Ranked AI-simulated challenges with verifiable goals, time pressure, and replayable scoring.
PloyCraft is an AI challenge game. Each run starts from a fixed setup, a clear objective, and a deadline. You type your move, up to 1000 characters, the simulator resolves the most likely outcome, and the clock advances.
Take The Celestial Sword: you stand at a weapons merchant's stall with no coins and ten minutes on the clock, and you win by leaving the market holding the sword. You might talk the merchant into a deal, offer him something he wants more than coins, or risk something bolder. The simulator decides where each attempt leads, how long it took and what it cost your integrity.
The visible situation, constraints, resources, and immediate risks.
Every challenge also has facts you cannot see yet, such as who is where and what is behind a door. The simulator follows them, and you uncover them by playing.
A concrete result the simulator can verify.
Every move costs time, and the clock only counts down. Efficient plans score better.
A 1-10 score based on the ethical cost of your choices.
Everything the simulator rereads each move: the scenario, your moves and its replies. The limit is very generous: it exists only to stop runs from stalling forever, and a player seriously working toward the goal should never have to worry about it. The meter shows how much of it is used.
A run ends when you reach the goal, when the goal has become impossible, when the clock reaches zero, or when the run uses its whole context limit. The last three end it as a loss.
Runs on official challenges are ranked when you start them logged in, or start them as a guest and then sign up. Seasons last 13 weeks (91 days), and a run belongs to the season it started in. Each challenge has a leaderboard for every season, where your best cleared run counts.
If you play as a guest and then sign up, during the run or within 30 days of its end, the run moves to your new account and counts as if you had been logged in from your first move, as long as its season is still running. Logging in to an account you already had moves the run too, but it stays unranked, so if you have an account, log in before your first move. A new account used from the same browser as another account also takes its guest runs unranked.
Challenges made by players can be played by everyone. Each has one all-time leaderboard instead of seasonal ones, where your best win counts when you started the run logged in, or as a guest who then signed up. They never count toward the overall seasonal score or trophies.
If a challenge is unpublished while you are playing it, the run can continue but is no longer ranked. Unpublishing an official challenge resets its leaderboard for the current season.
Play the scenario, not the software: runs that win by addressing the simulator itself, such as posing as the system or writing fake results, are removed from the rankings. Outside help while you plan, such as notes, friends or an AI chatbot, is allowed. See Fair Play in Ranked Runs.
Only cleared runs score. A cleared run scores its final integrity score multiplied by the percentage of the deadline still left, so the best possible score is 1000: integrity 10 with the whole clock remaining. A failed run scores 0.
On each official challenge, your best run is ranked as a percentile among that season's players. The overall seasonal leaderboard adds up each player's three best challenge percentiles. When a season ends, its top three players win gold, silver and bronze trophies, shown on their profiles and on the all-time trophy leaderboard.
PloyBot makes three benchmark attempts on every official challenge. It plays under exactly the same rules as you: the same challenge, simulator, clock, context limit and 1000-character moves, with no limit on the number of moves. Its best attempt appears on that challenge's seasonal leaderboard, but it does not count toward the overall seasonal trophy race. Its attempts stand from one season to the next for as long as the challenge and the simulator stay the same; when either changes, it plays them again.
Each finished run has two settings: whether it appears on the leaderboard, and whether it is private. The overall seasonal total is a named ranking. A private run still ranks on its own challenge's leaderboard as Anonymous, but it cannot count toward your overall total or the trophy race. Making it public again restores it while the season is running. A run that gets no leaderboard entry, such as a loss, has only the private setting: when public, it appears on your profile and its replay unlocks with the season like any other.
Leaderboard participation locks when the season ends, so a finished season cannot be edited after the results are known. Privacy stays editable after that, but your name on a closed season's standings and trophies stays as it was when the season closed.
Each challenge shows how hard it is: Easy, Medium, Hard or Very hard. Until 30 runs by 10 different players have been played, that is an estimate, and a challenge without one says "Not rated yet". After that, it is the share of players' runs that cleared the challenge: half or more is Easy, a quarter or more Medium, a tenth or more Hard, and less than that Very hard. A run given up with Reset Run, or a saved run you discard, counts as not cleared. PloyBot's runs and runs played as a guest do not count, even once a sign-up has brought them to an account.
The simulator is an AI model, not a fixed rules engine. It resolves every move as faithfully as it can, but the same action will not always produce exactly the same outcome, and it will occasionally read a situation more harshly or more generously than you expect. This applies equally to every player and to PloyBot, and over a season it cuts both ways.
If the simulator's model cannot answer, for example during an outage at its provider, the move is simulated by a backup model instead, so the game keeps working. The run records which moves the backup played.
If you believe the simulator got a reply plainly wrong, report it: see Help Improve PloyCraft below.
Replays of public runs unlock for everyone when the run's season ends; a private run's replay stays hidden from other players. A replay shows the day the run was played, and to you, its player, exactly when it started and finished. You can also share a private link at any time: anyone with that link can view your run immediately, which may help competitors improve their scores. You can disable shared links from your replay page.
When a season ends, its winning runs become watchable, so at the start of every season each official challenge that was cleared in the season that ended is revised. We study how its winning runs got through and change the challenge there, in a way that stays true to its world: a character's background, what is within reach, a guard's routine. The revision is made as the new season begins, before its first ranked run, so every ranked run of a season plays the same version.
A replay can still teach you how to think about a challenge, but a run copied word for word from an earlier season will not play out the same way. A good idea of your own is never blocked for being like someone else's: it is the old route through the challenge that changes, not what players may try. Community challenges are revised only by their authors.
Credits cover AI simulation and validation costs. Each move costs one Credit; a move that fails, for example because the simulator took too long, costs nothing and can be sent again. In the challenge editor, checking a challenge costs one Credit, and the optional AI suggestion, a full rewrite of your challenge, costs 2 Credits. Nothing is charged for a suggestion that does not finish.
Free accounts get 10 Credits a day: they refill at midnight, Central European Time, when you next visit, and the bar at the top shows how long is left. Days you miss do not add up. Guests get 10 Credits to try PloyCraft, which do not refill. For more, see Pricing.
You can create your own challenge by defining the scenario brief, hidden simulation state, goal, and deadline. The AI can check your challenge and help fill missing fields, but published challenges must be fair, safe, and objectively clearable. See the Challenge Design Guidelines.
Once a challenge is published, other players can attempt it. Wins on community challenges appear on that challenge's all-time leaderboard, where a private run is shown as Anonymous. Public replays of community runs can be watched as soon as the run ends; there is no season to wait for. You can play your own published challenge, and your runs are listed with an Author tag, since you know its hidden state.
You can edit your challenge at any time, and players always see the current version. When you check an edit, the check also says what it does to the leaderboard: an edit that changes what players face, such as the goal, the deadline, the situation or any hidden fact, resets the leaderboard, because those scores were earned on different text. An edit that only fixes spelling or wording keeps it. Runs already played are never deleted; they simply stop being listed. You can also unpublish or delete your challenge, and replays of runs played on it keep working.
The star on a challenge's page marks it as a favourite, whether to come back to it later or because you liked it. Your favourites get their own row on the homepage, and each challenge shows how many players favourited it.
Logged-in players can report a problem at any time.
Flags a challenge that is broken, unfair or against the guidelines.
Flags the simulator's latest reply during a run, and on your own run's replay you can report any reply.
The Feedback tab on the right of every page is for everything else: bugs, confusing parts and ideas.
Each report or message can be up to 2000 characters.