A trilogy in four parts. Part 5: I thought breaking Enigma would be easy.

by Christian Rudolph

When the fourth book of the Hitchhiker's Guide to the Galaxy was released in 1984, the official omnibus edition I own called it A Trilogy in Four Parts. Then, in 1992, Mostly Harmless, came out, making it the fifth part of a trilogy in four parts. This series of blog posts has developed a similar relationship with numbers. What began as a single post about a photograph taken at Bletchley Park has now reached Part 5, which, in the finest tradition of Douglas Adams, makes it the fifth part of a trilogy in four parts, with no sign of stopping.

To recap where we left off: the message was decrypted. The plaintext matched the Typex tape. Frode Weierud had been thanked. Bletchley Park had been thanked. The code had run, the simulator had confirmed it, and the settings were known.

But the original itch – the one that started all of this – had never been fully scratched. The goal was always to take a real message, find the correct machine settings, and watch German emerge letter by letter. Running code in a Linux container and clicking through a phone app were both remarkable in their own way, but neither was the thing itself. The thing itself was fingers on actual keys, and lamps lighting up one at a time.

Which is how a cell biologist ended up buying soldering equipment.

The search for a replica

The first instinct was to find a replica. Not a real Enigma – those fetch tens of thousands of dollars at auction, and in any case the point was never the object itself but the concepts behind it. A working replica would do perfectly.

It would not, as it turned out, do perfectly at all.

The market for Enigma replicas divides neatly into two disappointing categories. At the lower end, around £200, sits a range of convincing-looking machines that turn out on closer inspection to be 3D-printed decorative objects with no functional internals whatsoever. Pretty, but useless for the purpose. At the upper end, working electromechanical replicas hand-built to authentic specifications exist, but at prices that make a non-functioning £200 ornament look like a bargain. There is a certain irony in the fact that a fully working Enigma simulator is available for free on a phone, while a physical replica capable of doing the same thing costs several hundred pounds.

The solution, when it appeared, was an elegant middle path: the Enigma-E kit, produced by Crypto Museum. Not an electromechanical replica with moving rotors, but a fully functional electronic implementation of the Enigma cipher, with a proper keyboard, a lampboard, and accurate cryptographic behaviour. And crucially – a kit. Something to build.

The UKW-D question

Before ordering, there was one question worth asking. The message I had spent weeks decrypting used the UKW-D reflector – the rewirable reflector that the Luftwaffe introduced in 1944 and that had defeated every computational attack we had thrown at it. Would the Enigma-E support UKW-D?

The answer came from Paul Reuvers at Crypto Museum, and it was both honest and rather wonderful:

"The firmware was made by my colleague Marc Simons, and consists of hand-optimised assembly code, which barely fits the 2KB ROM space of the processor. I believe he only had 3 bytes of code space left ;-)"

Three bytes. In a machine designed to recreate one of the most sophisticated cipher devices of the Second World War. There is something entirely in the spirit of Bletchley Park about that.

Paul went further. It would not be possible to support UKW-D natively, but if I was interested, he would ask Marc whether a custom firmware version might be possible – one that replaced UKW-C with the specific UKW-D wiring from our message, hard-wired for that one configuration.

It was a generous offer, and after some thought the answer was no. A machine permanently wired for a single UKW-D configuration from a single day in November 1944 would be deeply satisfying for exactly one purpose and useful for nothing else. The standard kit supports both UKW-B and UKW-C, which covers the vast majority of documented historical Enigma traffic – including, as it happens, a range of messages made available by Frode Weierud himself. Versatility won.

The order was placed with Cor Moerman, who handles orders and whose customer service was excellent throughout – prompt, clear and immediately helpful in forwarding the UKW-D question to the right person.

The Build

What was not quite so straightforward was what happened when the kit arrived.

The Crypto Museum website states that basic soldering skills are required for the Enigma-E, and I rather naively thought I could build the Enigma-E together with my older son. In retrospect I do understand the statement, because in some way basic soldering skills were indeed all that were needed. However, the phrasing of basic soldering skills did not match my expectation when opening the kit revealed three bags full of individual components. My soldering experience but importantly also my tools were entirely inadequate. Before a single component was placed, money needed to be spent.

I only had a relatively cheap soldering iron. This was fine, as it was heating perfectly, but I already knew the tips were a problem. Just doing a very small project during lockdown in 2020 left one entirely degraded because the metal plating had gone, so the cheap tips that came with the iron were no good at all. So, the first thing I ordered was a set of quality tips. In addition, as part of the project in lockdown I had fought hard with lead-free solder, and lost on several occasions. I therefore made the deliberate choice to use 63/37 leaded solder I had already used, with more success, as a teenager. Lead-free solder is now standard in commercial electronics manufacturing for good environmental reasons, but the Enigma-E was designed in an era when leaded solder was the norm, and the difference in working properties is significant. 63/37 flows better, wets more reliably, and produces joints that are considerably easier to inspect. It remains entirely legal to purchase and use for hobby purposes.

Two other purchases proved their worth immediately. Flush cutters – which trim component leads cleanly and flush with the board – were a revelation to someone who had previously managed with standard wire cutters. And a lead bending tool, purchased for £3, turned out to be the single best investment of the entire project. A small plastic jig that bends component leads to the correct spacing and angle before insertion, it transformed a fiddly and inconsistent process into something precise and repeatable.

More properly equipped, the build began on the 5th of September.

Every evening from the 5th of September was devoted to the build. The process was to methodically follow the manual: diodes first, then resistors, ICs and so on. The kit instructions are clear and well-organised and I was able to follow them easily.

For someone whose eyesight no longer cooperates with small print without assistance, reading component markings became an unexpected challenge. This is where AI proved genuinely useful – photographs of components too small to read comfortably produced immediate and accurate identification. A resistor colour code, a capacitor marking, a transistor designation: all read without difficulty from a photograph when the naked eye could not manage it. The AI was cross-referenced with a magnifying glass in every case, but the identifications were consistently correct.

It proved rather less reliable for orientation verification. My understanding of electronics is not good enough to fully comprehend how the PCB works in detail. And even though the descriptions were very clear, there is something nice about a voice telling you: yes, you have put this component into the board in the correct orientation. Go on and solder it in.

Confidence that turned out to be misplaced: At one point, an integrated circuit was placed loosely in its socket the wrong way round – 180° from correct. As a matter of fact, I noticed that the IC was flipped while taking a photo to upload for confirmation with ChatGPT, so there was method to the madness. Just for fun I asked anyway and as I had feared the response was confident and enthusiastic: the notch was clearly visible, the orientation was correct, the joint was ready to solder.

It was not. The IC was backwards.

Upon querying this assessment, ChatGPT reconsidered and agreed that on reflection the IC was indeed incorrectly placed and that this was clearly visible in the photo. This is, it should be noted, precisely the behaviour described in Part 4 of this series: artificial enthusiasm in place of accurate analysis. A useful reminder that AI image verification is not a substitute for looking carefully at the thing itself. The magnifying glass, as ever, was more reliable than the algorithm.

The electronics were completed on the 10th of September. Approximately 800 solder joints, every one of them by hand, over six evenings.

800 joints. 800 opportunities for something to go wrong. A cold joint, a bridged connection, a component placed backwards, a lead trimmed too short. Any one of them capable of producing silence where there should be light.

The power was connected.

And the board lit up cleanly on the first attempt.

The Test

The first test was deliberately conservative. Rather than immediately attempting the full 173-character decrypt, a known configuration was used – the settings from the key sheet displayed at Bletchley Park, the very document that had started this entire adventure. A word was typed into the BP Enigma simulator for Android. The same word was typed into the Enigma-E. The outputs matched exactly.

The machine worked.

There was, at this point, only one thing left to do.

The Decrypt

The machine settings provided by Frode Weierud, as described in Part 3 – rotor order IV-II-V, ring settings G-M-Y, plugboard HY PV RG XQ WF KC JB TM IS ND – were programmed in. The reflector was set to UKW-C rather than the original UKW-D, for reasons explained in Part 3, but every other setting was identical to those used on 1 November 1944 over the Adriatic. The starting position was set to C-I-N.

The ciphertext – all 173 characters of it, generated during the investigation and verified against the known plaintext – was written out on a notepad.

And then, one key at a time, the message began to come out.

A U F K L

A V R U N

G R A U M

A D R I A

Eighty years after a Luftwaffe radio operator somewhere over the Adriatic made some typos, forgetting an S and adding a V where there should be an E, his message was coming out of a kit built on a dining room table in England, letter by letter, lamp by lamp. It kept going. All the way to UDASOXEINSZWO.

Every error faithful. Every substitution intact. DOTORSTOERUNG and SQLEQTWETTER and NIQT and all the rest of it, exactly as transmitted, exactly as decoded on the Typex tape at Bletchley Park.

The Photograph

The photographs that capture this moment show the notepad filled with letters – the full plaintext, line by line, all the way to EINSZW – and the Enigma-E, rotor windows showing the position after 172 keystrokes, the lamp O lit for the final character of EINSZWO.

It is not a dramatic photograph. It is a notepad and a circuit board. But what it represents took months of work, two letters to experts, a 1998 mailing list post, a Linux container, three bags of components and 800 solder joints to produce.

Is it the full scratch of the itch? Not entirely. Of course it came out. It was encoded using precisely these settings. With the machine working absolutely fine it had to come out exactly as encoded. And yet, it felt special. The settings are Frode Weierud's, the reflector is UKW-C rather than the original UKW-D and the letters of the ciphertext were generated by Claude rather than the one transmitted over the Adriatic in 1944. No kit currently supports UKW-D, and that particular chapter of the story lives in the code and the blog posts rather than on a lampboard.

But it is close enough. The plaintext is correct and, what makes it even more special, reconstructing the full message means that we can now encode – and decode – the letters that were missing from the original transcript, giving us the complete ciphertext for both UKW-D and UKW-C. The machine works. And in the process of building it, the original question – what would it actually feel like to watch a message come out – has been answered.

Similar and related blog posts:

• Part 1 of this series can be found here
• Part 2 of this series can be found here
• Part 3 of this series can be found here
• Part 4 of this series can be found here
• Part 6 of this series can be found here