Every screen in Dice Lock now closes the same way, without a drawn back arrowEvery screen in Dice Lock now closes the same way, without a drawn back arrow

November 5, 2026 · Dice Lock stopped drawing its own back button and let Escape, Android's back key, and a consistent X handle leaving every screen instead.Dice Lock's game screens stopped having their own back button, letting the phone's own back key, the Escape key, and a consistent close button handle leaving every screen instead.

Leaving a screen without an arrow

The chrome that frames every Dice Lock screen no longer draws a back arrow at all. chromeBar() stopped wiring one in, the old ARROW_UNTIL_REPLACED allowlist is gone, and every page now leaves the same way everyone else already left it: Escape on a keyboard, the Android back key, or an X on a sheet that closes the whole sheet. Checking all twenty pages for what that left behind turned up two real gaps, both fixed in the same change: Enter a Code now opens with the cursor already in its field, and a text field sitting outside the main menu or sheet area used to swallow the Escape key before it could close anything. A new probe, probe-back-keys.ts, now holds every page's exit live; an old one testing a mode that had already been cut months ago was deleted, since it could no longer run at all.

The same push folded Records, which used to be three separate screens behind three separate doors, into one page with BADGES, HISTORY and MILESTONES tabs across the top. There's no SCORES door any more: the RANK screen is the leaderboard now. Switching tabs redraws in place rather than opening a new screen, so Escape, the controller's B button and Android's back key still only need to go up one level to leave, and a finished or abandoned match now reopens straight onto the HISTORY tab, scrolled and filtered the way it was left.

The front door splits in two

The site's homepage used to be three doors into three parts of the game, crammed onto one page that also had to explain what Dice Lock was. It's now a single picture, one headline and two buttons that fit on one phone screen without scrolling: HELP TEST THE BETA, which goes to the signup page, and LEARN THE GAME, which goes to a separate /web page carrying everything the old homepage used to explain inline. Measured against the old page's own numbers, the new one loads in 848 milliseconds with 116.3 KB over 10 requests, and it was given its own speed budget rather than being allowed to just inherit the old page's ceiling. Old links built around the three-door layout, like the ones ending in #get-the-app, now forward straight to the sign-up page instead of landing on doors that no longer exist.

A quieter fix landed the same day in the app's build pipeline: an iOS archive can silently drop a feature. Capacitor regenerates its native plugin list from whatever is actually installed, not from what the app's package.json declares, so a plugin the code calls but npm never installed used to vanish without a warning, and the build still said BUILD SUCCEEDED. That's exactly what happened on 2026-09-28 with one plugin missing: four plugins synced instead of five, the archive signed and uploaded fine, and the feature that plugin provided was just gone from the running app. The archive step now checks every Capacitor package the app declares against what's actually installed and stops, by name, before it bundles or signs anything. The fix itself isn't new: it was written months ago on a branch that never reached main, and finally got ported across today.

Leaving a screen without an arrow

Dice Lock, the dice puzzle game this project builds, used to draw its own back arrow on screen so you could tap your way out of wherever you were. That arrow is gone now, everywhere. Instead every screen leaves the same way the rest of the phone already expects: the Escape key on a keyboard, the phone's own back button, or an X that closes a whole pop-up sheet in one go. Checking all twenty pages for anything this change might have broken turned up two real problems, both fixed at the same time: the Enter a Code screen now starts with the typing cursor already in its box, and a text box sitting in an odd corner of the screen used to swallow the back key before it could do anything. A new automatic check now opens every page and tries to leave it, to make sure this keeps working; an old check for a game mode that had already been removed months ago was deleted, since it could no longer run at all.

The same change also merged Records, which used to be three separate screens you reached through three separate doors, into one screen with three tabs across the top: Badges, History and Milestones. There's no separate scores screen any more, since the rank screen already shows that. Switching tabs just redraws the same screen rather than opening a new one, so the back key still only has to go up one level to leave, and finishing or giving up on a match now takes you straight back to the History tab, scrolled to where you left it.

The front door splits in two

The website's homepage used to be three separate doors into three different parts of the game, all crammed onto one page that also had to explain what Dice Lock even was. It's now just one picture, one line of text, and two buttons that fit on a single phone screen with no scrolling: one to sign up to test the game, and one to read about it, which takes you to its own separate page carrying everything the old homepage used to explain. Measured against the old page, the new one loads in well under a second, using less data over fewer requests, and it now has its own separate speed limit it has to meet rather than sharing one with a bigger page. Old links built for the three-door layout now send visitors straight to the sign-up page instead of to doors that don't exist any more.

A quieter fix landed in the app's build process the same day, for a problem that could have shipped an app missing a feature with nobody noticing. The tool that packages the game as an iPhone app builds its list of extra device features, like local storage, from whatever is actually installed on the computer doing the building, not from what the project's own records say should be there. So if a feature the game calls on was never actually installed on that computer, it used to just quietly vanish, and the build would still report success. That's exactly what happened in a test run: a feature that was never installed disappeared from the built app, and the tool said everything had worked. The packaging step now checks everything the project says it needs against what's actually installed, and stops by name, before it finishes building. The fix for this existed already, written months ago on a side branch that never made it into the main project, and it was finally brought across today.