Methodology

The numbers

The station has logged 15,889,392 contacts, 832,269 of which carry no identity: frames without a registration, kept in the database, absent from the public log.

Logged flights

the 15,057,123 real flights the public log lists: the registration is resolved when possible, otherwise shown as “unknown reg”. This is the traffic figure shown on the home page, the log and the statistics.

Logged contacts

the 15,889,392 contacts the database keeps, ghosts included: the raw measure, before any filter. It is cited only here.

Received positions

the 14,798,621,050 position points received, counted per flight: this total is their sum. The full track is not kept, only the first and last point of each flight.

The capture chain

The capture chain uses no off-the-shelf radar software. A program written for this station, in Python and with no external dependency whatsoever, continuously reads the messages decoded by the receiver. It first stores them in a local queue, then sends them to the server, outbound only. Started on its own when the machine powers up, it runs in the background unattended. If the network goes down, capture continues offline and catches up when it returns, never sending the same contact twice. A flight is rebuilt from its successive contacts, closed after thirty minutes of silence, and its identity is frozen at the first contact, so that one aircraft never becomes another along the way. On the server, each contact is de-duplicated, then routes and identities are resolved against reference data kept up to date locally.

The log is never corrected, it is annotated

No received line is modified or deleted. When a data point surprises, a marker explains it. A 1932 glider captured at 30,000 feet is no miracle: it is almost always a single bit flipped in the address of a nearby airliner, and the marker says so. Conversely, some oddities are perfectly normal: manufacturers' test registrations are slots reused from one airframe to the next, a Bermudan aircraft rightfully carries a British address, and airBaltic's A220 YL-BTA genuinely flies under Swiss, Air France or Brussels Airlines callsigns, a documented wet-lease.

The witness

The old system left two journals written between 2017 and 2026, never recomputed since. They serve as my witnesses: at every check, the site recounts the aircraft discovered day by day and must find exactly the figures of the time. A single day that moves, and the alarm rings.

The records

No record publishes itself. Every candidate is checked by hand, and a contact flagged as doubtful holds nothing without a written arbitration. Altitude belongs to a U-2S, range to TF-BBH on an evening of exceptional propagation. As for the longest contact, it does not come from an aeroplane: on the evening of 28 September 2018, the 62nd Coupe Aéronautique Gordon Bennett released twenty gas balloons from Bern. I hooked D-OWBA right from the launch wave and followed it for 22.4 hours straight, 77,863 positions, while Zenge and Eimers sailed towards third place, 840 km to Poland (sources: FAI, official race archives). The second longest contact in the log is a balloon too.

Two scores per contact

Every contact carries two distinct scores: did the contact really happen, and do we really know who it was. The two questions are independent; a contact can be certain while the identity is doubtful. Pure Mode S aircraft, which transmit no position, are scored on their own scale.

Availability

98.5% of days covered since April 2016. Interruptions are classified by cause when I know it: reception failure (sixteen days in January 2025, machine alive, antenna mute), full stops, and my own maintenance stops, documented like everything else.

The first second

The first second of the log is not a gradual start: on 5 April 2016 at 18:07:31, 128 aircraft enter the log at the same timestamp, the whole sky in range photographed at once as the station came online. None arrived before another. When they must still be ordered, for the list of firsts or the discovery rank, the log follows the insertion order, first_seen then id: a convention, not a measurement.