Ranking rules

Periods

• All periods run on Korea Standard Time (UTC+9) and flip at midnight.
• Weeks start on Monday. Week numbers follow ISO 8601.
• Months start on the 1st.
• A season is three months. Jan–Mar is Season 1, Oct–Dec is Season 4.
• Season numbers restart at 1 every year.

Titles

• When a period ends, ranks 1 to 100 get a title.
• Periods with fewer than 3 players give no titles.
• Each title shows the year, the period and how many played.
• Season #1 is shown as Champion.
• Titles are records. They never turn into money or items.
• Titles can't be sold, traded or given away.

Rank colors

Every color also comes with a label and a shape.

• ◆1st
• ◈2nd
• ◇3rd
• ▲Top 10
• •Top 100

How records are measured

• When you open a game, the server hands out a start ticket.
• Times are measured by the server, not your screen clock.
• For 2048 the server replays your moves and counts the score itself.
• For 15 Puzzle the server builds today's board, replays your slides and counts the moves. If the board doesn't end solved, the run is dropped.
• For Snake the server replays the direction you moved on every tick and counts the score.
• For Breakout the server replays your paddle position frame by frame through the physics and counts the score.
• For Puzzle Dungeon the server replays the dungeon rules (floors, counterattacks, relics) and counts floors and HP left.
• For Daily Solitaire the server replays your moves on today's verified deal and counts the actions.
• For vs AI games in Dwijib, Omok, Ojeom, Baduk, Janggi, Checkers and Paldo Yuram, the server takes your moves, replays the whole game with the AI's replies, and counts the result itself.
• In the Typing test daily race, the text you typed must match that day's text character for character. Runs faster than 30 ms per keystroke in Korean or 48 ms in English are filtered.
• Troy Tactics, Odyssey Quiz, The Capturing Move, Daily Solitaire and Sonpulgi only submit runs with Compete turned on. Regular runs send nothing.
• The stock rooms and the Gambitry ladder keep their own boards. When a month ends, the app hands over its final standings and we give titles from them.
• Only your best game per period counts.
• Ties go to whoever got there first.

What we can't stop

• We can't tell if a program solved the puzzle for you.
• We can't fully stop one person using several browsers.
• Puzzle Dungeon doesn't replay the moves inside each puzzle. It uses the puzzle results the app sends (mistakes, correct answers) and only drops values that puzzle can't produce.
• Final standings from the stock rooms and the Gambitry ladder are computed by those apps. We don't recompute them.
• For Troy Tactics, Odyssey Quiz, The Capturing Move and Sonpulgi, the server only measures time. Whether you won, got every answer, solved every problem or hit every target is what the app reports.
• The AI always plays the same moves, so memorizing a winning line or copying the AI's own moves wins every time. We don't filter that. Ratings stop at the AI ceiling.
• Server-measured time includes a little network delay. On short records like Sonpulgi's aim drill, that can show.
• Kankan's boards are baked at build time, so there's no way to replay and verify a solve. The server only measures time from the start ticket to submission; the screen decides whether the board is fully filled.
• Odd records get removed by hand after a report.
• The real-name filter only checks Korean surname patterns.

Filtered records

Monthly counts of records we filtered, by reason.

• too_fast Faster than a person can play
• too_slow Game left open too long
• moves_too_fast Moves came in too fast
• bad_seed Not today's puzzle
• bad_proof Moves or start ticket don't match
• score_mismatch Sent score differs from the replay
• copied_run Same moves as someone else's run
• expired Start ticket expired
• used Start ticket already used
• bad_value Value out of range

Rating (ranked)

Games with a winner also carry a rating, separate from the period leaderboards. The server computes it after every game, and the number you see is exactly what the server stored.

• It's Glicko-2 (Glickman's paper). Everyone starts at 1500, and your rating updates the moment a game ends.
• Settings: system constant τ 0.75, starting volatility 0.09, deviation (RD) 45 to 350. Same values as lichess.
• With RD 110 or more your rating is provisional. It shows a ? and no tier, and it stays off the rating leaderboard.
• While you're away your rating stays put and only RD grows a little (0.21436 rating periods a day).
• The paper recommends rating several games together per period. We rate each game on its own, like lichess. That's an approximation, and we'd rather say so.
• Tiers are just names for rating bands: Bronze below 1350, Silver 1350 to 1550, Gold 1550 to 1750, Platinum 1750 to 1950, Diamond 1950 to 2150, Master 2150 and up. They aren't a separate score. If your rating drops, your tier drops with it. No demotion shields.

Rating seasons

• Seasons are the same three months as the leaderboards (Korea Standard Time).
• A new season keeps your rating. Only RD goes up to at least 100, so the first games of a season move a bit more.
• Season titles go to players with 10+ rated games that season who aren't provisional when it ends. For games against people you also need 5+ different opponents.
• The title shows the tier of your end-of-season rating and your peak that season. Fewer than 3 qualifying players means no titles.

Rated games against people

• You need to be signed in to a Gwirae account. Both players take a seat before the game starts, or it isn't rated.
• The game server checks every move against the rules and then reports the result. We don't accept a result reported by a player's screen.
• The same two players can put 3 games a day (Korea time) into their rating. Games after that are recorded, not rated.
• Games between the same account, the same browser, or the same network (that day's IP) aren't rated.
• What we can't stop: one person with two accounts, two devices and two networks playing themselves. The daily cap and the opponent count only shrink what that gains.

Vs AI rating

• Games against the AI never mix with the rating against people. The screen always says Vs AI.
• Each AI level has a fixed rating that never moves. Only your rating moves.
• AI level ratings come from win rates measured over thousands of AI-vs-AI games. The app's standard AI is set to 1500 and other levels are placed by their measured win rates. Method and game counts are in the table below.
• Whether the server replays a vs AI game to count the result depends on the game. See the Result column below.
• Games where you used a helper (undo, suggested move) aren't rated. Rating them would show a number that isn't your skill.
• With one fixed AI as the opponent, we can't measure much beyond it. Once you're more than 400 above the strongest AI level, wins stop adding rating. Losses and draws still count (same idea as FIDE's 400-point rule).
• Vs AI ratings are per game. You can't compare them with another game's or with chess ratings.

AI level ratings

Why a game wasn't rated

• same_account Both seats were the same account
• same_browser Both seats came from the same browser
• same_network Both seats came from the same network (that day's IP)
• pair_daily Same two players past 3 games that day
• too_short Game ended too early
• assisted A helper was used
• seat_missing No seat taken, or a seat that was already used
• ai_ceiling A vs AI win while more than 400 above the strongest AI

Reports

Spot an odd record or name? Email us. ok981121@naver.com