Dice-lock's homepage reel plays at real speed and is 3 seconds shorterDice-lock's homepage reel plays at real speed and is 3 seconds shorter

September 23, 2026 · The homepage's demo video plays its lock and ability shots at real speed instead of slow motion and solves the puzzle three times faster, cutting the reel to 20.2 seconds.Dice-lock's homepage video now plays two clips at real speed instead of slow motion and solves its puzzle three times faster, cutting the video to 20.2 seconds.

Dice-lock's landing page carries a short video of the game in action, generated by a recorder that plays real gameplay through the engine and captures the frames. Two of its shots, LOCKED and ABILITIES, were recorded at double speed to compress them, the same trick used on the match and chain shots. On those two it read wrong. Instead of a quick demo it looked like slow motion, because a two-second lock and a five-die combo carry different amounts of visible action, and only some of that reads well sped up. The default recording speed is now 1x, with a shot free to set its own whole-number speed when it actually needs one. LOCKED is cut to 2.0 seconds, with the lock landing at 1250ms. ABILITIES goes to 2.7 seconds, with the fifth die landing at 2250ms.

The puzzle-solving shot had the opposite problem: it ran at the game's own pace, and watching a puzzle get solved one deliberate beat at a time is slower than anyone wants a demo to be. The recorder now runs three times faster through the wait between solve steps, but only inside the recorder itself. The function that actually advances a solve doesn't touch that wait at all, so a real player's puzzle still solves at the speed it always has. Only the video is sped up. That shot now runs 2.5 seconds. The whole reel, re-recorded end to end, comes to 20.2 seconds across 606 frames, down from 23.5 seconds before today's pass.

Separately, four tests in dice-lock's telemetry suite were one day from failing on a tree nobody had touched. They replay events through the real offline queue, which drops anything queued more than fourteen days ago, and the fixtures used a timestamp fixed at September 10th. That date would have aged past the fourteen-day limit at midnight on the 24th, failing four replay tests for no reason connected to any actual change. The fixtures now stamp their events a minute before the test runs instead of on a fixed date, so they can't age out from under a future run. Two tests that exist specifically to check the fourteen-day boundary still pass their own fixed time. Those are supposed to test aging, and would be pointless if it couldn't happen to them.

Dice-lock is a dice puzzle game, and its homepage carries a short video showing the game in action. That video is made automatically, by a program that plays real gameplay through the game itself and records what happens. Two of its clips, one showing a piece locking into place and one showing a special move, were recorded sped up to keep them short, the same trick used on two other clips. On those two it looked wrong: instead of a quick demo it looked like slow motion, because a two-second lock and a five-piece special move don't hold the same amount of visible action, and speeding up only works well for some of it. The video now records everything at its normal, real speed by default, and only a clip that genuinely needs to be sped up gets sped up on purpose. The locking clip is cut to 2.0 seconds, with the lock landing at 1250 milliseconds in. The special-move clip goes to 2.7 seconds, with the last piece landing at 2250 milliseconds in.

The clip showing a puzzle being solved had the opposite problem: it played at the game's normal pace, and watching a puzzle get solved one deliberate step at a time is slower than anyone wants a demo video to be. The recording program now waits three times less between each step of the solve, but only while it's making the video. The actual code that solves a puzzle during real play was left alone, so a real player's puzzle still moves at the speed it always has. Only the video is sped up. That clip now runs 2.5 seconds. The whole video, re-recorded from scratch, now runs 20.2 seconds and is made of 606 individual frames, down from 23.5 seconds before today's changes.

Separately, four tests that check dice-lock's usage-tracking system were about to fail on their own, on code nobody had touched. Those tests replay past events through the real system that queues data to send later, and that system throws away anything queued more than fourteen days ago. The test data used a fixed date of September 10th to stand in for "a little while ago." That date was about to cross the fourteen-day limit at midnight on the 24th, which would have failed four tests for no reason connected to any actual change in the code. The test data is now stamped one minute before each test runs instead of on a fixed date, so it can't go stale under a future run. Two other tests that exist specifically to check that fourteen-day limit still use their own fixed date on purpose. Those are supposed to test what happens when something goes stale, and would be pointless if it couldn't.