What can and cannot be known about the history of software

What survives from a decision

The manual, if somebody kept one. The standard, if there was one. The source, sometimes, and increasingly often now that preservation projects exist. A mailing list, patchily, where the discussion happened in public and somebody archived it.

That is a short list, and every entry in this document is written from it.

What was never written down

What the alternatives were. Why they were rejected. Who objected and on what grounds. What everybody in the room considered too obvious to state, which is usually the assumption that later turns out to have been the whole decision.

Those things existed in conversation and mostly nowhere else. No amount of archival work recovers them, because they were never in a form that could be archived.

The survivorship problem

The systems described in the literature are the ones that succeeded, or failed expensively enough to be investigated, or were built by people who wrote papers.

The ordinary majority — the business systems, the internal tools, the thousands of programs that worked adequately and were replaced — left almost nothing. Any general claim about how software was made in a period is a claim about the visible minority.

What oral history is worth

A great deal, and it is a specific kind of evidence. Recollection decades later is shaped by knowing how things turned out, by the narratives that formed since, and by the ordinary reconstruction memory performs.

It is also, frequently, the only account of intent that exists. This document uses it repeatedly and marks it wherever it does, which is the most that can be done: not discounting it, and not treating it as a document written at the time.

How folklore forms

A story that is memorable, explains something neatly, and has no traceable source. The origin of a card's dimensions in an earlier entry here is one. Several others in this field have the same shape: a satisfying causal chain, universally repeated, resting on nothing anybody can point at.

The tell is usually that the story is too tidy. Real decisions involve several small reasons pointing the same way, and folklore involves one large one.

2010s onwardsWhat is getting better

Source preservation at scale, archived discussion, oral history programmes recording participants while they are alive, and the ordinary fact that a great deal of current development happens in public with its argument attached.

A historian of the present period will have the opposite problem from one studying the nineteen-seventies: not a scarcity of record, but an unmanageable quantity of it, most of it noise.

The method this document used

Stated plainly at the end because it is the only thing here that transfers to the reader's own reading. Name the source and its year. Say what kind of source it is. Mark a recollection as a recollection. Give the popular version, then the document that contradicts it. And where the record does not settle something, say so and stop rather than choosing the better story.

None of that is difficult. It is simply slower than repeating what everybody knows, which is why what everybody knows is frequently wrong.

Reading a primary source is faster than it sounds

Most of the documents referred to in this work are short. A language report, a protocol specification, a conference paper of the period: these are typically a few dozen pages and written for people who had to implement them.

The barrier is not difficulty but the assumption that the summary is sufficient. In the majority of the entries here the summary was wrong in some specific way that the original settles in a paragraph.

What this document is for

Not nostalgia, and not the claim that older work was better. Several of the entries describe decisions that were correct then and are wrong now, and several describe designs that lost deservedly.

The use is predictive. A field that repeats its own arguments every fifteen years, under new names, is a field where reading the previous round tells you a great deal about the current one, including which objections were already answered and which were merely forgotten.

The limits of this document specifically

It is written in one language, from sources mostly in that language, about work done predominantly in a few countries. Substantial computing histories elsewhere are underrepresented here, not because they matter less but because the accessible record is what it is and the author reads what he reads.

It also selects for the legible: decisions with a document attached. The most consequential choices in software may well have been made in rooms nobody recorded, and this document would have no way of knowing.

What we cannot verify

Everything above is an argument about evidence rather than a claim about the world, and it rests on the experience of assembling the preceding forty-one entries. The preservation projects and oral history programmes named are public and their coverage can be examined. Whether the record misleads in the direction described here cannot itself be established from the record.