The Code War
Next to the atomic bomb, they were probably America’s most closely guarded secret of the entire war. Even the submachine-gun-toting Marines who patrolled the compound in northwest Washington, D.C., were forbidden to set foot in the building that housed the machines. Of the 4,000 Navy officers, enlisted men, WAVES, and specially cleared civilian workers who every day passed through the double fence of barbed wire surrounding the Naval Communications Annex, only a handful were even permitted to know of the machines’ existence.
Had any outsider managed to penetrate the inner sanctum of Building 4, he probably would not have known what to make of what was there: A collection of odd-shaped devices built of gears and rotors and sprockets and lamps and dials, some looking like the second cousin of a movie projector, some like a cross between an outsized lathe and a telephone switchboard, and some sprouting masses of coiling wires like apparatus from Dr. Frankenstein’s laboratory. Officially the devices were known only as Rapid Analytical Machinery, or RAM, a bland name that gave away little, if anything, of what they were used for. Un-officially they bore mysterious and whimsical nicknames like Copperhead, Gypsy, Icky, Tessie, and Mike. But to the chosen few who understood their design and purpose, the machines represented the key to America’s highly secret efforts to break the most difficult codes used by its Japanese and German foes—and, most secret of all, to break the supposedly unbreakable codes used by its Soviet ally.
Within a few years of the end of World War II, all these machines would be rendered obsolete by the electronic digital computer. In terms of the ancestry of the modern computer, most of the RAM devices were evolutionary dead ends, and very expensive ones: $50,000 apiece, which could buy a fighter plane in 1942. They were highly specialized calculators designed for a single and very unusual purpose, and they pushed to its very limits the computational power that could be extracted from the technology of the early 1940s. Without the urgent needs of wartime, it is unlikely that they would ever have been built at all. Yet they did the job, and they often did it remarkably well.
The code breakers of World War II faced a challenge unequaled in the history of warfare. The adoption of radio by the world’s armies and navies, beginning in World War I and continuing in the interwar years, meant that tens of thousands of signals or more were being sent over the airwaves each day, bearing movement orders, supply lists, even detailed plans of attack. All of this would have been useless, of course, if the enemy could have read it. So some sort of code was an absolute necessity. Over the centuries hundreds of ways have been found to scramble and disguise messages. The crux of the problem has always been to come up with a method that is complicated enough to baffle the enemy but simple enough to permit rapid encoding and decoding by its intended users.
One of the easiest ways to disguise a message is the straightforward cipher, in which one letter replaces another: A becomes (for example) M , B becomes F , and so forth. The resulting message can then be sent in ordinary Morse code. Ciphers of this sort require a bit of work to solve, but not too much, which is why they can be found on newspaper puzzle pages. Even with the spaces between words removed (which is how encoded messages are generally sent), a simple, unchanging cipher would have provided little more protection than transmitting the unencoded message. Word-based codes, in which each string of numbers represents a complete word, are a little harder to crack, but if the same string always represents the same word, they can also be broken without too much difficulty.
The next step in complexity is to shift the cipher with each new letter, so that A might become M the first time it appears, K the second time, and so on. The rules for creating this sort of cipher may be embodied in a table of numbers or in an encoding machine with wheels and gears that can be adjusted to different initial settings. As long as sender and recipient have identical tables or machines (and can specify where to start reading off numbers or how to initialize the dials), they will be able to encipher and decipher their messages with little trouble. A third party will have great difficulty making anything of such a message unless he gets hold of the key or deduces it.
The very large number of enciphering patterns that can be generated by such machines suddenly made the code breakers’ job much more difficult in the 1930s, when they entered widespread use. Traditional pencil-and-paper-plus-intuition methods were helpless against them. Perhaps the most famous of these devices was the German machine known as Enigma, which had millions of possible settings, each of which would encode a message in a different way.
Clever deductive work by Polish code breakers before the war yielded a detailed understanding of how Enigma encoded its messages. Without knowing the day’s setting, however, there were still millions of possibilities to try. But the brilliant British mathematician Alan Turing devised a way to greatly reduce the number of possibilities that had to be considered if one could guess the meaning of part of the underlying message. Then he designed an electromechanical machine called the bombe that would rapidly rotate through tens of thousands of settings to find one that was consistent with that guess. Several hundred of these were built during the war and used in Britain and the United States.
Since the late 1970s much has come out about Enigma and the bombes. But there were many other cryptanalytic machines, developed solely by the American code breakers, that played an equally important role in the war. Although the most advanced of these machines came almost too late to help, others proved extremely important in breaking German diplomatic codes, Japanese naval and military codes, and Soviet KGB codes.
These were word-for-word codes, rather than letter-for-letter ciphers, as with Enigma. Each word or phrase was assigned a number, which was disguised by adding another number taken from a list known as an additive book. The resulting strings were then transmitted in Morse code. Some Japanese codes were based on the kana, the 48 syllables that make up the Japanese language. A special Japanese Morse code was used in which each kana syllable is given its own unique pattern of dots and dashes. In 1928 the U.S. Navy began a secret training program that would eventually teach 176 enlisted men Japanese Morse. The three-month course took place in a 20-by-20-foot concrete blockhouse perched atop the old Navy Building in Washington. To get to class each morning, the students had to climb a ladder to the roof. They quickly started calling themselves the On-the-Roof Gang.
One weakness of disguising your code with numbers drawn from a list is that eventually you will reuse the same group of numbers (known as a stretch of key). Once that happens, clever cryptanalysis can go a long way toward extracting the message, and eventually toward cataloguing long sections of the codebook to simplify future operations. This process of accumulating knowledge about the encoding key is known as building up depth.
The main problem with this approach during World War II was that while enemy codes did repeat stretches of key, they did not do so very often. If an additive book has 30,000 numbers, for example, and a typical message is 15 words long, the probability that the stretches of key used to encode two messages will overlap by even a single number is about 1 in 1,000. Even if they do overlap, there is no guarantee that the overlapping portions will have any words in common, and without that, cryptanalysts have very little to work with. Finding those rare cases of overlap required comparing enormous numbers of messages. Fortunately, the demands of war gave cryptanalysts plenty of material to work with. That’s where the machines came in.
The U.S. Army and Navy had established their code-breaking offices back in the 1920s. They were tiny, poorly funded, and poorly equipped, and they operated in constant fear of being discovered—not so much by America’s potential enemies abroad as by opponents within the U.S. government. The State Department, in particular, objected to snooping on the communications of fellow diplomats as unethical at a time when the nations of the world were attempting to negotiate in good faith to limit arms and prevent war.
In 1934 the work of the code breakers actually became illegal, as the Communications Act of that year made it a crime for anyone to intercept and divulge the contents of private radio messages. The Army’s Signal Intelligence Service (SIS) was accordingly hidden away within the Signal Corps; officially it did not even exist. For most of the early 1930s the SIS consisted of its chief, William F. Friedman, plus three young mathematics teachers he had hired in 1930. The Navy’s code-breaking operation, opaquely designated OP-20-G, was not much larger and was, if anything, even more thoroughly buried in the obscurity of the Navy bureaucracy. It was hard to lobby for funding when you couldn’t even acknowledge your existence.
In the early 1930s Friedman and several of his Navy counterparts, most notably Lt. Joseph Wenger and Lt. Thomas Dyer, realized that IBM punch-card machines could automate some of the tedious operations involved in breaking codes. These operations included, for example, making counts of how often various words or combinations of letters were used in different languages. The resulting statistics were helpful in decoding foreign texts, just as it helps to know that E is the most common letter in English.
IBM machines were the closest thing to a general-purpose computer available at the time, but they were often cumbersome. Some code-breaking operations required millions of IBM cards. The machines could sort cards according to various criteria and perform simple comparisons between pairs of them, but sifting through vast amounts of data looking for repeated occurrences of code groups—a staple Operation in cryptanalysis—was an extremely slow process. Beyond a few counters in the tabulators, the machines had no real memory devices. In the hunt for repeats, the best that the IBM machines could do was to print out catalogues of code groups sorted into alphabetical or numerical order. These printouts would then have to be scanned by eye to hunt for repetitions.
A final drawback was that the machines were very expensive. IBM never sold its machines, even to the government; they could only be rented, and with a near-monopoly the company could charge what it liked. A printing tabulator went for $1,500 a year, and sorters and punches rented for $300 to $720 a year, so acquiring one of each—the basic data-processing setup of the 1930s—cost more than a full-time mathematician’s salary. By 1940 the Army’s SIS still could afford to rent only six IBM machines.
The Navy was slightly better off. Samuel Snyder of SIS recalled making regular trips from the Munitions Building to the adjacent Navy Building to beg stacks of fresh IBM cards from OP-20-G’s better-equipped machine room. The Navy had an unusually technical-minded captain as the head of its Office of Naval Communications, Stanford Hooper, who believed the Navy needed to forge much closer ties to industry and academia. Today that view is commonplace, but at the time it was revolutionary.
In the mid-1930s Hooper began meeting with Vannevar Bush of MIT to explore how the Navy could benefit from the latest in scientific research. Bush subsequently became the leading force in mobilizing scientists in support of the war effort. Bush was trying to develop a general-purpose computer that would use high-speed paper-tape readers to enter data, and fully electronic counting circuits to perform calculations. He knew that a device incorporating some of these principles could greatly speed up the task of combing through hundreds or thousands of coded messages. Bush somewhat grandly proposed that the Navy pay him a consulting fee of $10,000 to develop the idea.
OP-20-G was interested, but Navy Department bureaucrats were appalled. The Bureau of Ships, which handled all Navy contracting, wanted nothing to do with a college professor who demanded outrageous fees and was not even proposing to deliver a piece of equipment. Finally, in January 1937, a scaled-down contract was approved. Bush promised to build an actual machine and give it to the Navy at no cost except for shipping. Almost five years later, in December 1941, the device arrived in a huge crate at the Navy Building. In the confusion that followed Pearl Harbor, it sat around for three months gathering dust before Wenger finally assigned a lieutenant to get the thing working.
It was not easy. Bush’s Comparator and the later-generation machines it spawned looked like movie projectors gone slightly mad with sprockets and pulleys. The basic idea behind all the Comparator-type machines was to hunt for repeated strings in a pair of messages by feeding two loops of tape or photographic film simultaneously through a pair of scanning heads. The tape was punched with holes (or the film was marked with dots) that represented the encoded message. Photocells in each scanning head detected the holes or dots, and an electronic circuit compared the two outputs. If they matched, the machine would record the position of the match or (depending on what task the machine was performing) add it to a running tally or stop the tapes or film so that an operator could note manually where the coincidence had occurred.
There were many technical problems. Paper tapes tended to stretch, and the punches did not always work right. With photographic film or plates it was hard to align the dots with precision. But as these and other kinks were worked out over the following months, the machines played an increasing role in breaking some of the most difficult problems in cryptanalysis.
With a whole bank of photocells in the scanning heads, several rows on each tape could be read simultaneously, allowing a search for repeats that were several letters or groups long. The tapes were mounted as closed loops, and after both had gone around once, the first tape would be advanced one position and the process would begin again. That way every word in one message could be compared with every word in the other. Other variants of the machines operated on one tape at a time, searching for a specific, fixed sequence of cipher groups.
The Comparator, and other machines like it, made actual digital comparisons, using light or electronic devices to count the number of matches between one message and another. A second class of machines performed analog calculations, notably the IC machine, which played a pivotal role in breaking ciphers used by Japan’s naval attachés. The IC machine worked on poly alphabetic ciphers (shifting letter-for-letter substitutions of the sort that were used in Enigma) as well as the syllabic, kana-based ciphers used in some Japanese military systems.
While the mathematics behind this work is complicated, the principles can be readily understood. One of William Friedman’s 1930 mathematical recruits was Solomon Kullback, who in 1935 wrote the definitive theoretical analysis of poly alphabetic ciphers. Kullback considered what happens when you compare two messages that have been enciphered with the same stretch of key, whether that key is printed on paper or hard-wired into the rotors of a machine.
Suppose message 1 is printed on a strip of paper and placed above message 2. If the messages are aligned randomly, the frequency of matches—cases in which a letter in message 1 appears directly above the same letter in message 2 —should equal its random value, 1 in 26, or about 3.8 percent. But suppose the two messages are aligned so that a certain stretch of key kicks in at the same point in both—in other words, so that the letter encoded with a given additive in message 1 is above the letter encoded with the same additive in message 2. A pair of messages aligned this way are said to be “in depth.” In this situation the frequency of matches, known as the index of coincidence (IC), jumps to a higher value: the probability that two letters chosen randomly from a piece of English text (as opposed to two letters chosen randomly from the alphabet) will match. This value is equal to about 6.7 percent.
Therefore, if you take two messages that were encoded (in whole or in part) with the same stretch of key, place one on top of the other, and shift them until you get a high index of coincidence, you will have succeeded in placing them in depth. The trouble is that most pairs of messages do not use the same stretch of key, and the process of finding out whether they do or not is tedious in the extreme. This is why automating such comparisons was a high priority.
IBM machines could do the job, but the need for repeated sortings made the work slow and laborious. Bush-type machines, however, were tailor-made for determining indices of coincidence. In the IC machine, messages were recorded with clear dots on opaque plates. Each letter of the alphabet had an assigned position along the width of the plate. To run a comparison, one plate was placed on top of another, and a light was shone through. The position of the plates would be shifted, one letter at a time, and if the intensity of light transmitted through the pair of plates jumped, the cryptanalysts knew they probably had a match. In this analog device a single photocell measured the total light transmission through the pair of plates, eliminating the need for electronic counting circuits to carry out a row-by-row digital comparison.
Several other RAM devices used similar methods on a different problem: searching for depth in high-level Japanese military and naval and German and Soviet diplomatic codes. These were true codes, as opposed to ciphers, in the sense that code groups stood for whole words or phrases rather than individual letters. When cryptanalysts work on this type of code, finding the same string—say, 7364—in two messages is a mildly promising sign. It may indicate that the same word has been encoded with the same additive in both places, but it may be two different words with two different additives that happen to add up to the same value.
Much more promising, and correspondingly rarer, is a “double hit”: a pair of messages in which two identical numbers are repeated the same distance apart. For example: 3419 2100 7364 5642 9468 2316 7364 7130 0072 2316 0924 7464 when realigned as 3419 2100 7364 5642 9468 2316 7364 7130 0072 2316 0924 7464 yields a double hit. Such an occurrence is an almost certain sign that the two messages are in depth. The challenge in mechanizing such searches is that the telltale repeats can be spaced any number of groups apart.
During the war, the Army’s SIS worked out an IBM-based punch-card method to conduct such searches; it was known in the business as “brute forcing.” Frank Rowlett, another member of Friedman’s original “gang of three” from 1930, specialized in devising relays that could be wired into the tabulators and sorters to make such operations more efficient—in violation of IBM’s strict policy against customers opening the service panels of the machines. It was standard procedure for Rowlett to send an assistant to case the machine room to make sure the IBM service representative was not on-site. If the coast was clear, he and the others would descend on the machine, screwdrivers and soldering irons in hand, to install their latest invention.
A variation on this concept was the Slide Run machine, which was used to build up depth once a fragment of additive sequence had been recovered. First the cipher text from thousands of messages was punched on IBM cards. Then, using relays, the Slide Run unit automatically subtracted from every message a known sequence of five additive groups taken from this fragment. It compared the resulting code groups with a “library” of the 250 most commonly used groups in the codebook. If the code, after being stripped in this way, produced two or more matches with the library of common code groups, the machine would print out the results.
Several of the Comparator-type machines used light to perform this search for repeats more efficiently. The method of operation was somewhat similar to that of the IC machine, though these machines were looking for specific matches of individual words rather than a general index of overlap. One message would be represented with clear spots on opaque film, while a second would consist of opaque spots on clear film. (A message could be converted between these formats by simply making a photographic negative.) One of these films was placed on top of the other, and light from a photocell was shone upon them, one pair of numbers at a time, as they were run over a scanning head. The Navy’s Copperhead machine used a similar scheme, but with punched paper tape instead of film.
If the numbers did not match, there would be places where both films were opaque and places where both were clear. Therefore some light would shine through. But if the numbers on the two strips of film were identical, one would be opaque wherever the other was clear and vice versa. No light would shine through, and analysts would know they had a match. In this way a single photocell could examine a pair of messages for repeats step by step. To search for multiple repeats, as many as 100 cipher groups could be scanned simultaneously with 100 photo-cells. If two or more positions were blacked out, a double hit was registered.
Some machines could make thousands of comparisons per second as the tape or film zipped through them. The paper-tape speed record, however, was set by the British Heath Robinson and Colossus machines, built to help break the German teletype ciphers. A developer observed that one of the greatest secret inventions of the war was the discovery that ordinary teletype tape could be run at 30 miles per hour without tearing. It did, however, tend to stretch, which caused serious problems of alignment when two tapes were being compared. In Colossus this synchronization problem was solved by storing the data from one of the tapes electronically in an internal vacuum-tube memory. Thus, while Colossus was not a computer or even a direct ancestor of the computer, it did pioneer the use of fully electronic memory, a crucial component of the modern digital computer.
One of the odder machines, and one that underscored the inadequacies of the era’s technology for recording multiple channels of data, was the Navy’s Mike. This machine read two punched tapes simultaneously and tallied how often each of the 676 possible alphabetical bigraphs—combinations of two letters—occurred. It did this with a 26-by-26 bank of mechanical dials. When a run was completed, the most efficient method of recording the data was to photograph the dial board so the machine could be promptly reset and readied for another run. The photographs would then be developed and the data transcribed manually.
The ability of IBM and RAM equipment to make millions of comparisons automatically was the key to some of the most dramatic cryptanalytic successes of the war. IBM methods were indispensable in breaking JN-25, the main Japanese naval code. JN-25 signals revealed Japanese plans to attack Midway in June 1942, allowing two U.S. carrier task forces to move into position first and ambush the ambushers. A broken JN-25 message also gave away Adm. Isoroku Yamamoto’s itinerary for an inspection tour of the Solomon Islands in April 1943. His airplane was intercepted by 16 American P-38 fighters and shot from the sky.
To operate the machines, the Navy and Army recruited thousands of workers, most of them young women. The Navy scoured elite women’s colleges for recruits to handle mathematical tasks. After some reluctance, it also turned increasingly to women to perform routine operations and maintenance on the machinery itself. By February 1944 the staff of OP-20-G totaled 3,722, of whom 2,813 were WAVES (Women Accepted for Volunteer Emergency Service).
WAVES barracks were built across Nebraska Avenue from the Naval Communications Annex on a site that now includes part of the grounds of the Japanese Embassy. The Naval Communications Annex itself was a former girls’ school, Mount Vernon Seminary, that the Navy had seized in 1942, but other Navy facilities were not always set up to accommodate such a large number of women. Some of the first barracks had been built for men and were switched to women at the last minute. The women adapted to their new quarters with aplomb, decorating the urinals in the lavatories with potted geraniums.
The analysts’ work could be mind-numbingly boring, and it required absolute accuracy. An official history of the WAC (Women’s Army Corps) unit assigned to SIS after the war said: “It was proven over and over again that women were far better equipped than men for routine but detailed work.” The job also, of course, required absolute secrecy. Decades later the WAVES who had worked at the Naval Communications Annex vividly recalled the indoctrination lecture they had been given on arrival. They were ushered into the old school chapel and told that if they ever so much as hinted to anyone—their closest family members included—what they did, they would be put up against a wall and shot. For years afterwards many did not even tell their husbands what they had done during the war.
Decoding became considerably easier after Australian troops captured a complete set of Japanese army codebooks on New Guinea in January 1944. By this time SIS had also broken the indicator system, which told where in the additive book to start. From then on, the process of breaking Japanese army signals was completely automated. Radio teletype transmissions containing the intercepted Japanese signals were received at Arlington Hall, the new wartime headquarters of the SIS just outside Washington (which was also a former girls’ school). The teletype tape was fed into a machine that automatically converted the data to IBM cards, one card per code group. This stack of cards was fed through a series of sorters that matched the text against a library of cards on which the entire captured additive book had been punched. Starting at the proper place, the additives were automatically subtracted from message text, and the resulting code groups were punched onto cards. That deck was then sorted against a second library of cards containing the code groups and their plain Japanese meanings. The IBM printers then printed out the fully decoded text of the message.
The end result was a stack of cards that contained the broken message. There were many times when Arlington Hall read a Japanese message before its intended recipient, who had to look everything up by hand, could get it decoded. In the month after the New Guinea codebooks were in hand, SIS broke 36,000 Japanese army messages, more than a thousand a day. By the end of the war, as many as 2,500 Japanese army messages a day were being automatically decoded.
IBM and RAM methods were combined for what may have been the most astonishing cryptanalytic feat of the war. The extension of Lend-Lease to the Soviet Union following Hitler’s attack along its western border in June 1941 had brought about a huge increase in the Soviet presence in the United States—trade missions, purchasing commissions, inspectors stationed at factories filling Soviet orders. These representatives regularly communicated with Moscow via coded messages sent over commercial cable, which the SIS (renamed Signal Security Service in 1942 and Signal Security Agency the following year) routinely intercepted. It soon became apparent that five different code systems were in use. One, which made up about half of all the telegrams, clearly dealt with commercial matters, as they came from places where Soviet purchasing missions were located. The other four appeared to be purely diplomatic, originating at the embassy and consulates.
It had been known for decades that the Soviets were using a four—or five-digit code enciphered with a one-time pad, a system that in theory was unbreakable. A one-time pad works like a regular additive book except that each page is used just once and then destroyed, making the additive book, in effect, infinitely long. With no overlaps or duplication between messages, there is no way to build up depth. One-time pads offered a nearly perfect system of security, except that they would have had to be printed in impossibly large numbers to handle the volume of wartime communications. Thus, countries tended to reserve one-time pads for their most secret messages.
Lacking anything better to try, Richard Hallock, a member of the SIS team assigned to the Russian section, decided to begin a brute-force search of the Soviet trade traffic. Starting in the fall of 1943, he had the SIS machine branch punch the first five groups from 10,000 trade messages onto IBM cards and list them in numerical order. Further analysis was done by hand, and when all the sifting was done, Hallock had found seven double hits. That, of course, could easily have been the result of pure chance. But as Hallock and his team began working through the seven pairs of overlapping messages, they found to their surprise that the one-time pads were sometimes more like two-time pads. Every now and then a page had been reused.
Work progressed slowly through the following winter and spring. In July 1944, while the main effort continued on the trade code, Cecil Phillips, a 19-year-old college dropout from Asheville, North Carolina, was placed in charge of the rest of the Russian problem. It was a remarkable vote of confidence in Phillips, who had shown an extraordinary talent for cryptanalysis in his one year on the job, but it was also a statement of how little hope for success there seemed to be. Phillips sifted through the traffic and did not get anywhere until, in November 1944, he looked at one batch of messages from New York to Moscow. Dimly, through the sea of meaningless four-digit numbers, a shape began to loom.
An enciphered code should produce a purely random distribution of digits, but the first cipher group in each message Phillips was looking at was not random; the digit 6 appeared twice as often as it should have. That seemingly inconsequential discovery broke the problem wide open. The first group, it turned out, was an indicator; it consisted of the additive group found at the top left-hand corner of the one-time pad page being used. (The excess of 6’s was apparently just a fluke that resulted from a slight bias in the mechanical printing equipment the KGB had used to generate the pads.) From this Phillips quickly discovered that one-time pads that had been used in the trade code were also being used in the diplomatic codes.
The IBM and Copperhead machines were pressed into service to search for double hits in several thousand messages sent on the Soviet diplomatic systems. Over the next few years the Russian section searched through 750,000 telegrams and found 30,000 pages of reused one-time pad. It would later become clear that for a few months in early 1942 the KGB, under pressure of wartime shortages, had printed duplicate copies of pages and bound them, often with different page numbers, into separate one-time pads.
It also became clear that while one of the four diplomatic systems was being used for legitimate consular business, the other three were assigned to the KGB, Soviet military intelligence, and Soviet naval intelligence. The 2,900 messages transmitted from 1940 to 1948 that were eventually read by Arlington Hall—some as late as the 1970s—revealed a determined Soviet espionage operation in the United States. The cables showed that the Soviets had penetrated the Manhattan Project at multiple points, with at least four agents at Los Alamos reporting to a network controlled from New York. The cables pointed to highly placed and well-informed sources in the War Department’s general staff, in the OSS (predecessor to the CIA), and in the British Embassy in Washington. They told of a ruthless KGB operation to hunt down, kidnap, and return to the Soviet Union deserters from the Soviet merchant fleet living on the West Coast of the United States.
Within a few years of the war’s end, the Navy and Army code breakers, working together in a unit that would become known as the National Security Agency, would build some of the first programmable electronic digital computers. The RAM machines, representing the last gasp of the pre-electronic age, thousands of times slower and much less flexible, were dismantled and destroyed. Yet at the time they were built, they did what no other machines could do. With- out them the nearly impossible task of breaking enemy codes during wartime would surely have failed.
