekday and seasonal analysis
Flights are bucketed by their scheduled local departure time at the origin airport, which is how the source data records them, and which is what a traveler is actually choosing between when booking. Departure windows are: 5–9 AM, 9 AM–12 PM, 12–3 PM, 3–6 PM, 6–9 PM and 9 PM–5 AM.
Weekday analysis uses the scheduled departure date. Seasonal analysis pools each calendar month across every loaded year, and is only published where at least two years of that month are present — a single year of one month is weather, not a season.
Comparing airports that serve the same city
A traveler flying Atlanta to New York is choosing between LaGuardia, JFK and Newark, not between three unrelated destinations. Grouping those by city name does not work — Newark’s city is Newark — so FlyerIntel uses the US DOT’s own city market identifier, which is present in the source records and is the authoritative grouping. It places Newark with the New York airports, Baltimore with the Washington ones, and Burbank, Ontario and Santa Ana with Los Angeles.
Route profiles show every airport pairing serving the same city market so the comparison can be made directly. The differences are not small: Chicago to Los Angeles is a materially more reliable journey through some pairings than others, and nothing on a booking site surfaces that.
Current operational status
The status shown at the top of an airport profile is derived only from live inputs: FAA operational events and the current aviation weather observation. Historical performance never contributes to it.
- Significant disruption — an airport closure, ground stop, or a ground delay program with substantial delays
- Elevated disruption — a ground delay program, reported delays of 45 minutes or more, or severe weather conditions
- Minor disruption — shorter reported delays, or weather conditions associated with a higher chance of delay
- Normal operations — the FAA feed is current and reports nothing, and weather is not a factor
- No disruption reported — nothing in the inputs FlyerIntel can currently see indicates a problem, which is not the same as knowing everything is normal
- Not currently known — no live input is available
Where a live input is missing, the page does not fill the space with an apology or an estimate. It shows the historical record instead — what departures in that hour have actually done over the past three years — clearly labeled as history rather than as a current reading, and it states at the foot of the section which live inputs are absent.
Weather can raise the assessment but never on its own declares an airport significantly disrupted. Storms at an airport are a risk factor, not a delayed flight, and FlyerIntel does not present one as the other.
Weather impact
FlyerIntel decodes each aviation weather observation and assigns its own operational impact level from documented conditions: freezing precipitation, thunderstorms, snow, visibility, cloud ceiling and wind. Thresholds follow standard aviation flight-category boundaries. The assessment describes conditions associated with disruption; it never claims a specific flight will be delayed.
Data freshness
Different data ages differently, so there is no single freshness rule. A value is labeled Live, Recent, Stale or Unavailable according to its own class:
| Data | Live within | Recent within |
|---|---|---|
| Security wait times | 15 min | 30 min |
| Parking availability | 20 min | 60 min |
| FAA operational events | 10 min | 30 min |
| Weather observation (METAR) | 75 min | 150 min |
| Weather forecast (TAF) | 7 hr | 13 hr |
| Historical performance | Never labeled live; the source period is stated instead | |
Missing data
Where FlyerIntel cannot legitimately obtain a value, the section says so. It is never filled with an estimate, a placeholder or a figure carried over from elsewhere. An empty module means we do not have the data — which is more useful than a confident-looking number that is wrong.
Limitations
- Historical analysis covers scheduled domestic flights operated by carriers above the DOT reporting threshold. Smaller carriers and international flights are not included.
- Flights sold under a mainline brand but flown by a regional carrier are reported by the operating carrier. FlyerIntel presents them that way, which is accurate but can differ from the brand on the ticket.
- Delay-cause minutes are only reported for arrivals 15 or more minutes late, so cause breakdowns describe delayed flights rather than all flights.
- Source data is released monthly, two to three months in arrears. The analysis period is stated on every page that uses it.
- Live coverage varies by airport and depends on what each source permits.
Data sources · Download the numbers · Report a problem · Editorial policy