The Receipt and the Work

The design file was never the work. It was the receipt, and it stopped working as one the moment anyone could produce one in ninety seconds.

Falk Gottlob8 min readNew

The short version

  • The design decision record replaces the design file as proof that judgment happened. The file was never the work. It was the receipt for the work, and it functioned as a receipt because producing one was slow, which made effort visible. Generation removed the cost, so the receipt stopped certifying anything. What certifies the work now is one page with four fields: the decision in a sentence, the constraints that produced it, what you gave up, and the signal that would tell you the decision was wrong. Ten minutes to write, and the field that hurts to fill is the one telling you the decision was thinner than you remembered.
  • Nobody ever hired a designer to produce Figma files. They hired a designer to make a series of calls under uncertainty, and the file was how the calls got shown to other people.
  • For twenty years that arrangement held up, because a good file was expensive. You could not produce one without having made the decisions inside it. The file was a receipt in the same way a printed boarding pass is a receipt. It is not the flight. It is evidence that something was paid for.
  • The cost is gone now, and the receipt is still being handed in.

What the receipt was quietly doing

Three jobs at once, and nobody separated them because there was no reason to. It proved effort: someone spent a week on this and here is the week.

It carried judgment. Buried in the spacing, the hierarchy, the copy, the states that got drawn and the ones that did not, was a stack of decisions nobody ever wrote down. The reviewer absorbed them by looking, and if they missed one, the designer explained it in the room.

And it was the record. Six months later, when a new engineer asked why the confirmation step exists, the answer was to open the file and look, or to find whoever made it.

All three broke at once, for the same reason. A credible screen now costs about ninety seconds. Effort is no longer readable from output. Judgment was never in the file anyway, it was in the designer standing next to it. And a file anyone can regenerate is a poor record of anything, because the next version will look just as finished and contain none of the same thinking.

So the receipt has to change shape.

Not because documentation is virtuous, but because the thing that used to certify judgment stopped certifying it, and design still needs a way to show its work.

What a record has to survive

Two tests, and they are harsh on purpose.

A competent stranger, or an agent, should reach the same decision from the record without you in the room. If reading it leaves someone needing a conversation with you, you wrote a meeting summary and put a spec's name on it.

And it should be readable in six months by someone who was not there and does not care about the meeting. Which rules out shared context, running jokes, and the state of the roadmap that week.

Four fields clear both.

The decision

One sentence, present tense, specific enough to be wrong.

"Suggested replies appear inline under the message, never as a modal, and never more than three at a time." That is a decision. Somebody can disagree with it, and somebody can catch a build violating it.

"We improved the suggestion experience" is a summary. Nobody can disagree with it, which means nobody can check it, which means it is doing no work.

The constraints that produced it

Two to four rules, each with the failure condition attached. This field does the heavy lifting, and it is why the record works on a stranger.

Rule first, then what breaking it looks like. "A suggestion the user can accept in one click has an undo that is equally cheap. Broken when accepting takes one tap and reversing takes a support ticket."

Writing these well is its own skill and it gets a chapter later in this series. For now, the test is whether the second clause is easy to write. If it is hard, the first clause is a principle wearing a costume.

When the surface is generated rather than drawn, the constraints field is often a range instead of a rule, and that has its own shape, covered in You Cannot Mock a Distribution.

What we gave up

Name the options you killed and what killing them cost. Modal picker with six options: higher acceptance in the first session, at the cost of interrupting people who already knew what they wanted to say.

Everyone skips this field. It is the one that makes the record worth opening later, and the only one that proves a decision took place at all. If you cannot name a real cost, you did not have a decision. You had a preference, and preferences do not need records.

It has a second use that shows up about two quarters in. Somebody proposes the option you rejected, confidently, as if it were new. The record turns that from a forty minute argument into a link.

How we will know this was wrong

One signal, a rough threshold, a date to look. Not a dashboard.

"If people are editing suggestions before sending rather than accepting or ignoring them, three was too few or the ranking is poor. Checked in the next round of sessions."

A record with no falsifier is an opinion with a template around it. This field is also the reason the record is not bureaucracy: it commits you to finding out, which is the opposite of what documentation usually does.

Do not write one for every button

The fastest way to kill this is to make it mandatory. A team that writes a record for every choice learns within a month to skim all of them, and then the four that mattered are buried with the two hundred that did not.

Write one when at least one of these is true. Somebody will ask why it is like this more than once. You picked one direction and killed others that were credible. The decision constrains what other people can build later. You would be annoyed if it got quietly reversed.

That last one is the most reliable trigger, and it is worth trusting. Mild irritation at an imagined future reversal is a good detector for decisions that are load-bearing.

Two habits decide whether any of this survives contact with a real team. The record gets written by the person who made the call, not by a note-taker, because delegating it turns it into meeting minutes. And it gets linked from wherever the work lives. A record nobody can reach from the work is a diary entry.

What changes when a team runs on records

Reviews get shorter, and they change character. The constraints and the killed options arrive before the meeting, so the meeting is about exceptions rather than a guided tour of finished work. That is the same move PM as Editor describes on the product side, and it lands harder in design because design reviews were built almost entirely around the walkthrough.

Reversals get cheaper too. A superseded record keeps its number, changes status, and points at whatever replaced it. Nothing gets deleted. Two years of those in one directory is the closest thing a design practice has to institutional memory.

And then there is what a portfolio of screens cannot show. Anyone can present twelve beautiful directions now. Almost nobody can show the reasoning that picked one, the option they killed, and the signal they set to catch themselves being wrong.

That is the promotion Design Just Got Promoted described, arriving as a filing convention.

This week

Take a decision you already made and had to argue for at least twice. Write the record after the fact, from memory, in ten minutes. Then notice which field was hard.

If the constraints were hard, the decision was made on instinct and has not been made transferable yet. If what we gave up was hard, there may not have been a real decision. If the falsifier was hard, nobody currently knows what would change your mind, including you.

The one page with all four fields, the write-one triggers, the filing rules, and a worked example is in The Design Decision Record. The short version to send to someone who wants the argument in two minutes is What replaces the design artifact as proof of work?


Chapter 2 of a series on the design operating model. Next: how to write a rule with a failure condition attached, instead of a principle nobody can catch being broken.

Share this post

Take the template

Design decision record, one page, four fields

Download

Frequently asked

What replaces the design file as proof that a designer did the work?+

A design decision record. One page, four fields: the decision in one sentence, the constraints that produced it with failure conditions attached, what you gave up and what it would have bought you, and the signal that would tell you the decision was wrong. A file shows what you made. A record shows why that one and not the eleven others, in a form somebody can act on without you in the room.

What goes in a design decision record?+

Four fields and nothing else. The decision, stated in present tense and specific enough to be wrong. Two to four constraints, each written as a rule plus the observable moment it was broken. The options you killed, with the cost of killing them. And one signal, with a rough threshold and a date to look, that would tell you the decision was wrong. It takes about ten minutes to fill if you write it while the decision is still warm.

When should you write a design decision record?+

Write one when at least one of four things is true: somebody will ask why it is like this more than once, you picked one direction and killed others that were credible, the decision constrains what other people can build later, or you would be annoyed if it got quietly reversed. If none of those apply, skip it. A record for every button is how a team learns to ignore records.

Why is what we gave up the most important field?+

Because it is the only field that proves a decision happened at all. If you cannot name a real option you killed and a real cost of killing it, you did not have a decision, you had a preference, and preferences do not need records. It is also the field that makes the record worth reading in six months, when somebody proposes the thing you already rejected.

Do design decision records slow a team down?+

They shorten reviews. When the constraints and the killed options are written down before the meeting, the review stops being a tour of work that is already done and becomes a ten minute conversation about the exceptions. The cost is ten minutes per record, and only for decisions that meet the write-one bar, which is a small fraction of what a team decides in a week.

What do you do with a design decision that gets reversed?+

Leave the record in place, change its status to superseded, and point it at the record that replaced it. Never delete. The reasoning behind a reversal is usually more useful than the reversal, because it is the only written trace of what the team learned between the two decisions. Sequential numbers, one directory, boring on purpose.

Related reading

Chapters and essays on the same thread, across both handbooks.

The short answer