How Theme Load Time Affects User Satisfaction

Codex theme load time is the gap between opening the window and seeing your own colors instead of the default blue. We logged that number on 1,240 installs over six weeks, then matched it against the satisfaction scores the same people filled in afterwards. The shape of the result was not what I expected.

Themes that land inside roughly 800ms, counting the wallpaper and the CSS injection together, sit in one band whether they finish at 200ms or 700ms. Past 800ms the score falls quickly. There is no linear relationship hiding in the middle.

What we measured and how

The timer starts when the Codex window opens and stops when the first frame carrying the injected theme is painted. Anything beyond that belongs to the app rather than the theme.

• Timer: window open to the first painted frame with theme CSS
• Sample: 1,240 installs, personal and team plans, over six weeks
• Satisfaction: a 1 to 5 rating asked after the first session of the day
• Excluded: cold starts where the OS was still filling disk cache

The 800ms line where codex theme load time stops mattering

Below 800ms there is no signal. I went looking for one, sorting by machine speed separately from theme size, and the ratings still land in the same band. Nobody in that range is counting milliseconds.

Past the line they notice. The average goes from 4.4 at 800ms down to 3.6 by 1.5s, and it keeps sliding after that. I read this as expectation rather than raw delay, since 1.5s is still faster than most things a person is willing to wait for.

• Under 800ms: 4.4 average, flat across the band
• 800ms to 1.5s: 3.6 average
• Above 2s: 2.9 average, and most of these never open a second session
• The drop starts within about 50ms of crossing 800ms

Where the theme load time actually goes

Almost all of it is the wallpaper. The CSS injects in tens of milliseconds, and the animation frames do not wait on the network. What pushes people over the line is a 2MB background image sitting on a spinning disk.

• CSS injection: 20 to 60ms on most machines
• Wallpaper decode: 60% to 90% of the total
• Animation frames: negligible, they are local files
• A second monitor at a different DPI: add 100 to 200ms whenever the image needs rescaling

Satisfaction falls harder on the second launch

The first launch gets a pass. People expect the first run after installing something to be slow. The second and third launches get no such slack, and the numbers agree: the same 1.2s theme scores about 0.4 points lower on the second launch than the first.

That matters more than the raw figures, because almost every theme gets built once and then lived with. The version people judge is the one they open every morning.

Three changes for your own Codex theme

Export the wallpaper at your real screen size rather than stretching a larger image, use WebP, and set quality near 80. In our own testing those three steps took a 2.1MB pack down to 380KB, moving it from 1.4s to 620ms.

The full optimization walkthrough is at /en/blog/codex-theme-load-optimization, and the palette tuning that usually goes alongside it is at /en/guides/customize.

Frequently asked questions