The waterfall paper that argued against waterfall

1970The paper, and what is in it

The document everybody cites for the sequential model contains the diagram, and contains, on the same page, the author's statement that implementing it that way is risky and invites failure.

The remainder of the paper is a description of what to do instead. The word by which the model is universally known does not appear in it.

What he actually recommended

Build it twice: a pilot version to discover what the difficult parts are, then the real one. Involve the customer throughout rather than at the ends. Plan for iteration between adjacent stages, and expect changes to reach back further than one stage.

That is not a modern iterative method. It is a heavily documented, specification-driven process with feedback loops and a prototype, written for large defence and aerospace systems. Claiming the paper as an early argument for present-day practice overstates it in the opposite direction.

1970s and 1980sHow the diagram became the doctrine

Subsequent papers cited it for the picture. A defence procurement standard published in the middle nineteen-eighties then required a sequential process with defined documents at each boundary, which made the model contractual for a very large body of work.

Once a model is written into procurement, arguments about its merits become irrelevant to whether it is used, and that is the mechanism by which it spread.

Why the diagram was attractive

It is legible to people who are not building the thing. Stages with boundaries can be scheduled, reported and paid against, and a percentage complete can be stated with a straight face.

Any process that admits it will revisit earlier decisions is harder to schedule and much harder to report, and the reporting audience is usually the one holding the budget.

What it got right

Worth defending, because the reaction against it overcorrected. Thinking before building is not a mistake. Writing down what is to be built is not a mistake. For systems where the requirements genuinely are fixed and the cost of a late change is enormous, a sequential process with heavy verification is the correct answer and remains in use for exactly those cases.

What it got wrong

The assumption that requirements can be known in advance for systems whose users have never seen anything like them. That is not a failure of discipline; it is a fact about the kind of thing being specified. People discover what they want by using something.

The model treats a change of requirement as a defect in the earlier stage, which makes it expensive to admit, which makes it arrive later, which makes it more expensive, which is the loop the paper itself warned about.

The citation pattern

A paper cited for the opposite of its argument, by people who reproduced its figure and not its text, for fifty years.

The mechanism is worth understanding rather than mocking. The diagram is portable and the argument is not: a picture can be lifted into a slide and a caveat cannot. Which is a general fact about how ideas travel, and it appears again in the next part of this document.

The other document that shaped practice by procurement

The pattern is not unique. Several working methods spread because they were written into standards, contracts or certification schemes rather than because they were shown to work, and the same is true of some current ones.

Which suggests a useful question when a method appears to be universal: whether it is used because it is effective, or because somebody has to sign something. The two produce identical adoption figures.

1990s onwardsWhat replaced it, and what did not change

The methods that displaced it share the paper's actual recommendation, which is short cycles with feedback, and add practices it did not discuss. They also acquired their own certifications, their own procurement language and their own diagrams.

The durable observation from this entry is not about either camp. It is that a portable picture defeats a careful argument, and that a method sufficiently widely adopted will be adopted for the wrong reasons by most of the people adopting it.

How to read any famous diagram

The practical habit this entry recommends is short: when a figure is cited as the source of a method, find the paragraph immediately after it.

Diagrams are extracted and captions are not, so the sentence explaining what the author thought of their own picture is exactly the part that does not travel. In this case it was one page away for fifty years.

What to do with the paper now

Read it, because it is short and because the experience of finding the caveat under the famous picture is the fastest available lesson in checking sources.

The general practice it recommends is the useful residue: identify the parts you do not understand, build those first at small scale, and expect the specification to change as a result. That advice is unaffected by anything that happened to its reputation.

What we cannot verify

The paper is published and can be read in full. The procurement standard is public. The path by which the model became doctrine is reconstructed from citation patterns and from accounts by people who were there, and the relative weight of the standard against ordinary citation drift is not established.