Dice Lock's desktop window now plays the phone's layout, scaled upDice Lock's desktop window now plays the phone's layout, scaled up
October 31, 2026 · Dice Lock now shows the same single board on desktop as on a phone, just bigger, the same day a usage check that kept freezing started fixing itself.Dice Lock now shows the same single board on desktop as on a phone, just bigger, the same day a usage check that kept freezing started fixing itself.
last updated
One board layout now covers every window size
Dice Lock on a desktop browser used to draw a different arena entirely: two boards side by side, a status card bolted on, dice capped at 64 pixels no matter how big the window got. At 2560x1440 the board filled about 57% of the window's height. Today that layout is gone, and desktop plays the same single board the phone does, centred in a 460-pixel column and magnified by a zoom factor tied to the window's own height: 1 at 720 pixels tall, 1.28 at 1080, 1.4 at 1440, so a die that is 48 pixels on a laptop is 97 on a big monitor. The deck, menus, sheets and the Tower all moved into that same column over the day's run of tickets, and Large Board and its walkthrough are turned off there, since neither has a reason to exist in a window that already zooms.
The old two-board code did not get flagged off, it got deleted outright: 39 CSS rules for status cards, meter columns and the desktop duel, the restart button on the top bar, the two player cards, 243 net lines gone. Restart still lives on the pause menu, nothing was left half-working behind a switch, and the 2,391 tests covering the old shape were replaced one for one rather than dropped. A headless-Chrome check now measures the column at five window sizes from 1280x720 up to 2560x1440, in both light and dark, with no scrolling and no snap when a window resizes.
A null reading from an upstream endpoint froze every account's usage status
At 19:01 UTC the usage endpoint this project reads started sending one of its fields, a five-hour usage window, as null instead of a number. The check that reads it treated null as a stray zero, passed that through its percent test, then crashed reading a field that did not exist on a null value. That one crash took the whole refresh down with it: every account's status froze at that timestamp, and the jiolab account started showing usage unknown. The fix is one line: skip a window sent as null instead of reading it as zero percent, so a gap in the upstream data is treated as missing rather than crashed on.
The companion piece is what happens the next time it stalls. The launcher that starts a Claude session now notices when an account has no usage reading, or a stale one, and kicks off that same check in the background instead of leaving the number stuck. Launching does not wait on it, median launch time moved from 51 to 54 milliseconds across ten timed runs, and a five-minute lock stops two launches in a row from firing off two fetches at once. Separately, the two jobs that publish this blog and this project's generated documentation used to fast-forward their shared checkout before pushing it, which meant a push rejected by a moved-on origin left that checkout ahead of a remote nobody else could see. That had happened 19 times in Dice Lock's repository since October 16. Both jobs now push from their own worktree first and only fast-forward the shared checkout once that push has actually landed.
Dice Lock looks the same on a desktop browser as on your phone now, just bigger
Dice Lock used to look like a different game on a desktop browser than it does on a phone: two game boards side by side, an extra status card bolted onto the screen, and dice stuck at a small fixed size no matter how big the window got. On a large monitor the board filled only about 57% of the window's height, leaving the rest empty. As of today that separate desktop layout is gone. A desktop browser now shows the exact same single board a phone shows, centred in the window and scaled up by an amount tied to how tall the window is: no scaling at all at 720 pixels tall, about 28% bigger at 1080, about 40% bigger at 1440, so a die that is 48 pixels across on a small laptop is 97 pixels across on a big monitor. The card deck, the menus, the pop-up panels and the Tower mode all moved into that same scaled column, and a feature called Large Board, which makes the dice extra large on a phone, is turned off there, since a window that already scales everything up has no need for it.
The old two-board version was deleted outright, not switched off behind a setting: 39 separate style rules for the status cards and the old side-by-side layout, the restart button that used to sit at the top of the screen, and the two player name cards, 243 lines of code gone in total. The restart button still works, it just lives on the pause menu now. All 2,391 automated checks covering the old two-board layout were rewritten to cover the new single column instead of just deleted, and a new check now screenshots the column at five window sizes, from a small laptop up to a large monitor, in light and dark, confirming nothing scrolls off screen and nothing jumps when a window is resized.
A missing number from an outside service froze everyone's usage status, and the system now catches itself
This project tracks how much of its own Claude usage allowance each account has used, by reading a report from an outside service. At 19:01 UTC that service started leaving one number out of its report entirely, instead of sending a number for it. The code reading that report treated the missing number as if it were a zero, which then made it try to read a second piece of information that only exists when there actually was a number, and crashed. That one crash took down the status check for every account at once: the status froze at that exact timestamp, and the account named jiolab started showing its usage as unknown. The fix is small: when that number is missing, the code now skips it instead of treating it as a zero and crashing, so a gap in what the outside service sends back is treated as missing rather than something to crash on.
Alongside that fix, the program that starts a new Claude working session no longer just shows a frozen or missing number. It now notices, when it starts up, that an account's usage number is missing or stale, and quietly re-fetches it in the background while the session starts anyway: across ten timed launches, the median start time went from 51 to 54 milliseconds, and a lock stops two sessions started close together from both re-fetching at once. Separately, the two automated jobs that publish this blog and this project's internal documentation used to update their own local copy of the live website before sending their finished work to the shared copy everyone else works from. When that send failed because someone else's work had landed first, the shared copy was left ahead of what everyone else saw, which made the next ordinary save by a person or another job bounce for the same reason, something that had already happened 19 times in Dice Lock's codebase since October 16. Both jobs now send their work up first, confirm it arrived, and only then update their own copy.