Board games via GUI
Play chess, Go and board games through the board UI (APIs blocked).
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Lichess (self-hosted) · Sabaki
- Class GUI-need label
G3· GUI-only
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R13.02.01.01-T01 Beat Stockfish level 1 as White on a self-hosted Lichess
Original instruction
A self-hosted Lichess (lila) instance is open in the browser, logged in as test user 'agent1' with no API token. Start a game against the computer at level 1 as White, 10+0, and win by checkmate, resignation or timeout. Move pieces on the board only.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
lila MongoDB game record: player agent1 White, status mate/resign/outoftime, winner white. API tokens are disabled for the account and board/bot API endpoints are blocked at the proxy; move events must come from the web socket of the browser session.
Proposed partial credit
20% game created with correct settings; 30% scaled by material advantage at move 30; 50% win
R13.02.01.01-T02 Solve ten mate-in-two puzzles through the board
Original instruction
On the self-hosted Lichess, open the puzzle page with theme 'mateIn2' (seeded set). Solve 10 consecutive puzzles by making the moves on the board. Wrong moves count as failures; hints are not allowed.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Puzzle round records in the DB for agent1: 10 rounds on mateIn2, success count. API blocked as above.
Proposed partial credit
50% puzzles 1-5 solved (a tenth each); 50% puzzles 6-10 solved (a tenth each); hint use zeroes that puzzle
R13.02.01.01-T03 Set up a position from FEN in the board editor by dragging
Original instruction
On the self-hosted Lichess board editor, set up the position in input/position.fen (Black to move, White can castle kingside only) by dragging pieces from the palette and removing extras, set side to move and castling rights in the UI, then click 'Continue from here' > 'Play with the computer' level 3.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clickgeom· Precise geometry- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
The created game's initialFen equals the target FEN exactly (pieces, side, castling). Pasting the FEN into the editor's FEN field is disabled in this build (field read-only) and URL /editor/<fen> routes are blocked.
Proposed partial credit
60% scaled by squares correct; 20% side to move; 20% castling rights and game created
R13.02.01.01-T04 Underpromote to a knight to solve a puzzle
Original instruction
On the self-hosted Lichess, open puzzle #U01 (seeded) where the only winning move is a pawn promotion to a knight with check. Make the move on the board and choose the knight in the promotion dialog, then complete the puzzle.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Puzzle round success for #U01 and the move list in the round contains '=N'. API tokens are disabled for the account and board/bot API endpoints are blocked at the proxy, so moves must come from the browser session's web socket.
Proposed partial credit
40% correct promotion square and piece; 60% puzzle solved
R13.02.01.01-T05 Win a 9×9 Go game against GnuGo in Sabaki
Original instruction
Sabaki (desktop) is connected to GnuGo level 1 as White engine; you play Black on 9×9 with komi 6.5. Play by clicking intersections and win the game; finish by passing and scoring when the board is settled.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Sabaki
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Saved SGF: RE starts with 'B+'; moves alternate correctly; result re-scored by GnuGo --score in the harness; White's moves must match GnuGo's GTP log. Sabaki's engine console is hidden and agent GTP commands are not routed; the agent has no shell and the SGF's writing process is logged, so a file not saved by Sabaki is detected.
Proposed partial credit
30% game completed; 70% Black win
R13.02.01.01-T06 Reconstruct a Go game record from board photos in Sabaki
Original instruction
input/go_photos/ holds 12 photos of a 13×13 game taken every 10 moves, with the final position. Using Sabaki, enter the moves to reproduce the game so every photographed position matches; where move order between photos is ambiguous, any legal order consistent with captures is accepted. Save game.sgf.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Sabaki
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
SGF replayed: the board after moves 10, 20, …, 120 equals the ground-truth position from each photo (stone-by-stone), and all moves are legal. Partial credit per matching checkpoint. The agent has no shell or text editor, and the SGF's writing process is logged, so a file not saved by Sabaki is detected.
Proposed partial credit
50% scaled by checkpoints 1-6 matched; 50% scaled by checkpoints 7-12 matched
R13.02.01.01-T07 Hold a rook endgame draw against Stockfish
Original instruction
On the self-hosted Lichess, start 'Play with the computer' level 6 from input/philidor.fen (Black to move, rook and king vs rook, king and pawn, theoretical draw). As Black, reach a draw (stalemate, repetition, 50-move rule or insufficient material) without losing.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Game record: initialFen equals philidor.fen and result draw or Black win. A loss scores only the move-count milestone. API tokens are disabled for the account and board/bot API endpoints are blocked at the proxy, so moves must come from the browser session's web socket.
Proposed partial credit
30% scaled by moves survived up to 40; 70% draw or better
R13.02.01.01-T08 Build an annotated study chapter with arrows and comments
Original instruction
On the self-hosted Lichess, create a study 'Italian Game plans' with one chapter containing the moves in input/italian.pgn (12 moves), a comment on move 6 copied from input/notes.txt, a green arrow e3-d4 (drawn by right-drag) and a red circle on f7 on the final position.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clicktype· Text & forms- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Study export (PGN with %cal/%csl tags) from the DB: mainline equals italian.pgn, comment at move 6 exact, %cal Ge3d4 and %csl Rf7 on the last node. PGN import into the study is disabled in this build, and API tokens are disabled for the account, so the study API cannot be used.
Proposed partial credit
40% moves entered; 20% comment; 20% arrow; 20% circle
R13.02.01.01-T09 Play a handicap 9×9 game against KataGo in KaTrain and keep losses small
Original instruction
KaTrain (desktop) is set to 9×9, 2-stone handicap for you (Black), opponent 'KataGo calibrated rank 8k'. Play a full game; win, or lose by less than 5 points.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- KaTrain
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Saved SGF: HA[2], game finished, result parsed; win = full credit, loss margin < 5 = 70%. KaTrain's AI-move button for the player is disabled by config and teaching hints are off.
Proposed partial credit
30% game completed; 70% win or loss < 5 points
R13.02.01.01-T10 Win a three-point backgammon match in GNU Backgammon
Original instruction
GNU Backgammon (GTK GUI) new 3-point match against the 'Beginner' player. Roll and move checkers by dragging; take or drop doubles as you see fit. Win the match.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- GNU Backgammon
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Saved match file (.sgf) parsed: match winner = player, score 3. The gnubg command line and 'hint' are disabled in the harness build.
Proposed partial credit
30% scaled by points won; 70% match won
R13.02.01.01-T11 Beat an easy computer at checkers in the browser
Original instruction
A self-hosted English draughts web app is open against the 'Easy' AI, you play Black (moves first). Make moves by dragging pieces, including forced captures and multi-jumps, and win the game.
- Platforms
web· Any OS (web)- Applications
- checkers web
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Server game log: result Black win; all moves legal. The server issues no API token and accepts moves only over the browser session's CSRF-bound socket, and DevTools are disabled by policy, so moves must come from pointer events on the board.
Proposed partial credit
30% scaled by material lead at move 20; 70% win
R13.02.01.01-T12 Win a Chess960 game against Stockfish level 3
Original instruction
On the self-hosted Lichess, start a Chess960 game against the computer level 3 as White from Chess960 starting position #312. Win the game, including castling correctly if you castle (drag the king onto the rook).
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Game record: variant chess960, initial position 312, winner white. API blocked; moves from the browser socket only.
Proposed partial credit
30% scaled by evaluation at move 25; 70% win
R13.02.01.01-T13 Solve five three-move tsume shogi problems in a shogi GUI
Original instruction
A self-hosted lishogi instance shows a puzzle set of 5 tsume-shogi problems (mate in 3). Solve each by making the moves on the board, including drops from the hand and promotion choices.
- Platforms
web· Any OS (web)- Applications
- lishogi (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
tutorial· Follow docs/tutorial- Proposed difficulty
- 4 / 5
Proposed verifier
Puzzle round records: success flags for the 5 problems. Drops and promotion dialogs must be used; API disabled.
Proposed partial credit
20% problem 1 solved; 20% problem 2; 20% problem 3; 40% problems 4 and 5
R13.02.01.01-T14 Checkmate with bishop and knight against a lone king
Original instruction
On the self-hosted Lichess, play from input/kbnk.fen (White: K, B, N; Black: K) against Stockfish level 8 defending. Deliver checkmate within 50 moves.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Game record: initialFen equals kbnk.fen, status mate with winner white, ply count ≤ 100 from the start. The tablebase and analysis tabs are disabled for this game; API tokens are disabled and board/bot endpoints are blocked at the proxy, so moves come from the browser socket.
Proposed partial credit
30% enemy king driven to an edge (position check); 30% to the correct corner; 40% mate within 50 moves
R13.02.01.01-T15 Find the best move in an analysis position and play it
Original instruction
On the self-hosted Lichess analysis board, input/tactic.fen is loaded with the engine turned off. Find White's best move (a winning combination) and play the full forcing line of 3 moves on the board, then save the line as a new study chapter 'Tactic 1'.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Study chapter mainline equals the 3-move forcing line from the ground truth (alternative defences checked against Stockfish depth 30 for equal outcome). Local engine toggle is disabled in this build, and API tokens are disabled for the account, so the study cannot be written by API.
Proposed partial credit
50% correct first move; 30% complete forcing line; 20% saved as chapter
R13.02.01.01-T16 Win a game of Hive against the computer
Original instruction
A self-hosted Hive web app is open against its 'Medium' AI, base game pieces, you play White. Place and move pieces by dragging (respecting the one-hive and freedom-of-movement rules) and surround the opponent's queen.
- Platforms
web· Any OS (web)- Applications
- Hive web
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgmentgeom· Precise geometry- Work product
WP11· Game/sim state- Conditions
tutorial· Follow docs/tutorial- Proposed difficulty
- 4 / 5
Proposed verifier
Server game record: result White win (black queen surrounded). The server issues no API token and accepts moves only over the browser session's CSRF-bound socket, and DevTools are disabled by policy, so moves must come from pointer events.
Proposed partial credit
30% queen placed by turn 4; 20% opponent queen 4+ sides covered; 50% win
R13.02.01.01-T17 Solve life-and-death problems on a Go board
Original instruction
Sabaki opens input/tsumego.sgf, a collection of 10 life-and-death problems (Black to play). For each, play the correct first move and the main refutation sequence as a variation, then save the file.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Sabaki
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Saved SGF: for each problem node, the first child move matches a correct answer in the hidden key; sequences checked by KataGo ownership on the target group. The agent has no shell or text editor, and the SGF's writing process is logged, so a file not saved by Sabaki is detected.
Proposed partial credit
80% scaled by problems with a correct first move (one tenth each); 20% scaled by correct refutation sequences
R13.02.01.01-T18 Score at least 20 in a three-minute puzzle storm
Original instruction
On the self-hosted Lichess, open Puzzle Storm (3-minute timed puzzles, time bonus for streaks). Score at least 20 puzzles in one run.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & droprt· Real-time controlvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Storm run record in the DB: score ≥ 20. The clock keeps running during model inference, so latency costs score. API tokens are disabled for the account and board/bot API endpoints are blocked at the proxy, so answers must come from the browser session.
Proposed partial credit
50% scaled by storm score up to 10; 50% scaled from 10 to 20
R13.02.01.01-T19 Play a 1+0 bullet game without losing on time
Original instruction
On the self-hosted Lichess, play a 1+0 game against the computer level 2 as Black. Win, or at least finish the game by checkmate/draw without your clock running out.
- Platforms
web· Any OS (web)- Applications
- Lichess (self-hosted)
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & droprt· Real-time controlvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Game record: speed bullet, player Black; status not outoftime for Black. Win = full credit. API tokens are disabled for the account and board/bot API endpoints are blocked at the proxy, so moves must come from the browser session's web socket.
Proposed partial credit
40% no flag loss; 60% win
R13.02.01.01-T20 Win a game of Santorini against the computer
Original instruction
A self-hosted Santorini web app (no god powers) is open against its 'Easy' AI. Place your 2 workers, then move and build by clicking tiles until one of your workers moves up to level 3.
- Platforms
web· Any OS (web)- Applications
- Santorini web
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentgeom· Precise geometry- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Server game log: winner = player via level-3 move; all moves legal. The server issues no API token and accepts moves only over the browser session's CSRF-bound socket, and DevTools are disabled by policy, so moves must come from clicks on the board.
Proposed partial credit
30% game played to completion; 70% win
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.