The Ladder replaces the Daily, and six AI testers now try every build firstA new ranked puzzle mode replaces the old one, and six AI players now test every new version first

October 27, 2026 · Dice Lock's Daily puzzle gives way to the Ladder, a 90-day ranked season, while six AI testers now play through every new build before a person does.Dice Lock's old daily puzzle mode is replaced by a 90-day ranked ladder, while six computer-controlled testers now play through every new version before a person does.
last updated

The Ladder is live

Dice Lock's puzzle tab has had one ranked mode for a long time: the Daily, played once a day with a leaderboard. Today it has a different one. The Ladder opened on the live site, and Season 1, named Square One, started at 00:00 UTC, the same clock the server already used for streaks and the Daily's own days. It runs ninety days, ending 2027-01-25, and the puzzle page now says so: "Season 1 · Square One · 90 days left."

Climbing it means solving boards pulled from a bank built for the purpose, 7,500 puzzles a season spread across the difficulty range, with the server rating every solve and a level that moves one step per two score points. Three lives regenerate over time, a season rank tracks everyone against everyone, and anything past level 30 comes from that server-side bank rather than the phone generating its own board, so no player sees the same puzzle twice in a season.

None of this reached a real player the moment the code was ready. The server side, the rating and its event log, had already been live in production, waiting on three things: the hard-board bank held only 50 preview puzzles, nowhere near enough to survive real play; the Ladder's screens still needed a last look against the approved design before they shipped; and Season 1 had no start date, so its countdown had nothing to count from. All three landed today, in that order, and only once they had did the Daily come down: its engine, its routes, its thirteen error codes, its home-screen tile and its drawer entry, removed in the same push. The streaks it tracked carry over rather than reset: the badge for a first clean solve now fires on a Ladder board, and the seven-day streak counts clean Ladder days. An old link to the Daily still works; it lands on the Ladder now, with a notice for anyone on an app old enough to need one.

Six AI testers now try every build before I do

The second thing that shipped today runs in the background rather than on the puzzle tab. Checking a new build against real play used to mean someone sitting down and playing it, or waiting for production numbers to come in. Today's commits built a harness that runs a build through six blind testers instead: a novice, a quick player, a strategic one, a legend-tier player, an explorer and an everyday player, each its own separate session with no memory of the game beyond a short persona description, driving its own browser against a copy of the build running on a scratch database.

Each tester sees only what the screen actually shows: the visible text, the buttons, a plain description of the board while a match is in progress, and a screenshot only when something changes, a new screen, a popup, a tutorial step, or the end of a match. Time runs on a fake clock rather than the real one, so a match can play at a quarter, half, or full speed depending on how fast that persona is meant to think, or one move at a time for the legend persona. Each tester writes down what it notices as it plays, and the strategic and legend personas keep notes across runs, so something one of them flagged against an earlier build gets checked again on the next one rather than forgotten. A run is budgeted by how many turns a tester takes, not by how many times the harness is called, and a screenshot counts as a fraction of a turn rather than a whole one, so a tester that stops to look doesn't quietly cost more than one that just plays. The explorer persona works differently again: it keeps exploring until something tells it to stop, rather than stopping at a fixed count.

Every run writes its findings to the same page format used for the reports shown after any other change, so the result is something to open and read rather than a log to piece together by hand.

A new ranked puzzle mode replaces the old one

Dice Lock, the dice puzzle game, has had one puzzle mode you play for rank: the Daily, a single puzzle each day with a leaderboard. Today that mode is gone, replaced by a new one called the Ladder. Season 1, named Square One, started today at midnight, by the one clock the game uses no matter where a player lives, and runs for ninety days, ending on 2027-01-25.

Climbing the Ladder means solving puzzles pulled from a big pre-made set built just for this, 7,500 puzzles for the season, spread from easy to hard. Every solve moves a rating up or down, and a level climbs with it, one step for every two points scored. Players get three lives that slowly refill, plus a season-long rank against everyone else playing. Past a certain level, the harder puzzles come from that pre-made set rather than being made up on the spot, so a player never sees the same puzzle twice in one season.

None of this reached a player the moment it was built. The behind-the-scenes half of the Ladder, the part that stores ratings and hands out puzzles, had already been live and waiting on three things: the set of harder puzzles had only 50 preview ones ready, nowhere near enough for real players; the Ladder's screens still needed a last look against the approved design before anyone could see them; and Season 1 had no official start date yet, so there was nothing for its countdown to count from. All three got finished today, in that order, and only once they had did the old Daily mode get removed, along with its screen, its error messages, and its icon on the home screen, all at once. Streaks carry over instead of resetting: the badge for solving a puzzle cleanly now counts a Ladder puzzle, and the seven-day streak counts clean Ladder days. An old link to the Daily still works, it just sends the player to the Ladder now, with a message for anyone on a copy of the app old enough to need an update.

Six computer-controlled testers now try out every new version before a person does

The second thing built today works behind the scenes. Checking whether a new version plays well used to mean someone sitting down and playing it, or waiting on real player data afterward. Today's work built a setup that instead runs a new version through six computer-controlled testers: a novice, a quick player, a careful strategic player, a top-tier "legend" player, an explorer who wanders rather than aims for score, and an everyday casual player. Each one is its own AI session, starting with only a short description of that kind of player, and each controls its own browser against a copy of the game on a throwaway practice database.

Each tester only sees what a real player's screen would show: the visible words, the buttons, a plain written description of the board while a match is happening, and a picture of the screen only when something changes, such as a new screen, a popup, or the end of a match. Time itself is faked rather than real, so a match runs at a quarter, half, or full speed depending on how fast that tester is meant to think, or move by move for the legend tester. Each tester writes down anything worth noticing, and the two toughest testers keep notes from one run to the next, so something flagged earlier gets checked again rather than forgotten. How much a tester can do in one run is measured in moves taken, and a look at the screen counts as part of a move rather than a whole one, so pausing to look doesn't cost more than playing straight through. The explorer tester works differently again: it keeps wandering until told to stop, rather than stopping after a fixed number of moves.

Every test run writes up what it found as a finished page, so the result is something to read, not raw notes to sort through by hand.