Two Lieutenants of the Royal Canadian Navy were posted to our party: Alan Roffey and Wally Fraser. They joined seamlessly both in the work and in off-duty visits that a number of us made together by train and bicycle to the Shakespeare Memorial Theatre at Stratford-upon-Avon whose building was then quite new.
We did not move geographically but remained where we had been in Italian Naval on the upper floor of Block B. The internal partitions have all gone but it is easy to reconstruct them. There is a staircase at the East end of the Block, directly overlooking Block C: the mathematicians had the room at the head of it. Next to it a very small room which Commander Belfield, in recognition of his seniority, had to himself. Next, on the side of the central corridor facing the lake, was our headquarters room occupied by myself with our Japanese linguists and Olive Thorogood who worked with them in producing in-team intelligence: this group took the name GEN. Across the corridor in the Big Room, Adelaide Dickens had day-to-day charge of the main additive recovery work, ranging from servicing the paperwork to the cryptanalysis itself.
In passing, the room next to our headquarters was occupied by two Army Intelligence Captains who ran JN-11, the system in use by the Japanese Auxiliary Fleet. One of them, Brian Augarde, a professional jazz clarinettist, created the BP Jazz Quintet which contributed joyfully to the Park’s rich musical life but is usually overlooked by those who write about the classical music performances and operas. The range and celebrity quality of these was astonishing, as has often been observed. They took place in the Assembly Hall which had been built outside the perimeter fence so that the people of Bletchley could use it too.
In July 1943 a performance of ‘Acis and Galatea’ produced within the Park had Peter Pears and Elizabeth Welch as visiting singers in the title roles. Myra Hess gave a recital in November. These were followed by Denis Matthews, the Jacques String Orchestra, Isobel Baillie, Leon Goossens, Maggie Teyte, Francis Poulenc and Pierre Bernac. The Ballet Rambert were dancing on 7 May 1945 as Germany surrendered. We had equally distinguished lectures: three talks in September 1943 on invasions of Sicily from the Normans through Garibaldi (the Allies had landed there in July); the Marquess of Linlithgow, former Viceroy, on India; Arthur Bryant; Emil Cammaerts (whose daughter Jeanne was in the Park and son Francis was among the most distinguished of the Special Operations Executive) on Rubens. These are occasions that I attended; it does not pretend to be a complete list.
Unmentioned elsewhere, to my knowledge, is a special occasion in the cinema which also operated in the Assembly Hall. In April 1945 it showed Münchhausen, in Agfacolor with German soundtrack and no subtitles: this had been released in Germany in March 1943. It was shortly followed by a showing of the less distinguished Die Goldene Stadt (1942). There must be a good story as to how these came to be shown at Bletchley Park, but I do not remember any comment on this at the time.
Encoding and enciphering by JN-25 is described in Appendix V. Very briefly, the message text was first transformed into a series of 5-digit codegroups. These were then disguised by adding to each, using non-carrying addition, 5-digit ‘additives’ taken consecutively from a table of many pages, with ten rows and ten columns of additives on each page. An ‘indicator’ system showed where to start. The cryptanalysis of JN-25 was no different in principle from that of the Italian Libia with which we were familiar. True, the code-groups had five digits instead of four, they ‘scanned’ (as a garble check, the sum of the five digits was divisible by three in virtually all the JN-25 codes) and the language was totally foreign whereas Italian had never seemed forbidding even to those who had not studied it. But we were taking aboard variations, not starting something new.
The JN-25 ‘traffic’ – the intercepted messages – was by now divided into four channels sharing the same codebook but each with its own additive table and indicators. The Channel labelled A, which was used by nearly all vessels for important routine communications, had about 300 messages daily. The B Channel was used by most vessels, except submarines, for strategic communications and had about a quarter of that number: enough in each case to make the cipher readable. Two other channels had too little traffic to approach readibility. It did not take us long to decide that the A Channel was too big for us and that we would concentrate on the B Channel. As each new codebook came into use it was given a capital letter designation, and each new additive table an Arabic numeral. For example, the B Channel system in force when we started was J-24.
Allied cryptanalysts had been working on JN-25 for a long time so we had at our disposal at once a substantial body of established ‘good groups’: groups known to be in the codebook because they had appeared in bits of message text previously deciphered. Far from starting from scratch, we were straight into the systematic recovery of new additives. At first we regarded ourselves as experimenting and learning, but by 6 October we felt ready to tell Washington we had been filling in gaps and had recovered over 250 additives: would they like them? When they replied ‘please’, we were officially in business.
On 16 October Table 30 came into use in the B Channel; it was not known whether the codebook had changed too. We attacked the traffic using the J codebook anyway, and on 12 November, before any additives had been published elsewhere, we signalled seven additives and the conclusion that codebook J was still in use. Washington agreed that they ‘looked good’. On 6 November we had a visit from Commander Howard T. Engstrom, US Navy, who saw what we were doing and hazarded that, as we grew, there might have to be a partition of work between us and the American stations. With these two encouragements we ended November with high hopes.
I will return to the history later. First, there is much more to be said about the circumstances and the methods of the cryptanalysis.
The enciphered messages came to us from two sources. Motorcycle despatch riders, who are an abiding memory for anyone who worked at the Park, brought bags of it two or three times daily from Flowerdown, who also teleprinted it. These were current messages, but constituted only some 15 per cent of the total available. Much bigger deliveries came by bag two or three times a week from Washington, taking us to some 85 per cent of the total available. But the Washington material was two or three weeks old when it reached us: a serious handicap we were never to overcome.
On arrival the traffic went first to the Processing Party which served all the Japanese Naval parties. This was led by Mrs E. Parsons and it moved in September 1943 from Hut 7 to occupy the lower floor of Block B. At its peak it had some 200 staff, three-quarters of them Wrens. The Processing Party registered all incoming messages, sorted them by type and distributed them to where they were needed. I return later to their other functions and ways in which we worked with them.
Our share of the incoming messages went first to the Big Room where they were individually registered with details of their date, time and call signs of origin and address. As the indicator system was known, new messages could be placed immediately at their correct starting points on the additive table. They were written on to ‘depth sheets’ (also called ‘Tiltman sheets’) prepared for the purpose. These were something like 50 inches wide, ruled vertically into 100 columns so that each sheet corresponded to all ten lines and ten columns of one page of the additive table, and each additive position was identified by its page, row and column numbers. The sheets were deep enough to provide for (I guess) some 30 messages to be written one below the other, with room for another couple of lines left blank for working below each message. As new messages were added, each from its own starting-point, they created ‘depth’ where they overlapped: because all the enciphered groups in a column had been enciphered by the same additive. Depth was the life-blood of the additive recovery process. Providing on the sheets for up to 30 messages in depth was an aspiration and an act of faith. Most often we had to work on depths of as little as five: a depth of eight or ten was luxury. To fit the 5-digit groups within the columns called for a small neat hand, so those without that talent were politely excused that t
ask.
The routine attack on a column, given a depth of some five or more, was through ‘differencing’. This is explained in detail in Appendix VI. To summarise, differencing down a column of enciphered groups would hopefully produce one or more speculative additives for it. Subtracting each of these in turn down the column – ‘stripping’ it – gave speculative deciphered codegroups, to be tested first to see if they scanned and – if they did – for good groups. A sufficiency of good groups (I return later to what ‘sufficiency’ meant) would confirm that speculative additive as probably correct. To make this practicable we relied heavily on the ‘Freebornery’ to provide us with Indexes of Differences and of Good Groups. This was the massive Hollerith installation in Block C, serving the whole of Bletchley Park and named after its head ‘Mr Freeborn’.
Once a first break had been made on a stretch of additive, usually by differencing, and several more or less adjoining additives recovered, the cryptanalysts moved on to a subtler approach. This was based on study of the good groups, their behaviour and the company they kept: did they have favourite positions in the message, did they tend to recur, were they regularly associated with other good groups preceding or following them? For some of these the bookbuilders, whose job was to work out the textual meaning of each codegroup, would have worked out the meaning, for others not. For the cryptanalysts the meaning was useful but not essential. A good group, though only a five-digit number, could become a good friend.
For instance, the feature of a pair of codegroups regularly appearing together with a few others in between could be produced by meanings as different as ‘(….)’ and ‘today… (weather)… tomorrow… (weather)’. Such a feature could be exploited irrespective of meaning: if one of the pair appeared, the other was searched for. Of course if some of the meanings were known, the search was better informed. When one of the codegroups for Maru (ship) appeared it would very probably be preceded by a ship-name, flanked by the ‘arrive’ or ‘depart’ codegroups and accompanied by the names of the ports identified by the message’s callsigns. If any of these details were known, that was a ‘crib’ worth trying out.
Each possibility produced by this kind of reasoning had then to be followed up by inserting the suggested new codegroup into its adjoining column, deriving the speculative additive it implied there by subtracting it from the corresponding enciphered group and testing that speculative additive as before. This procedure, called ‘horizontal stripping’ to distinguish it from differencing, was more art than arithmetic and was capable of limitless improvement in knowledge and technique. In the internal history of the work on JN-25 which I compiled in August 1945, I wrote:
The most important asset of a first-rate stripper is a ‘faculty of association’, a kind of photographic memory which not only stores each incident but equips it with ‘hooks and eyes’ by which the whole can be brought to the surface again, consciously or subconsciously, by the recurrence of some part of it. This faculty builds up for the stripper such a store of experience that eventually almost every group in the codebook calls up some image of a phrase previously seen which included it.
The Big Room covered all the functions from the clerical routine writing of new messages on depth sheets and preparation of material to go to the Freebornery to the cryptanalysis itself. Its members did not divide into servicers and strippers by rank or experience, and everyone had to share the clerical work when the pressure required it, with Adelaide Dickens steering them when necessary to ensure a satisfactory spread of effort across all the tasks. Those with the aptitude and the opportunity for stripping had a degree of flexibility as to how they went about it. Racks held the depth sheets in order, offering a choice: whether to have first go at a depth newly enhanced by new messages or to persevere with a stubborn one, now dog-eared, which really ought to be made to yield. Copies of the Indexes of Good Groups and of Differences, and of the codebook as so far recovered, were all around for reference. And when a certain point was reached – which could be either success or the need to get help – the stripper would move around to deliver or to consult. It was essentially tedious work but performed with great spirit and good humour, encouraged always by the prospect that the next column might be the one that would come out sensationally right.
If the work of the Big Room was spread fairly thinly over very many depth sheets, that of the group called GEN was quite the opposite: concentrated attention to individual problems. It had begun with Gladys Gibson working on routine messages such as fuel returns or weather reports. Our break into J-30 had indeed come through a routine, not through differencing. Later she led GEN comprising Henry Reed, Commander Belfield and Olive Thorogood to provide in-team intelligence, starting with little more than the Rose Innes and Kenkyusha Japanese dictionaries and a map. Their job was to develop knowledge and techniques, contexts and cribs, to find a way round when the frontal attack on a column had got stuck. A Big Room stripper with a possible but still dodgy speculative additive which had thrown up some familiar good groups could walk across the corridor and ask GEN for suggestions for extending the text to right or to left. Each new codegroup that GEN suggested was taken back to the depth sheet and used to produce more speculative additives to be tested as before. If one of them proved good, that was both a new recovered additive and confirmation of the dodgy one that had led to it.
Assembling and organising the rich mixture of crib and context information and devising means of getting into it quickly called for a certain ingenuity, and it was skilfully done. Of course formal cataloguing and indexing were only part of it; as with skilled stripping, acute observation, a good memory, lateral thinking and imaginative hunches founded in experience often found the way in. This small team contributed substantially to the overall success and patently enjoyed doing it.
At one stage when we were struggling to keep up with the increasing quantity of incoming traffic Washington sent us a dozen or so calculating machines made by the National Cash Register Company. We called them the ‘fruit machines’. They were big, perhaps four feet tall, and floor-standing; mechanical, not electrical. They were set up to do non-carrying addition and subtraction of five-digit numbers. For example, all the enciphered groups of a column of depth could be entered and then a speculative additive placed at their head: one pull of the handle stripped the additive from all the groups simultaneously to reveal all the speculative codegroups, which could be tested immediately for scanning. The enciphered groups stayed on the machine until all the speculative additives for that column had been tried. At first and at their best the machines were a great help, faster than manual stripping and easier on the eyes. The trouble was that they proved prone to mechanical breakdown and, as there seemed to be few technicians around capable of mending them, they were out of action for long periods. Confidence in them waned and eventually we gave them up altogether.
In discussing differencing above I mentioned ‘a sufficiency of good groups’ but slid by the question that this begs. It was not just the number of good groups that carried conviction: some were patently better evidence than others. How did a speculative additive producing one very frequently-used codegroup rate against another which produced a string of middling ones? As more and more Wrens joined the Big Room and there were big differences in their experience, there was an evident need to simplify the answering of that question, to standardise the judgements that had to be made right across the team and to enable new recruits to be initiated more quickly.
The answer that the mathematicians, with Ian Cassels in the lead, came up with applied Bayes’ Theorem to combine the weights of the evidence that each speculative codegroup provided, then used logarithms to substitute addition for multiplication and finally rounded the logarithms to two-digit scores which could be added mentally or by pencil jotting. This is explained more fully in Appendix VII.
The system performed very successfully its intended purpose of quickly and systematically testing the masses of speculative additives that differencing
produced, and picking out those that were probably true. It was never seen as doing the whole job. It could not help with horizontal stripping. Borderline cases, or messages of particular concern, could be handed over for more intensive study elsewhere in the party. And, as always, the imaginative hunch grounded in experience could often find significance where the arithmetic alone failed to.
Our party’s output of recovered additives went daily to the Processing Party. For them it was but a supplement to the much greater number of additives they received from Washington. Their central task was to apply all the known additives to all the messages enciphered by them, deciphering them to their codegroups, and to write in the codegroup meanings where known to reproduce the original message texts. They sent this deciphered material in two directions: by teleprinter to Bletchley Park’s outstations at Colombo and elsewhere, and to our book-builders and translators in Block B. They fulfilled these functions for other Japanese systems, and after September 1943 for all of them, not just JN-25.
The relationship between our cryptanalytic party and the Processing Party was essentially a logistical one, keeping the paper flowing smoothly. They did not concern themselves with the cryptanalysis, nor we with their routine deciphering, decoding and teleprinting. We supported each other well, however, in staffing, each recognising that we shared a chronic understaffing problem. There were occasions when we lent strippers (whom we could ill spare) to help the Processing Party to keep abreast of their essential work when the pressure was extreme, and other occasions when they could spare time to help us. Mrs Parsons was happy to recommend some of her particularly able Wrens (whom she had no means of promoting) for a trial run on stripping with us; and they always made good.
The Bletchley Park Codebreakers Page 17