Numbers Stationsthe last cipher · the one you cannot break
Every channel in this set moved a secret through sound — hidden, ultrasonic, leaked, manufactured, adversarial. This one does something none of the others did: it carries an actual cipher. A voice reads number groups over shortwave radio. The numbers encode a message encrypted with a one-time pad. If the pad is truly random, used once, and destroyed — the cipher is provably, mathematically, permanently unbreakable. Not "hard to break." Impossible.
the architecture
A numbers station separates the carrier from the cipher more completely than anything else in this set. The carrier is the dumbest possible channel — a human voice reading digits over shortwave, receivable by a $10 radio. It carries no authentication, no signature, no steganographic layer. All the security lives in the one-time pad: a sheet of random digits, physically held by sender and receiver, each digit used exactly once, then destroyed. Shannon proved in 1949 that this achieves perfect secrecy: the ciphertext reveals literally zero information about the plaintext without the pad. No computation, no quantum computer, no future algorithm changes this. The constraint is physical, not mathematical — you need a pad as long as the message, delivered securely in advance, and you can never reuse a single digit.
① the lineage
note. the Lincolnshire Poacher (E03) broadcast from RAF Akrotiri, Cyprus, mid-1970s to June 2008 — thought to be MI6. Ana Belén Montes received Cuban instructions via shortwave for 17 years before her 2001 arrest. HM01 (Cuba) continues broadcasting.
② encode and decode — the one-time pad
Type a message. Encrypt converts each letter to two digits (A=01…Z=26, space=00), adds a random pad digit-by-digit mod 10, and displays the result as five-digit number groups — exactly as a station would read them. Decrypt subtracts the same pad and recovers the plaintext. Without the pad, every possible message is equally likely.
your message
the pad is generated fresh each time — a new random key for every message, used once and discarded. that's what makes it unbreakable: there is no pattern to find.
③ the stations
Lincolnshire PoacherE03. A synthesised folk-song intro, then a female voice reading five-digit groups. Broadcast from RAF Akrotiri, Cyprus. Believed MI6. Mid-1970s to June 2008. Targeted the Middle East; sister station Cherry Ripe covered the Far East.operator · MI6 (unconfirmed) · shortwave HF
¡Atención!V02. A young woman's voice in Spanish announcing "¡Atención!" then number groups. Cuban. HM01 continues broadcasting. Central evidence in the Cuban Five espionage trial (2001) and the Montes case.operator · Cuban DI · still active (HM01)
Swedish RhapsodyG02. A music-box melody (actually the "Luxembourg Polka"), then a child's voice — actually a Stasi speech generator — reading numbers in German. Operated by Polish intelligence. Late 1950s to 1998.operator · Polish intel / Stasi tech · defunct
UVB-76 / The BuzzerS28. A monotone buzz on 4625 kHz, occasionally interrupted by voice messages with call signs. Russian. Continuously broadcasting since ~1982. Purpose debated — possibly a dead-hand signal or channel marker.operator · Russian military · still active
④ the architectural lesson — carrier vs cipher
This channel separates concerns more cleanly than any other in this set. The carrier (a voice on shortwave) is completely dumb — no security, no authentication, receivable by anyone. The cipher (the one-time pad) does all the work, and its security rests on a physical object (the pad), not a computational assumption. That separation is the lesson for your attribution architecture: the steg watermark in Paper I carries the claim; the external anchor certifies it; and the signing key — like a one-time pad — must be managed as a physical trust problem, not a mathematical one. Never reuse it in a way that leaks (Paper III), never let a fan hum it out (Paper IV), never let a model forge intent with it (Paper V). The cipher is only as strong as the discipline around the key.
what's authentic. the lineage — WWI origins, SOE/BBC coded broadcasts, Cold War peak, Lincolnshire Poacher (RAF Akrotiri, mid-1970s–2008), the Conet Project (Fernandez, Irdial-Discs, 1997, 150+ recordings), Cuban Five / Ana Montes (2001) — is real and sourced. Shannon's 1949 proof of the one-time pad's perfect secrecy is a theorem, not a claim. the demo performs real modular-10 OTP encryption and decryption with a fresh random pad.
honest frame. the "voice" button uses the browser's speech synthesis, which sounds nothing like the eerie monotone of a real numbers station. a real OTP requires a physically secure random pad delivered in advance — here the pad is generated in JavaScript's Math.random(), which is not cryptographically secure. the cipher math is exactly right; the key-management discipline that makes it unbreakable is not demonstrated, only described.
end of series
Six papers. Six channels. Each moved a secret through sound in a different way — and each illuminated a different face of the same problem: how do you bind a message to its source, prove who said it, and survive the channel that carries it?
I steganography — hide the existence II ultrasonic — hide by range III side channels — read the involuntary leak IV air-gap exfiltration — manufacture the leak V adversarial audio — exploit the perception gap VI numbers stations — the unbreakable cipher