What it can do is not what it says

A lesson from Why Good Design Dies. Play it above, or read it through below.

• It can be pressed, or it cannot. Nobody's eyes are in that sentence.

  The common read: Affordance is the visual signal that an element is interactive, so adding a shadow to a card adds an affordance to it.

  A shadow adds no affordance. The word names what an object's actual properties let a person do with it, and the property is there whether or not anybody perceives it: a borderless region of a header that opens the account menu on click affords that click at three in the morning, with the lights off and nobody looking. This is a framework somebody proposed, not a result somebody measured. James Gibson put it forward in 1979. It isolates one variable.

  Take the provenance honestly, because provenance is what this course sells. Gibson's own sentences were not read for this lesson. The only copy reachable here is a publisher's page confirming the book, the 1979 date and a chapter titled The Theory of Affordances, with no definition in Gibson's words anywhere on it. What reaches you is Norman's characterisation of Gibson, published under Norman's name. That is a weaker thing than Gibson and a stronger thing than the summary you have read elsewhere.

  So the test has a yes and a no in it, and no observer. Can the action happen. A settings row that opens its detail screen on tap can, with no chevron, no divider and no fill anywhere near it. An order button that the address form has disabled cannot, in the full primary colour, with a hover state. Both sentences stay true when the person looking at the screen changes, which is the property the word was coined to carry.

  It arrives as a ticket titled "the order button does nothing", filed against the front-end build by somebody who read the colour and not the state, closed with a comment saying the address had not validated, and filed again by four more people the same week. Split the question before anybody argues about the fix. Ask what the element can do, then ask what it shows. The first has an answer somebody can check in the build; the second is the one the room has opinions about.

• The cue is a second thing, and it is the only one anybody acts on.

  The common read: The tap still works and nothing in the build changed, so taking the borders off the settings list took nothing away.

  Taking the borders off took something away, and it was not the tap. Norman's word for what left is signifier, and the definition he published is deliberately wide: "A 'signifier' is some sort of indicator, some signal in the physical or social world that can be interpreted meaningfully." Read what is not in that sentence. Not a shadow. Not a border. Not a control, and not a screen. A signal, and a person who can read it.

  This is a choice Norman made in 2008, not a result he measured, and it is worth taking on those terms. He needed a word for the perceivable half, and his line for how the two connect is the one to hold onto: "The perceivable part of an affordance is a signifier, and if deliberately placed by a designer, it is a social signifier." Deliberately placed. That is the designer's half of the sentence, and it is the only half a designer can do anything about.

  The second variable has its own yes and no, and the first one does not answer it. Is there something a person can perceive that announces the action. A chevron is one. A divider is one. A fill is one. A hover state is one on a pointer and nothing whatsoever on a phone. A help page is one, badly placed, reaching only the people who already went looking for it. None of them changes whether the tap works, which is exactly why the two questions come apart.

  It arrives as a support thread saying the settings do not open any more, filed against a release that changed no behaviour at all, and closed by somebody who re-tested the tap and found it fine. Then it arrives again the next quarter. Say the missing half out loud in the review, in those words: the action is there and nothing announces it. The sentence costs three seconds. Not saying it costs the next hour, spent on the half that was already working.

• Two variables, four cells. Most reviews have a word for one of them.

  The common read: A thing that looks pressable and a thing that is pressable are one claim, so one fix covers both.

  The order button sits in full primary colour with a live hover state, and the address form has disabled it, so pushing its contrast and weight up until nobody can miss it

  They are two claims and they fail separately. Can the action happen, and does something perceivable announce it: two yes-or-no questions over one element, which is four states and not two. Both present is the case nobody files a ticket about. The other three are the whole of the job, and only one of them has a name a review thread can reach for without stopping to explain itself first.

  Action present and announcement absent is the invisible one. The row taps, the header region opens the menu, and almost nobody finds either. Announcement present and action absent is the false promise, and it is the one that reads as finished in a static mockup: the order button in the full primary colour that the address form has disabled, the drag handle whose behaviour did not ship, the full-fidelity picture of a search field inside an onboarding overlay that people keep tapping.

  The false promise has a property worth its own sentence, because it inverts the instinct. The obvious fix makes it worse. Raise the disabled button's contrast and weight and the refusal is advertised more loudly than before. A component library's variant list reads "Button / Primary / Disabled" directly above "Button / Primary / Disabled / Hover". The fourth cell, neither present, is a line of body copy with no behaviour and no cue, and it is the only one of the four that is usually correct.

  It arrives as a review thread where three people agree the card needs work, a shadow ships, and nothing about the click rate moves, because the route behind that card was never wired and nobody in the thread asked. Name the cell before anybody names a fix. It is one sentence, it is checkable against the build rather than arguable across a table, and it is the difference between repairing the half that is broken and polishing the half that is not.

• Norman took the word back. The one everybody uses is the one he replaced.

  The common read: Affordance means the visual cue that shows something is interactive, so "this card needs more affordance" is a sentence about a drop shadow.

  That sentence is about a drop shadow, and the word in it is the wrong one, and the person who says so is the man who popularised it. Don Norman published "Signifiers, Not Affordances" in Interactions in 2008, and his account of what happened to his own term is one sentence: "Since then, the term has been widely used and misused. The result has been confusions and a goldmine for academic scholars who get to write learned articles about the true meaning of the term."

  The correction is not a tightening of the old word. It is a second word, taking over the job the misuse had handed to the first: "The perceivable part of an affordance is a signifier, and if deliberately placed by a designer, it is a social signifier." Perceivable part. The affordance is the whole thing, the signifier is the half you can see, and collapsing them leaves a vocabulary with a name for the visible half and no name whatsoever for the other one.

  The usage you hold is not carelessness and nobody in your review invented it. It is the dominant usage in design practice, and Norman's own 1988 popularisation is what set it running. What it costs is small and specific. When the missing half is the capability, a vocabulary with one word in it produces a thread about shadows, and that thread has no way to say the thing that is actually wrong.

  It arrives as a critique thread where somebody writes that the card needs more affordance and means it needs a drop shadow, filed against a visual mockup with no working build behind it, and approved by two people who read the word the same way. Say the other half out loud. "Do you mean the cue, or do you mean the click?" costs four awkward seconds in a review, and not asking it costs a sprint spent on the half that was already there.