Methodology
BoardFlight is new. Asking someone to plan a flight around a number from a product they have never heard of is a real thing to ask. So instead of claiming an accuracy figure nobody can check, this page sets out how each arrive-by time is built, what data stands behind each step, and where the model falls short of our own checks.
How accurate is this?
The honest answer has two parts: what is checked today, and what can't be measured yet.
What is checked. Each airport's hour-by-hour curve is compared with a year of TSA's published hourly checkpoint throughput. To pass, it has to match when that airport's day peaks and beat a generic average of the33 airports it was scored on. It passes at 16 of 61 airports; the 28 profiled since the last run are not scored yet. Everywhere else, the hourly chart is withheld rather than shown with a caveat. The check covers the shape of the day, not the exact minutes.
How the plan handles a bad day. Each step of the journey is a range, not a single number. The arrive-by time is read at the 88th percentile of the combined range, so a slower-than-usual day still fits.
What we can't tell you yet. How often a plan matches a real trip. That needs real trips. Travelers can report their actual security wait after they fly, but reports are just starting, too few to calibrate any airport yet. There is no live queue data behind any number on this site.
If an airport you know looks wrong, tell us at contact@boardflight.com.
1. Resolving your flight
A commercial flight-data provider resolves your flight number into the four facts that change how early you need to be: the departure airport, the terminal, the scheduled departure time, and whether the route is domestic or international. Nothing is inferred from the airline code alone. If a flight number matches more than one departure that day, BoardFlight shows both and lets you pick, rather than choosing for you.
2. Modeling the journey, step by step
The arrive-by time is not a buffer subtracted from departure. It is the sum of the steps you perform: arriving at the curb, reaching check-in, dropping a bag, clearing the security lane, and walking to the gate. Each step is estimated for that airport, that terminal, and that hour. Boarding lead is handled separately, as an airline policy rather than a queue, because you are not standing in line during it.
Those per-airport estimates are researched and recorded, not assumed. Every airport-specific cell in a profile carries a stated source and a dated verification. For composite walk and wait times, no external source often exists; in that case the cell records the absence explicitly instead of staying silent. A profile with gaps is marked as having gaps.
3. Planning for a slightly bad day
Each step is a distribution, not a single number. They are combined by simulation rather than by adding worst cases together, because stacking every pessimistic estimate produces a time so early it is useless. The plan is then read off at the 88th percentile of that simulated distribution: enough margin to absorb a slower-than-usual day at that airport and hour. That is a deliberate trade. Aiming higher means long, wasted waits at the gate for a margin that matters a few times a decade. Aiming lower makes normal congestion a missed flight.
4. Checking the model against outside data
A model that only agrees with itself proves nothing. The hour-by-hour demand curve behind each airport page is compared against independent published airport throughput data across a full year of it, and the comparison is a gate rather than a report card. Clearing it takes two things: the curve has to track the real hourly pattern, and it has to beat a generic average of every airport the gate scored at the specific airport it claims to describe. The second test is there because the first one alone is nearly free — every US airport is quiet at 3am and busy at 7am, so a generic curve passes it almost everywhere. On the curve those pages normally serve it passes at 16 of 61 airports, and at the other 45 the hourly chart is suppressed rather than shown with a caveat. The gate was last run on the 33 airports profiled at the time; the 28 added since have not been scored yet, and an airport that has not been scored is treated as failing until it is. Wrong information is worse than no information, so those airports show their representative arrive-by times and no curve.
When live schedule data is unavailable for a date, those pages fall back to a curve built from that airport’s own bundled history instead, and that curve is held to the same two tests separately — it clears them at 12 of 61. A chart appears only where the day being shown and the weekend it is compared against have both cleared the bar, so on a day where only part of the schedule is available fewer airports show one, never more. Both numbers are published here for the same reason they are enforced in the code: a validation step you only mention when it passes is marketing, not validation.
5. How deeply each airport is mapped
Coverage is not uniform, and the product says so on every page rather than averaging the difference away. Of the 61 airports profiled, 54 have a full terminal-by-terminal profile with worked reasoning behind the numbers, 7 are mapped the same way but still have source dates to log, and 0 rely on the general security-difficulty curve for airports of their size. A badge on every result tells you which one you are looking at before you rely on it.
Flights from airports outside this set still get a plan, built from a conservative profile for that airport’s size and difficulty tier. It is labeled as such.
6. Delays are handled asymmetrically, on purpose
A posted delay does not push your arrive-by time later. Delays get rolled back, gates change, and boarding often runs close to the original schedule. Treating an unconfirmed delay as extra time is a reliable way to miss a flight. BoardFlight shows the delay, distinguishes an airline-confirmed one from an unconfirmed report, and keeps the plan anchored to the scheduled departure. The asymmetry is the point: being an hour early costs an hour, being ten minutes late costs the trip.
7. What we do not claim
BoardFlight has no live camera feeds and no real-time checkpoint queue lengths. No consumer product in the US has those at national scale for free. It models historical and scheduled patterns, and says so. It has no corpus of traveler reports yet either. The feedback loop is built and open, and every figure it will eventually correct stays marked as an estimate until it does. Reports are meant to correct those estimates over time, not to turn a single traveler report into a measured number; when enough of them correct a figure, its label changes. That is the whole mechanism, and it runs in the direction of admitting things, not asserting them.
It also does not cover the flight itself, connections, customs on arrival, or non-US airports. It covers curb to gate, departure-side, in the United States.
Questions about any of this
The how it works page answers the common ones in short form, and each airport page shows the representative times and profile depth for that specific airport.