An instruction is a sentence that tells someone what to do. Most instructions work because the reader already understands the point of the job and quietly fills in whatever is missing. This lesson looks at what happens when the reader does not do that.

A literal follower is a reader who does exactly what the words say, adds nothing, and never guesses what the words were for.

Take the job of making a jam sandwich, written as instructions for a literal follower who has never seen a sandwich and does not know what one is for. The first instruction most people write is put the jam on the bread. Figure 1 shows three ways of obeying it.

The first follower places the closed jar on top of the loaf. The second empties the jar over the whole loaf, bag and all. The third puts one drop of jam on the outside of the crust. Every one of these is a correct reading of the words. The jar is jam. The loaf is bread. Placing one on the other is putting it on. If you objected, each follower could truthfully reply “that is what you asked for”, and none of them has disobeyed you.

This gap is the subject of the whole course. You wrote down what you meant. The follower did what you said. Those are two different things, and almost every mistake in computing lives in the distance between them.

When someone says a computer does exactly what it is told, the word “exactly” is not an exaggeration. Nothing inside the machine does the job a person does automatically when reading instructions.

A person reading your instructions runs a second process alongside the reading. They ask themselves what the instructions are for. They work out that you want a sandwich, and from then on they interpret every line in the light of that goal. If you forget to mention a knife, they fetch one. A machine has no such second process. If no knife is mentioned, there is no knife.

Figure 2 shows the instruction cut the bread given to both kinds of reader. The person picks a bread knife and cuts two even slices. That is helpful, and it is also unpredictable, because a different person would make a different choice. The machine finds no knife specified and stops with nothing to do. That is unhelpful, and it is exactly the same every time.

The literal column is the one we want. Consider what is lost with a helpful interpreter. If your instructions are filled in using somebody’s judgment, then their judgment becomes part of the result. The next person’s judgment will be slightly different, and so will the result. You could never be sure what your instructions actually do.

Predictability is the property that the same instructions always produce the same result. A machine that does precisely what it is told, every time, with no opinions of its own, has this property.

Predictability sounds dull, and it is the reason enormous things can be built out of small reliable pieces. Each piece can be trusted to do the same thing every time, so pieces can be stacked without the lower ones shifting. Literalness is not a limitation to work around. It is the foundation.

An instruction is pinned down when there is only one way to follow it. Three questions catch most of the gaps that let in extra readings.

Worked example. Apply the three questions to an instruction that sounds simple: put the socks into pairs.

1. Which one exactly? Which socks: all of them, or only the ones in this basket?
2. In what order? Paired by what: color, size, or both?
3. When do I stop? When the basket is empty, or when there are no unmatched socks left? These are different, because there may be a sock with no partner.

The third question is the one people most often forget, and it causes the worst trouble. An instruction with no clear stopping point can be followed forever, and a machine will do exactly that.

A pinned-down version reads: take socks one at a time from this basket. For each one, look through the socks already on the table for a matching one. If you find a match, put the pair in the drawer. If not, leave the sock on the table. Stop when the basket is empty.

This version is longer and duller, and there is only one way to follow it. Notice that it also handles the odd sock without being told to, because the stopping rule is about the basket rather than about pairs.

There are two reasons to practice writing instructions for a literal follower, even though almost nobody writes sock instructions for a machine.

The first reason is practical. This is what programming is, underneath every language and tool. Being able to say precisely what should happen, in what order, until when, is the skill. Everything else is notation.

The second reason is where this course is heading. Writing a task down as instructions is the hardest test of whether you really understand it. It is possible to be fluent at something and still be unable to say how you do it. Pairing socks felt obvious until it was written down, and then three decisions appeared that nobody had noticed making.

If you cannot write it down, you do not know how you do it — you just know that you can.

For socks, this is a small surprise, and a little care fixes it. For some other tasks the gap turns out to be permanent, and no amount of care will close it. The second half of this course is about those tasks.

The ordinary skills come first. Lesson 2 takes a job far too big to write in one go and breaks it into pieces small enough to be followed.