Dark Welcome illustration is a forecast scene #44

New Issue

Closed

opened 2026-08-19 01:02:36 -05:00 by null · 0 comments

null commented

2026-08-19 01:02:36 -05:00

Owner

In dark mode, onboarding step 1 shows a calendar with a predicted-days arc — the forecast picture — where light mode shows overlapping cycle rings.

What is true now

docs/design/dist/Dark/01_WelcomeIllustration_dark.webp draws a rounded calendar, an arc of day-beads running into dashed circles, and an arrow. That is the subject 07_Forecast is for. Its light twin, Light/01_WelcomeIllustration_light.webp, correctly draws three overlapping rings over a dawn landscape.

So art_welcome in drawable-night-* is a forecast scene, and the first screen a dark-mode user ever sees illustrates the wrong idea — and the same idea step 7 will show them six screens later.

How it got here, because the earlier diagnosis was wrong

An earlier check found Dark/01 and Dark/07 byte-identical and recorded it as "the forecast's dark illustration is a copy of the welcome one". That was backwards. Both were forecast scenes. Replacing 07 with a correctly-drawn forecast fixed the duplicate hash and left 01 still wrong, which is why the set passed a uniqueness check afterwards.

The lesson worth keeping: a hash check proves two files differ, not that either is right.

What it costs

Light mode is correct, so this is invisible unless somebody opens the app in dark mode — the same silent shape as a missing -night asset, which is what #42 is about. It is cosmetic rather than harmful, but it is the product's first impression and it undercuts the step's own copy: "Know what's coming" over a picture of a prediction the user has not made yet.

What to do

Re-render 01_Welcome dark only, to match its light twin: overlapping translucent cycle rings over a night landscape, crescent moon, botanicals to one side, one rose bead on the outer ring. 3:2, 2048 × 1365, composed natively wide. The base prompt and per-scene line are in docs/design/dist/README.md and BRAND_GUIDE.md §22.

Traps

Do not re-render 07 as well. Its dark version is good and states "predicted" more clearly than its light twin does — if the two are ever reconciled, light 07 is the one to change.
The set must stay free of text, step numbers, and named forecast fields.
Check the result against Light/01 side by side, not on its own. This defect survived because each file was judged alone.

Verify: docs/design/dist/Dark/01_WelcomeIllustration_dark.webp draws overlapping cycle rings and no calendar, its hash differs from every other file in the set, and onboarding step 1 shows the same subject in both themes on a device.

In dark mode, onboarding step 1 shows a calendar with a predicted-days arc — the forecast picture — where light mode shows overlapping cycle rings. ## What is true now `docs/design/dist/Dark/01_WelcomeIllustration_dark.webp` draws a rounded calendar, an arc of day-beads running into dashed circles, and an arrow. That is the subject `07_Forecast` is for. Its light twin, `Light/01_WelcomeIllustration_light.webp`, correctly draws three overlapping rings over a dawn landscape. So `art_welcome` in `drawable-night-*` is a forecast scene, and the first screen a dark-mode user ever sees illustrates the wrong idea — and the same idea step 7 will show them six screens later. ## How it got here, because the earlier diagnosis was wrong An earlier check found `Dark/01` and `Dark/07` byte-identical and recorded it as "the forecast's dark illustration is a copy of the welcome one". That was backwards. **Both were forecast scenes.** Replacing `07` with a correctly-drawn forecast fixed the duplicate hash and left `01` still wrong, which is why the set passed a uniqueness check afterwards. The lesson worth keeping: a hash check proves two files differ, not that either is right. ## What it costs Light mode is correct, so this is invisible unless somebody opens the app in dark mode — the same silent shape as a missing `-night` asset, which is what #42 is about. It is cosmetic rather than harmful, but it is the product's first impression and it undercuts the step's own copy: "Know what's coming" over a picture of a prediction the user has not made yet. ## What to do Re-render `01_Welcome` dark only, to match its light twin: **overlapping translucent cycle rings** over a night landscape, crescent moon, botanicals to one side, one rose bead on the outer ring. 3:2, 2048 × 1365, composed natively wide. The base prompt and per-scene line are in `docs/design/dist/README.md` and `BRAND_GUIDE.md` §22. ## Traps - **Do not re-render 07 as well.** Its dark version is good and states "predicted" more clearly than its light twin does — if the two are ever reconciled, light 07 is the one to change. - The set must stay free of text, step numbers, and named forecast fields. - Check the result against `Light/01` side by side, not on its own. This defect survived because each file was judged alone. Verify: `docs/design/dist/Dark/01_WelcomeIllustration_dark.webp` draws overlapping cycle rings and no calendar, its hash differs from every other file in the set, and onboarding step 1 shows the same subject in both themes on a device.