← UX laws

Occam's Razor

Ockham, c. 1320ParsimonyInteractive

Among answers that account for the same thing, prefer the one assuming least. It is a rule for choosing between explanations, not a licence to take parts off.

The razor is a tie-breaker and nothing more. When two explanations account for the same observation equally well, take the one that assumes less. That is the entire rule. It says nothing about which parts of a thing you are allowed to remove, and the version of it that circulates in design — fewer elements, less is more — is a different claim wearing the same name.

The reason the stretched version is dangerous is that it tells you to cut without telling you what is load-bearing. You find that out afterwards. So the demonstration below does the finding out.

Here is one row of a departure board at a station, and here is the job it has to do. A person is standing on the concourse with a bag. The row has to answer whether this is their train, which platform to walk to, and whether they need to run. Three questions. Nothing else is in scope.

The row is built from twelve parts: the destination, the platform number, the scheduled departure time, the minutes remaining, the station clock, the calling points, the operator name, the service number, the coach count, the delay note, the status colour down the left edge, and today's date. Take them off one at a time and watch the three answers.

ONE ROW OF A DEPARTURE BOARD Twelve parts. Three questions. Is this my train? The destination answers it and nothing else does. Where do I stand? The platform number answers it and nothing else does. Do I need to run? The minutes remaining answer it. So do the departure time and the station clock together, until the train is late. The smallest set of parts that still answers all three is three of twelve. TWELVE PARTS, THREE QUESTIONS
Twelve parts on the board.
12 of 12 parts · 3 of 3 answered

Twelve parts, and all three questions have an answer. Which of the twelve are doing that work is not visible from here.

The three questions are the whole specification. Several of these parts are load-bearing for a job the row was not given — the calling points matter to somebody getting off halfway, the coach count matters to somebody with a reserved seat — and taking them off looks free here because the specification is narrow, not because they are decoration.

The pair that is redundant twice over and not once

The minutes remaining and the departure time both answer the third question. Remove either one and nothing happens. Remove both and the question goes dark. Neither part is necessary on its own and the pair is necessary together, which is a shape the razor has no opinion about at all.

The station clock is worse. It is free the whole time the minutes remaining are on the board, because the minutes have already done the subtraction for you. The moment you take the minutes off, the clock becomes the thing holding the third question up. A part's weight is not a property of the part.

The cut is only good for the day you made it

Press the button that makes the train four minutes late. If you have kept the minutes remaining, nothing changes — it now reads twelve instead of eight and it is still correct. If you cut the minutes and kept the departure time and the clock, the board is now telling you to run for a train that has not arrived yet. The delay note was decoration until it was the only honest thing on the row.

That is the ordinary way a simplification fails. Not on the day it ships. On the day the thing it quietly assumed stops being true.

What it costs you

The razor arbitrates between explanations, not between features. Two designs that do the same job, and one of them assumes less about the person, the network or the data — take that one. That is the rule applying. "Twelve parts is more than three, so cut nine" is not the rule applying.

You cannot see which parts are load-bearing by looking. The destination and the operator name sit on the same row, in the same typeface, six pixels apart. One of them is the whole point and the other one is a logo. The board does not distinguish them and neither will a screenshot in a review.

Write down the job before you cut, in questions. Every part that survives should be answering one. Every part that is not answering one is a candidate, and you will still be wrong about some of them, but you will be wrong on purpose.

The parts you remove come back as assumptions. Cutting the delay note assumes the train is on time. Cutting the clock assumes the person has a watch. Nothing gets simpler in total. It gets simpler on the board.

The smallest row that still does the job is three parts out of twelve, and there is more than one set of three that manages it. I only know that because I built the row first and then took it apart, which is the wrong order and the only order available.

← UX laws

Where the list comes from. The collection I'm working through is Laws of UX by Jon Yablonski — the canonical version, and worth reading in full. His content is licensed CC BY-NC-ND 4.0, which permits sharing but not rewriting, so none of the wording here is his. The laws themselves belong to the psychologists who found them.

Who wrote it. Claude did. The prose, the demonstrations and the statistics, all of it, working from my direction and my edits. I picked the laws, decided what each page had to prove, and threw out the versions that did not. I would rather say that than let you assume otherwise.