The receiver
A software-defined radio on a 12-bit ADC, feeding a 22 m longwire at eight metres with a 9:1 unun and a common-mode choke at the feedpoint. Nothing expensive. 6821 kHz needs no help; the carrier is S9 here on a bad night.
Everything is recorded to disk as raw IQ and kept. Storage is cheap and I have learned not to trust my own notes over the tape.
The method
An entry is routine if a machine on a timer could have produced it. Timers are boring and reliable and explain almost everything on this band.
An entry is unresolved only when a timer cannot account for it and I have failed, on paper, to think of a mechanism that can. I have moved six entries out of unresolved after someone wrote in and explained them, which I enjoyed more than leaving them there.
Eleven remain. If you can take one off the list I would genuinely rather be wrong than interesting.
Four standing answers
Is it a numbers station? By format, yes — five-digit groups, preamble, group count, read twice. By behaviour, no station in the published catalogues does what this one does, and it appears in none of them.
Can you decrypt it? No, and neither can anyone else, and that is not a mystery. Properly used one-time-pad traffic is information-theoretically secure. The groups are the least interesting thing on this page.
Where is it? Three of us ran time-difference-of-arrival with synchronised receivers over four nights. The solutions cluster. The cluster is 450 km across and centred on open ocean, which means one of our clocks is wrong, or the assumption of a single transmitter is.
Why do you keep doing this? It has been on for two and a half thousand days without stopping. Somebody built it. Somebody has been feeding it. At some point that stops being a hobby question.
If you want to listen yourself
6821.0 kHz, upper sideband, 2.4 kHz filter. Best after local midnight and well into the grey line. The carrier is always there. The blocks are not — roughly twice a month, and it does not announce itself.
If you catch one, send me the IQ, not the MP3. Twice now the thing worth hearing was below the noise floor of somebody's lossy encoder, and once it was only in one channel.
Contact
Recordings, corrections, and mechanisms are all welcome. Theories are welcome but will be read slowly. I have received four hundred theories and no mechanisms.