What a Logo Has to Survive

Duskglow had a sparkle for a logo. Four-pointed star, a small plus beside it, cream on brown. Tonight it is gone, replaced by an eclipse, and the part worth writing down is how I got there, because I did not pick the mark that looked best on a big screen. I picked the one that held up everywhere it would actually appear.

Why the sparkle had to go

The sparkle is the icon half the AI products I use put on their AI features. That is exactly the problem. It has come to mean "something magic happens here," and the one thing I do not want a nightly journal to promise is magic.

Duskglow tells you plainly, in the conversation, that Luna is an AI and not a person or a therapist. I think that plain version is the right one for an app people open at the end of a hard day, and a logo that leans on the AI look pulls the other way. It brands the app by its technology instead of by what it is for. Polishing the sparkle would not have fixed that, because the trouble was in what it said.

I asked for questions instead of more options

The first round was six mock-ups, drawn in the app's real colours and sent to my phone, since that is where I judge anything visual. I liked two of them. A solid lantern, which matches the lantern on the home screen, and a sun sitting on the horizon, which pictures the name itself.

I could not choose, and another round of options would not have helped, so I asked for questions instead. They were good ones. What should the icon tell someone before they open the app? Should the logo be the same lantern that shows your progress? If clear-at-tiny-sizes and unlike-any-other-app pull apart, which wins?

My answers surprised me a little. I liked the sun because it describes Duskglow, and I was worried it would read as a weather app. I had no strong feeling about the lantern's role. And I picked clear at tiny sizes over distinctive, which I would not have predicted.

Then I contradicted myself, and it is worth being honest about. Asked whether people would describe Duskglow as a journal they write in or someone they talk to, I said a journal. Then I picked the plain horizon to refine, which says nothing about journals. The response pointed out the mismatch instead of quietly following the last thing I said. I kept the horizon anyway, with the journal part left to the words beside the mark, and I am glad the contradiction was said out loud before I locked anything.

The answer was not on the list

Nineteen mock-ups in, I suggested something none of them were. An eclipse, with a spark in the corner.

That idea came straight out of the questions. An eclipse is light at the edge of darkness, which is close to what a nightly journal is for. It cannot be mistaken for a weather icon. And a ring is about the simplest shape there is, so it survives at sixteen pixels.

The spark needed work, and this is where listing failure conditions earned its keep. Drawn as a four-pointed star, the spark was the sparkle again, the exact thing I was retiring. Drawn as a round dot in the corner, it had a different problem: on Android a dot in an icon's top corner means you have an unread notification, so the app would look permanently like it was waiting for you. That is the kind of quiet pressure I have spent months keeping out of this product.

The version that survived puts the spark on the ring itself. A single bright bead of light on the edge of the eclipse, which is also what a real eclipse looks like a moment before totality. I tried a faint moon surface inside the ring and took it back out. The empty middle was stronger.

The mock-ups drew a glow the phone never would

One correction came from my own side of the process. For three rounds every mark sat on a soft glow on the "launch screen," the brief screen you see while an installed app opens. The real launch screen is one flat colour with the icon placed on top. There is no glow unless the icon image carries it. So for three rounds I was judging a launch moment that would never happen.

It mattered because the launch screen was half of why the logo was on my list at all. The old one showed a lighter square around the sparkle when the app opened, and that square turned out to be the icon's own background colour, sitting on a slightly different launch colour behind it. Two near-identical browns that did not quite match. The new icons are painted in exactly the launch colour, so the mark sits on the screen with nothing around it.

The first time I installed the test version on my phone, it still showed the old sparkle. The second time, the eclipse. I never fully isolated why. The most likely answer is that I was looking at the real app's icon, which keeps its old image until the phone refreshes it, beside the test one with the same name. I now ask how to tell two same-named things apart before I trust a check like that.

Mental Models

Pick a visual mark by where it can fail. A logo spends its life at sixteen pixels in a browser tab, in a circle on a home screen, on a dark launch screen and on someone else's sign-in page. Judge it there first, and judge it big last.

When the options stop helping, ask for questions. Another round of choices anchors on the choices. Questions about what the thing has to do pulled out constraints I had not said, and the best answer came from me once they were on the table.

A mock-up is only as honest as what it leaves out. If it draws something the real screen will not, it is showing you a different product. Check the mock-up against how the platform actually renders before trusting your own reaction to it.

Decoration says something too. A logo that borrows the industry's AI shorthand makes a promise the product does not mean to make. For an app that tries to be plain about what it is, the mark should be plain about what it is for.