Datasets:
accepted_answers listlengths 1 1 | artifact_hash stringlengths 64 64 | completion listlengths 1 1 | metadata unknown | privileged_context stringlengths 48 1.55k | prompt listlengths 2 2 | split stringclasses 1
value | target dict |
|---|---|---|---|---|---|---|---|
[
"move queen c2 -> b3"
] | 277223cb080bd6c92494771f9defc616c510881b87da4c0883cb63678c8ae620 | [
{
"content": "Final answer: \\boxed{move queen c2 -> b3}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "277223cb080bd6c92494771f9defc616c510881b87da4c0883cb63678c8ae620",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 250025,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__tic_tac_ch... | Slides the queen from c2 to b3, and the position is playable.
So we play move queen c2 -> b3.
Final answer: \boxed{move queen c2 -> b3} | [
{
"content": "You are an expert player of training_variants__tic_tac_chess__opposite_corners_locked. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Corner Locked Line Chess\n\n## Board\nThe board has 3 rows a... | preliminary | {
"accepted_answers": [
"move queen c2 -> b3"
],
"canonical": "move queen c2 -> b3",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"a6"
] | c04db64db56e91ad379b92adfcf6e274de6c52b876a1af19efced18bb5ec628b | [
{
"content": "Final answer: \\boxed{a6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "c04db64db56e91ad379b92adfcf6e274de6c52b876a1af19efced18bb5ec628b",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220034,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The tempting move is f1 but it comes out worse for us, while a6 comes out edging our way.
Against the opponent's best reply, the position is well on top, and that is their sharpest reply, so it holds without our reaching for more.
So a6.
Final answer: \boxed{a6} | [
{
"content": "You are an expert player of training_variants__simplified__othello__late_diagonal_unlock. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Late Diagonal Othello\n\n## Board\nThe board has 6 rows a... | preliminary | {
"accepted_answers": [
"a6"
],
"canonical": "a6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place vertical b2-b1"
] | 01ed6566c6cbccbb23f6f40854ec27a40a84614450a922ff866acfb73694fc8d | [
{
"content": "Final answer: \\boxed{place vertical b2-b1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "01ed6566c6cbccbb23f6f40854ec27a40a84614450a922ff866acfb73694fc8d",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220001,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__domineerin... | Covers b2 and b1, and place vertical b2-b1 comes out clearly on top.
Against the opponent's best reply, the position is level, and that is their sharpest reply, so it holds without our reaching for more.
So we play place vertical b2-b1.
Final answer: \boxed{place vertical b2-b1} | [
{
"content": "You are an expert player of training_variants__domineering__alternating_orientation. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Alternating Domino Blockade\n\n## Board\nThe board has 6 rows ... | preliminary | {
"accepted_answers": [
"place vertical b2-b1"
],
"canonical": "place vertical b2-b1",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"d6"
] | 945260ea98bca150c257ce71780b8b9d93297f55f02401f933af8de597625807 | [
{
"content": "Final answer: \\boxed{d6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "945260ea98bca150c257ce71780b8b9d93297f55f02401f933af8de597625807",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220005,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The opponent is a move from winning, so we have to answer it.
D6 comes out worse for us here.
The move is d6.
Final answer: \boxed{d6} | [
{
"content": "You are an expert player of training_variants__simplified__othello__edge_bonus_scoring. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Edge Bonus Othello\n\n## Board\nThe board has 6 rows and 6 ... | preliminary | {
"accepted_answers": [
"d6"
],
"canonical": "d6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place a6"
] | 847d5648c65b1dcb8b287ed59ebd13705dd3bf2ac22208929bf8f156fea1ab2e | [
{
"content": "Final answer: \\boxed{place a6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "847d5648c65b1dcb8b287ed59ebd13705dd3bf2ac22208929bf8f156fea1ab2e",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220001,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_remaining_20260609/training_variants__... | Place a6 comes out edging our way here.
So we play place a6.
Final answer: \boxed{place a6} | [
{
"content": "You are an expert player of training_variants__simplified__first_attack__edge_opening. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Edge First Safe Placement\n\n## Board\nThe board has 6 rows ... | preliminary | {
"accepted_answers": [
"place a6"
],
"canonical": "place a6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"e4"
] | 286e8736489040a68256647edc6da2c98856d4b700501eacbb4c82ec4e3b5e6a | [
{
"content": "Final answer: \\boxed{e4}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "286e8736489040a68256647edc6da2c98856d4b700501eacbb4c82ec4e3b5e6a",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220028,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The tempting move is b1 but it comes out about even, while e4 comes out clearly on top.
Against the opponent's best reply, we are still a little ahead.
So we play e4.
Final answer: \boxed{e4} | [
{
"content": "You are an expert player of training_variants__simplified__othello__max_one_flip. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Quiet Flip Othello\n\n## Board\nThe board has 6 rows and 6 column... | preliminary | {
"accepted_answers": [
"e4"
],
"canonical": "e4",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"0"
] | 9d099fd0e5add6288b2ef7768f262f32c6f0666001d9e64e6775882b966628b7 | [
{
"content": "Final answer: \\boxed{0}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "9d099fd0e5add6288b2ef7768f262f32c6f0666001d9e64e6775882b966628b7",
"asserted_predicates": [
{
"arguments": {
"cell": 3,
"target_action": 14
},
"checker": "successor_cell_value",
"expected": 0,
"id": "qa1",
"s... | Replayed place horizontal e4-f4 and verified the changed cells: e4, f4. Inspected successor cell d6: encoded value 0, meaning no piece (it is empty).
Final answer: \boxed{0} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Edge Opening Dominoes\n\n## Board\nThe board has 6 rows and 6 columns. E... | preliminary | {
"accepted_answers": [
"0"
],
"canonical": "0",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:successor_cell_value"
} |
[
"place horizontal c6-d6"
] | f46fa4fde177c3c1b3925e5b43756ec0ca6257653c2151ccd0fe8712098b72d0 | [
{
"content": "Final answer: \\boxed{place horizontal c6-d6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "f46fa4fde177c3c1b3925e5b43756ec0ca6257653c2151ccd0fe8712098b72d0",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220023,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__domineerin... | The tempting move is place horizontal b6-c6, which covers b6 and c6, but it comes out edging our way, while place horizontal c6-d6 comes out winning.
Against the opponent's best reply, the position is well on top, and that is their sharpest reply, so it holds without our reaching for more.
The move is place horizontal ... | [
{
"content": "You are an expert player of training_variants__domineering__misere_blocked. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Misere Domino Blockade\n\n## Board\nThe board has 6 rows and 6 columns.... | preliminary | {
"accepted_answers": [
"place horizontal c6-d6"
],
"canonical": "place horizontal c6-d6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place e5"
] | 732c77db72b05c3721715fc1a9b0eaa8a1c16ad1a0441dd624d25572d94233e5 | [
{
"content": "Final answer: \\boxed{place e5}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "732c77db72b05c3721715fc1a9b0eaa8a1c16ad1a0441dd624d25572d94233e5",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220003,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The opponent is a move from winning, so we have to answer it.
The natural move is place b5, but it does not finish the game.
The move that does is place e5, and it takes the game.
So place e5.
Final answer: \boxed{place e5} | [
{
"content": "You are an expert player of training_variants__simplified__first_attack__one_diagonal_disabled. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Single Diagonal Safe Placement\n\n## Board\nThe boa... | preliminary | {
"accepted_answers": [
"place e5"
],
"canonical": "place e5",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"1"
] | 2309aa7d69357537a738ff7f0569f75dc8650d48204f27eb579689453589ece6 | [
{
"content": "Final answer: \\boxed{1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "2309aa7d69357537a738ff7f0569f75dc8650d48204f27eb579689453589ece6",
"asserted_predicates": [
{
"arguments": {
"target_action": 6
},
"checker": "opponent_immediate_win_count",
"expected": 1,
"id": "qa1",
"source_state_... | Enumerated opponent replies and checked each for an immediate terminal win: column 1, column 2, column 3, column 4, column 5, column 6, column 7. Verified winning replies: column 6. Count: 1.
Final answer: \boxed{1} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Delayed Center Gravity Four\n\n## Board\nThe board has 6 rows and 7 colu... | preliminary | {
"accepted_answers": [
"1"
],
"canonical": "1",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:opponent_immediate_win_count"
} |
[
"column 4"
] | 2750cbfb21e969ba233aeaba700dfda65c436c7b3ed9320b0e5277e6d80e30d7 | [
{
"content": "Final answer: \\boxed{column 4}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "2750cbfb21e969ba233aeaba700dfda65c436c7b3ed9320b0e5277e6d80e30d7",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220032,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__connect4__... | Drops a disc into column 4, and the position is playable.
So column 4.
Final answer: \boxed{column 4} | [
{
"content": "You are an expert player of training_variants__connect4__outer_opening. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Outer Opening Gravity Four\n\n## Board\nThe board has 6 rows and 7 columns.... | preliminary | {
"accepted_answers": [
"column 4"
],
"canonical": "column 4",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place vertical f4-f3"
] | 8d94a8d0a194a67a462cbaddfc4e9d4b0ccbe568289c05cea0f5e373635fd327 | [
{
"content": "Final answer: \\boxed{place vertical f4-f3}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "8d94a8d0a194a67a462cbaddfc4e9d4b0ccbe568289c05cea0f5e373635fd327",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220001,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__domineerin... | Covers f4 and f3, and the position is playable.
So we play place vertical f4-f3.
Final answer: \boxed{place vertical f4-f3} | [
{
"content": "You are an expert player of training_variants__domineering__alternating_orientation. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Alternating Domino Blockade\n\n## Board\nThe board has 6 rows ... | preliminary | {
"accepted_answers": [
"place vertical f4-f3"
],
"canonical": "place vertical f4-f3",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"3"
] | 5c8c4d96b16f2e80e810d9285f1a4d8c7c4b6260804d557adf7922de80146bf1 | [
{
"content": "Final answer: \\boxed{3}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "5c8c4d96b16f2e80e810d9285f1a4d8c7c4b6260804d557adf7922de80146bf1",
"asserted_predicates": [
{
"arguments": {},
"checker": "legal_action_count",
"expected": 3,
"id": "qa1",
"source_state_hash": "80e3d1ac7e621a76f58a64d29f0e7c691f3c... | Enumerated legal handles before counting: e2, f2, a1. The verified list contains 3 entries.
Final answer: \boxed{3} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# No Pass Othello\n\n## Board\nThe board has 6 rows and 6 columns. Empty c... | preliminary | {
"accepted_answers": [
"3"
],
"canonical": "3",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:legal_action_count"
} |
[
"column 2"
] | 10453283da495adfb4d9056f1d81eea5d65d50dbc87d4672bb4021459addc6e4 | [
{
"content": "Final answer: \\boxed{column 2}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "10453283da495adfb4d9056f1d81eea5d65d50dbc87d4672bb4021459addc6e4",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 255005,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__connect4__... | Drops a disc into column 2, and the position is roughly level.
So we play column 2.
Final answer: \boxed{column 2} | [
{
"content": "You are an expert player of training_variants__connect4__one_diagonal_disabled. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Single Diagonal Gravity Four\n\n## Board\nThe board has 6 rows and ... | preliminary | {
"accepted_answers": [
"column 2"
],
"canonical": "column 2",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"e1"
] | 967e84b587cccec81add0dcadf0743a2afcffb8a8b84c12dcf9531f5eb6d2cb8 | [
{
"content": "Final answer: \\boxed{e1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "967e84b587cccec81add0dcadf0743a2afcffb8a8b84c12dcf9531f5eb6d2cb8",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220034,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The tempting move is b2 but it comes out a shade against us, while e1 comes out roughly level.
Against the opponent's best reply, the position is about even, and that is their sharpest reply, so it holds without our reaching for more.
The move is e1.
Final answer: \boxed{e1} | [
{
"content": "You are an expert player of training_variants__simplified__othello__late_diagonal_unlock. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Late Diagonal Othello\n\n## Board\nThe board has 6 rows a... | preliminary | {
"accepted_answers": [
"e1"
],
"canonical": "e1",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"9"
] | 9a9f6c807bce6f27ed970bb328b818ff41690b5d0c9d09b276249ce1a37c224b | [
{
"content": "Final answer: \\boxed{9}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "9a9f6c807bce6f27ed970bb328b818ff41690b5d0c9d09b276249ce1a37c224b",
"asserted_predicates": [
{
"arguments": {},
"checker": "legal_action_count",
"expected": 9,
"id": "qa1",
"source_state_hash": "4043c5480c437971bfc2d5465510010da10c... | Enumerated legal handles before counting: place vertical a6-a5, place vertical b6-b5, place vertical a5-a4, place vertical b5-b4, place vertical d5-d4, place vertical d4-d3, place vertical a2-a1, place vertical b2-b1, place vertical c2-c1. The verified list contains 9 entries.
Final answer: \boxed{9} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Edge Opening Dominoes\n\n## Board\nThe board has 6 rows and 6 columns. E... | preliminary | {
"accepted_answers": [
"9"
],
"canonical": "9",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:legal_action_count"
} |
[
"e1"
] | bb4026e425817b08a6398b917978db9298c12298123d507f564b664c23d16eb4 | [
{
"content": "Final answer: \\boxed{e1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "bb4026e425817b08a6398b917978db9298c12298123d507f564b664c23d16eb4",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220045,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The opponent is a move from winning, so we have to answer it.
The tempting move is a5 but it comes out about even, while e1 comes out a little ahead.
Against the opponent's best reply, the position is clearly on top, and that is their sharpest reply, so it holds without our reaching for more.
So e1.
Final answer: \box... | [
{
"content": "You are an expert player of training_variants__simplified__othello__corner_required_when_available. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Corner Forced Othello\n\n## Board\nThe board ha... | preliminary | {
"accepted_answers": [
"e1"
],
"canonical": "e1",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"column 7"
] | 8ba2290e0ea96992874ba5e9aed58c479c915976c75df6aaf4e583348694dc85 | [
{
"content": "Final answer: \\boxed{column 7}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "8ba2290e0ea96992874ba5e9aed58c479c915976c75df6aaf4e583348694dc85",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220029,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__connect4__... | The opponent is a move from winning, so we have to answer it.
Drops a disc into column 7, and the position is roughly level.
So we play column 7.
Final answer: \boxed{column 7} | [
{
"content": "You are an expert player of training_variants__connect4__orthogonal_only. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Orthogonal Gravity Four\n\n## Board\nThe board has 6 rows and 7 columns. ... | preliminary | {
"accepted_answers": [
"column 7"
],
"canonical": "column 7",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"0"
] | ecdb21bddf0f425b8e42144a04aead2e57927725642494ba90c5ecdf7601ce8a | [
{
"content": "Final answer: \\boxed{0}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "ecdb21bddf0f425b8e42144a04aead2e57927725642494ba90c5ecdf7601ce8a",
"asserted_predicates": [
{
"arguments": {
"cell": 27,
"target_action": 3
},
"checker": "successor_cell_value",
"expected": 0,
"id": "qa1",
"s... | Replayed place vertical d6-d5 and verified the changed cells: d6, d5. Inspected successor cell d2: encoded value 0, meaning no piece (it is empty).
Final answer: \boxed{0} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Swapped Domino Blockade\n\n## Board\nThe board has 6 rows and 6 columns.... | preliminary | {
"accepted_answers": [
"0"
],
"canonical": "0",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:successor_cell_value"
} |
[
"place horizontal a2-b2"
] | b04ff077d62ecf32919c9401c017da659b6561b41d59354544ee86169f578aae | [
{
"content": "Final answer: \\boxed{place horizontal a2-b2}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "b04ff077d62ecf32919c9401c017da659b6561b41d59354544ee86169f578aae",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220033,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__domineerin... | Covers a2 and b2, and the position is ahead.
So place horizontal a2-b2.
Final answer: \boxed{place horizontal a2-b2} | [
{
"content": "You are an expert player of training_variants__domineering__both_horizontal. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Horizontal Domino Blockade\n\n## Board\nThe board has 6 rows and 6 col... | preliminary | {
"accepted_answers": [
"place horizontal a2-b2"
],
"canonical": "place horizontal a2-b2",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"move queen a2 -> a1"
] | 63238bdf3aa744927c0a4b212d18a8d4d5d49a42276b1442af164037461fb2c4 | [
{
"content": "Final answer: \\boxed{move queen a2 -> a1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "63238bdf3aa744927c0a4b212d18a8d4d5d49a42276b1442af164037461fb2c4",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 250025,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__tic_tac_ch... | The opponent is a move from winning, so we have to answer it.
Slides the queen from a2 to a1, and the position is playable.
The move is move queen a2 -> a1.
Final answer: \boxed{move queen a2 -> a1} | [
{
"content": "You are an expert player of training_variants__tic_tac_chess__opposite_corners_locked. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Corner Locked Line Chess\n\n## Board\nThe board has 3 rows a... | preliminary | {
"accepted_answers": [
"move queen a2 -> a1"
],
"canonical": "move queen a2 -> a1",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place horizontal a6-b6, place horizontal b6-c6, place horizontal c6-d6, place horizontal d6-e6, place horizontal e6-f6, place horizontal c5-d5, place horizontal b4-c4, place horizontal c4-d4, place horizontal d4-e4, place horizontal e4-f4, place horizontal e3-f3, place horizontal d2-e2, place horizontal e2-f2, pla... | c76ef5d5dc48e4d2806f06b818d89212bdac805ecd2100f4377fb40c453387f9 | [
{
"content": "Final answer: \\boxed{place horizontal a6-b6, place horizontal b6-c6, place horizontal c6-d6, place horizontal d6-e6, place horizontal e6-f6, place horizontal c5-d5, place horizontal b4-c4, place horizontal c4-d4, place horizontal d4-e4, place horizontal e4-f4, place horizontal e3-f3, place horizo... | {
"algorithm": "efficientzero",
"artifact_hash": "c76ef5d5dc48e4d2806f06b818d89212bdac805ecd2100f4377fb40c453387f9",
"asserted_predicates": [
{
"arguments": {},
"checker": "replay_legal_actions",
"expected": [
0,
1,
2,
3,
4,
7,
11,
... | Matched replay-verified action ids [0, 1, 2, 3, 4, 7, 11, 12, 13, 14, 19, 23, 24, 28, 29] to the Legal Options in their displayed order: place horizontal a6-b6, place horizontal b6-c6, place horizontal c6-d6, place horizontal d6-e6, place horizontal e6-f6, place horizontal c5-d5, place horizontal b4-c4, place horizonta... | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Alternating Domino Blockade\n\n## Board\nThe board has 6 rows and 6 colu... | preliminary | {
"accepted_answers": [
"place horizontal a6-b6, place horizontal b6-c6, place horizontal c6-d6, place horizontal d6-e6, place horizontal e6-f6, place horizontal c5-d5, place horizontal b4-c4, place horizontal c4-d4, place horizontal d4-e4, place horizontal e4-f4, place horizontal e3-f3, place horizontal d2-e2, p... |
[
"a6"
] | 14ffaaec0fda614625c4a4c62a5b458a5a700413a27004d5fd1e7ad659fba9c3 | [
{
"content": "Final answer: \\boxed{a6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "14ffaaec0fda614625c4a4c62a5b458a5a700413a27004d5fd1e7ad659fba9c3",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220004,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | The tempting move is e6 but it comes out losing, while a6 comes out clearly on top.
Against the opponent's best reply, we are still ahead.
So a6.
Final answer: \boxed{a6} | [
{
"content": "You are an expert player of training_variants__simplified__othello__no_pass_blocked_loses. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# No Pass Othello\n\n## Board\nThe board has 6 rows and 6 ... | preliminary | {
"accepted_answers": [
"a6"
],
"canonical": "a6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"2"
] | 45b99ea6bc94c0f2987b667e1ff85262b4a5f2a4212906d30bd2245ee06f6a34 | [
{
"content": "Final answer: \\boxed{2}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "45b99ea6bc94c0f2987b667e1ff85262b4a5f2a4212906d30bd2245ee06f6a34",
"asserted_predicates": [
{
"arguments": {
"cell": 13,
"target_action": 7
},
"checker": "successor_cell_value",
"expected": 2,
"id": "qa1",
"s... | Replayed place vertical b5-b4 and verified the changed cells: b5, b4. Inspected successor cell b4: encoded value 2, meaning a player-2 piece.
Final answer: \boxed{2} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Contact Domino Blockade\n\n## Board\nThe board has 6 rows and 6 columns.... | preliminary | {
"accepted_answers": [
"2"
],
"canonical": "2",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:successor_cell_value"
} |
[
"move queen b3 -> a3"
] | 1981ec71f2ff653d04fed1e520c670693ddcd52dcd57abbbf3832bba87530b6f | [
{
"content": "Final answer: \\boxed{move queen b3 -> a3}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "1981ec71f2ff653d04fed1e520c670693ddcd52dcd57abbbf3832bba87530b6f",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 235002,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__tic_tac_ch... | Slides the queen from b3 to a3, and the position is playable.
The move is move queen b3 -> a3.
Final answer: \boxed{move queen b3 -> a3} | [
{
"content": "You are an expert player of training_variants__tic_tac_chess__capturing_hops. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Capture Hop Line Chess\n\n## Board\nThe board has 3 rows and 3 column... | preliminary | {
"accepted_answers": [
"move queen b3 -> a3"
],
"canonical": "move queen b3 -> a3",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"place d4"
] | aedd376878d627cbfc1b2cee26f68d8602a2a0327d121568485a4f4b14f59976 | [
{
"content": "Final answer: \\boxed{place d4}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "aedd376878d627cbfc1b2cee26f68d8602a2a0327d121568485a4f4b14f59976",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220011,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_remaining_20260609/training_variants__... | The tempting move is place c6 but it comes out on the ropes, while place d4 comes out winning.
The move is place d4.
Final answer: \boxed{place d4} | [
{
"content": "You are an expert player of training_variants__simplified__first_attack__corners_locked. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Corner Closed Placement\n\n## Board\nThe board has 6 rows ... | preliminary | {
"accepted_answers": [
"place d4"
],
"canonical": "place d4",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"2"
] | 928f32657a24f0da5ae74cb441e721bbaae840c5aa1e68d14c0edd0f123c70f7 | [
{
"content": "Final answer: \\boxed{2}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "928f32657a24f0da5ae74cb441e721bbaae840c5aa1e68d14c0edd0f123c70f7",
"asserted_predicates": [
{
"arguments": {
"cell": 20,
"target_action": 6
},
"checker": "successor_cell_value",
"expected": 2,
"id": "qa1",
"s... | Replayed column 7 and verified the changed cells: g4. Inspected successor cell g4: encoded value 2, meaning a player-2 piece.
Final answer: \boxed{2} | [
{
"content": "You are a rigorous board-state analyst. Use only the supplied rules, history, board, state, and Legal Options as evidence. Explain the verified state facts briefly and answer the stated analysis question.\n\n# Game Rules\n\n# Delayed Center Gravity Four\n\n## Board\nThe board has 6 rows and 7 colu... | preliminary | {
"accepted_answers": [
"2"
],
"canonical": "2",
"kind": "state_answer",
"verifier": "opsd_replay_qa_v2:successor_cell_value"
} |
[
"place vertical c2-c1"
] | 2dad9d77211106876f5fd1845c539b9f01af3343433ec1d0174384eb3dd3b687 | [
{
"content": "Final answer: \\boxed{place vertical c2-c1}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "2dad9d77211106876f5fd1845c539b9f01af3343433ec1d0174384eb3dd3b687",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220024,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__domineerin... | The tempting move is place vertical b2-b1, which covers b2 and b1, but it comes out a little ahead, while place vertical c2-c1 comes out ahead.
Against the opponent's best reply, the position is roughly level, and that is their sharpest reply, so it holds without our reaching for more.
So we play place vertical c2-c1.
... | [
{
"content": "You are an expert player of training_variants__domineering__edge_opening. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Edge Opening Dominoes\n\n## Board\nThe board has 6 rows and 6 columns. Em... | preliminary | {
"accepted_answers": [
"place vertical c2-c1"
],
"canonical": "place vertical c2-c1",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
[
"e6"
] | 68a52aba7f8dbdc07e19c9415d174475513d7d9ab7c1f461063e6ba93f80b20b | [
{
"content": "Final answer: \\boxed{e6}",
"role": "assistant"
}
] | {
"algorithm": "efficientzero",
"artifact_hash": "68a52aba7f8dbdc07e19c9415d174475513d7d9ab7c1f461063e6ba93f80b20b",
"checkpoint_alias": "ckpt_last.pth.tar",
"checkpoint_envstep": 220045,
"checkpoint_metadata_source": "models/simplified5_othello_variants45_bot_superhuman_20260526/training_variants__simplified... | Neither e5 nor e6 clearly leads, but e6 keeps the position in hand.
Against the opponent's best reply, the position is on the ropes, and that is their sharpest reply, so we stay worse, but it is the steadiest defence we have.
So e6.
Final answer: \boxed{e6} | [
{
"content": "You are an expert player of training_variants__simplified__othello__corner_required_when_available. You and the user will play a game. When it is your turn, explain your reasoning, then end with your chosen move inside \\boxed{}.\n\n# Game Rules\n\n# Corner Forced Othello\n\n## Board\nThe board ha... | preliminary | {
"accepted_answers": [
"e6"
],
"canonical": "e6",
"kind": "move_choice",
"verifier": "checkpoint_replay_legal_move_v1"
} |
End of preview. Expand in Data Studio
SPSD-Variants-opsd
Grounded on-policy self-distillation (OPSD) teacher-context dataset over 45 board-game rule variants (5 families × 9: connect4, domineering, simplified_first_attack, simplified_othello, tic_tac_chess), derived from trained MuZero/EfficientZero checkpoints (plan-528 v2).
Each row is a decision-state task (a move choice or one of six auxiliary
state-QA tasks). The privileged_context is the teacher signal: grounded
natural-language reasoning that discovers the best move over the tempting
alternative and verifies it against the opponent's best reply, grounded in fresh
MuZero-net values that choose the wording but are never printed, and never
naming the search engine. The student-facing prompt, the completion, and the
target are unchanged from the underlying pools — only privileged_context
carries the grounded reasoning.
Contents
Files are gzip-compressed JSON Lines (datasets loads .jsonl.gz transparently).
| file | rows | purpose |
|---|---|---|
train.jsonl.gz |
213,406 | training split (170,725 move + 42,681 state-QA) |
frozen_probes/move.jsonl.gz |
6,840 | held-out move probe (disjoint from train) |
frozen_probes/state_qa.jsonl.gz |
40,841 | held-out six-task state-QA probe |
manifest.json |
— | full provenance, per-variant counts, balance, census fingerprint |
Per-family subsets are obtained by filtering metadata.game_name /
metadata.family (the five families are connect4, domineering,
simplified_first_attack, simplified_othello, tic_tac_chess).
State-QA tasks (six): occupancy, legality, threat_count,
legal_action_count, legal_action_enumeration, successor_state.
Row schema
prompt— student-facing board task (chat messages).completion—Final answer: \boxed{…}.privileged_context— grounded teacher reasoning + the boxed answer (teacher-only).target/accepted_answers— verifier + canonical answer.metadata—task,game_name,decision_state_hash, checkpoint provenance (path/alias/SHA/env-step, algorithm,num_simulations, backend), andreplay_evidence_horizon(richvs a conservative-fallback mode).
Provenance & method
- Sources: the trained
simplified5othello-variant EfficientZero checkpoints; 36 variants viarandom-prefix-full-gamecheckpoint self-play, 9 via exhaustive real-env enumeration. - Grounding: one fresh 50-simulation MCTS search per decision state; a separation gate (min root gap 0.05) refuses a distinction the root value does not support and falls back to conservative prose (no row is ever dropped).
- Move coverage: ~54% rich grounded traces, ~46% conservative fallback.
- Seed 0. Verified corpus-wide: no raw search stats or engine names leak into
privileged_context; state-QA rationales are array-free.
Caveats
- Text only (
tokenizer_hash=not_tokenized): tokenize and build the OPSD-ready rows for your trainer. - 6 connect4 variants (
exact_four,holey_four,no_center_opening,one_diagonal_disabled,orthogonal_only,outer_opening) carry a declaredlegalityshortfall (positive-skewed legality QA — illegal-handle states are rare there); the pooled answer-class balance stays ≤ 0.65. - Global per-task counts are balanced per family, not equal across the corpus (reported, not enforced).
Private research dataset.
- Downloads last month
- 15