How FlyVerdict works
Every score on FlyVerdict comes from what actually happened to flights and passengers, using rules we publish here.
Where the flight data comes from
We collect departure and arrival times from several commercial flight-data providers and from ADS-B tracking, then merge them. When providers disagree, the most reliable source wins, and a final status (landed, cancelled, diverted) is never overwritten by an older reading. Codeshare flight numbers, such as a partner airline selling seats on another airline's flight, are attached to the flight that actually operated so each flight is counted once, for the airline that flew it.
What counts as late
A flight is late when it arrives more than 15 minutes after its scheduled arrival time. That is the standard used by airlines, airports and regulators worldwide. Cancelled flights count against the airline: the on-time rate is on-time arrivals divided by all flights that either arrived or were cancelled.
The Verdict Score for an airline
Each airline gets a score out of 10 on each route and overall:
- On-time performance: 50%
- Overall passenger rating: 30%
- Cabin crew rating: 20%
If a part has too little data, it's left out and the other weights are scaled up. On-time performance needs at least 10 flights, and ratings need at least 5 passengers before they show.
Recommending a journey
For a route, we look at every direct flight and every one-stop connection that actually operated, day by day. A connection counts only if the onward flight left at least 60 minutes after the first flight was due to land and within 24 hours. If the first flight landed too late to make the onward flight, that journey counts as late, because that's what happened to the passengers on it.
Each journey is scored on:
- Arriving on time at the final destination: 50%
- Journey time: 20% (the fastest journey on the route gets full marks)
- Passenger rating of the airlines flown: 20%
- Cabin crew rating: 10%
Journey statistics are kept per week, so a period counts whole weeks: “past 3 months” covers the 13 or 14 most recent weeks.
We deduct a little for connections on separate airlines (if you miss the connection, the second airline may not rebook you) and for overnight stopovers. Reliability is judged by the on-time rate we can be confident of given how many journeys are on record, using the lower bound of a 90% confidence interval. A short lucky streak can't beat a long, solid record. When two journeys score the same, the more reliable one wins.
The written explanation under each recommendation is generated from these numbers. Where it is written by AI, the AI only explains our pick; it can't change it, and any explanation containing a figure that isn't in our data is discarded.
Passenger ratings
Only signed-in travellers who name a flight we can confirm actually operated can rate it, and only once per flight. A flight can be rated from departure until 90 days later. Phone numbers and email addresses are removed from comments automatically.
Crew service reports
Travellers can report on cabin crew service on a specific flight. A moderator reviews every report before it counts. Reports are published as totals per airline, route and flight; we don't publish individual crew members' names. Each report records the flight number and date, which is what an airline needs to follow it up with the crew involved.
What we don't do
We don't sell rankings, airlines can't pay to change their scores, and we don't compare fares.