Methodology

JetSweep is built from two public ingredients: the radio signals aircraft broadcast about themselves, and the records governments and companies are required to disclose. Everything on the site is reconstructed from those two sources. We do not access private flight plans, passenger manifests, or any non-public FAA system.

We use the public broadcast, not the FAA feed

Since 2020, most aircraft operating in controlled U.S. airspace are required to carry ADS-B Out, a transponder that continuously broadcasts the aircraft’s identity, position, altitude, and speed over a public radio frequency (1090 MHz). These signals are unencrypted by design. Air traffic control, other aircraft, and anyone with an inexpensive receiver can hear them, and a worldwide community of volunteers collects them.

We draw those broadcasts from community-run, unfiltered feeds: live positions come from airplanes.live and adsb.fi, and our historical archive is built from adsb.lol data dumps. We are grateful to the volunteers who run the receivers behind them.

People often confuse two different things: the signal an aircraft transmits about itself, and the data feed the FAA redistributes to commercial vendors. JetSweep works from the former, the open broadcast, not from any FAA feed. Some operators enroll their aircraft in FAA privacy programs to stay off tracking sites; those programs limit the FAA’s feed and do not silence the broadcast. We cover how that works, and where it leaves a real blind spot, in how aircraft try to hide.

Turning a stream of signals into flights

The raw feed is a sparse, uneven stream of position points with frequent gaps. The first job is to cut that stream into discrete flights: a takeoff, a route, a landing. We key on the altitude profile rather than on raw low points: a climb away from the ground, a cruise, then a sustained descent toward a field.

Why not simply count every point below a few thousand feet? Because small airports sit in coverage holes. A jet drops off the receivers on final approach and never broadcasts those low points at all, so counting low points alone misses a lot of small-field landings. Reading the whole descent profile recovers them. Each completed flight is then re-checked against the dense archive and tagged with the evidence behind its endpoints, so a quick call in the live view is replaced by a firmer one once the flight is done.

When coverage drops

Volunteer coverage is uneven, so low legs, remote areas, and brief outages can leave gaps in a track. When the signal disappears mid-flight, two patterns usually tell us what happened:

• Long silence after a descent. The aircraft descended and then went quiet, which points to a landing at a field with no nearby receiver.
• Same-field bracketing. The jet vanished near an airport and reappeared near the same one a while later. It could not have gone far and come back in that time, so it landed there.
• Helicopters. They routinely drop below the floor at small fields. We infer a pickup only when the inbound heading points at the field, the timing fits a round trip, and ideally the aircraft leaves the way it came. Otherwise we do not guess.

For the map, when a track ends mid-descent we draw the missing approach as a dashed line, picking the likely runway end from the wind and the aircraft’s own past approaches. We abstain at busy fields where airliner-style traffic flow would override a light wind. These connectors are always dashed: a drawn inference, labeled as one.

Predicting a flight in progress

While a jet is still airborne, we predict where it is going and roughly when it will land. The prediction draws on the aircraft’s heading and reachable range, the routes this specific tail tends to fly, the time of year, and the shape of the descent once it begins. We also flag novelty: whether the likely destination is somewhere this jet has never been, which is the signal worth watching.

We say how sure we are, and we abstain when we are not. Confidence grows as a flight goes on: a guess early in a long flight is weak, while a jet on final approach is all but settled. Early in a flight to a genuinely new place we often cannot call it, and we say so rather than bluff. This is the same engine that explains a finished flight, run forward in time: as evidence accumulates, the live estimate converges on the final read.

Judging what a flight was for

Tracking the route is the straightforward part. The harder question is what a flight was for. We sort each completed flight into one of five purposes (recruiting, a meeting or event, game-day team travel, personal or contractual-perk travel, or routine and unexplained), and we separately judge how strongly the aircraft ties to a program.

We weigh evidence rather than read minds. The signals include:

• where the jet went relative to its own normal corridor: a trip that breaks the pattern is more interesting than one more shuttle run;
• how close stops fall to documented recruiting targets, whether the date lands in an NCAA contact period, and whether the trip has a multi-stop tour shape;
• several programs’ jets converging on one airport, which points to a meeting;
• aircraft size and game-day timing, which point to a team charter; and
• resort destinations and dead-period weekends, which point to personal travel.

Each signal abstains when its data is missing. The result is a calibrated confidence, not a yes or no. When we tag a flight likely recruiting, we mean it in a measured way, checked against documented cases rather than asserted. And a documented record always overrides the model. If a manifest or filing says what a flight was, that wins.

Public records

To turn a tail number into a story, we rely on disclosed records: the FAA Aircraft Registry to map tails to owners, state corporate filings and SEC filings to look behind holding companies and booster LLCs, and travel and charter manifests released under the Freedom of Information Act and state open-records laws. How we obtain those records, how they corroborate a flight track, and the line between a documented fact and a guess, has its own page: how we source flight records.

How confident we are

Two different things carry a confidence label.

Aircraft identity, who the airplane belongs to:

• Registry-confirmed. The broadcast identity matches a tail in the FAA registry, and that registration ties to the owner we name.
• Ownership-resolved. The registered owner is a holding company or LLC, and we have traced it to the underlying institution, booster, or person through primary filings.
• Pattern-attributed (probable). A privacy-masked aircraft is linked to a likely operator through route and schedule analysis. Labeled probable, not proven.

A flight’s story, what we can say about a given trip:

• Documented. A manifest, filing, FOIA record, or news report confirms it.
• Likely or possible. Scored by the model, not yet matched to a document.
• Tracked. An owned jet’s flight we have not scored yet.

Claims about who was on board a given flight sit at a higher bar still. See our Editorial & Ethics policy.

What we can’t do

• Coverage depends on volunteer receivers. Low-altitude legs, remote areas, and brief gaps can be missing from a track.
• Owner and aircraft details reflect records as we obtained them and can lag real-world changes such as a sale or re-registration.
• A jet attributed to a program does not mean any particular person was aboard on any particular flight.
• Early in a flight to a new destination, prediction is weak by nature. That is a limit of information, not of effort, and we abstain rather than guess.
• Purpose confidence is calibrated against documented cases, which skew toward trips that got reported. Treat it as a calibrated estimate, not a guarantee.

If you believe something here is inaccurate, please tell us, and see corrections.