FEATURE ARTICLE: PROGRAMMABILITY

Colossus and
Programmability
Thomas Haigh We analyze the capabilities of the Colossus codebreaking
University of Wisconsin
devices, built in 1943–1945 under the direction of
Milwaukee, Universitat
Siegen Tommy Flowers of the UK General Post Office.
Mark Priestley Colossus is often described as a programmable
Independent Scholar computer, a misconception we trace to old battles
about the “first computer” and to former secrecy about
its actual capabilities. In fact, Colossus was not called
a computer at the time, and does not meet later definitions because it carried out no
mathematical operation other than counting. Colossus automatically executed a program,
i.e., performed series of discrete operations, but it was not programmable because that
program could not be fundamentally modified by its users. Instead of the old focus on
allocating “firsts,” we argue for new perspectives on Colossus focused on its use and
impact, its relationship to other early machines, and its place in the history of digital
communications engineering.

The original Colossus was built by the British General Post Office at the end of 1943, under the
direction of telecommunications engineer Tommy Flowers, to assist in British attacks on certain
German codes. Although it was recently honored on a Royal Mail stamp as “world’s first electronic
digital computer,” its place within the history of computing remains ambiguous. Colossus is also often
said to have been a “programmable electronic computer,” again typically in order to claim it as the first
such computer. Despite its fame, accounts of what Colossus was and what it could do are contradictory
and no clear description of its control capabilities has been published. This is in large part the result
of residual secrecy that clung to Colossus during the early 1980s when the nascent history of
computing community hashed out the allocation of “firsts” among early machines before moving on
to more productive matters.
In this paper, we present a clear and concise account of the architecture and control capabilities of
Colossus, grounded in primary sources declassified in recent decades. We are particularly concerned
with the concept of programmability, which has never been properly defined in the history of
computing literature. To establish broad and historically grounded definitions of “program” and

IEEE Annals of the History of Computing Published by the IEEE Computer Society
October/December 2018 5 1058-6180/18/$33.00 ß2018 IEEE
 IEEE ANNALS OF THE HISTORY OF COMPUTING

“programmability,” we looked at the contemporaneous emergence of the idea of “programming” in

the ENIAC project and at the everyday use of “program” in other contexts. This led us to a sense of
“program” as a series of operations carried out by a computer, which is applicable to Colossus even
though these operations were embodied in its structure rather than being encoded symbolically as
instructions. For this reason, the program of operations executed by the machine could not be
fundamentally changed by users. We argue that it was not, therefore, programmable. We also challenge
the idea of Colossus as a computer, which we suggest has more to do with a need to justify its
importance in the context the discourse of the 1970s than with its actual capabilities. Despite making
extensive use of digital electronics Colossus fits neither 1940s definitions of the term nor more modern
ones. Statements made by Flowers himself, and some of those who worked with him, made more
nuanced characterizations, for example, as an “electronic processor.”
Our conclusion that Colossus was neither programmable nor a computer, reached rather to our own
surprise, does not diminish its historical importance.1 Rather, the special pleading previously
undertaken to shoehorn Colossus into the role of programmable computer reflects the concerns of the
history of computing community in its early days, and indeed the limitations of “history of computing”
as an analytical frame. Colossus fits within that frame only to the extent to which it was a computer.
Today, as our reliance on digital communications grows and computers vanish from view into digital
devices and data centers, we are better able to appreciate this remarkable machine on its own terms.

SITUATING COLOSSUS
Dozens of unique electronic and mechanical computers were built during the 1940s. A handful, such
as the Harvard Mark 1, ENIAC, and EDSAC have clear and prominent places in the history of
computing. They consistently appear in overview histories, such as Martin Campbell-Kelly and
Willian Aspray’s Computer and Walter Isaacson’s The Innovators, in television documentary series,
and in comprehensive museum exhibitions such as those at the Computer History Museum and the
Heinz Nixdorf Museums Forum. Each is remembered as the “first” machine to reach one or another
historical milestone, agreed upon after a long and messy battle to anoint the “first computer.”
The historical place of Colossus is less clear. Most other pioneering computers were publicized during
their operational lifetimes. ENIAC, for example, was announced to the world with a front page story
in the New York Times and installed in a showpiece facility where it was frequently displayed for
visitors.2 The Colossus machines were designed in secret, deployed as a vital part of one of the war’s
most militarily sensitive operations, and kept confidential for decades afterward. From the 1940s to the
1970s, as teams of lawyers gathered records concerning other early machines and subjected their
designers to repeated rounds of deposition and testimony, those responsible for Colossus remained
quiet about the machine’s capabilities and even its existence.
The word of Colossus began to spread in the 1970s after Brian Randell, a computer scientist with an
interest in the early history of electronic computing, gathered testimony from veterans of the project
and persuaded the U.K. government to acknowledge its existence. In 1976, he shocked the computer
pioneers at a seminal computer history meeting at Los Alamos National Laboratory with news that
“a series of programmable electronic digital computers was built in Britain during World War II, the
first being operational in 1943.” He characterized Colossus as “a special-purpose program-controlled
electronic digital computer” that could “most aptly be compared” to ENIAC in its flexibility and
programming method.3
Particularly in Britain, the high public profile of Colossus comes in large part from its connection with
the work of Bletchley Park, which has become one of the most celebrated facets of the war effort.
Colossus even makes a brief appearance in Cryptonomicon, Neal Stephenson’s hugely popular novel
of cryptography and the wartime origins of information technology. For this reason, Colossus is more
often considered as part of the history of codebreaking than of the history of computing or of
telecommunications. It was recently honored with a postage stamp (see Figure 1).
Within the history of computing, Colossus is celebrated by a small community of enthusiasts, some of
whom view it as the most important of all the early computers. For example, Jack Copeland has
claimed that if the Colossus machines had been preserved as “the heart of a scientific research facility”

October/December 2018 6 www.computer.org/annals

 PROGRAMMABILITY

Figure 1. Colossus commemorative stamp, issued by the Royal Mail in 2015, calls it the
“world’s first electronic digital computer.”

then “the Internet—and even the personal computer—might have been developed a decade or more
earlier.”4 Fans of Colossus, like those of other early computers, sometimes strike a partisan tone.5
Copeland, for example, attributes to Flowers a remark that “Colossus was ‘much more of a computer
than ENIAC,’” which was “just a ‘number cruncher.’”6 Tony Sale, who devoted his retirement to the
construction of a new Colossus to replace those disassembled at the end of the war, was motivated by
a sense that “for far too long the Americans have got away with the myth that the ENIAC was the first
large-scale electronic digital calculator in the world.” Following the completion of the reconstruction,
he claimed, “There has been a stunned silence from across the water.”7
It is far from clear that the silence noted by Sale is the result of stunned acquiescence to his claims.
Colossus has been marooned between the praise of boosters, who believe it to have a singularly important
place in the history of computing, and the scholarly authors of overview histories of computing, who
have politely ignored these claims. In Computer, Campbell-Kelly and Aspray note merely that Bletchley
Park’s work on mechanical devices to attack the Enigma code “was followed by an electronic machine,
the Colossus, in 1943,” as a result of which several people who would later work on computer projects
were exposed to electronic technologies.8 The other standard scholarly history of computing, Paul E.
Ceruzzi’s A History of Modern Computing, does not mention Colossus at all.9
The current situation, then, is one in which Colossus is largely ignored by historians shaping broader
narratives on the emergence of modern computing but lavishly, if sometimes shrilly, praised by its fans
and increasingly embraced by the British public as a symbol of national greatness. This reflects an
enduring vagueness about what Colossus actually did. Randell dug up an impressive amount of
information, but without access to original documents his account was unavoidably speculative.
Randell’s revelations led to considerable interest in Tommy Flowers, who gave several public talks and
interviews in the 1970s and published his own technical article on Colossus and its history in 1983.
Yet even Flowers was working from memory, and his description turned out to have several significant
historical and technical inaccuracies.10 More recent investigations have provided masses of detail but
not a clear and concise overall description of the architecture and capabilities of Colossus separate
from the codebreaking applications for which it was designed, or of the methods by which it was
prepared for use.11 To bridge this gap, we have gone back to the primary sources to clarify what
Colossus did and to probe its similarities and differences with early electronic computers.

WAS COLOSSUS A COMPUTER?

There are two senses in which a historical object might be considered to be a computer. One is if
people at the time called it a computer, even if definitions have subsequently shifted. That would apply,

October/December 2018 7 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

for example, to the human computers in Philadelphia calculating firing tables for the U.S. Army in
1942, or the analog torpedo computer used in submarine warfare.12 Historians call a term used in
original sources an “actors’ category” and would typically look at when people first started using it,
and how its meanings were contested and changed over time. The second sense in which we could
interpret the term is to ask whether an object, though not necessarily called a computer at the time,
would meet a later definition of computer. For example, Babbage’s analytical engine would, if
constructed, have had comparable capabilities to some of the computers of the 1940s. Historians
refer to a term used in this way as an “analysts’ category.” This requires us to endorse a particular
definition of the term, or formulate a new one, and apply it consistently across time regardless of
actual historical usage.
Colossus was not a computer in the first sense. Only many years after the Colossus machines were shut
down did anyone begin to call them computers or even calculators. This sets Colossus apart from most
of the pioneering electronic computers of the 1940s, which were usually named either as computers (the
“C” in ENIAC stood for “Computer” as did the “C” in EDVAC) or calculators (the “C” in machines such
as EDSAC and IBM’s SSEC). This is because the new machines replaced the labor of humans, whose
job title was “computers.”13 They were often known as “automatic computers,” just as machinery that
could direct the flying of a plane was called the “automatic pilot” because it carried out some of the tasks
of a human pilot. In fact, Colossus was sometimes called a “counting machine” by its users.
Neither was Colossus a computer in the second sense, as it does not meet any plausible subsequent
definition of a computer. Computers carried out lengthy mathematical tasks, often involving thousands
of individual mathematical operations. This drew attention to their ability to move from one operation
to the next without human intervention. George Stibitz built a series of pioneering tape-controlled
computers at Bell Labs during the 1940s. His 1945 definition captures the contemporary
understanding of a computer as something able to perform automatically a sequence of operations
(“some or any of” multiplications, divisions, additions, and subtractions), storing the intermediate
results from earlier operations so they could be further manipulated by later ones.14 More complex
operations, such as square roots, logarithms, and trig functions, were handled in some early machines
with special hardware and in others by specifying the appropriate sequence of elementary operations.
Many of these machines were designed with table making in mind and made it easy to compute a
result for one set of parameters after another by constantly repeating the same processes.
Colossus, unlike these other machines, was not built to carry out numerical computations and we know
of no evidence that it was used to carry out calculations, or could have been usefully applied to them.15
Flowers did not see Colossus as a computer and was never particularly interested in computing,
despite the popular misconception that he was keen to work with computers after the war but was
somehow thwarted. His career trajectory confirms that electronic telephone exchanges, not computers,
were his passion.16 Flowers eventually left the Post Office when promised he could develop his
exchange technology elsewhere—a conspicuous contrast with his decision to remain there after work
stalled on the ACE computer he was supposed to be building for the National Physical Laboratory
shortly after the war.17
Flowers himself was hesitant to call Colossus a computer. When, in 1977, he gave one of his first
public talks about Colossus, after news of the machine had begun to reach the public, he related that
“it is now said that during the Second World War I was responsible for the production of the world’s
first electronic digital computer,” yet cautioned that “if so, that was an accident incidental to the
solution of a problem.”18 Flowers complained that telecommunications companies had focused on
using general-purpose computers to control the switches connecting together telephone lines. He had
spent his career urging that electronic “processors” should be used to replace electromagnetic switches,
rather than control them, so that the “structure” of the exchange would itself become electronic.
Flowers noted that claims for Colossus as the first computer came “to the surprise of those concerned
who thought of it as just another processor” or “a new-fangled processor,” seeing it as continuous with
his work before and after the war on electronic telephone exchanges.
Perceptions of what people did and why it mattered change over time, even in their own minds.
Flowers showed remarkable restraint in continuing to nuance his language decades after others won
him recognition as an inventor of the computer. The closest he came to claiming Colossus as a
computer seems to have been this 1983 passage: “Colossus had features now associated with digital

October/December 2018 8 www.computer.org/annals

 PROGRAMMABILITY

computers—semipermanent and temporary data storage, arithmetic and logic units including
branching logic, and variable programming—that may justify its being regarded as the first digital
computer.” In the rest of that paper Flowers consistently calls Colossus a “machine” rather than a
“computer.”19
Flowers stuck with this position to the end of his life, even as he heard and saw others claim Colossus
as the first computer. Even his posthumously published chapter “Colossus” opened with the sentence
“Machines such as counters, computers, and Colossus process information.” This positioned Colossus
as related to both counters and computers, but not as itself a computer. In the rest of this paper, he uses
the word “machine” rather than “computer” when talking about Colossus.20 In another posthumous
publication, “D-Day at Bletchley Park,” Flowers consistently used phrases such as “electronic machine”
and “processor” to describe Colossus and attributed to others the idea that Colossus was a computer:

[A]cademics interested in the history of computing have recognized that Colossus was the
world’s first electronic computer. It was not designed as a computer: computers had not yet
been invented. It resembled a modern computer about as much as George Stephenson’s
Rocket locomotive of 1829 resembled the Royal Scot and other steam locomotives of the
twentieth century. The basic technology used in a modern computer—data storage and
retrieval, ultrafast processing, variable programming, the printing out of the results of the
processing, and so forth—were [sic.] all anticipated by Colossus, some of it by as much as
ten years.21
Even here Flowers remains reluctant to call Colossus a computer or himself the inventor of the
computer, though he does, properly, claim credit for the development of many of the digital electronic
techniques later used to build computers.22

COLOSSUS PROGRAM
The term “program” can be treated either as an actors’ category or an analysts’ category. We know of
no evidence that the word “program” was applied during the mid-1940s to any part of Colossus, or to
anything it did. It was not an actor’s category at Bletchley Park or Dollis Hill, though on the other side
of the Atlantic the term was being adopted by members of the ENIAC Group at the University of
Pennsylvania. Thus, we use “program” here as an analysts’ category, imposing our own definition.
However, we attempt to do this in a way that is sympathetic to 1940s usage, not just with regard to
automatic computers but also in other areas.
To begin with a conclusion: Colossus did execute a program. We can describe that program using a
flowchart (see Figure 4). Many years later, Harry Fensom, a senior member of the team that designed
Colossus, reconstructed from memory the series of human and automatic actions it took to guide
Colossus through a typical run. As he mentioned, “One panel of Colossus contained the so-called
‘master control.’ This acted as a program sequencer, guiding the run through all its steps, from switch-
on, to print-out, and then on to the end of the run. Flowers designed the routine, or program, carried out
by the master control, using a timing diagram and logic diagrams that had almost a modern flavor.”23
Before proceeding, we need so say a little bit about what Colossus did. It was designed to attack the
teleprinter encryption produced by a German device, known as the Lorenz SZ40 by the Germans and
codenamed “Tunny” by the British. This held 12 rotating code wheels, each studded with a different
number of configurable pins. With each new character of the message, some of the wheels turned to
their next positions. The encryption process is shown in Figure 2.
Colossus was designed to dramatically speed one particular part of the code-breaking process, shown
in Figure 3 as “chi wheel setting.” This was the most time-consuming of all the tasks to accomplish
with manual methods. Decrypting the message required knowledge of the bit patterns set on each
wheel and the position to which each wheel should be turned at the beginning of the message.
Unencrypted text is made of words and, as any Scrabble player knows, the distribution of letters in
natural language words is highly irregular. In a well-encrypted bitstream, however, all codes are
equally likely. A simple statistical test on decrypted text could easily identify the correct set of wheel
start positions, assuming the pin settings were already known. But someone who sat down to try
decrypting every possible combination of start positions for the 12 wheels would still be working

October/December 2018 9 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

Figure 2. Logical representation of the action of the Lorenz machine, dubbed “Tunny” by the
British. The gap of four characters shown between the chi and psi wheels is to symbolize the
idea of two logically independent transformations applied to a five channel bitstream, rather
than a representation of the actual inner working of the machine.

when the war finished, and indeed when the earth was swallowed up by the sun. Colossus was
designed to exploit a subtle flaw in the design of the Lorenz equipment used by the Germans. Because
of this flaw, it was possible to obtain statistical evidence of correct wheel settings by looking only at
correspondences between two of the chi wheels and two of the five bit channels that composed the
intercepted message. This method, devised by mathematician Bill Tutte, involved comparing the deltas
(changes) between successive bits generated by the code wheels and read from the intercepted
message tape. Setting the first two chi wheels involved 1271 trial decryptions of the message with all
possible start positions. In the best case, two more runs (or in the case of the later versions of Colossus,
three runs carried out in parallel) would identify the start positions of the other three wheels.24
Making thousands of complete trial decryptions for each message remained almost as impractical for
cryptographers without electronic assistance. With Colossus, however, they could get the job done in
less than an hour. The message was read again and again from a loop of paper tape rotated at up to
5000 characters a second, while the machine automatically tallied similarities between the deltas
generated by the code wheels and read from the tape. Once the message tape completed a revolution it
reset the counters, incremented the start position of one of the code wheels, and started again. To speed
operation, Colossus used electronics to simulate the revolution of the code wheels, rather than the
physical cogs of the real Lorenz machine.
Colossus switched between operations based on the interaction of its control circuits with the contents
of the message tape. A special code punched at the end of the message triggered control signals to reset
its counters and, if a predefined threshold had been reached, to print the code wheel settings being
evaluated and the counts obtained. Each message was followed by a blank sequence in the tape, which
gave Colossus time to increment the uniselectors holding the code wheel settings currently being
scored. These settings were then used to fix the positions of the electronic code wheels, so that when

October/December 2018 10 www.computer.org/annals

 PROGRAMMABILITY

Figure 3. Overall work flow needed to produce decrypted Tunny messages. This shows a
typical configuration as for late-1944, in which Colossus was to set and verify the start position
of chi wheels for each message and to initially “break” the same wheels by determining
appropriate bit patterns. Techniques were identified to use Colossus machines to the psi
wheels but because of the limited supply of machine time this was usually done manually.

the tape spun round again to the special character that marked the beginning of the message, the
machine was ready to evaluate the message against the next possible combination of wheel settings.
Fensom documented 13 manual actions to get the machine ready—loading a message tape,
configuring the plug board with the appropriate logical inputs, setting the wheel start positions, etc.
These were followed by 20 automatic steps, its program, such as resetting wheel positions, waiting for
the message start signal, and comparing the counts to the thresholds selected by the operator. The
sequence included inner and outer loops. The inner loop was followed each time a character was read
from the message. The outer loop repeated each time the entire message had been read to reset the
totals and increment the wheel start position.25 Building on Fensom’s description, modified by other
sources and extended to include the additional capabilities of the second and subsequent Colossus
machine, we prepared a flowchart (see Figure 4) to visualize the program carried out by Colossus.
For the job we mentioned earlier, setting the first two chi wheels, Colossus would be configured to
generate the difference between two consecutive message bits on two of the five channels of the tape
and compare this to the signals coming from two simulated code wheels. In that case, the inner loop
would cycle each time the message tape finished, the middle loop would increment the start position
for one of the code wheels being tested, and the outer loop would increment the position for the other
code wheel. When both code wheels had returned to their start positions a light would illuminate to
tell the operators that the job was finished.

WHAT IS A PROGRAM, THEN?

How can we argue that Colossus was not a computer, but nevertheless carried out a program? In
discussion of computing these things are often treated as inseparable: something is a computer because

October/December 2018 11 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

Figure 4. Program of operations performed by Colossus.

it can store and run a program, and a program is a series of instructions for a computer. We are using
a more general definition of program: a series of discrete operations carried out over time. We should
stress that Colossus did not store an encoded program in any single part of its apparatus, or read it from
a medium. Rather, its machinery embodied a single, largely fixed program of operations. In this sense,
a program is something enacted. Our definition fits not just computer programs but also other common
uses current in the 1940s. It is also sympathetic to the original meaning of “program” in computing,
which originated as a simple extension of its everyday meaning.
To begin with the more general meanings of program, current in the 1940s and today. A concert
program, for example, specifies a sequence of musical works to be performed by an orchestra on a

October/December 2018 12 www.computer.org/annals

 PROGRAMMABILITY

particular evening. A television network programmer is responsible for choosing and sequencing
shows (also called “television programs”) to produce a schedule. A washing machine fills with water,
soaks, agitates, empties, rinses, and spins under the control of its “programmer.” In each of these cases,
the program consists of actions to be performed in a particular order. We believe that the application
of the word to computers began as a simple extension of this ordinary sense, and only later evolved
into a distinct meaning of its own. This is a contrast to some previous work on the topic, which has
made rather more convoluted arguments.26
The defining characteristic of the automatic computing machine was its ability to carry out one
mathematical operation after another without human intervention, which fits naturally with the
established idea of a program as a sequence of actions. The first groups attempting to build automatic
computers did not use the term program to describe these sequences, though it would be entirely
reasonable for a historian to apply the term as an analysts’ category. Charles Babbage followed a
standard mathematical terminology in calling the discrete actions carried out by his planned Analytical
Engine “operations.” It would have been reasonable and consistent with contemporary English for
Babbage to say that his machine would carry out a program of operations, though as far as we know
he never used that phrase. Most of the first generation of automatic computers, built during the early
1940s, relied on paper tape to control their operation. The patterns punched onto the control tape of
the Harvard Mark 1 computer, built by IBM for Howard Aiken’s Computing Laboratory, were called
“codes,” and its staff talked about “coding” rather than programming. The word “sequence” was
often used to describe the content of a particular strip of paper tape, usually a subroutine. The ability
to automatically perform sequences of operations was central to the new machines, something
recognized in the titles IBM gave to this machine (the Automatic Sequence Controlled Calculator)
and to its more ambitious successor (the Selective Sequence Electronic Calculator). A phrase like
“Program Controlled Calculator” would have also fit with contemporary usage.
The earliest application of the idea of a program to an automatic computer was to describe the control
mechanisms of ENIAC. ENIAC used vacuum tubes rather than electromechanical relays for its
arithmetic and memory circuits. These could switch thousands of times faster than the relays used in
earlier computers. To exploit this speed, its designers fully automated its control. Rather than read
control sequences from tape, its operations were sequenced by a network of wires carrying “program
pulses” between different parts of the machine. Their arrival of a pulse at a particular input terminal
triggered whatever action had previously been set using the unit’s control switches and knobs. These
controls “programmed” the operations carried out by that unit’s circuits. As well as discussing
“program pulses,” a June 1944 progress report described two ENIAC accumulator units as being
“automatically programmed to receive the multiplier and multiplicand” when a program pulse
triggered the multiplier unit to which they were attached. ENIAC’s most complex unit, the “master
programmer” controlled the overall computation, counting off loops and branching between different
sequences when it was time to move on to a new task.
Interestingly, the terms “programming” and “program” were not originally used to describe the
ENIAC’s closest analogs to their modern senses: the act of configuring the machine to carry out a
particular problem and the resulting configuration of wires and switches. Instead these were called,
respectively, “setting up” ENIAC and a “set-up.” By late 1945, however, the ENIAC team was
beginning to use “programming” in something much closer to its modern meaning. The new meaning
of program seems to have been connected to the new approach to automatic control formulated for
EDVAC, the follow-on to ENIAC commissioned in the summer of 1944. John von Neumann’s
celebrated the “First Draft of a Report on the EDVAC,” circulated within the ENIAC team in April
1945, and combined the established approach of controlling a computation by reading a sequence of
coded instructions with the novel idea of storing these instructions in a large, addressable memory
using the same mechanisms employed to store and manipulate data.
Von Neumann himself followed the Harvard group by calling these instructions “code” and the process
of producing them “coding,” but others adapted the Moore School’s existing vocabulary of
“programming” to the new approach, altering its meaning in the process. A letter from one of the
project’s leaders noted that “the EDVAC will contain a large number of units capable of remembering
programming instructions,” to be copied from tape “before the actual program is started.”27 Similar
terminology was soon applied to ENIAC: a report described the practices used in “planning a set-up

October/December 2018 13 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

for the ENIAC” as “programming techniques.”28 Rather than describe the action of the control circuits
responsible for triggering operations at the correct time, the verb “to program” now referenced the
work of the humans devising sequences of operations.
Since then the idea of a program in computing has continued to develop a specialized set of meanings,
particularly in the discourse around the so-called “stored program concept.”29 Here, we are using it in
its more general sense. Even this broader sense is useful in distinguishing the capabilities of Colossus
from those of other contemporary machines which did not sequence distinct operations over time. For
example, in analog computers, such as differential analyzers, each part of the machine carried out the
same operation throughout the course of the computation. There was no sense in which the machine
executed a program of operations, or in which one part of the machine was “programming” another
to stop what it was doing and to start something different.

DEFINING PROGRAMMABILITY
The history of computing community has generally observed a distinction between “general purpose”
computers that can be reprogrammed by their users to do different kinds of tasks and “special purpose”
computers executing a “fixed program.” This distinction was already commonplace within the
computing field by the 1950s. Early special purpose computers were built for tasks such as missile
guidance or toll collection.
Colossus is, to our knowledge, unique in being widely characterized as both “special purpose” and
“programmable,” a startling formulation derived from Randell’s early work. He called Colossus both
“programmable” and “program controlled,” apparently interchangeably, but also described it as
“special-purpose.” The legacy of this characterization endures, for example, on the Wikipedia page for
Colossus which describes it as a “special-purpose electronic digital programmable computer.” In his
history of the relationship between computers and cryptography, Paul Gannon reshuffled the adjectives
slightly: “Colossus can be defined as an, (sic.) electronic, binary/logic-processing, programmable,
specific-purpose machine.”30 Yet to the best of our knowledge, neither Randell nor any other scholar has
attempted to define specifically what “programmable” means as a historical term, or exactly what degree
of configurability would qualify a device like Colossus as programmable but not as general purpose.31
We may gain insight from a parallel discussion underway in the mid-1970s. By the 1960s, as anything
described as a computer was understood to be programmable, people usually just talked about
computers rather than “programmable computers.” The word “programmable” gained new currency in
the 1970s following the introduction of powerful electronic calculators, where users could specify and
store sequences of operations to be carried out automatically. Were these computers? The concept of a
“programmable calculator” was introduced to describe a class of portable, personal machines that
could be programmed by their users but that were more limited than true computers.32 For example,
according to a 1976 report “Calculators and the Computer Science Curriculum” cheaper calculators
were not “programmable by the user” even though “they do contain stored programs and can execute
these programs” for example, when a user pushes the square root button. Thus, “there seems to be a
clear difference between these calculators and what most computer scientists commonly think of as
computers.” In contrast, “programmable calculators” had “sufficient memory to store a series of key
strokes (that is, a sequence of machine language instructions) and then to execute the program.” In a
revealing nod to computer history the author continued, “at the higher price levels (but well under
$1000) such machines approach the ENIAC in capability, and will soon exceed it.”33 That was the
same year in which Randell originally described Colossus as “programmable,” and it seems reasonable
to suppose that he had this discourse in mind when invoking the idea that a machine without the full
capabilities of a general purpose computer could still be programmable. (At the time Randell was
writing the documents that would have made clear whether Colossus actually had capabilities similar
to a programmable calculator were still classified).
Having failed to find a useful and relevant definition of “programmable” in the existing literature, we
will instead attempt to create our own. Our definition of a program as a series of operations carried out
over time helps us to separate the concepts of program and programmability. Whereas “program” has
a long history in many different contexts, “programmable” appeared only after the spread of the
electronic computer and can thus be applied to Colossus only as an analysts’ category. The Oxford

October/December 2018 14 www.computer.org/annals

 PROGRAMMABILITY

English Dictionary shows no usage prior to 1953, in which year it documents the appearance of two
distinct but related meanings: “Capable of being scheduled in accordance with a programme of events”
and “Of an apparatus, operation, etc.: capable of being programmed.” Our definition of programmable
is fairly similar. A program is a sequence of operations performed over time. Programming is the act
of specifying those actions, and a device is programmable to the extent that its users can program it.
The simplicity of that definition conceals an important point: Invoking the action of users means that a
device will be programmable under some circumstances but not others. Any device that performs a
program was designed and built by a group of people. Those same people could have chosen to design
the machine to perform a different program, and with sufficient resources and motivation could
presumably have taken it apart and rebuild it to do something different. For example, the claim of
programmability is at the heart of the case traditionally made for the importance of Colossus to the
history of computing, distinguishing it from the slightly earlier “Atanasoff Berry Computer.” The ABC
was likewise driven by the rotation of a physical medium, in this case a rotating capacitor drum
memory rather than the paper tape used with Colossus. It carried out a sequence of mathematical
operations, performing some steps on a conditional basis depending on the results obtained. ABC and
Colossus have traditionally both been granted “firsts” as follows: The ABC was a special purpose
fixed program digital electronic computer whereas Colossus, built a few years later, was also an
electronic digital computer but was in addition programmable.
The ABC is a famous example of a fixed program machine. Its design embodied a single program,
which could not be fundamentally changed except by redesigning and rebuilding it. That had the
advantage, for its intended users, that the steps needed to solve systems of linear algebraic equations
were built into the machine and did not need to be specified. In contrast ENIAC, a general-purpose
computer, had no built-in program. When used as originally intended, and documented in its
instruction manual, users had to go through a long and elaborate process of deciding on the appropriate
series of operations and figuring out the appropriate network of wires and switch settings to trigger
those operations. Its program changed entirely from one job to another.
Later computers, modeled after the 1945 design for the EDVAC, essentially combined these
approaches by adopting what we have called the “modern code paradigm.”34 On the most fundamental
level they ran a single fixed program, like the ABC. But this program was, to borrow a term that
became popular a little later, an interpreter. It fetched, decoded, and executed instructions held as data
in an addressable memory. This meant that, like ENIAC, the computers could carry out whatever series
of operations their users specified.35 As we have described elsewhere, in 1948 ENIAC itself was
reconfigured for this new programming paradigm, with a fixed program wired on its controls and user
programs coded as numerical sequences on its function tables. Later systems added further levels of
fixed programs underlying the changeable program—microcode, operating systems, interpreters,
BIOS code, and so on. But as long as their users can specify some new programs, we would consider
those aspects of the system programmable.
A device that is programmable by one user with one tool will not necessarily be programmable by
other users with other tools. For example, when the iPhone was first released, its operating system was
deliberately locked down to prevent the creation or installation of any programs other than those Apple
had created for it. For users outside Apple’s own development teams, it was not programmable. Some
users wrote new code to replace parts of the standard operating system to “jailbreak” it, allowing the
installation of other programs. Soon Apple itself created its “App Store” and tools to allow users
outside the company to create their own programs for the iPhone, which quickly became a crucial
feature. Most users still do not create or modify its programs, but the machine is programmable for
those with the skills, motivation, and tools to do so.
We should also distinguish between creating programs and choosing between preexisting programs.
For example, by our definition, a player piano follows a program when it plays notes from a music roll.
If its user had a machine able to punch notes onto a blank roll it would be programmable. Most users
did not. They could select between programs by changing rolls, but not alter them. Likewise, Atari’s
1970s home videogame console, the VCS, was a simple unit that ran code from ROM cartridges.
Ordinary users purchased, selected, and ran programs but did not create them. For someone with the
appropriate development system and the ability to burn programs to ROM, the VCS was
programmable.

October/December 2018 15 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

Figure 5. On this typical washing machine dial, users select one of three predefined
programs, such as the normal program, which runs as follows: hot, warm, cold, rinse, spin, off.
Users can also skip to a particular point within each program, for example, starting the normal
wash program at the beginning for heavily soiled clothes or omitting the hot wash operation for
regular soilage.
Other systems provided the ability to choose between built-in programs, but unlike the player piano or
the Atari VCS were not designed to let users swap in additional programs. An example is the washing
machine. As we mentioned above, an automatic washing machine incorporates a “programming unit”
to control the sequence of operations it carries out. Washing machine users choose between several
different programs by turning a control dial (see Figure 5) to the desired starting point, and can also
push buttons to set parameters, such as “light load,” that modify the washing operations. Yet it would
seem very unusual to talk about “programming” a washing machine.36 The conventional term is that a
user “selects” a program cycle. Likewise, in the computing context one would not usually call the act
of choosing a predefined program and triggering its execution “programming.”37
We conclude that the concept of “programmability” as applied to a device requires not only that the
device carries out a sequence of distinct operations over time, i.e., that it follows a program, but also
that it allows a given user to define new sequences of operations. We see programmability as a
relational rather than absolute property. One user with one set of tools will be able to program aspects
of the machine’s behavior that other users, with other tools, cannot program. So in asking whether
Colossus was programmable, we must be a little more specific. We suggest that the definition of
programmability should be inclusive, to define a machine as programmable if any actual group of
users with any set of existing tools could program its overall behavior, excluding only the special case
of engineers rebuilding it. If the cryptanalysts and operators at Bletchley Park were able to set up
Colossus to carry out new sequences of operations then Colossus was programmable. If changing the
program carried out by Colossus (and diagrammed by us as Figure 4) would have required calling
Flowers and his team to take it apart then Colossus was not programmable. As in the case of the
washing machine and the ABC, allowing users to select between predefined sequences of actions, skip
steps, or apply parameters to alter the behavior of particular steps does not constitute programmability.
On the other hand, because we have a broad sense of program our definition of programmability is not
encumbered with requirements for Turing completeness, conditional branching, and so on. By our
definition, a player piano is programmable by someone with a tape punching machine, and so were
relay computers like Harvard Mark I.

PARTS AND CONTROLS OF COLOSSUS

To characterize Colossus as “programmable,” then, is to suggest not only that it followed a program
but also that users could fundamentally change the program of operations it performed without
rebuilding its hardware. Colossus was indeed highly configurable, so that its users applied it to many
different tasks—a tribute to the foresight of Flowers’ team and its close liaison with Newman and his

October/December 2018 16 www.computer.org/annals

 PROGRAMMABILITY

Figure 6. Reconstructed Colossus 2 at the National Museum of Computing. By TedColes -

Own work, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid ¼ 38648141.

group at Bletchley Park. But did these configuration capabilities change the program itself, or just set
parameters for a fixed program? To answer those questions, we must retreat temporarily from
historiography to materiality, to talk a little bit about what Colossus did and the extent to which its
program could be changed by expert users.
Thanks to the declassification of archival records and the work done by Tony Sale and his team in
rebuilding a Colossus at the U.K.’s National Museum of Computing (see Figure 6), we are now
able to document exactly what it could and could not do. The overall structure of the machines
and details of electronics and circuitry are not well described in the surviving primary sources
available to us, and here we have relied on the work of the rebuild team. Their depictions in the
earlier secondary literature are not always accurate. The detailed functionality and intended use of
most of the machines’ controls, however, are described in the 1945 “General Report on Tunny”
particularly in chapter 53.38 We have also made use of the substantial number of printouts from
Colossus runs reproduced in that report and others dating from 1945; these have enabled us to
confirm many details of the machines’ functionality independent of the reconstruction. In
combination, these sources allow a more complete and accurate description of the functionality
and use of the machines than might be expected given the secrecy that surrounded them. We
present a summary of these findings here, but readers who want more detail on the analysis and
behind them should read our full technical report.39
Many of the racks (see Figure 7) combined several distinct functions, so we find it useful to
conceptualize Colossus as being composed of three main subunits (see Figure 8). Its forerunner, Heath
Robinson, was literally designed and manufactured in three separate parts, interconnected for the first
time at Bletchley Park. These were the tape unit, the combining unit, and the counter. Colossus
followed the same pattern, though the subunits were less physically distinct.
The reading and generating unit consisted of a tape drive (often used to mount encrypted text) and a set
of 12 electronic ring counters that simulated physical code wheels in the Lorenz cipher machines used
by the Germans. Physically, this functional unit included “bedstead” tape equipment and reader, and
lots of electronics for the simulated wheels in racks W, M, and R at the back of the machine. For
convenience, most of the controls for these circuits were mounted on the front of Colossus on the S
rack and included:

“Setting jacks” to set the initial wheel start positions,

Stepping controls, to govern the stepping of these initial positions each time the message
was restarted,

Pins to set the bit patterns on the simulated wheels (known as “triggers”),

“Multiple testing” controls to determine which code wheel’s output was buffered so that up
to five consecutive wheel positions could be evaluated simultaneously, and

Span controls, used to count only part of the input tape when producing totals.

October/December 2018 17 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

Figure 7. Physical layout of Colossus, showing its main user-configurable controls.

The combining unit combined selected channels from the various tape and wheel bitstreams. The
results were further combined by using them as inputs to use-configurable networks of logic gates.
The equipment for these functions centered on a plugboard in Rack J and a large switch panel in Rack
Q (the “Q Panel”), both used to define the logical tests applied to inputs and specify the counters to
which the results should be sent. User-configurable controls included:

“Q selection switches” to determine which of the many possible inputs were routed onward
to the Q panel and whether to supply it with raw values or “deltas” between successive bits,

Wires set on the plug board (“jack field”) to run selected inputs through particular
combinations of logic gates and route them to particular counters, and

Keys set on the Q panel to run selected inputs through particular combinations of logic
gates.

The counting unit included the electronically controlled typewriter used for printing and, in the C
Rack, the electronics that drove it and buffered output so that Colossus could continue to work while
the results of the previous set of counts were output. Only one aspect of this was configurable:

Rotary switches to set the threshold above which the contents of a counter, and
corresponding initial wheel positions, should be printed.

WHAT COULD THE CONTROLS CONFIGURE

These various controls did give Colossus considerable flexibility. It was designed with wheel setting in
mind but codebreakers found many other ways to use Colossus for different parts of the decoding job.
For example, one way to determine whether decryption settings were correct was to count the number
of times each letter appeared in a message. Colossus could be set to look only at the characters on the
tape (ignoring the simulated cipher wheels entirely) and to count the frequency with which they
appeared. A run of that kind would finish after a single pass through the inner loop in our flow

October/December 2018 18 www.computer.org/annals

 PROGRAMMABILITY

Figure 8. Logical architecture of Colossus, broken down into its three main function units and
showing data and control flows between them as well as the main configurable settings
available to control them.

diagram, having immediately met the termination conditions for the middle and outer loops. In our
technical report, we discuss several common Colossus configurations and exactly how the machine
would be set up for each.
Those describing Colossus as programmable have generally rested their case on the combining unit
controls. For example, according to Benjamin Wells “tightly refined codebreaking algorithms were
implemented in plug-wiring and switches. But the crucial story is that the same machine supported
many different algorithms via flexible programming.”40
We will therefore describe these controls in a little more detail. Plugboard cables and, on later models,
switches were used to select inputs and run them through logic gates to generate pulses for the

October/December 2018 19 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

counters. Users had a great deal of flexibility in configuring its circuits to run inputs from the message
tape and electronic code wheels though different logic gates to combine them in different ways for
different purposes. Signals from the electronic simulated code wheels and the message tape were
available as inputs. The message tape appeared as five separate binary channels, as did each of the two
main sets of code wheels. The Colossus machines also provided inputs representing the differences
between two consecutive character positions and, in later models, five consecutive positions of one of
the electronic code wheels (the “R stream” on our diagram). The results of these logical combinations
could be fed into different electronic counters, the contents of which would be printed or not printed
after the message was fully read according to thresholds set by the user. After each reading of the
message tape, the initial code wheel settings would be stepped to a different combination according to
switches configured by the user.
Operators plugged wires or set switches to combine these inputs. This let them specify up to five truth
tables, making particular logical connections between the input pulses and the output fed to each
counter. Colossus included circuits to implement a variety of Boolean logic operations, including XOR
and NOT. By running pulses through several of these circuits other logical conditions could be
specified.41 Each of these truth tables specified which combinations of inputs should trigger an output
pulse for a counter. The logic circuits then reset, and processed the next combination of inputs. This
was analogous to the function of a traditional tabulating machine, in that both processed a stream of
records (characters from paper tape for Colossus, punched cards for the tabulating machine) and
incremented the appropriate counter whenever the input data satisfied a specified condition. Colossus
allowed for rather more complex logical conditions than traditional tabulating machines, but it
similarly transformed and counted streams of input data. The only conditional action that Colossus
users could specify via the logic circuits of the combining unit was to either increment or not
increment each counter in response to a particular combination of input received from the tape and
code wheels. Data pulses output from the combining unit were routed only to counters, and the
contents of the counters went only to the printer.
Computer scientists use automata theory to distinguish between the fundamental capabilities of
different kinds of automatic devices. The most advanced, including programmable computers from
ENIAC onward, are equated with Turing Machines. Push down automata are less powerful than
Turing machines, and finite state automata are less powerful than push down automata. As their
name suggests, even finite state automata preserve state information from each time step to the
next. In contrast, the “combining unit” of Colossus did not preserve state information. The
technical term for this kind of system is “combinational logic”42 (which is echoed in the term
“combining unit” applied by Flowers and his team to the corresponding part of Heath Robinson)
or “time-independent logic.”43
The logic performed one complex step repeatedly, but did not sequence operations or maintain any
state information from one input character to the next. Counter totals were retained from one character
of the tape to the next, and the uniselectors held current wheel start positions from one entire cycle of
the tape to the next. In an abstract sense both held state information, and thus constituted a kind of
memory, but these data could not be fed into the combining unit as inputs. Counter values were not
available as inputs for the configurable logic in the combining unit, eliminating the possibility of using
counters as control flags. We do not believe that the ability to reconfigure combinational logic to
implement an arbitrary truth table is evidence of “programmability” because it does not truly change
the sequence of operations performed by the machine. It merely provides a logic expression evaluated
in one of those steps.
It has sometimes been claimed that Colossus possessed the capability for conditional branching.
Colossus certainly included circuits that would trigger an action only under particular conditions—for
example, resetting the machine when the end of message code was read from the tape but not when
ordinary data were read. During the reset process at the end of each loop of the message tape,
dedicated circuits would compare the counter values against a user selected threshold to determine
whether the results of that cycle should be printed. If that was “conditional branching” then so is any
control circuit, even an adder that carries digits from one position to the next, or a thermostat that turns
on a heater once the temperature falls below a user-specified level.

October/December 2018 20 www.computer.org/annals

 PROGRAMMABILITY

Colossus was not programmable, according to our definition, because the basic sequence of operations
performed, the program, could not be changed by the users. That judgment involves answering a
difficult question: what kinds of modification to the program executed by a machine are sufficient to
make that machine “programmable” as opposed to “configurable”? What is a new program and what is
just a parameter? We have shown that the Colossus controls could not alter the basic sequence of
operations shown in our flow diagram, though with certain switch settings some steps would be
skipped. To us, that kind of configuration seems more like setting a parameter than defining a new
program.
The actual programming capabilities of Colossus turn out to be much more limited than those claimed
by its boosters. We stridently dispute the suggestion of Barry Cooper that Colossus “may well be
Turing complete” (i.e., could handle any problem solvable by any computer if given sufficient time
and storage space).44 Wells’ claim that different “algorithms” could be set up is similarly misleading.
An algorithm expresses a computational procedure as a step-by-step series of operations carried out
over time. Colossus could carry out only one algorithm. Fensom characterized it as carrying a program
designed by Flowers, rather than running many different programs written by its users.45 We think he
was right.

WAS COLOSSUS BINARY?

Having described the architecture of Colossus we can clarify a final point. Colossus is often described
in the secondary literature as a binary machine, sometimes with the implication that this made it closer
to modern computers, and hence more advanced, than computers using decimal number
representations. Paul Gannon made this explicit: because the “modern computer can be defined as an
electronic, binary/logic-processing, conditional-programming, stored-program-control, general-
purpose machine” and Colossus, unlike ENIAC, was also a “binary/logic-processing, programmable”
machine “Colossus was closer to the modern concept of the computer than ENIAC was in some
significant ways.”46 Echoing this, in The Innovators, Walter Isaacson noted that “well before
ENIAC. . . the British code breakers had built a fully electronic and digital (indeed binary) computer.”
ENIAC, in contrast, “was not like a modern computer” to the extent that it used a decimal system.47
In popular discourse of this kind, binary is often associated with the idea of representing information as
a series of electrical pulses, conceptualized as 0s and 1s. Colossus certainly did that—its users spoke of
“impulses,” which they represented as “dots” and “crosses” rather than the 1 and 0 of abstract logic.
But all digital electronic machines transmit pulses, whether they use binary or not. ENIAC, for
example, transmitted the decimal digit four as four pulses.
“Decimal” and “binary” properly refer to the number bases used by computers when representing and
manipulating numbers: 10 for decimal and 2 for binary. The difference between binary and decimal
computers came in three main areas: the format used to store numbers in memory, the format used to
transmit them, and the design of the arithmetic circuits used to manipulate them. It is hard to define
Colossus as either binary or decimal, because most of the electronic pulses manipulated by Colossus
did not represent numbers. It had no hardware to interpret input patterns as binary numbers, or to carry
out any arithmetic other than incrementing a counter. A computer reading a five-channel paper tape
would interpret the five channels as a five-digit binary number, assigning a different value to a 1
depending on its position. For example, 10011 might be interpreted by its hardware as the number 19.
In contrast, Colossus treated the five channels as synchronized but independent bitstreams. It
performed logical transformations, rather than arithmetic, on the data it read.
Three parts of Colossus counted each in a different way but none used a binary representation.

1. The units that emulated the encoding wheels of the Lorenz machines were a set of electronic
ring counters of different lengths (and thus neither binary nor decimal).
2. The electronic counters tallied pulses output from the combining unit. These used decimal
arithmetic, but stored each digit as a bit pattern using a biquinary representation.
3. The electromechanical uniselectors that held the current start positions for the code wheels
were stepper counters of various periods, advancing with each cycle of the message tape.48

October/December 2018 21 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

CONCLUSIONS: COLOSSUS AND THE EARLY DIGITAL

Computers were far more visible than other forms of digital electronics during the 1970s and 1980s,
and inventing the computer a much more visible honor than inventing the digital signal processor. For
decades, from the late-1940s to the 1990s, specialists and the public shared a fairly clear sense of what
“the computer” was. It consisted of a box holding a processor and memory, to which were connected
storage devices such as disks and tapes, and input and output devices such as keyboards and printers.
Over time the boxes shrank and got cheaper, and they spread from data processing centers into
schools, offices, and homes. Making a case for the historical importance of Colossus meant
shoehorning it into this tradition, as Brian Randell did when he won it a prominent place at the seminal
1976 Los Alamos meeting where the early agenda for the history of computing field was set.
Early computing innovation is still often discussed from the viewpoint of logic, stressing mathematical
and theoretical insights behind the development of computer architecture.49 The progression from one
early computer to the next has been depicted as a series of abstract architectural innovations to be
checked off and annotated with dates and machines—first conditional branch, first stored program, and
so on. In turn, these innovations have sometimes been represented as mere practical instantiations of
the work of Alan Turing on the mathematical logic of computation.50
This obsession with firsts has limited our understanding of all early machines, by reducing each to a
date and a couple of approved adjectives.51 Even ENIAC, with a long and productive life, has been
remembered only as a single historical point in 1945. During this research, we realized that the
tradition has imposed even more serious limitations on general understanding of Colossus. Within it,
the implicit reason to care about Colossus has been that it was the first programmable electronic
computer. We do not believe it was either of those things. Colossus certainly followed a program, in
that it sequenced different operations over time. But contrary to previous claims, it was not
programmable, because this sequence could not be changed in any fundamental way. We think that
Flowers’ own, carefully nuanced, characterization of Colossus as an “electronic processor” rather than
a “computer” is essentially correct. Colossus did not have any capabilities for numerical operations.
This sets it aside from every other machine celebrated in the history of computing.
As Flowers proudly noted, Colossus certainly had many elements in common with early computers,
but its architecture and purpose were fundamentally different. In his public statements on Colossus,
Flowers situated Colossus within the context of early work on digital communication. The recent
proliferation of digital devices and the vanishing of “the computer” as a distinct and coherent thing
makes us better able to appreciate Colossus, as Flowers did himself, more as a contribution to the
development of digital communication than to computing. Colossus was a digital electronic device
able to generate bitstreams electronically, combine those bitstreams with others read from paper tape,
apply logical transformations, and count the results.
Claims for the historical importance of Colossus can and should be about more than the exact string of
adjectives to insert between “first” and “computer.” This challenges us to articulate its importance in
more productive terms, for example, by its achievements in use. From the traditional historiographic
perspective, the only thing separating Colossus from the Atanasoff Berry Computer is the insertion of
the word “programmable” between “first” and “electronic computer.” That distinction does not hold
up: both machines could be configured by setting parameters, but not by defining new sequences of
operations. As with Colossus, the basic sequence of operations the ABC performed was fixed, in this
case to solve systems of linear equations, but its behavior would change based on the parameters
supplied when the machine was configured by setting switches and rewiring plug boards.52
Looking at the use of the two machines, rather than their claims to “firstness,” is a much more
revealing way of separating them. The ABC incorporated many novel features, but its homebrewed
input/output system (using sparks to burn paper) never worked reliably enough for it to tackle the
problems it had been built to solve. (It managed some useful work on much simpler statistical
problems, but nothing that would justify the creation of such an elaborate machine). Colossus, in
contrast, proved spectacularly effective when applied as intended. It facilitated the reading by Allied
commanders of some of Nazi Germany’s highest level military communications, including messages
written by Hitler himself. It has been credited with a crucial role in revealing German troop positions
and plans in Normandy before and after the D-day landings. Its often claimed that breaking Enigma, or

October/December 2018 22 www.computer.org/annals

 PROGRAMMABILITY

sometimes even the personal contributions of Turing, shortened the war by two years.53 We find that
unconvincing, as did Max Hastings and John Keegan in overview histories of the contributions of
intelligence work.54 Still, if access to Tunny intercepts shortened the war by even a month, that would
make Colossus one of the best investments in history. The ABC spent the war abandoned in a
basement.
Flowers’s partnership with the codebreakers at Bletchley Park predated his work on Colossus, and he
had enough sense of the operational context to optimize the usefulness and versatility of his machines.
Not only did he lead a project that quickly built a reliable machine from exotic parts, but he also built
exactly the machine that was needed by Britain’s codebreakers to deliver intelligence to Allied leaders.
In doing so, it made an appreciable contribution to the course of world history. Less than a year went
by between the request from Newman for Dollis Hill to work on a machine to tackle Tunny and the
delivery of the first Colossus to Bletchley Park, and for much of that time Flowers was focused on
other machines.
Colossus also stood out for its reliability in comparison with other similarly complex electronic digital
machines of the 1940s and early 1950s. All the others all seem to have spent a year or so between
being finished and being reliable enough to carry out useful work. For example, after being moved
to Aberdeen Proving Ground at the start of 1947 ENIAC relapsed into unreliability, and for more than
a year struggled to carry out any useful work. As Colin Burke’s recently declassified work has shown,
Colossus was one of many wartime designs for complex devices, sometimes called “rapid analytical
machines” intended to speed codebreaking. No other machine of comparable technical ambition was
ready in time to assist with the war, and many of them never worked reliably enough to be useful.55
Colossus, in contrast, was handling a full load of production work by March 1944, less than two
months after being delivered to Bletchley Park.56
This reminds us of the enormous importance of professional engineering work in distinguishing the
relatively small number of successful projects from the greater number of failures. Flowers and his
colleagues were part of a telecommunications engineering institution. They could draw on experience,
internal engineering talent, and existing relationships with component suppliers. Colossus made
extensive use of electronics for counters and logic. Flowers insisted that these technologies were not
unproven to him, even if the rest of the world was skeptical, because of the prewar work he had been
doing on electronic telephone switching.57
Colossus challenges us to look beyond the traditional reading of the history of computing in the 1940s
as a series of incremental steps leading to the “modern computer.” Colossus is an exemplary artifact
of what we like to call the “early digital” because of, not despite, its fundamental lack of architectural
resemblance to a computer.58 Our analysis underlines the need for a historiography of the early digital
that is concerned with use as much as invention and for a history of computing fully integrated with
broader historical analyses, such as social, business, labor, and military history. If there has been such
a thing as a digital revolution then it involved much more than just computers.

ACKNOWLEDGMENTS
This project was generously supported by Mrs. L. D. Rope’s Second Charitable Settlement. We
extend our thanks to C. Rope and to all those at Lucy House and also to those who commented
on drafts prior to submission including B. Randell, M. Campbell-Kelly, Q. DuPont, J. Reeds, and
others at the 2016 SIGCIS workshop and the National Museum of Computing. We appreciate the
helpful suggestions and careful analysis of anonymous reviewers 2 and 3, which strengthened
the paper significantly, and the editorial guidance of D. Hemmendinger.

REFERENCES
1. We previously endorsed the idea of Colossus as a digital electronic computer based on
claims in the secondary literature, but had not been able to form a very clear impression of
what it actually did or how it was used. Cf. T. Haigh, M. Priestley, and C. Rope, ENIAC In
Action: Making and Remaking the Modern Computer. Cambridge, MA, USA: MIT Press,
2016. Realizing this ignorance inspired the research described here.
2. Ibid.

October/December 2018 23 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

3. A revised version was published in the seminal volume B. Randell, “The Colossus,” in A
History of Computing in the Twentieth Century, N. Metropolis, J. Howlett, and G.-C. Rota,
Eds. New York, NY, USA: Academic, 1980, pp. 47–92.
4. B. J. Copeland, Turing: Pioneer of the Information Age. New York, NY, USA: Oxford
Univ. Press, 2013, p. 119.
5. This is particularly noticeable with fans of computer pioneer J. Atanasoff, such as A. R.
Burks, Who Invented the Computer: The Legal Battle That Changed Computing. New
York, NY, USA: Prometheus Books, 2003.
6. J. Copeland, “Colossus and the rise of the modern computer,” in Colossus: The Secrets of
Bletchley Park’s Codebreaking Computers, J. Copeland Ed. New York, NY, USA: Oxford
Univ. Press, 2006, pp. 101–115.
7. T. Sale, “The Colossus rebuild project,” Codes Ciphers, Accessed: Jul. 4, 2018. [Online].
Available: http://www.codesandciphers.org.uk/lorenz/rebuild.htm
8. M. Campbell-Kelly and W. Aspray, Computer: A History of the Information Machine. New
York, NY, USA: Basic Books, 1996, pp. 99–100. The more recent third edition retains the
same text (p.82).
9. P. E. Ceruzzi, A History of Modern Computing. Cambridge, MA, USA: MIT Press, 1998.
Ceruzzi gives Colossus about a page in his more recent Computing: A Concise History.
He asks why Colossus is not “more heralded,” and concludes that its focus on textual rather
than numerical operations had combined with the long prevalent secrecy to marginalize it.
10. T. H. Flowers, “The design of Colossus,” IEEE Ann. History Comput., vol. 5, no. 3,
pp. 239–252, Jul.–Sep. 1983. For example, Flowers remembered the “shift register”
introduced on Colossus 2 as buffering the outputs of five channels of simulated cipher
wheel output, but later analysis showed that it buffered only one channel. A detailed
analysis of this episode can be found in J. A. Reeds, W. Diffie, and J. V. Field, Breaking
Teleprinter Ciphers at Bletchley Park. (An Edition of General Report on Tunny With
Emphasis on Statistical Methods (1945), Piscataway, NJ, USA: Wiley, 2015, pp. 606–608.
In subsequent notes we cite this as an accessible version of the 1945 report, whose original
authors were I. J. Good, D. Michie, and G, Timms.
11. Colossus has been explored in J. Copeland, Colossus: The First Electronic Computer,
New York, NY, USA: Oxford Univ. Press, 2006; P. Gannon, Colossus: Bletchley Park’s
Greatest Secret. London, U.K.: Atlantic Books, 2006; H. G. Cragon, From Fish to
Colossus,. Dallas, TX, USA: Cragon Books, 2003; and Reeds et al. (2015).
12. D. A. Mindell, Between Human and Machine: Feedback, Control, and Computing Before
Cybernetics. Baltimore, MD, USA: The Johns Hopkins Univ. Press, 2002.
13. While we know of no evidence that Colossus was ever called a “computer” during the
1940s, some British codebreakers were known as “Computor Clerks,” or sometimes
simply as “computors.” According to M. Smith, “This curious title had nothing to do with
electronic computers, but was an echo of an old War Office covername for cryptanalysts –
Signal Computor.” M. Smith, The Secrets of Station X: How the Bletchley Park
Codebreakers Helped Win the War. London, U.K.: Biteback Publishing, 2011. A 1940
memo noted that “The air ministry has already instituted the use of the word ‘Computor’ in
all cases in lieu of “cryptographer.” (Illegible) to Denniston, July 14, 1940, HW 14/6, TNA.
Another document noted that secrets sometimes had to be discussed over the telephone, and
mandated substitutions which “bear no relation to the real nature of the intelligence.” It
specified “special computer” as a substitute for cryptographer, and “special computation” as
a substitute for cryptography. “Draft,” July 28, 1940, HW 14/6, TNA. This reminds us of
other deliberately misleading wartime cover names, like the “tank” in World War 1, or the
“tube alloys” codename for British atomic research. However, a subtly different pattern of
usage occurred inside Bletchley Park itself, where such subterfuge was unnecessary. Later
documents suggest that itself the term was most commonly used for workers carrying out
more routine tasks, particularly those focused on counting and tallying operations. For
example, within the Newmanry where Colossus was used, the term “computer” was applied
not to the male cryptographers who oversaw the attack but to the “Wrens who enter, flag,
and converge rectangles” (a laborious process of tabulating and adding data). Reeds et al.
(2015, Sections 31B & 71). Thus, the term “computer,” while mandated for broad use
outside Bletchley Park, may have sometimes been used within its walls in a sense
paralleling the scientific use of human, usually female, “computers” discussed in D. A.
Grier, When Computers Were Human. Princeton, NJ, USA: Princeton Univ. Press, 2006.

October/December 2018 24 www.computer.org/annals

 PROGRAMMABILITY

Given the prevalence of scientists and mathematicians in senior roles within this part of
Bletchley Park the adoption of this terminology is unsurprising.
14. G. Stibitz, 1945 AMP report “Relay computers,” ML27-b3 in his papers at Dartmouth
College, p. 2.
15. Jack Good, a veteran of Colossus practice at Bletchley Park, later claimed that, if
appropriately configured, Colossus could almost have carried out a multiplication but that
this would not have been possible in practice because of constraints on what could be
accomplished in a processing cycle. (We have not been able to figure out what he had in
mind, but then we are not as clever as him). This claim has been offered as proof of the
flexibility of Colossus, which in a sense it does attest to: a device designed without any
attention to numerical computations could almost, but not quite, have multiplied thanks to
the flexibility with which logical conditions could be combined. Yet it also proves the very
real differences between Colossus and devices designed for scientific computation.
Multiplications were vital to computations, and a device that could not multiply would not,
by the standard of the 1940s, be termed a “computer “or “calculator.” See I. J. Good, “From
hut 8 to the newmanry,” in Copeland (2006, pp. 204–222).
16. Flowers’ career is examined T. Haigh, “Thomas Harold (‘Tommy’) Flowers: Designer of
the Colossus codebreaking machines,” IEEE Ann. History Comput., vol. 40, no. 1,
pp. 72–78, Jan.–Mar. 2018.
17. B. J. Copeland, Ed. Alan Turing’s Automatic Computing Engine: The Master
Codebreaker’s Stuggle to Build the Modern Computer. New York, NY, USA: Oxford Univ.
Press, 2005.
18. T. H. Flowers, “Electronic computers and telephone exchanges,” Nat. Phys. Lab.,
Teddington, U.K., NPL Rep. DNACS 24/80, 1980.
19. T. H. Flowers, “The design of Colossus,” IEEE Ann. History Comput., vol. 5, pp. 239–252,
1983. His assertion that Colossus incorporated “branching logic” is questionable, and the
intended meaning of “variable programming” is unclear.
20. T. H. Flowers, “Colossus,” in Copeland (2006, pp. 91–100).
21. T. H. Flowers, “D-day at Bletchley Park,” in Copeland (2006, pp. 78–83).
22. While we could find no article or talk by Flowers in which he directly calls Colossus a
computer, Jack Copeland has made prominent use of quotations from unpublished
interviews in which Flowers appears less ambivalent.
23. H. Fensom, “How Colossus was built and operated—one of its engineers reveals its
secrets,” in Copeland (2006, pp. 297–306).
24. A variation on this description occurs somewhere in almost every one of the previous
descriptions of Colossus we have already cited. Our own attempt at a somewhat more
complete description came in T. Haigh, “Colossal genius: Tutte, Flowers, and a bad
imitation of Turing,” Commun. ACM, vol. 60, no. 1, pp. 29–35, Jan. 2017.
25. Fensom described the first version of Colossus. One of the features added in second and
subsequent machines was the ability to set wheels to step either “fast” or “slow.” Reflecting
this, our diagram breaks his outer loop into middle and outer loops—once the fast stepping
wheel (or wheels) had been tried in every possible start position (middle loop) Colossus
would increment the start position of the slow stepping wheel.
26. D. A. Grier, “The ENIAC, the verb ‘to program’ and the emergence of digital computers,”
IEEE Ann. History Comput., vol. 18, no. 1, pp. 51–55, Jan. 1996; D. A. Grier,
“Programming and planning,” IEEE Ann. History Comput., vol. 33, no. 1, pp. 86–88,
Jan.–Mar. 2011. We share Grier’s general sense that the idea of “programming” was
introduced to computing in the ENIAC project, but are not convinced by his suggestions
adoption of the term reflected a desire to “establish the economic importance of automatic
computing” or that the alternative term “planning” reflected a specific commitment to
Taylorist conceptions of production engineering.
27. H. Goldstine and H. Curry, “ENIAC Patent Trial Collection,” Univ. Pennsylvania,
Philadelphia, PA, USA, Oct. 3, 1945.
28. J. P. Eckert et al., “Description of the ENIAC and comments on electronic digital
machines,” Distributed by the Appl. Math. Panel, Nat. Def. Res. Committee, Philadelphia,
PA, USA: Moore School Elect. Eng., AMP Rep. 171.2R, Nov. 30, 1945.
29. The “stored program concept” has recently been examined at length in T. Haigh,
M. Priestley, and C. Rope, “Reconsidering the stored program concept,” IEEE Ann. History

October/December 2018 25 www.computer.org/annals

 IEEE ANNALS OF THE HISTORY OF COMPUTING

Comput., vol. 36, no. 1, pp. 4–17, Jan.–Mar. 2014; Haigh et al. (2016); and D. Swade,
“Inventing the user: EDSAC in context,” Comput. J., vol. 54, no. 1, pp. 143–147, 2011.
30. Gannon, Colossus, 435.
31. Randell points out to us that he had, however, tried previously to rigorously define
programs and programmability in his nonhistorical work with J. J. Horning on
computational processes, in dialog with Dijkstra’s contemporaneous work on structured
programming. They describe a “general-purpose computer” as “a processor with the
outstanding characteristic that program instructions, as well as program status, are
represented by values of its (changeable) state variables.” They then introduce “the term
programmable processor to describe an interpreter with the property that some significant
variables of the underlying process are instruction variables, and are not observable at the
higher level.” It is not obvious how to map the concepts of instructions or variables onto the
capabilities of Colossus. A program is defined by reference to programmability: “Given a
programmable processor, it becomes convenient to define a process by means of a program.
A program for a particular processor is a set of initial values for its program status and
instruction variables.” J. J. Horning and B. Randell, “Process structuring,” ACM Comput.
Surv., vol. 5, no. 1, pp. 5–30, 1973.
32. This was more of a practical and marketing distinction than a theoretical one, as some
programmable calculators had Turing complete programming languages.
33. D. Moursund, “Calculators and the computer science curriculumn,” ACM SIGCSE Bull.,
vol. 8, no. 2, pp. 21–23, 1976.
34. Haigh et al. (2014)
35. At this level of abstraction, one might depict Turing’s “universal machine” as doing
something similar, as it could carry out the same work as any other machine without
requiring modification to its control table if provided with the appropriate sequence of
symbols on tape.
36. Google currently finds 94 uses of the exact phrase “programmable washing machine”
which suggests that it is rarely used.
37. Such a usage might resonate with the original 1944 ENIAC sense of programming, in
which directing the operation of a piece of machine was programming it, but the concept of
“programmability” appeared only after our modern sense of a computer program was well
established. Thus, the idea of “programmability” has historically been applied only to a
specific, narrower, and later sense of “program” and, as the OED suggests, means that a
user can set up a schedule of events.
38. This report was recently published as Reeds et al. (2015).
39. The report has the working title “Colossus Configuration and Capabilities” and is expected
to appear in the Media of Cooperation working paper series of Siegen University during
2019.
40. B. Wells, “The PC-user’s guide to Colossus,” in Copeland (2006, pp. 116–140).
41. Reeds et al. (2015, p. 323).
42. G. A. Maley and J. Earle, The Logic Design of Transistor Digital Computers. Englewood
Cliffs, NJ, USA: Prentice-Hall, 1963.
43. C. J. Savant, M. S. Roden, and G. L. Carpenter, Electronic Design: Circuits and Systems.
Redwood City, CA, USA: Benjamin/Cummings, 1991.
44. S. B. Cooper, “The machine as data: A computational view of emergence and definability,”
Synthese, vol. 192, no. 7, pp. 1955–1988, Jul. 2015.
45. Flowers himself, in 1983, described Colossus as following a “master control program”
which was fixed by the control unit rather than defined by the operator. However, he then
muddied this distinction by calling Colossus configurations set up with jacks and switches
“programs” and the person configuring Colossus a “cryptanalyst-programmer.”
46. P. Gannon, Colossus, circa 435.
47. W. Isaacson, The Innovators: How a Group of Hackers, Geniuses, and Geeks Created the
Digital Revolution. New York, NY, USA: Simon and Schuster, 2014, pp. 75–79. This
connection of binary arithmetic with progress is rather dubious. Decimal arithmetic has
some advantages for floating point and remained a feature of many later computer
architectures.
48. Biquinary representation, as used in the Colossus counters, took advantage of vacuum tube
circuits that had not just two stable positions but five. Using a two-position counter and a
five-position counter for each decimal digit allowed for ten possible combinations,

October/December 2018 26 www.computer.org/annals

 PROGRAMMABILITY

49. The classic contribution of this kind is M. Davis, Engines of Logic: Mathematicians and the
Origin of the Computer. New York, NY, USA: Norton, 2001.
50. This assumption is critiqued in T. Haigh, “Actually, Turing did not invent the computer,”
Commun. ACM, vol. 57, no. 1, pp. 36–41, Jan. 2014.
51. M. R. Williams, “A preview of things to come: Some remarks on the first generation of
computers,” in The First Computers: History and Architectures, R. Rojas and U.
Hashagen, Ed. Cambridge, MA: MIT Press, 2000, pp. 1–16
52. A. R. Burks and A. W. Burks, The First Electronic Computer: The Atanasoff Story. Ann
Arbor, MI, USA: Univ. Michigan Press, 1989.
53. The claim is often made that the Second World War was shortened by two years either by
Alan Turing’s personal contributions to codebreaking, or by breaking Enigma, or by
Colossus. The original source of the idea seems to be the introduction to E. F. Hinsley and
A. Stripp, Eds. Code Breakers: The Inside Story of Bletchley Park. New York, NY, USA:
Oxford Univ. Press, 1993, where it is made for Ultra intelligence, i.e., the Bletchley Park
operation as a whole. Specifically, the authors suggest that without the benefit of Ultra, the
battle of the Atlantic and the campaign in North Africa would both have taken significantly
longer to win, with the result that the D-Day landings could not have been attempted in
1944. They also believe that the landings would then have failed or been impossible in
1945 without Ultra intelligence, in part because of attacks by V weapons and new weapons
that in our timeline Germany was not able to bring into operation. Thus, D-Day would not
take place until 1946, lengthening the war by two years. This rather Britain-centric view
seems to us to neglect the overall trend of the conflict. In particular, the rapid progress of
Soviet troops westward during the first half of 1944 suggests that the Red Army would
have reached Berlin well before mid-1947 with or without the D-Day landings. We also
note that the atomic weapons used in Japan in 1945 could have been deployed against
Germany if the war had lasted even a few months longer.
54. M. Hastings, The Secret War. New York, NY, USA: Harper Collins, 2016.
55. C. B. Burke, It Wasn’t All Magic: The Early Struggle to Automate Cryptanalysis,
1930s-1960s. National Security Agency, MD, USA, 2002. We expand on this point in
another paper, working title “Colossus in Context,” forthcoming in 2020 with Technology
& Culture.
56. de Grey to Radley, March 29, 1944, HW 62/6, TNA.
57. T. H. Flowers, “The design of Colossus,” IEEE Ann. History Comput., vol. 5, no. 3,
pp. 239–252, Jul. 1983.
58. T. Haigh, Ed. Exploring the Early Digital. New York, NY, USA: Springer, 2019.

ABOUT THE AUTHORS

Thomas Haigh is currently an Associate Professor of History at the University of
Wisconsin—Milwaukee and Comenius Visiting Professor for the History of Computing at
Siegen University. He has published widely on many aspects of the history of computing.
Read more at www.tomandmaria.com/tom.

Mark Priestley is a scholar of the history and philosophy of computing. He is the author of
A Science of Operations and Routines of Substitution and the coauthor (with Thomas
Haigh and Crispin Rope) of ENIAC in Action. Read more at www.markpriestley.net.

October/December 2018 27 www.computer.org/annals