Datasets:
scenario_id stringlengths 20 25 | tier stringclasses 3
values | phase stringclasses 3
values | pos_id stringclasses 75
values | judge stringclasses 3
values | n_present int64 20 20 | label_to_model dict | grades dict | note stringlengths 24 300 | prompt_tokens int64 2.94k 6.45k | completion_tokens int64 339 2.23k | ts stringlengths 32 32 |
|---|---|---|---|---|---|---|---|---|---|---|---|
26MZFDEq_141#beginner | beginner | endgame | 26MZFDEq_141 | claude | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "ours_v5",
"D": "mistral3",
"E": "gemma3_27b",
"F": "base",
"G": "kimi25",
"H": "pbase_4b",
"I": "glm5",
"J": "ours",
"K": "gpt",
"L": "claude",
"M": "base_4b",
"N": "dsv32",
"O": "llama33_70b",
"P": "gemini",
"Q": "ours_v4",
"R": "q3_next80b",
... | {
"pbase_4b": {
"move": 9,
"instr": 8
},
"ours": {
"move": 8,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 8
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 2,
"instr": 2
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Rf6/Rf7+ responses mostly strong; C endorses the poor move, Q fabricates squares. | 4,788 | 414 | 2026-07-09T02:34:03.868889+00:00 |
26MZFDEq_141#intermediate | intermediate | endgame | 26MZFDEq_141 | claude | 20 | {
"A": "gpt",
"B": "claude",
"C": "ours_v4",
"D": "ours",
"E": "base_4b",
"F": "llama33_70b",
"G": "glm5",
"H": "kimi25",
"I": "ours_4b",
"J": "q3_next80b",
"K": "dsv32",
"L": "base",
"M": "ours_v5",
"N": "pbase_4b",
"O": "dsr1",
"P": "gemini",
"Q": "ours_v3",
"R": "mistral3",
"S":... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 8,
"instr": 2
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 8,
"instr": 4
},
"ours_v5": {
"move": 8,
"instr": 6
},
"dsr1": {
"move": 8,
"instr": 6
},
"ours_v3": {
"mov... | Rf6 and d5 responses generally strong; D/L/R lose points for inventing pieces or false threats. | 5,115 | 413 | 2026-07-09T02:34:03.943585+00:00 |
26MZFDEq_141#intermediate | intermediate | endgame | 26MZFDEq_141 | gemini | 20 | {
"A": "gpt",
"B": "claude",
"C": "ours_v4",
"D": "ours",
"E": "base_4b",
"F": "llama33_70b",
"G": "glm5",
"H": "kimi25",
"I": "ours_4b",
"J": "q3_next80b",
"K": "dsv32",
"L": "base",
"M": "ours_v5",
"N": "pbase_4b",
"O": "dsr1",
"P": "gemini",
"Q": "ours_v3",
"R": "mistral3",
"S":... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 2
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 7
},
"dsr1": {
"move": 10,
"instr": 4
},
"ours_v3": {
... | Most responses correctly suggest a verified move, but scores differentiate those with accurate explanations from those hallucinating threats or board features. | 3,550 | 615 | 2026-07-09T02:34:05.709404+00:00 |
26MZFDEq_141#advanced | advanced | endgame | 26MZFDEq_141 | claude | 20 | {
"A": "gpt",
"B": "pbase_4b",
"C": "ours_v3",
"D": "q3_32b",
"E": "gemma3_27b",
"F": "dsr1",
"G": "dsv32",
"H": "ours_v5",
"I": "mistral3",
"J": "kimi25",
"K": "q3_next80b",
"L": "ours",
"M": "ours_4b",
"N": "base_4b",
"O": "glm5",
"P": "llama33_70b",
"Q": "claude",
"R": "base",
"... | {
"pbase_4b": {
"move": 8,
"instr": 4
},
"ours": {
"move": 7,
"instr": 4
},
"q3_32b": {
"move": 9,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 8
},
"dsr1": {
"move": 8,
"instr": 4
},
"ours_v3": {
"mov... | Rf6 responses that stick to verified facts (A, J, D, E, K, O, I) score highest; those inventing d5 knight traps or Nd2+ threats lose instr points. | 5,136 | 435 | 2026-07-09T02:34:09.633677+00:00 |
26MZFDEq_141#beginner | beginner | endgame | 26MZFDEq_141 | gemini | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "ours_v5",
"D": "mistral3",
"E": "gemma3_27b",
"F": "base",
"G": "kimi25",
"H": "pbase_4b",
"I": "glm5",
"J": "ours",
"K": "gpt",
"L": "claude",
"M": "base_4b",
"N": "dsv32",
"O": "llama33_70b",
"P": "gemini",
"Q": "ours_v4",
"R": "q3_next80b",
... | {
"pbase_4b": {
"move": 10,
"instr": 8
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 5
},
"mistral3": {
"move": 10,
"instr": 4
},
"ours_v5": {
"move": 0,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Evaluated based on soundness of moves and accuracy of the board geometry and instruction. | 3,378 | 1,473 | 2026-07-09T02:34:11.867441+00:00 |
26MZFDEq_141#beginner | beginner | endgame | 26MZFDEq_141 | gpt | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "ours_v5",
"D": "mistral3",
"E": "gemma3_27b",
"F": "base",
"G": "kimi25",
"H": "pbase_4b",
"I": "glm5",
"J": "ours",
"K": "gpt",
"L": "claude",
"M": "base_4b",
"N": "dsv32",
"O": "llama33_70b",
"P": "gemini",
"Q": "ours_v4",
"R": "q3_next80b",
... | {
"pbase_4b": {
"move": 10,
"instr": 6
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 0,
"instr": 1
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Best responses recommend a verified sound move and explain a simple double-duty or pawn-pushing endgame idea without inventing tactics. | 3,247 | 919 | 2026-07-09T02:34:15.069098+00:00 |
26MZFDEq_141#intermediate | intermediate | endgame | 26MZFDEq_141 | gpt | 20 | {
"A": "gpt",
"B": "claude",
"C": "ours_v4",
"D": "ours",
"E": "base_4b",
"F": "llama33_70b",
"G": "glm5",
"H": "kimi25",
"I": "ours_4b",
"J": "q3_next80b",
"K": "dsv32",
"L": "base",
"M": "ours_v5",
"N": "pbase_4b",
"O": "dsr1",
"P": "gemini",
"Q": "ours_v3",
"R": "mistral3",
"S":... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 6,
"instr": 1
},
"q3_32b": {
"move": 8,
"instr": 4
},
"mistral3": {
"move": 6,
"instr": 2
},
"ours_v5": {
"move": 7,
"instr": 5
},
"dsr1": {
"move": 8,
"instr": 5
},
"ours_v3": {
"mov... | Best responses recommend a verified quiet improving move and explain concrete attack/defense without inventing tactics. | 3,398 | 982 | 2026-07-09T02:34:16.985702+00:00 |
0edKejOp_89#beginner | beginner | endgame | 0edKejOp_89 | claude | 20 | {
"A": "base",
"B": "ours_v3",
"C": "ours",
"D": "llama33_70b",
"E": "gemini",
"F": "q3_32b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "gpt",
"J": "claude",
"K": "mistral3",
"L": "ours_v4",
"M": "ours_v5",
"N": "dsr1",
"O": "q3_next80b",
"P": "dsv32",
"Q": "glm5",
"R": "base_4b",
"S":... | {
"pbase_4b": {
"move": 8,
"instr": 7
},
"ours": {
"move": 6,
"instr": 2
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 8,
"instr": 7
},
"ours_v5": {
"move": 6,
"instr": 2
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | c4 was sound, so responses framing it as a mistake lose some credit; queen-defense and Qf5-attack ideas are well explained by G, I, L, O, Q. | 4,883 | 431 | 2026-07-09T02:34:17.360704+00:00 |
0edKejOp_89#beginner | beginner | endgame | 0edKejOp_89 | gemini | 20 | {
"A": "base",
"B": "ours_v3",
"C": "ours",
"D": "llama33_70b",
"E": "gemini",
"F": "q3_32b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "gpt",
"J": "claude",
"K": "mistral3",
"L": "ours_v4",
"M": "ours_v5",
"N": "dsr1",
"O": "q3_next80b",
"P": "dsv32",
"Q": "glm5",
"R": "base_4b",
"S":... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 0,
"instr": 0
},
"ours_v5": {
"move": 0,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 6
},
"ours_v3": {
"... | Evaluated all responses against verified reference facts, heavily penalizing hallucinations of non-existent threats, pieces, or attacks. | 3,473 | 346 | 2026-07-09T02:34:18.785814+00:00 |
26MZFDEq_141#advanced | advanced | endgame | 26MZFDEq_141 | gemini | 20 | {
"A": "gpt",
"B": "pbase_4b",
"C": "ours_v3",
"D": "q3_32b",
"E": "gemma3_27b",
"F": "dsr1",
"G": "dsv32",
"H": "ours_v5",
"I": "mistral3",
"J": "kimi25",
"K": "q3_next80b",
"L": "ours",
"M": "ours_4b",
"N": "base_4b",
"O": "glm5",
"P": "llama33_70b",
"Q": "claude",
"R": "base",
"... | {
"pbase_4b": {
"move": 10,
"instr": 2
},
"ours": {
"move": 10,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 9
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 9
},
"ours_v3": {
... | Scores reflect penalties for hallucinations regarding pieces and attacks (such as White having a knight or d5 attacking f1) while rewarding accurate, tier-appropriate endgame plans. | 3,517 | 1,386 | 2026-07-09T02:34:20.608392+00:00 |
0edKejOp_89#intermediate | intermediate | endgame | 0edKejOp_89 | claude | 20 | {
"A": "base_4b",
"B": "gemini",
"C": "ours_v4",
"D": "gpt",
"E": "q3_next80b",
"F": "base",
"G": "dsv32",
"H": "gemma3_27b",
"I": "mistral3",
"J": "ours_v3",
"K": "ours",
"L": "claude",
"M": "q3_32b",
"N": "ours_v5",
"O": "pbase_4b",
"P": "ours_4b",
"Q": "dsr1",
"R": "kimi25",
"S"... | {
"pbase_4b": {
"move": 8,
"instr": 6
},
"ours": {
"move": 9,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 5
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 5
},
"ours_v3": {
"mov... | Most correctly recommend Qd3 with sound reasoning; some invent Black threats (I, Q) or wrong squares/pieces (C, P) which lowers instr. | 5,222 | 433 | 2026-07-09T02:34:23.361900+00:00 |
26MZFDEq_141#advanced | advanced | endgame | 26MZFDEq_141 | gpt | 20 | {
"A": "gpt",
"B": "pbase_4b",
"C": "ours_v3",
"D": "q3_32b",
"E": "gemma3_27b",
"F": "dsr1",
"G": "dsv32",
"H": "ours_v5",
"I": "mistral3",
"J": "kimi25",
"K": "q3_next80b",
"L": "ours",
"M": "ours_4b",
"N": "base_4b",
"O": "glm5",
"P": "llama33_70b",
"Q": "claude",
"R": "base",
"... | {
"pbase_4b": {
"move": 9,
"instr": 4
},
"ours": {
"move": 8,
"instr": 3
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 8,
"instr": 4
},
"ours_v3": {
"m... | Best responses recommend a verified sound move and explain either Rf6's double-purpose role or d5's quiet conversion without inventing tactics. | 3,372 | 1,022 | 2026-07-09T02:34:23.989190+00:00 |
0edKejOp_89#beginner | beginner | endgame | 0edKejOp_89 | gpt | 20 | {
"A": "base",
"B": "ours_v3",
"C": "ours",
"D": "llama33_70b",
"E": "gemini",
"F": "q3_32b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "gpt",
"J": "claude",
"K": "mistral3",
"L": "ours_v4",
"M": "ours_v5",
"N": "dsr1",
"O": "q3_next80b",
"P": "dsv32",
"Q": "glm5",
"R": "base_4b",
"S":... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 8,
"instr": 1
},
"q3_32b": {
"move": 9,
"instr": 4
},
"mistral3": {
"move": 9,
"instr": 4
},
"ours_v5": {
"move": 8,
"instr": 3
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | All recommended moves were sound, but several explanations invented pressure, targets, or piece activity not supported by the position. | 3,333 | 947 | 2026-07-09T02:34:25.543525+00:00 |
0edKejOp_89#advanced | advanced | endgame | 0edKejOp_89 | claude | 20 | {
"A": "mistral3",
"B": "dsr1",
"C": "ours_v5",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "ours",
"J": "gemini",
"K": "ours_v3",
"L": "kimi25",
"M": "llama33_70b",
"N": "dsv32",
"O": "gemma3_27b",
"P": "glm5",
"Q": "base",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 7,
"instr": 4
},
"ours": {
"move": 8,
"instr": 6
},
"q3_32b": {
"move": 8,
"instr": 7
},
"mistral3": {
"move": 8,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | T is garbled with fabricated squares; all four moves are engine-sound, best responses combine correct defense with clear tier-appropriate method. | 5,690 | 420 | 2026-07-09T02:34:28.785525+00:00 |
05minwyU_68#beginner | beginner | middlegame | 05minwyU_68 | claude | 20 | {
"A": "ours",
"B": "ours_v4",
"C": "mistral3",
"D": "q3_32b",
"E": "claude",
"F": "dsr1",
"G": "llama33_70b",
"H": "dsv32",
"I": "base_4b",
"J": "glm5",
"K": "ours_v5",
"L": "ours_4b",
"M": "gemma3_27b",
"N": "gpt",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "base",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 9
},
"ours_v5": {
"move": 7,
"instr": 4
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Rc3 responses generally strong; R falsely suggests capturing e6 and is misleading. | 5,136 | 406 | 2026-07-09T02:34:33.948173+00:00 |
0edKejOp_89#intermediate | intermediate | endgame | 0edKejOp_89 | gemini | 20 | {
"A": "base_4b",
"B": "gemini",
"C": "ours_v4",
"D": "gpt",
"E": "q3_next80b",
"F": "base",
"G": "dsv32",
"H": "gemma3_27b",
"I": "mistral3",
"J": "ours_v3",
"K": "ours",
"L": "claude",
"M": "q3_32b",
"N": "ours_v5",
"O": "pbase_4b",
"P": "ours_4b",
"Q": "dsr1",
"R": "kimi25",
"S"... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 10
},
"q3_32b": {
"move": 10,
"instr": 5
},
"mistral3": {
"move": 10,
"instr": 0
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 0
},
"ours_v3": {
... | Evaluated based strictly on adherence to verified facts, penalizing hallucinations of moves like Rc5 or Re6, and rewarding correct incorporation of the verified defensive properties of each candidate move. | 3,643 | 1,709 | 2026-07-09T02:34:37.141117+00:00 |
0edKejOp_89#advanced | advanced | endgame | 0edKejOp_89 | gemini | 20 | {
"A": "mistral3",
"B": "dsr1",
"C": "ours_v5",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "ours",
"J": "gemini",
"K": "ours_v3",
"L": "kimi25",
"M": "llama33_70b",
"N": "dsv32",
"O": "gemma3_27b",
"P": "glm5",
"Q": "base",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 0,
"instr": 0
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Responses with hallucinations about phantom knights, non-existent rooks, imaginary pins, or incorrect square control were penalized. | 3,925 | 1,361 | 2026-07-09T02:34:37.938653+00:00 |
0edKejOp_89#intermediate | intermediate | endgame | 0edKejOp_89 | gpt | 20 | {
"A": "base_4b",
"B": "gemini",
"C": "ours_v4",
"D": "gpt",
"E": "q3_next80b",
"F": "base",
"G": "dsv32",
"H": "gemma3_27b",
"I": "mistral3",
"J": "ours_v3",
"K": "ours",
"L": "claude",
"M": "q3_32b",
"N": "ours_v5",
"O": "pbase_4b",
"P": "ours_4b",
"Q": "dsr1",
"R": "kimi25",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 5
},
"ours": {
"move": 9,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 7,
"instr": 2
},
"ours_v5": {
"move": 10,
"instr": 10
},
"dsr1": {
"move": 7,
"instr": 3
},
"ours_v3": {
"m... | Most recommended moves were sound, but several explanations invented pressure, loose pieces, or tactics not supported by the verified position. | 3,529 | 1,202 | 2026-07-09T02:34:39.095751+00:00 |
0edKejOp_89#advanced | advanced | endgame | 0edKejOp_89 | gpt | 20 | {
"A": "mistral3",
"B": "dsr1",
"C": "ours_v5",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "ours_4b",
"H": "pbase_4b",
"I": "ours",
"J": "gemini",
"K": "ours_v3",
"L": "kimi25",
"M": "llama33_70b",
"N": "dsv32",
"O": "gemma3_27b",
"P": "glm5",
"Q": "base",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 9,
"instr": 3
},
"ours": {
"move": 8,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 3
},
"ours_v5": {
"move": 9,
"instr": 7
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mo... | Best coaching stayed concrete about queen moves defending key pawns or Qf5 attacking the rook, while weaker answers invented threats, pieces, or plans not in the position. | 3,801 | 1,108 | 2026-07-09T02:34:42.685776+00:00 |
05minwyU_68#intermediate | intermediate | middlegame | 05minwyU_68 | gemini | 20 | {
"A": "ours_4b",
"B": "kimi25",
"C": "base",
"D": "gpt",
"E": "claude",
"F": "ours_v3",
"G": "base_4b",
"H": "q3_32b",
"I": "q3_next80b",
"J": "mistral3",
"K": "pbase_4b",
"L": "dsv32",
"M": "llama33_70b",
"N": "glm5",
"O": "gemini",
"P": "dsr1",
"Q": "gemma3_27b",
"R": "ours_v5",
... | {
"pbase_4b": {
"move": 10,
"instr": 1
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 1
},
"mistral3": {
"move": 10,
"instr": 2
},
"ours_v5": {
"move": 0,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Evaluated based on correct identification of piece roles, sound candidate moves, and instructive value for an intermediate player, penalizing hallucinations of pieces or threats. | 3,905 | 356 | 2026-07-09T02:34:43.268739+00:00 |
05minwyU_68#beginner | beginner | middlegame | 05minwyU_68 | gpt | 20 | {
"A": "ours",
"B": "ours_v4",
"C": "mistral3",
"D": "q3_32b",
"E": "claude",
"F": "dsr1",
"G": "llama33_70b",
"H": "dsv32",
"I": "base_4b",
"J": "glm5",
"K": "ours_v5",
"L": "ours_4b",
"M": "gemma3_27b",
"N": "gpt",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "base",
... | {
"pbase_4b": {
"move": 6,
"instr": 3
},
"ours": {
"move": 8,
"instr": 3
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 4
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Sound active-rook advice scored best, while invented tactics or confusion about the e6 pawn was penalized. | 3,461 | 1,035 | 2026-07-09T02:34:43.724216+00:00 |
05minwyU_68#intermediate | intermediate | middlegame | 05minwyU_68 | claude | 20 | {
"A": "ours_4b",
"B": "kimi25",
"C": "base",
"D": "gpt",
"E": "claude",
"F": "ours_v3",
"G": "base_4b",
"H": "q3_32b",
"I": "q3_next80b",
"J": "mistral3",
"K": "pbase_4b",
"L": "dsv32",
"M": "llama33_70b",
"N": "glm5",
"O": "gemini",
"P": "dsr1",
"Q": "gemma3_27b",
"R": "ours_v5",
... | {
"pbase_4b": {
"move": 6,
"instr": 3
},
"ours": {
"move": 8,
"instr": 3
},
"q3_32b": {
"move": 6,
"instr": 5
},
"mistral3": {
"move": 7,
"instr": 5
},
"ours_v5": {
"move": 3,
"instr": 2
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | R and T invent a queen/e5 fiction; C/G/K misdescribe threats; D and E give best sound tier-appropriate coaching. | 5,676 | 424 | 2026-07-09T02:34:43.928198+00:00 |
05minwyU_68#beginner | beginner | middlegame | 05minwyU_68 | gemini | 20 | {
"A": "ours",
"B": "ours_v4",
"C": "mistral3",
"D": "q3_32b",
"E": "claude",
"F": "dsr1",
"G": "llama33_70b",
"H": "dsv32",
"I": "base_4b",
"J": "glm5",
"K": "ours_v5",
"L": "ours_4b",
"M": "gemma3_27b",
"N": "gpt",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "base",
... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 9
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Responses that hallucinate capturing one's own pawn on e6 or claim Rc3 defends e6 received zeros for contradicting board facts. | 3,632 | 1,341 | 2026-07-09T02:34:48.400836+00:00 |
05minwyU_68#advanced | advanced | middlegame | 05minwyU_68 | claude | 20 | {
"A": "gemma3_27b",
"B": "ours_4b",
"C": "dsr1",
"D": "pbase_4b",
"E": "claude",
"F": "base_4b",
"G": "q3_32b",
"H": "gpt",
"I": "ours_v4",
"J": "dsv32",
"K": "ours_v5",
"L": "ours",
"M": "base",
"N": "kimi25",
"O": "q3_next80b",
"P": "gemini",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 8,
"instr": 2
},
"ours": {
"move": 8,
"instr": 3
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 7,
"instr": 4
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Rc3 answers are strong; several Rc4 replies invent a bogus Rxe6 tactic, hurting instr. | 5,584 | 416 | 2026-07-09T02:34:48.716045+00:00 |
0bFEeDLi_49#beginner | beginner | middlegame | 0bFEeDLi_49 | claude | 20 | {
"A": "llama33_70b",
"B": "base_4b",
"C": "base",
"D": "dsv32",
"E": "q3_32b",
"F": "gpt",
"G": "q3_next80b",
"H": "dsr1",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "kimi25",
"L": "gemini",
"M": "ours_4b",
"N": "mistral3",
"O": "ours",
"P": "claude",
"Q": "pbase_4b",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 8,
"instr": 6
},
"ours": {
"move": 8,
"instr": 7
},
"q3_32b": {
"move": 8,
"instr": 4
},
"mistral3": {
"move": 8,
"instr": 3
},
"ours_v5": {
"move": 8,
"instr": 6
},
"dsr1": {
"move": 8,
"instr": 8
},
"ours_v3": {
"mov... | Kf1 and a3 are both sound; responses fabricating pins/threats (L, G, C, N) scored lower on instruction. | 4,936 | 418 | 2026-07-09T02:34:53.664745+00:00 |
05minwyU_68#advanced | advanced | middlegame | 05minwyU_68 | gemini | 20 | {
"A": "gemma3_27b",
"B": "ours_4b",
"C": "dsr1",
"D": "pbase_4b",
"E": "claude",
"F": "base_4b",
"G": "q3_32b",
"H": "gpt",
"I": "ours_v4",
"J": "dsv32",
"K": "ours_v5",
"L": "ours",
"M": "base",
"N": "kimi25",
"O": "q3_next80b",
"P": "gemini",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 10,
"instr": 2
},
"ours": {
"move": 10,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Responses that correctly identify the active targets of Rc3 and acknowledge Rc6 as a solid alternative scored highest in instruction, while those fabricating captures or non-existent threats were heavily penalized. | 3,782 | 1,184 | 2026-07-09T02:34:55.713363+00:00 |
05minwyU_68#intermediate | intermediate | middlegame | 05minwyU_68 | gpt | 20 | {
"A": "ours_4b",
"B": "kimi25",
"C": "base",
"D": "gpt",
"E": "claude",
"F": "ours_v3",
"G": "base_4b",
"H": "q3_32b",
"I": "q3_next80b",
"J": "mistral3",
"K": "pbase_4b",
"L": "dsv32",
"M": "llama33_70b",
"N": "glm5",
"O": "gemini",
"P": "dsr1",
"Q": "gemma3_27b",
"R": "ours_v5",
... | {
"pbase_4b": {
"move": 9,
"instr": 2
},
"ours": {
"move": 7,
"instr": 1
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 4
},
"ours_v5": {
"move": 5,
"instr": 1
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Best responses recommend an active rook move like Rc3 with correct targets and avoid inventing non-existent pieces or threats. | 3,729 | 932 | 2026-07-09T02:34:56.006008+00:00 |
05minwyU_68#advanced | advanced | middlegame | 05minwyU_68 | gpt | 20 | {
"A": "gemma3_27b",
"B": "ours_4b",
"C": "dsr1",
"D": "pbase_4b",
"E": "claude",
"F": "base_4b",
"G": "q3_32b",
"H": "gpt",
"I": "ours_v4",
"J": "dsv32",
"K": "ours_v5",
"L": "ours",
"M": "base",
"N": "kimi25",
"O": "q3_next80b",
"P": "gemini",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 6,
"instr": 1
},
"ours": {
"move": 9,
"instr": 1
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 6
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
"m... | Best coaching emphasized active rook counterplay with Rc3 or Rc4 without inventing impossible captures or nonexistent targets. | 3,619 | 1,091 | 2026-07-09T02:35:01.616785+00:00 |
0bFEeDLi_49#beginner | beginner | middlegame | 0bFEeDLi_49 | gpt | 20 | {
"A": "llama33_70b",
"B": "base_4b",
"C": "base",
"D": "dsv32",
"E": "q3_32b",
"F": "gpt",
"G": "q3_next80b",
"H": "dsr1",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "kimi25",
"L": "gemini",
"M": "ours_4b",
"N": "mistral3",
"O": "ours",
"P": "claude",
"Q": "pbase_4b",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 8,
"instr": 2
},
"mistral3": {
"move": 8,
"instr": 2
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mo... | Sound quiet moves scored well, while fabricated threats or confusing explanations reduced instruction scores. | 3,371 | 994 | 2026-07-09T02:35:01.676058+00:00 |
0bFEeDLi_49#intermediate | intermediate | middlegame | 0bFEeDLi_49 | claude | 20 | {
"A": "ours_v5",
"B": "ours_v3",
"C": "ours_v4",
"D": "q3_32b",
"E": "gemini",
"F": "q3_next80b",
"G": "ours_4b",
"H": "mistral3",
"I": "glm5",
"J": "base_4b",
"K": "gemma3_27b",
"L": "base",
"M": "gpt",
"N": "dsv32",
"O": "claude",
"P": "pbase_4b",
"Q": "kimi25",
"R": "dsr1",
"S"... | {
"pbase_4b": {
"move": 6,
"instr": 3
},
"ours": {
"move": 9,
"instr": 7
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Kf1 and Kg2 are both sound; L fabricates a queen and is worst, while Kg2 responses claiming Kf1 loses to a nonexistent pin overstate but recommend a sound move. | 5,529 | 445 | 2026-07-09T02:35:01.967931+00:00 |
0bFEeDLi_49#advanced | advanced | middlegame | 0bFEeDLi_49 | gemini | 20 | {
"A": "gemma3_27b",
"B": "llama33_70b",
"C": "gemini",
"D": "ours",
"E": "ours_v3",
"F": "ours_4b",
"G": "mistral3",
"H": "claude",
"I": "ours_v5",
"J": "ours_v4",
"K": "pbase_4b",
"L": "kimi25",
"M": "q3_32b",
"N": "base_4b",
"O": "glm5",
"P": "base",
"Q": "dsr1",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 10,
"instr": 1
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 1
},
"mistral3": {
"move": 10,
"instr": 4
},
"ours_v5": {
"move": 10,
"instr": 2
},
"dsr1": {
"move": 10,
"instr": 4
},
"ours_v3": {
... | Evaluated based on adherence to facts, appropriate move recommendation, and instructional quality. | 3,914 | 344 | 2026-07-09T02:35:05.671904+00:00 |
0bFEeDLi_49#advanced | advanced | middlegame | 0bFEeDLi_49 | claude | 20 | {
"A": "gemma3_27b",
"B": "llama33_70b",
"C": "gemini",
"D": "ours",
"E": "ours_v3",
"F": "ours_4b",
"G": "mistral3",
"H": "claude",
"I": "ours_v5",
"J": "ours_v4",
"K": "pbase_4b",
"L": "kimi25",
"M": "q3_32b",
"N": "base_4b",
"O": "glm5",
"P": "base",
"Q": "dsr1",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 7,
"instr": 4
},
"ours": {
"move": 8,
"instr": 8
},
"q3_32b": {
"move": 4,
"instr": 2
},
"mistral3": {
"move": 6,
"instr": 4
},
"ours_v5": {
"move": 4,
"instr": 2
},
"dsr1": {
"move": 6,
"instr": 4
},
"ours_v3": {
"mov... | Kg2 responses (D,L) strongest; several fabricate threats or wrong squares (I,J,M,P) lowering both scores. | 5,726 | 420 | 2026-07-09T02:35:07.344235+00:00 |
0bFEeDLi_49#intermediate | intermediate | middlegame | 0bFEeDLi_49 | gemini | 20 | {
"A": "ours_v5",
"B": "ours_v3",
"C": "ours_v4",
"D": "q3_32b",
"E": "gemini",
"F": "q3_next80b",
"G": "ours_4b",
"H": "mistral3",
"I": "glm5",
"J": "base_4b",
"K": "gemma3_27b",
"L": "base",
"M": "gpt",
"N": "dsv32",
"O": "claude",
"P": "pbase_4b",
"Q": "kimi25",
"R": "dsr1",
"S"... | {
"pbase_4b": {
"move": 10,
"instr": 1
},
"ours": {
"move": 10,
"instr": 1
},
"q3_32b": {
"move": 10,
"instr": 3
},
"mistral3": {
"move": 10,
"instr": 1
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Responses that correctly validate the student's sound move (Kf1) and provide grounded reasoning scored highest; responses hallucinating queens, phantom pins, or impossible defensive relationships were heavily penalized. | 3,862 | 1,389 | 2026-07-09T02:35:09.726067+00:00 |
0bFEeDLi_49#beginner | beginner | middlegame | 0bFEeDLi_49 | gemini | 20 | {
"A": "llama33_70b",
"B": "base_4b",
"C": "base",
"D": "dsv32",
"E": "q3_32b",
"F": "gpt",
"G": "q3_next80b",
"H": "dsr1",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "kimi25",
"L": "gemini",
"M": "ours_4b",
"N": "mistral3",
"O": "ours",
"P": "claude",
"Q": "pbase_4b",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 10,
"instr": 3
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 1
},
"mistral3": {
"move": 10,
"instr": 1
},
"ours_v5": {
"move": 10,
"instr": 3
},
"dsr1": {
"move": 10,
"instr": 4
},
"ours_v3": {
... | Responses were heavily penalized for hallucinating tactical motives, non-existent threats, or mentioning unverified facts, with K providing the best straightforward explanation for a3. | 3,505 | 2,166 | 2026-07-09T02:35:11.416914+00:00 |
2FDDWj7A_19#beginner | beginner | opening | 2FDDWj7A_19 | claude | 20 | {
"A": "gpt",
"B": "dsr1",
"C": "glm5",
"D": "q3_next80b",
"E": "ours_v5",
"F": "ours_v4",
"G": "base_4b",
"H": "q3_32b",
"I": "pbase_4b",
"J": "gemini",
"K": "kimi25",
"L": "dsv32",
"M": "mistral3",
"N": "ours",
"O": "gemma3_27b",
"P": "llama33_70b",
"Q": "ours_v3",
"R": "claude",
... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Most recommend sound Re1 well; F fabricates squares, S praises Bh3+ with weak reasoning. | 5,062 | 353 | 2026-07-09T02:35:12.777214+00:00 |
0bFEeDLi_49#intermediate | intermediate | middlegame | 0bFEeDLi_49 | gpt | 20 | {
"A": "ours_v5",
"B": "ours_v3",
"C": "ours_v4",
"D": "q3_32b",
"E": "gemini",
"F": "q3_next80b",
"G": "ours_4b",
"H": "mistral3",
"I": "glm5",
"J": "base_4b",
"K": "gemma3_27b",
"L": "base",
"M": "gpt",
"N": "dsv32",
"O": "claude",
"P": "pbase_4b",
"Q": "kimi25",
"R": "dsr1",
"S"... | {
"pbase_4b": {
"move": 7,
"instr": 1
},
"ours": {
"move": 6,
"instr": 3
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 6,
"instr": 2
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 4
},
"ours_v3": {
"mov... | Best responses praised the sound Kf1 with simple coordination logic; many others chose sound moves but invented threats or misstated what the moves defend. | 3,734 | 1,110 | 2026-07-09T02:35:15.317792+00:00 |
2FDDWj7A_19#intermediate | intermediate | opening | 2FDDWj7A_19 | claude | 20 | {
"A": "mistral3",
"B": "ours_v3",
"C": "gpt",
"D": "glm5",
"E": "q3_next80b",
"F": "base_4b",
"G": "ours_v4",
"H": "q3_32b",
"I": "claude",
"J": "base",
"K": "ours",
"L": "dsr1",
"M": "llama33_70b",
"N": "kimi25",
"O": "gemma3_27b",
"P": "ours_v5",
"Q": "ours_4b",
"R": "gemini",
"... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 9,
"instr": 7
},
"q3_32b": {
"move": 9,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 9,
"instr": 7
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Most recommend sound Re1; G has fabricated squares/pieces, J wrongly praises the student's move, S/T overstate winning the knight. | 5,406 | 423 | 2026-07-09T02:35:18.599771+00:00 |
2FDDWj7A_19#beginner | beginner | opening | 2FDDWj7A_19 | gemini | 20 | {
"A": "gpt",
"B": "dsr1",
"C": "glm5",
"D": "q3_next80b",
"E": "ours_v5",
"F": "ours_v4",
"G": "base_4b",
"H": "q3_32b",
"I": "pbase_4b",
"J": "gemini",
"K": "kimi25",
"L": "dsv32",
"M": "mistral3",
"N": "ours",
"O": "gemma3_27b",
"P": "llama33_70b",
"Q": "ours_v3",
"R": "claude",
... | {
"pbase_4b": {
"move": 10,
"instr": 8
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 3
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Scores reflect accurate use of verified facts, proper move soundness, and appropriate instructional value for a beginner. | 3,598 | 847 | 2026-07-09T02:35:20.054126+00:00 |
2FDDWj7A_19#beginner | beginner | opening | 2FDDWj7A_19 | gpt | 20 | {
"A": "gpt",
"B": "dsr1",
"C": "glm5",
"D": "q3_next80b",
"E": "ours_v5",
"F": "ours_v4",
"G": "base_4b",
"H": "q3_32b",
"I": "pbase_4b",
"J": "gemini",
"K": "kimi25",
"L": "dsv32",
"M": "mistral3",
"N": "ours",
"O": "gemma3_27b",
"P": "llama33_70b",
"Q": "ours_v3",
"R": "claude",
... | {
"pbase_4b": {
"move": 10,
"instr": 8
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 6
},
"ours_v5": {
"move": 9,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
"... | Best responses chose a sound simple developing move and explained it without inventing tactics or overstating the check. | 3,434 | 897 | 2026-07-09T02:35:20.341423+00:00 |
0bFEeDLi_49#advanced | advanced | middlegame | 0bFEeDLi_49 | gpt | 20 | {
"A": "gemma3_27b",
"B": "llama33_70b",
"C": "gemini",
"D": "ours",
"E": "ours_v3",
"F": "ours_4b",
"G": "mistral3",
"H": "claude",
"I": "ours_v5",
"J": "ours_v4",
"K": "pbase_4b",
"L": "kimi25",
"M": "q3_32b",
"N": "base_4b",
"O": "glm5",
"P": "base",
"Q": "dsr1",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 8,
"instr": 2
},
"ours": {
"move": 9,
"instr": 7
},
"q3_32b": {
"move": 7,
"instr": 0
},
"mistral3": {
"move": 9,
"instr": 1
},
"ours_v5": {
"move": 7,
"instr": 0
},
"dsr1": {
"move": 8,
"instr": 3
},
"ours_v3": {
"mov... | Sound move choices were often undermined by fabricated threats, pieces, or plans not present in the position. | 3,784 | 1,267 | 2026-07-09T02:35:25.036773+00:00 |
2FDDWj7A_19#advanced | advanced | opening | 2FDDWj7A_19 | claude | 20 | {
"A": "ours_v4",
"B": "gemma3_27b",
"C": "dsr1",
"D": "q3_next80b",
"E": "ours_4b",
"F": "ours_v3",
"G": "mistral3",
"H": "ours",
"I": "claude",
"J": "kimi25",
"K": "dsv32",
"L": "pbase_4b",
"M": "glm5",
"N": "gemini",
"O": "q3_32b",
"P": "base",
"Q": "base_4b",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 8,
"instr": 6
},
"ours": {
"move": 8,
"instr": 8
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 8,
"instr": 7
},
"ours_v5": {
"move": 8,
"instr": 8
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | Bh3+ was actually sound; most Re1 responses are good coaching, though F wrongly claims b4 attacks f8 bishop and K claims a pin that doesn't exist. | 5,776 | 435 | 2026-07-09T02:35:26.082464+00:00 |
2FDDWj7A_19#advanced | advanced | opening | 2FDDWj7A_19 | gemini | 20 | {
"A": "ours_v4",
"B": "gemma3_27b",
"C": "dsr1",
"D": "q3_next80b",
"E": "ours_4b",
"F": "ours_v3",
"G": "mistral3",
"H": "ours",
"I": "claude",
"J": "kimi25",
"K": "dsv32",
"L": "pbase_4b",
"M": "glm5",
"N": "gemini",
"O": "q3_32b",
"P": "base",
"Q": "base_4b",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 6
},
"q3_32b": {
"move": 10,
"instr": 5
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 5
},
"dsr1": {
"move": 10,
"instr": 6
},
"ours_v3": {
... | Evaluated based on reference facts and instructional value. | 4,003 | 670 | 2026-07-09T02:35:27.788380+00:00 |
2FDDWj7A_19#intermediate | intermediate | opening | 2FDDWj7A_19 | gemini | 20 | {
"A": "mistral3",
"B": "ours_v3",
"C": "gpt",
"D": "glm5",
"E": "q3_next80b",
"F": "base_4b",
"G": "ours_v4",
"H": "q3_32b",
"I": "claude",
"J": "base",
"K": "ours",
"L": "dsr1",
"M": "llama33_70b",
"N": "kimi25",
"O": "gemma3_27b",
"P": "ours_v5",
"Q": "ours_4b",
"R": "gemini",
"... | {
"pbase_4b": {
"move": 10,
"instr": 0
},
"ours": {
"move": 10,
"instr": 7
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Responses that hallucinated pieces, squares, or attacks (like G, N, S, T) were penalized in instruction. | 3,766 | 1,071 | 2026-07-09T02:35:28.302375+00:00 |
2FDDWj7A_19#intermediate | intermediate | opening | 2FDDWj7A_19 | gpt | 20 | {
"A": "mistral3",
"B": "ours_v3",
"C": "gpt",
"D": "glm5",
"E": "q3_next80b",
"F": "base_4b",
"G": "ours_v4",
"H": "q3_32b",
"I": "claude",
"J": "base",
"K": "ours",
"L": "dsr1",
"M": "llama33_70b",
"N": "kimi25",
"O": "gemma3_27b",
"P": "ours_v5",
"Q": "ours_4b",
"R": "gemini",
"... | {
"pbase_4b": {
"move": 10,
"instr": 4
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Most Re1 recommendations were sound, with instruction lowered when responses invented concrete tactics or contradicted the verified facts. | 3,606 | 914 | 2026-07-09T02:35:28.979613+00:00 |
2FDDWj7A_19#advanced | advanced | opening | 2FDDWj7A_19 | gpt | 20 | {
"A": "ours_v4",
"B": "gemma3_27b",
"C": "dsr1",
"D": "q3_next80b",
"E": "ours_4b",
"F": "ours_v3",
"G": "mistral3",
"H": "ours",
"I": "claude",
"J": "kimi25",
"K": "dsv32",
"L": "pbase_4b",
"M": "glm5",
"N": "gemini",
"O": "q3_32b",
"P": "base",
"Q": "base_4b",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 4
},
"q3_32b": {
"move": 9,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 3
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Most recommendations chose sound moves, but several explanations invented threats, pins, loose pieces, or follow-ups not supported by the position. | 3,814 | 1,167 | 2026-07-09T02:35:42.466778+00:00 |
3ScuG09o_76#beginner | beginner | endgame | 3ScuG09o_76 | claude | 20 | {
"A": "ours_v5",
"B": "q3_next80b",
"C": "ours_v4",
"D": "dsv32",
"E": "gpt",
"F": "mistral3",
"G": "q3_32b",
"H": "pbase_4b",
"I": "ours_4b",
"J": "glm5",
"K": "ours_v3",
"L": "gemini",
"M": "base",
"N": "claude",
"O": "ours",
"P": "llama33_70b",
"Q": "kimi25",
"R": "gemma3_27b",
... | {
"pbase_4b": {
"move": 9,
"instr": 8
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 6
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | b5 was actually sound, so responses treating it as a real mistake (M) lose instr points; Ra4/Rc3+ recommendations are all valid. | 4,739 | 427 | 2026-07-09T02:36:02.797383+00:00 |
3ScuG09o_76#intermediate | intermediate | endgame | 3ScuG09o_76 | claude | 20 | {
"A": "ours_4b",
"B": "dsv32",
"C": "gpt",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "dsr1",
"H": "ours_v3",
"I": "pbase_4b",
"J": "ours_v4",
"K": "q3_32b",
"L": "gemini",
"M": "glm5",
"N": "ours_v5",
"O": "llama33_70b",
"P": "ours",
"Q": "gemma3_27b",
"R": "base",
"S... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 8,
"instr": 4
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 3
},
"dsr1": {
"move": 8,
"instr": 8
},
"ours_v3": {
"mov... | Ra4/Rb3 responses cleaner; several Rc3+ ones fabricate king squares (c3/c6) or false threats. | 5,218 | 422 | 2026-07-09T02:36:03.042578+00:00 |
3ScuG09o_76#advanced | advanced | endgame | 3ScuG09o_76 | claude | 20 | {
"A": "pbase_4b",
"B": "glm5",
"C": "llama33_70b",
"D": "ours",
"E": "kimi25",
"F": "q3_32b",
"G": "ours_v5",
"H": "gemma3_27b",
"I": "base_4b",
"J": "gemini",
"K": "mistral3",
"L": "base",
"M": "gpt",
"N": "claude",
"O": "q3_next80b",
"P": "ours_4b",
"Q": "dsv32",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 8,
"instr": 4
},
"ours": {
"move": 9,
"instr": 9
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 7
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Most correctly recommend sound Rc3+/Ra4/Rh3; L and I contain fabricated/confused claims lowering scores. | 5,345 | 421 | 2026-07-09T02:36:03.330010+00:00 |
P8kFy28M_82#beginner | beginner | endgame | P8kFy28M_82 | claude | 20 | {
"A": "mistral3",
"B": "ours",
"C": "gemma3_27b",
"D": "q3_32b",
"E": "kimi25",
"F": "base_4b",
"G": "gemini",
"H": "llama33_70b",
"I": "q3_next80b",
"J": "pbase_4b",
"K": "dsv32",
"L": "glm5",
"M": "ours_4b",
"N": "base",
"O": "ours_v3",
"P": "dsr1",
"Q": "ours_v4",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 9,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 7
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Most recommend sound Rh5/Rh7/Rh8; Q fabricates h7/d6 squares and N gives vague king-safety advice. | 5,015 | 424 | 2026-07-09T02:36:08.839419+00:00 |
3ScuG09o_76#beginner | beginner | endgame | 3ScuG09o_76 | gemini | 20 | {
"A": "ours_v5",
"B": "q3_next80b",
"C": "ours_v4",
"D": "dsv32",
"E": "gpt",
"F": "mistral3",
"G": "q3_32b",
"H": "pbase_4b",
"I": "ours_4b",
"J": "glm5",
"K": "ours_v3",
"L": "gemini",
"M": "base",
"N": "claude",
"O": "ours",
"P": "llama33_70b",
"Q": "kimi25",
"R": "gemma3_27b",
... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 10,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Responses were graded based on their alignment with verified facts, avoiding hallucinations, and providing clear, tier-appropriate instruction. | 3,343 | 926 | 2026-07-09T02:36:09.073919+00:00 |
3ScuG09o_76#advanced | advanced | endgame | 3ScuG09o_76 | gemini | 20 | {
"A": "pbase_4b",
"B": "glm5",
"C": "llama33_70b",
"D": "ours",
"E": "kimi25",
"F": "q3_32b",
"G": "ours_v5",
"H": "gemma3_27b",
"I": "base_4b",
"J": "gemini",
"K": "mistral3",
"L": "base",
"M": "gpt",
"N": "claude",
"O": "q3_next80b",
"P": "ours_4b",
"Q": "dsv32",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 5
},
"mistral3": {
"move": 10,
"instr": 4
},
"ours_v5": {
"move": 10,
"instr": 7
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Evaluated based on adherence to verified facts and instructiveness for an advanced player. | 3,689 | 607 | 2026-07-09T02:36:10.015275+00:00 |
3ScuG09o_76#intermediate | intermediate | endgame | 3ScuG09o_76 | gemini | 20 | {
"A": "ours_4b",
"B": "dsv32",
"C": "gpt",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "dsr1",
"H": "ours_v3",
"I": "pbase_4b",
"J": "ours_v4",
"K": "q3_32b",
"L": "gemini",
"M": "glm5",
"N": "ours_v5",
"O": "llama33_70b",
"P": "ours",
"Q": "gemma3_27b",
"R": "base",
"S... | {
"pbase_4b": {
"move": 9,
"instr": 8
},
"ours": {
"move": 9,
"instr": 0
},
"q3_32b": {
"move": 9,
"instr": 3
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 9,
"instr": 0
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Responses with factual errors regarding piece squares and illogical moves were penalized heavily. | 3,642 | 1,177 | 2026-07-09T02:36:10.347067+00:00 |
3ScuG09o_76#advanced | advanced | endgame | 3ScuG09o_76 | gpt | 20 | {
"A": "pbase_4b",
"B": "glm5",
"C": "llama33_70b",
"D": "ours",
"E": "kimi25",
"F": "q3_32b",
"G": "ours_v5",
"H": "gemma3_27b",
"I": "base_4b",
"J": "gemini",
"K": "mistral3",
"L": "base",
"M": "gpt",
"N": "claude",
"O": "q3_next80b",
"P": "ours_4b",
"Q": "dsv32",
"R": "ours_v3",
... | {
"pbase_4b": {
"move": 10,
"instr": 3
},
"ours": {
"move": 10,
"instr": 5
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 4
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Most recommendations chose a sound move, but instruction was reduced where responses overstated b5 as bad or added unsupported concrete follow-ups. | 3,528 | 881 | 2026-07-09T02:36:13.983015+00:00 |
P8kFy28M_82#intermediate | intermediate | endgame | P8kFy28M_82 | claude | 20 | {
"A": "ours_4b",
"B": "dsr1",
"C": "kimi25",
"D": "gemma3_27b",
"E": "mistral3",
"F": "base",
"G": "base_4b",
"H": "ours_v4",
"I": "claude",
"J": "dsv32",
"K": "ours_v5",
"L": "q3_32b",
"M": "llama33_70b",
"N": "ours_v3",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "glm5",
"R": "gemini... | {
"pbase_4b": {
"move": 8,
"instr": 7
},
"ours": {
"move": 8,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 3,
"instr": 2
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | K badly misstates squares (g2/d4/d6), F is vague and partly wrong; Rh5/Rh7 multitask explanations are strongest and accurate. | 5,250 | 429 | 2026-07-09T02:36:15.727554+00:00 |
3ScuG09o_76#beginner | beginner | endgame | 3ScuG09o_76 | gpt | 20 | {
"A": "ours_v5",
"B": "q3_next80b",
"C": "ours_v4",
"D": "dsv32",
"E": "gpt",
"F": "mistral3",
"G": "q3_32b",
"H": "pbase_4b",
"I": "ours_4b",
"J": "glm5",
"K": "ours_v3",
"L": "gemini",
"M": "base",
"N": "claude",
"O": "ours",
"P": "llama33_70b",
"Q": "kimi25",
"R": "gemma3_27b",
... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 7,
"instr": 3
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 7,
"instr": 3
},
"ours_v5": {
"move": 8,
"instr": 4
},
"dsr1": {
"move": 8,
"instr": 4
},
"ours_v3": {
"mov... | Sound move choices were mostly good, but several explanations invented attacks, defenses, forced king moves, or incorrect board details. | 3,159 | 1,036 | 2026-07-09T02:36:16.426924+00:00 |
P8kFy28M_82#advanced | advanced | endgame | P8kFy28M_82 | claude | 20 | {
"A": "dsv32",
"B": "llama33_70b",
"C": "pbase_4b",
"D": "q3_next80b",
"E": "ours_4b",
"F": "base",
"G": "mistral3",
"H": "ours",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "gemini",
"L": "ours_v5",
"M": "glm5",
"N": "base_4b",
"O": "claude",
"P": "ours_v3",
"Q": "kimi25",
"R": "gpt",
... | {
"pbase_4b": {
"move": 7,
"instr": 5
},
"ours": {
"move": 8,
"instr": 6
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 7,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 8
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | F is weakest with fabricated bishop; Rh7/Rh5/Ke6/b5 answers are sound with solid tier-appropriate instruction. | 5,751 | 428 | 2026-07-09T02:36:20.962870+00:00 |
3ScuG09o_76#intermediate | intermediate | endgame | 3ScuG09o_76 | gpt | 20 | {
"A": "ours_4b",
"B": "dsv32",
"C": "gpt",
"D": "claude",
"E": "base_4b",
"F": "q3_next80b",
"G": "dsr1",
"H": "ours_v3",
"I": "pbase_4b",
"J": "ours_v4",
"K": "q3_32b",
"L": "gemini",
"M": "glm5",
"N": "ours_v5",
"O": "llama33_70b",
"P": "ours",
"Q": "gemma3_27b",
"R": "base",
"S... | {
"pbase_4b": {
"move": 9,
"instr": 5
},
"ours": {
"move": 7,
"instr": 2
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 8,
"instr": 4
},
"ours_v5": {
"move": 8,
"instr": 5
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
"mo... | Best responses recommend a verified active rook move while accurately explaining simple rook-endgame pressure without inventing tactics or mislocating pieces. | 3,475 | 1,179 | 2026-07-09T02:36:21.228469+00:00 |
P8kFy28M_82#beginner | beginner | endgame | P8kFy28M_82 | gpt | 20 | {
"A": "mistral3",
"B": "ours",
"C": "gemma3_27b",
"D": "q3_32b",
"E": "kimi25",
"F": "base_4b",
"G": "gemini",
"H": "llama33_70b",
"I": "q3_next80b",
"J": "pbase_4b",
"K": "dsv32",
"L": "glm5",
"M": "ours_4b",
"N": "base",
"O": "ours_v3",
"P": "dsr1",
"Q": "ours_v4",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 10,
"instr": 7
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Most chose a sound rook move, but several explanations lost credit for inventing threats, wrong defended pawns, or calling a5 acceptable. | 3,349 | 1,061 | 2026-07-09T02:36:23.897827+00:00 |
P8kFy28M_82#beginner | beginner | endgame | P8kFy28M_82 | gemini | 20 | {
"A": "mistral3",
"B": "ours",
"C": "gemma3_27b",
"D": "q3_32b",
"E": "kimi25",
"F": "base_4b",
"G": "gemini",
"H": "llama33_70b",
"I": "q3_next80b",
"J": "pbase_4b",
"K": "dsv32",
"L": "glm5",
"M": "ours_4b",
"N": "base",
"O": "ours_v3",
"P": "dsr1",
"Q": "ours_v4",
"R": "gpt",
"... | {
"pbase_4b": {
"move": 10,
"instr": 4
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 8
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Responses that hallucinated board facts, such as pieces on wrong squares or threats not present, were heavily penalized. | 3,527 | 1,431 | 2026-07-09T02:36:24.166493+00:00 |
P8kFy28M_82#advanced | advanced | endgame | P8kFy28M_82 | gemini | 20 | {
"A": "dsv32",
"B": "llama33_70b",
"C": "pbase_4b",
"D": "q3_next80b",
"E": "ours_4b",
"F": "base",
"G": "mistral3",
"H": "ours",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "gemini",
"L": "ours_v5",
"M": "glm5",
"N": "base_4b",
"O": "claude",
"P": "ours_v3",
"Q": "kimi25",
"R": "gpt",
... | {
"pbase_4b": {
"move": 10,
"instr": 4
},
"ours": {
"move": 10,
"instr": 7
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 10,
"instr": 2
},
"ours_v3": {
... | Scores reflect the soundness of the chosen move and the accuracy/utility of the explanations for an advanced player. | 3,976 | 580 | 2026-07-09T02:36:24.279510+00:00 |
P8kFy28M_82#intermediate | intermediate | endgame | P8kFy28M_82 | gemini | 20 | {
"A": "ours_4b",
"B": "dsr1",
"C": "kimi25",
"D": "gemma3_27b",
"E": "mistral3",
"F": "base",
"G": "base_4b",
"H": "ours_v4",
"I": "claude",
"J": "dsv32",
"K": "ours_v5",
"L": "q3_32b",
"M": "llama33_70b",
"N": "ours_v3",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "glm5",
"R": "gemini... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 10,
"instr": 3
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 9
},
"ours_v3": {
... | Grades reflect accurate use of the verified facts and clear, instructive explanations without hallucinating piece positions or faulty logic. | 3,683 | 1,578 | 2026-07-09T02:36:26.245997+00:00 |
P8kFy28M_82#intermediate | intermediate | endgame | P8kFy28M_82 | gpt | 20 | {
"A": "ours_4b",
"B": "dsr1",
"C": "kimi25",
"D": "gemma3_27b",
"E": "mistral3",
"F": "base",
"G": "base_4b",
"H": "ours_v4",
"I": "claude",
"J": "dsv32",
"K": "ours_v5",
"L": "q3_32b",
"M": "llama33_70b",
"N": "ours_v3",
"O": "q3_next80b",
"P": "pbase_4b",
"Q": "glm5",
"R": "gemini... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 6,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 9
},
"ours_v3": {
"m... | Best responses recommend a verified active rook move and explain the simple attack-and-defend plan without inventing nonexistent targets. | 3,499 | 1,071 | 2026-07-09T02:36:26.909272+00:00 |
ZNWZX66g_82#beginner | beginner | endgame | ZNWZX66g_82 | claude | 20 | {
"A": "claude",
"B": "llama33_70b",
"C": "dsr1",
"D": "base_4b",
"E": "dsv32",
"F": "ours_v5",
"G": "q3_next80b",
"H": "base",
"I": "gemma3_27b",
"J": "ours_4b",
"K": "ours",
"L": "q3_32b",
"M": "gemini",
"N": "mistral3",
"O": "ours_v4",
"P": "kimi25",
"Q": "glm5",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 8,
"instr": 6
},
"ours": {
"move": 7,
"instr": 5
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 8,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 3
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Kf6/g6/Rh6 responses generally strong; E recommends unsound Rxg4 losing active rook, H/F/O invent knight or wrong squares. | 4,979 | 432 | 2026-07-09T02:36:27.085446+00:00 |
ZNWZX66g_82#intermediate | intermediate | endgame | ZNWZX66g_82 | claude | 20 | {
"A": "ours_v3",
"B": "dsr1",
"C": "gpt",
"D": "base",
"E": "ours",
"F": "gemma3_27b",
"G": "kimi25",
"H": "ours_v5",
"I": "mistral3",
"J": "q3_32b",
"K": "q3_next80b",
"L": "ours_4b",
"M": "llama33_70b",
"N": "gemini",
"O": "ours_v4",
"P": "pbase_4b",
"Q": "glm5",
"R": "base_4b",
... | {
"pbase_4b": {
"move": 7,
"instr": 4
},
"ours": {
"move": 6,
"instr": 4
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 8,
"instr": 5
},
"ours_v5": {
"move": 7,
"instr": 6
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | Kf6 and g6 responses tend to be most instructive; several Rh8/Rh6 responses fabricate threats or misstate the board (O especially, E claims Rh8 defends g7 wrongly). | 5,176 | 444 | 2026-07-09T02:36:30.042787+00:00 |
P8kFy28M_82#advanced | advanced | endgame | P8kFy28M_82 | gpt | 20 | {
"A": "dsv32",
"B": "llama33_70b",
"C": "pbase_4b",
"D": "q3_next80b",
"E": "ours_4b",
"F": "base",
"G": "mistral3",
"H": "ours",
"I": "ours_v4",
"J": "gemma3_27b",
"K": "gemini",
"L": "ours_v5",
"M": "glm5",
"N": "base_4b",
"O": "claude",
"P": "ours_v3",
"Q": "kimi25",
"R": "gpt",
... | {
"pbase_4b": {
"move": 9,
"instr": 4
},
"ours": {
"move": 9,
"instr": 4
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 9,
"instr": 4
},
"ours_v3": {
"m... | Sound move choices scored well, while unsupported tactics, nonexistent pieces, or contradictions reduced instructional value. | 3,799 | 1,097 | 2026-07-09T02:36:32.810432+00:00 |
ZNWZX66g_82#advanced | advanced | endgame | ZNWZX66g_82 | claude | 20 | {
"A": "claude",
"B": "dsv32",
"C": "base_4b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "dsr1",
"G": "gemini",
"H": "gpt",
"I": "pbase_4b",
"J": "ours",
"K": "ours_v3",
"L": "kimi25",
"M": "q3_32b",
"N": "ours_4b",
"O": "q3_next80b",
"P": "mistral3",
"Q": "base",
"R": "ours_v5",
"S"... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 7,
"instr": 4
},
"ours_v5": {
"move": 8,
"instr": 4
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | Rh8 is sound so praise for it is fine; J and H best combine correct facts with clear reusable ideas, while N/O/Q/R/S invent threats or misdescribe the board. | 5,597 | 436 | 2026-07-09T02:36:33.434112+00:00 |
NxmDDPKw_97#beginner | beginner | endgame | NxmDDPKw_97 | claude | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "q3_next80b",
"D": "glm5",
"E": "mistral3",
"F": "base",
"G": "gpt",
"H": "gemini",
"I": "ours_v5",
"J": "llama33_70b",
"K": "dsv32",
"L": "pbase_4b",
"M": "kimi25",
"N": "claude",
"O": "base_4b",
"P": "ours",
"Q": "q3_32b",
"R": "dsr1",
"S": ... | {
"pbase_4b": {
"move": 8,
"instr": 7
},
"ours": {
"move": 7,
"instr": 7
},
"q3_32b": {
"move": 9,
"instr": 8
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 5
},
"dsr1": {
"move": 8,
"instr": 8
},
"ours_v3": {
"mov... | Ke6 is sound; best coaching (B,C,G,Q) shows more active alternatives while acknowledging Ke6, whereas F invents false h7 promotion threats and S misstates d5. | 4,830 | 442 | 2026-07-09T02:36:39.547596+00:00 |
ZNWZX66g_82#beginner | beginner | endgame | ZNWZX66g_82 | gpt | 20 | {
"A": "claude",
"B": "llama33_70b",
"C": "dsr1",
"D": "base_4b",
"E": "dsv32",
"F": "ours_v5",
"G": "q3_next80b",
"H": "base",
"I": "gemma3_27b",
"J": "ours_4b",
"K": "ours",
"L": "q3_32b",
"M": "gemini",
"N": "mistral3",
"O": "ours_v4",
"P": "kimi25",
"Q": "glm5",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 8,
"instr": 2
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 7,
"instr": 2
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
"mo... | Sound simple endgame moves scored well, while fabricated squares, pieces, or defensive claims sharply reduced instruction scores. | 3,340 | 1,110 | 2026-07-09T02:36:40.338552+00:00 |
ZNWZX66g_82#intermediate | intermediate | endgame | ZNWZX66g_82 | gemini | 20 | {
"A": "ours_v3",
"B": "dsr1",
"C": "gpt",
"D": "base",
"E": "ours",
"F": "gemma3_27b",
"G": "kimi25",
"H": "ours_v5",
"I": "mistral3",
"J": "q3_32b",
"K": "q3_next80b",
"L": "ours_4b",
"M": "llama33_70b",
"N": "gemini",
"O": "ours_v4",
"P": "pbase_4b",
"Q": "glm5",
"R": "base_4b",
... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Checked for factual inaccuracies regarding move functions and pieces. | 3,563 | 1,256 | 2026-07-09T02:36:41.089337+00:00 |
ZNWZX66g_82#intermediate | intermediate | endgame | ZNWZX66g_82 | gpt | 20 | {
"A": "ours_v3",
"B": "dsr1",
"C": "gpt",
"D": "base",
"E": "ours",
"F": "gemma3_27b",
"G": "kimi25",
"H": "ours_v5",
"I": "mistral3",
"J": "q3_32b",
"K": "q3_next80b",
"L": "ours_4b",
"M": "llama33_70b",
"N": "gemini",
"O": "ours_v4",
"P": "pbase_4b",
"Q": "glm5",
"R": "base_4b",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 10,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 9,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"... | Sound move choices were common, but several explanations invented defenses, threats, or follow-up lines not supported by the position. | 3,409 | 1,008 | 2026-07-09T02:36:42.144749+00:00 |
ZNWZX66g_82#advanced | advanced | endgame | ZNWZX66g_82 | gemini | 20 | {
"A": "claude",
"B": "dsv32",
"C": "base_4b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "dsr1",
"G": "gemini",
"H": "gpt",
"I": "pbase_4b",
"J": "ours",
"K": "ours_v3",
"L": "kimi25",
"M": "q3_32b",
"N": "ours_4b",
"O": "q3_next80b",
"P": "mistral3",
"Q": "base",
"R": "ours_v5",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 0,
"instr": 0
},
"q3_32b": {
"move": 8,
"instr": 7
},
"mistral3": {
"move": 0,
"instr": 0
},
"ours_v5": {
"move": 0,
"instr": 0
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Responses were penalized for hallucinating pieces, diagonals for rooks, or incorrect material evaluations. | 3,796 | 1,112 | 2026-07-09T02:36:42.643245+00:00 |
ZNWZX66g_82#beginner | beginner | endgame | ZNWZX66g_82 | gemini | 20 | {
"A": "claude",
"B": "llama33_70b",
"C": "dsr1",
"D": "base_4b",
"E": "dsv32",
"F": "ours_v5",
"G": "q3_next80b",
"H": "base",
"I": "gemma3_27b",
"J": "ours_4b",
"K": "ours",
"L": "q3_32b",
"M": "gemini",
"N": "mistral3",
"O": "ours_v4",
"P": "kimi25",
"Q": "glm5",
"R": "gpt",
"S"... | {
"pbase_4b": {
"move": 10,
"instr": 4
},
"ours": {
"move": 10,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 9
},
"mistral3": {
"move": 10,
"instr": 9
},
"ours_v5": {
"move": 10,
"instr": 0
},
"dsr1": {
"move": 10,
"instr": 10
},
"ours_v3": {
... | Responses with hallucinations (knights, wrong pawn squares, calling g6 a rook move, claiming d6 is on the edge, or inventing unlisted moves/responses like Rxg4+ or b6) were heavily penalized in the instruction score. | 3,491 | 2,152 | 2026-07-09T02:36:44.410427+00:00 |
NxmDDPKw_97#intermediate | intermediate | endgame | NxmDDPKw_97 | claude | 20 | {
"A": "base",
"B": "gemini",
"C": "llama33_70b",
"D": "ours",
"E": "ours_v5",
"F": "ours_4b",
"G": "dsv32",
"H": "gpt",
"I": "mistral3",
"J": "glm5",
"K": "kimi25",
"L": "gemma3_27b",
"M": "ours_v4",
"N": "pbase_4b",
"O": "q3_next80b",
"P": "dsr1",
"Q": "q3_32b",
"R": "ours_v3",
"... | {
"pbase_4b": {
"move": 8,
"instr": 7
},
"ours": {
"move": 8,
"instr": 9
},
"q3_32b": {
"move": 8,
"instr": 4
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 9
},
"dsr1": {
"move": 8,
"instr": 7
},
"ours_v3": {
"mov... | Ke6 and Kc7 both sound; D and E give best tier-calibrated reasoning, A fabricates knight tactics, M and Q have errors. | 5,206 | 426 | 2026-07-09T02:36:46.710726+00:00 |
ZNWZX66g_82#advanced | advanced | endgame | ZNWZX66g_82 | gpt | 20 | {
"A": "claude",
"B": "dsv32",
"C": "base_4b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "dsr1",
"G": "gemini",
"H": "gpt",
"I": "pbase_4b",
"J": "ours",
"K": "ours_v3",
"L": "kimi25",
"M": "q3_32b",
"N": "ours_4b",
"O": "q3_next80b",
"P": "mistral3",
"Q": "base",
"R": "ours_v5",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 7
},
"ours": {
"move": 9,
"instr": 1
},
"q3_32b": {
"move": 8,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 2
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mo... | Sound move choices were common, but several explanations invented false defenses, targets, or plans. | 3,633 | 1,083 | 2026-07-09T02:36:46.839317+00:00 |
NxmDDPKw_97#beginner | beginner | endgame | NxmDDPKw_97 | gemini | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "q3_next80b",
"D": "glm5",
"E": "mistral3",
"F": "base",
"G": "gpt",
"H": "gemini",
"I": "ours_v5",
"J": "llama33_70b",
"K": "dsv32",
"L": "pbase_4b",
"M": "kimi25",
"N": "claude",
"O": "base_4b",
"P": "ours",
"Q": "q3_32b",
"R": "dsr1",
"S": ... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 10,
"instr": 6
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 10,
"instr": 9
},
"ours_v3": {
... | Evaluated responses based on move soundness and instructional value without hallucinations. | 3,394 | 708 | 2026-07-09T02:36:48.108899+00:00 |
NxmDDPKw_97#advanced | advanced | endgame | NxmDDPKw_97 | claude | 20 | {
"A": "base_4b",
"B": "gemma3_27b",
"C": "q3_32b",
"D": "ours_v4",
"E": "kimi25",
"F": "q3_next80b",
"G": "dsr1",
"H": "dsv32",
"I": "ours",
"J": "gemini",
"K": "glm5",
"L": "ours_4b",
"M": "mistral3",
"N": "claude",
"O": "base",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 8,
"instr": 4
},
"ours": {
"move": 8,
"instr": 7
},
"q3_32b": {
"move": 8,
"instr": 4
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 7,
"instr": 3
},
"dsr1": {
"move": 8,
"instr": 3
},
"ours_v3": {
"mov... | Kc7 is the strongest tier-appropriate active king move; several responses invented Ne5+ lines or misdescribed the position. | 5,673 | 420 | 2026-07-09T02:36:49.997122+00:00 |
NxmDDPKw_97#intermediate | intermediate | endgame | NxmDDPKw_97 | gemini | 20 | {
"A": "base",
"B": "gemini",
"C": "llama33_70b",
"D": "ours",
"E": "ours_v5",
"F": "ours_4b",
"G": "dsv32",
"H": "gpt",
"I": "mistral3",
"J": "glm5",
"K": "kimi25",
"L": "gemma3_27b",
"M": "ours_v4",
"N": "pbase_4b",
"O": "q3_next80b",
"P": "dsr1",
"Q": "q3_32b",
"R": "ours_v3",
"... | {
"pbase_4b": {
"move": 10,
"instr": 6
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Evaluated based on the verified facts and pedagogical value for an intermediate player. | 3,609 | 590 | 2026-07-09T02:36:50.104958+00:00 |
NxmDDPKw_97#beginner | beginner | endgame | NxmDDPKw_97 | gpt | 20 | {
"A": "ours_v3",
"B": "ours_4b",
"C": "q3_next80b",
"D": "glm5",
"E": "mistral3",
"F": "base",
"G": "gpt",
"H": "gemini",
"I": "ours_v5",
"J": "llama33_70b",
"K": "dsv32",
"L": "pbase_4b",
"M": "kimi25",
"N": "claude",
"O": "base_4b",
"P": "ours",
"Q": "q3_32b",
"R": "dsr1",
"S": ... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 9,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 8
},
"mistral3": {
"move": 8,
"instr": 3
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 10,
"instr": 9
},
"ours_v3": {
"... | Best responses recommend a verified sound move while accurately explaining simple endgame activity without inventing threats. | 3,265 | 1,000 | 2026-07-09T02:36:50.813461+00:00 |
QZtyICkY_73#beginner | beginner | endgame | QZtyICkY_73 | claude | 20 | {
"A": "q3_next80b",
"B": "mistral3",
"C": "gemini",
"D": "base_4b",
"E": "gpt",
"F": "ours",
"G": "ours_4b",
"H": "claude",
"I": "q3_32b",
"J": "pbase_4b",
"K": "glm5",
"L": "ours_v4",
"M": "ours_v5",
"N": "llama33_70b",
"O": "dsr1",
"P": "dsv32",
"Q": "kimi25",
"R": "base",
"S": ... | {
"pbase_4b": {
"move": 8,
"instr": 5
},
"ours": {
"move": 8,
"instr": 5
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 5
},
"dsr1": {
"move": 8,
"instr": 6
},
"ours_v3": {
"mov... | Qe8+ responses emphasizing defend-with-tempo scored best; L invented a c6 target and P fabricated rooks. | 4,874 | 418 | 2026-07-09T02:36:53.683404+00:00 |
QZtyICkY_73#intermediate | intermediate | endgame | QZtyICkY_73 | claude | 20 | {
"A": "llama33_70b",
"B": "ours",
"C": "kimi25",
"D": "q3_next80b",
"E": "glm5",
"F": "dsr1",
"G": "ours_v5",
"H": "gpt",
"I": "gemma3_27b",
"J": "claude",
"K": "pbase_4b",
"L": "ours_v3",
"M": "base",
"N": "dsv32",
"O": "ours_v4",
"P": "q3_32b",
"Q": "base_4b",
"R": "gemini",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 6
},
"dsr1": {
"move": 8,
"instr": 5
},
"ours_v3": {
"mov... | Most recommend sound Qe8+ with the e4-defense theme; some fabricate false claims (Qd6+ hanging e4, escape differences) lowering instr. | 5,484 | 432 | 2026-07-09T02:36:56.469990+00:00 |
NxmDDPKw_97#advanced | advanced | endgame | NxmDDPKw_97 | gemini | 20 | {
"A": "base_4b",
"B": "gemma3_27b",
"C": "q3_32b",
"D": "ours_v4",
"E": "kimi25",
"F": "q3_next80b",
"G": "dsr1",
"H": "dsv32",
"I": "ours",
"J": "gemini",
"K": "glm5",
"L": "ours_4b",
"M": "mistral3",
"N": "claude",
"O": "base",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 10,
"instr": 6
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 5
},
"dsr1": {
"move": 10,
"instr": 4
},
"ours_v3": {
... | Evaluated responses strictly against verified facts, penalizing hallucinations of pieces or variations. | 3,898 | 1,039 | 2026-07-09T02:36:57.837652+00:00 |
NxmDDPKw_97#intermediate | intermediate | endgame | NxmDDPKw_97 | gpt | 20 | {
"A": "base",
"B": "gemini",
"C": "llama33_70b",
"D": "ours",
"E": "ours_v5",
"F": "ours_4b",
"G": "dsv32",
"H": "gpt",
"I": "mistral3",
"J": "glm5",
"K": "kimi25",
"L": "gemma3_27b",
"M": "ours_v4",
"N": "pbase_4b",
"O": "q3_next80b",
"P": "dsr1",
"Q": "q3_32b",
"R": "ours_v3",
"... | {
"pbase_4b": {
"move": 10,
"instr": 6
},
"ours": {
"move": 10,
"instr": 10
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Most recommendations were sound, but several explanations lost credit for inventing threats, pieces, or continuations not supported by the position. | 3,488 | 1,032 | 2026-07-09T02:36:59.128103+00:00 |
QZtyICkY_73#beginner | beginner | endgame | QZtyICkY_73 | gemini | 20 | {
"A": "q3_next80b",
"B": "mistral3",
"C": "gemini",
"D": "base_4b",
"E": "gpt",
"F": "ours",
"G": "ours_4b",
"H": "claude",
"I": "q3_32b",
"J": "pbase_4b",
"K": "glm5",
"L": "ours_v4",
"M": "ours_v5",
"N": "llama33_70b",
"O": "dsr1",
"P": "dsv32",
"Q": "kimi25",
"R": "base",
"S": ... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 10,
"instr": 6
},
"q3_32b": {
"move": 10,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 4
},
"dsr1": {
"move": 10,
"instr": 6
},
"ours_v3": {
... | Grades prioritize adherence to verified facts and beginner-friendly instruction. | 3,476 | 762 | 2026-07-09T02:36:59.536686+00:00 |
NxmDDPKw_97#advanced | advanced | endgame | NxmDDPKw_97 | gpt | 20 | {
"A": "base_4b",
"B": "gemma3_27b",
"C": "q3_32b",
"D": "ours_v4",
"E": "kimi25",
"F": "q3_next80b",
"G": "dsr1",
"H": "dsv32",
"I": "ours",
"J": "gemini",
"K": "glm5",
"L": "ours_4b",
"M": "mistral3",
"N": "claude",
"O": "base",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "llama33_70b",
... | {
"pbase_4b": {
"move": 9,
"instr": 4
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 4
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 5
},
"dsr1": {
"move": 9,
"instr": 5
},
"ours_v3": {
"mov... | Sound move choices were common, but instruction was penalized for fabricated threats, illegal follow-ups, or contradicting that Ke6 is playable. | 3,761 | 965 | 2026-07-09T02:36:59.988356+00:00 |
QZtyICkY_73#advanced | advanced | endgame | QZtyICkY_73 | claude | 20 | {
"A": "gemma3_27b",
"B": "ours",
"C": "ours_v5",
"D": "kimi25",
"E": "llama33_70b",
"F": "mistral3",
"G": "gpt",
"H": "claude",
"I": "pbase_4b",
"J": "q3_next80b",
"K": "base",
"L": "ours_v3",
"M": "base_4b",
"N": "ours_v4",
"O": "ours_4b",
"P": "dsr1",
"Q": "q3_32b",
"R": "dsv32",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 6
},
"ours_v5": {
"move": 8,
"instr": 5
},
"dsr1": {
"move": 8,
"instr": 6
},
"ours_v3": {
"mov... | Qe8+ answers dominate; N and K contain fabricated squares/pieces; T/H falsely claim up material. | 5,619 | 416 | 2026-07-09T02:37:03.393467+00:00 |
QZtyICkY_73#intermediate | intermediate | endgame | QZtyICkY_73 | gemini | 20 | {
"A": "llama33_70b",
"B": "ours",
"C": "kimi25",
"D": "q3_next80b",
"E": "glm5",
"F": "dsr1",
"G": "ours_v5",
"H": "gpt",
"I": "gemma3_27b",
"J": "claude",
"K": "pbase_4b",
"L": "ours_v3",
"M": "base",
"N": "dsv32",
"O": "ours_v4",
"P": "q3_32b",
"Q": "base_4b",
"R": "gemini",
"S"... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 10,
"instr": 5
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 10,
"instr": 7
},
"ours_v3": {
... | Evaluated responses based on move soundness and instructive value without hallucinating. | 3,875 | 841 | 2026-07-09T02:37:03.667448+00:00 |
QZtyICkY_73#intermediate | intermediate | endgame | QZtyICkY_73 | gpt | 20 | {
"A": "llama33_70b",
"B": "ours",
"C": "kimi25",
"D": "q3_next80b",
"E": "glm5",
"F": "dsr1",
"G": "ours_v5",
"H": "gpt",
"I": "gemma3_27b",
"J": "claude",
"K": "pbase_4b",
"L": "ours_v3",
"M": "base",
"N": "dsv32",
"O": "ours_v4",
"P": "q3_32b",
"Q": "base_4b",
"R": "gemini",
"S"... | {
"pbase_4b": {
"move": 9,
"instr": 5
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 6
},
"mistral3": {
"move": 10,
"instr": 9
},
"ours_v5": {
"move": 9,
"instr": 5
},
"dsr1": {
"move": 8,
"instr": 5
},
"ours_v3": {
"m... | Best coaching clearly compared the two sound queen checks by their extra defensive job without inventing unsupported continuations. | 3,771 | 859 | 2026-07-09T02:37:05.114271+00:00 |
QZtyICkY_73#beginner | beginner | endgame | QZtyICkY_73 | gpt | 20 | {
"A": "q3_next80b",
"B": "mistral3",
"C": "gemini",
"D": "base_4b",
"E": "gpt",
"F": "ours",
"G": "ours_4b",
"H": "claude",
"I": "q3_32b",
"J": "pbase_4b",
"K": "glm5",
"L": "ours_v4",
"M": "ours_v5",
"N": "llama33_70b",
"O": "dsr1",
"P": "dsv32",
"Q": "kimi25",
"R": "base",
"S": ... | {
"pbase_4b": {
"move": 10,
"instr": 7
},
"ours": {
"move": 8,
"instr": 2
},
"q3_32b": {
"move": 10,
"instr": 4
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 9,
"instr": 7
},
"dsr1": {
"move": 8,
"instr": 6
},
"ours_v3": {
"... | Best responses recommend a verified sound move and explain the simple multi-purpose idea without inventing tactics or pieces. | 3,356 | 987 | 2026-07-09T02:37:05.224867+00:00 |
HF8jqo1b_71#beginner | beginner | endgame | HF8jqo1b_71 | claude | 20 | {
"A": "q3_32b",
"B": "ours_4b",
"C": "base_4b",
"D": "llama33_70b",
"E": "dsv32",
"F": "claude",
"G": "ours",
"H": "q3_next80b",
"I": "mistral3",
"J": "base",
"K": "ours_v4",
"L": "glm5",
"M": "dsr1",
"N": "kimi25",
"O": "ours_v5",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "gpt",
"S":... | {
"pbase_4b": {
"move": 7,
"instr": 6
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 9
},
"mistral3": {
"move": 9,
"instr": 9
},
"ours_v5": {
"move": 5,
"instr": 5
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Most correctly recommend Nd6 with sound double-purpose explanation; J,O,Q weaker or slightly off, K has a d7 typo, N invents a c8 threat. | 5,121 | 436 | 2026-07-09T02:37:05.773985+00:00 |
HF8jqo1b_71#beginner | beginner | endgame | HF8jqo1b_71 | gemini | 20 | {
"A": "q3_32b",
"B": "ours_4b",
"C": "base_4b",
"D": "llama33_70b",
"E": "dsv32",
"F": "claude",
"G": "ours",
"H": "q3_next80b",
"I": "mistral3",
"J": "base",
"K": "ours_v4",
"L": "glm5",
"M": "dsr1",
"N": "kimi25",
"O": "ours_v5",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "gpt",
"S":... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 10,
"instr": 5
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
... | Evaluation of responses based on verified sound moves and instructive value for beginners. | 3,667 | 341 | 2026-07-09T02:37:07.068895+00:00 |
HF8jqo1b_71#intermediate | intermediate | endgame | HF8jqo1b_71 | claude | 20 | {
"A": "q3_next80b",
"B": "base_4b",
"C": "llama33_70b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "gemini",
"G": "ours",
"H": "pbase_4b",
"I": "q3_32b",
"J": "dsv32",
"K": "claude",
"L": "mistral3",
"M": "kimi25",
"N": "dsr1",
"O": "gpt",
"P": "ours_v5",
"Q": "base",
"R": "ours_4b",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 8
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 6
},
"ours_v3": {
"mov... | Most recommend sound Nd6 with good tier coaching; N and R invent shaky tactical lines, B overstates 'long-range piece' and Q's justification is muddled. | 5,054 | 432 | 2026-07-09T02:37:10.952520+00:00 |
QZtyICkY_73#advanced | advanced | endgame | QZtyICkY_73 | gemini | 20 | {
"A": "gemma3_27b",
"B": "ours",
"C": "ours_v5",
"D": "kimi25",
"E": "llama33_70b",
"F": "mistral3",
"G": "gpt",
"H": "claude",
"I": "pbase_4b",
"J": "q3_next80b",
"K": "base",
"L": "ours_v3",
"M": "base_4b",
"N": "ours_v4",
"O": "ours_4b",
"P": "dsr1",
"Q": "q3_32b",
"R": "dsv32",
... | {
"pbase_4b": {
"move": 10,
"instr": 5
},
"ours": {
"move": 10,
"instr": 9
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 7
},
"ours_v5": {
"move": 10,
"instr": 3
},
"dsr1": {
"move": 10,
"instr": 6
},
"ours_v3": {
... | Evaluated based on adherence to facts and instructional value for an advanced player. | 3,920 | 1,129 | 2026-07-09T02:37:11.358483+00:00 |
HF8jqo1b_71#advanced | advanced | endgame | HF8jqo1b_71 | claude | 20 | {
"A": "ours",
"B": "ours_v3",
"C": "claude",
"D": "q3_32b",
"E": "base",
"F": "ours_4b",
"G": "gemini",
"H": "base_4b",
"I": "dsr1",
"J": "kimi25",
"K": "ours_v4",
"L": "gemma3_27b",
"M": "llama33_70b",
"N": "ours_v5",
"O": "pbase_4b",
"P": "q3_next80b",
"Q": "glm5",
"R": "dsv32",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 9,
"instr": 9
},
"q3_32b": {
"move": 9,
"instr": 7
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 9,
"instr": 8
},
"dsr1": {
"move": 9,
"instr": 7
},
"ours_v3": {
"mov... | Most recommend sound Nd6 well; E is confused and wrong, R/T misstate material and outpost claims. | 5,796 | 412 | 2026-07-09T02:37:12.277343+00:00 |
HF8jqo1b_71#intermediate | intermediate | endgame | HF8jqo1b_71 | gemini | 20 | {
"A": "q3_next80b",
"B": "base_4b",
"C": "llama33_70b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "gemini",
"G": "ours",
"H": "pbase_4b",
"I": "q3_32b",
"J": "dsv32",
"K": "claude",
"L": "mistral3",
"M": "kimi25",
"N": "dsr1",
"O": "gpt",
"P": "ours_v5",
"Q": "base",
"R": "ours_4b",
... | {
"pbase_4b": {
"move": 10,
"instr": 6
},
"ours": {
"move": 10,
"instr": 3
},
"q3_32b": {
"move": 10,
"instr": 9
},
"mistral3": {
"move": 10,
"instr": 9
},
"ours_v5": {
"move": 10,
"instr": 10
},
"dsr1": {
"move": 10,
"instr": 3
},
"ours_v3": {
... | Responses were penalized for hallucinating geometry, such as claiming a4 defends c4 or creating impossible forks. | 3,543 | 1,025 | 2026-07-09T02:37:17.311990+00:00 |
HF8jqo1b_71#intermediate | intermediate | endgame | HF8jqo1b_71 | gpt | 20 | {
"A": "q3_next80b",
"B": "base_4b",
"C": "llama33_70b",
"D": "gemma3_27b",
"E": "ours_v4",
"F": "gemini",
"G": "ours",
"H": "pbase_4b",
"I": "q3_32b",
"J": "dsv32",
"K": "claude",
"L": "mistral3",
"M": "kimi25",
"N": "dsr1",
"O": "gpt",
"P": "ours_v5",
"Q": "base",
"R": "ours_4b",
... | {
"pbase_4b": {
"move": 9,
"instr": 6
},
"ours": {
"move": 7,
"instr": 6
},
"q3_32b": {
"move": 10,
"instr": 8
},
"mistral3": {
"move": 8,
"instr": 6
},
"ours_v5": {
"move": 10,
"instr": 9
},
"dsr1": {
"move": 7,
"instr": 5
},
"ours_v3": {
"m... | Most Nd6 recommendations are sound, with penalties for invented threats, illegal/unsupported plans, or misstating what a4/c5 accomplishes. | 3,366 | 725 | 2026-07-09T02:37:17.899861+00:00 |
QZtyICkY_73#advanced | advanced | endgame | QZtyICkY_73 | gpt | 20 | {
"A": "gemma3_27b",
"B": "ours",
"C": "ours_v5",
"D": "kimi25",
"E": "llama33_70b",
"F": "mistral3",
"G": "gpt",
"H": "claude",
"I": "pbase_4b",
"J": "q3_next80b",
"K": "base",
"L": "ours_v3",
"M": "base_4b",
"N": "ours_v4",
"O": "ours_4b",
"P": "dsr1",
"Q": "q3_32b",
"R": "dsv32",
... | {
"pbase_4b": {
"move": 10,
"instr": 4
},
"ours": {
"move": 10,
"instr": 8
},
"q3_32b": {
"move": 10,
"instr": 7
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 6
},
"dsr1": {
"move": 10,
"instr": 4
},
"ours_v3": {
... | All recommended moves are sound, but instruction scores vary sharply due to unsupported or incorrect concrete claims. | 3,788 | 1,116 | 2026-07-09T02:37:17.983464+00:00 |
EYrWAPbF_139#beginner | beginner | endgame | EYrWAPbF_139 | claude | 20 | {
"A": "dsv32",
"B": "ours_v3",
"C": "gemini",
"D": "mistral3",
"E": "ours_v4",
"F": "ours",
"G": "llama33_70b",
"H": "glm5",
"I": "ours_v5",
"J": "q3_next80b",
"K": "gpt",
"L": "gemma3_27b",
"M": "q3_32b",
"N": "kimi25",
"O": "claude",
"P": "base",
"Q": "pbase_4b",
"R": "ours_4b",
... | {
"pbase_4b": {
"move": 9,
"instr": 4
},
"ours": {
"move": 9,
"instr": 6
},
"q3_32b": {
"move": 9,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 6
},
"ours_v5": {
"move": 2,
"instr": 3
},
"dsr1": {
"move": 9,
"instr": 6
},
"ours_v3": {
"mov... | Be6 is sound; deductions for fabricated forks/Ba5 threats (N,P,Q) and vague or wrong claims (I,F center). | 4,845 | 427 | 2026-07-09T02:37:18.041407+00:00 |
HF8jqo1b_71#beginner | beginner | endgame | HF8jqo1b_71 | gpt | 20 | {
"A": "q3_32b",
"B": "ours_4b",
"C": "base_4b",
"D": "llama33_70b",
"E": "dsv32",
"F": "claude",
"G": "ours",
"H": "q3_next80b",
"I": "mistral3",
"J": "base",
"K": "ours_v4",
"L": "glm5",
"M": "dsr1",
"N": "kimi25",
"O": "ours_v5",
"P": "pbase_4b",
"Q": "ours_v3",
"R": "gpt",
"S":... | {
"pbase_4b": {
"move": 7,
"instr": 4
},
"ours": {
"move": 7,
"instr": 5
},
"q3_32b": {
"move": 10,
"instr": 9
},
"mistral3": {
"move": 10,
"instr": 8
},
"ours_v5": {
"move": 8,
"instr": 5
},
"dsr1": {
"move": 10,
"instr": 8
},
"ours_v3": {
"... | Best responses recommended a sound simple move and explained Nd6's two jobs without inventing extra tactics. | 3,481 | 1,236 | 2026-07-09T02:37:20.753650+00:00 |
EYrWAPbF_139#intermediate | intermediate | endgame | EYrWAPbF_139 | gemini | 20 | {
"A": "mistral3",
"B": "ours_v5",
"C": "kimi25",
"D": "gemini",
"E": "glm5",
"F": "ours_v4",
"G": "gpt",
"H": "q3_32b",
"I": "gemma3_27b",
"J": "ours",
"K": "ours_4b",
"L": "base_4b",
"M": "q3_next80b",
"N": "pbase_4b",
"O": "dsv32",
"P": "ours_v3",
"Q": "base",
"R": "dsr1",
"S": ... | {
"pbase_4b": {
"move": 10,
"instr": 3
},
"ours": {
"move": 10,
"instr": 0
},
"q3_32b": {
"move": 10,
"instr": 2
},
"mistral3": {
"move": 10,
"instr": 5
},
"ours_v5": {
"move": 10,
"instr": 7
},
"dsr1": {
"move": 10,
"instr": 6
},
"ours_v3": {
... | Evaluated based on soundness of move and lack of hallucinations in instruction. | 3,698 | 342 | 2026-07-09T02:37:21.723970+00:00 |
EYrWAPbF_139#intermediate | intermediate | endgame | EYrWAPbF_139 | claude | 20 | {
"A": "mistral3",
"B": "ours_v5",
"C": "kimi25",
"D": "gemini",
"E": "glm5",
"F": "ours_v4",
"G": "gpt",
"H": "q3_32b",
"I": "gemma3_27b",
"J": "ours",
"K": "ours_4b",
"L": "base_4b",
"M": "q3_next80b",
"N": "pbase_4b",
"O": "dsv32",
"P": "ours_v3",
"Q": "base",
"R": "dsr1",
"S": ... | {
"pbase_4b": {
"move": 8,
"instr": 6
},
"ours": {
"move": 7,
"instr": 4
},
"q3_32b": {
"move": 8,
"instr": 5
},
"mistral3": {
"move": 9,
"instr": 7
},
"ours_v5": {
"move": 9,
"instr": 9
},
"dsr1": {
"move": 9,
"instr": 8
},
"ours_v3": {
"mov... | Q fabricates queens/knights/forks; J misstates f4 attacking e5; K/L invent Ba5-plan narratives; Kd1/Be6 responses are soundest. | 5,421 | 435 | 2026-07-09T02:37:24.044334+00:00 |
End of preview. Expand in Data Studio
Chess Coach — Grand Eval (comprehensive leaderboard)
One fresh, apples-to-apples comparison of every model in the chess move-review coaching project — our tuned specialists, the untuned baselines, and the full frontier lineup — on the same held-out validation slice (120 positions × 3 tiers = 360 scenarios), scored with two independent layers:
- Deterministic moat metrics (free,
python-chessover pre-computed Stockfish/Maia facts): tier-fit, distinct-moves-per-level, move-soundness, raw faithfulness (verify-pass on draft 1), tier-coherence, and shipped-gate soundness. - Blinded cross-family frontier council (GPT-5.5 + Claude Opus 4.8 + Gemini 3.1 Pro via the TrueFoundry gateway), grading each anonymised response 0–10 on move and instructiveness, with 95 % cluster-bootstrap CIs. Council: 225 items × 3 judges = 675 gradings.
Every gateway (TFY) model was regenerated fresh on these exact positions; our Modal/MLX tuned models are deterministic given their adapter (reused where noted — see the "How each row was generated" table below).
Files
| File | What |
|---|---|
GRAND_EVAL_LEADERBOARD.md |
the human-readable leaderboard (rendered below) |
report.json |
every metric per model + per-tuned-model moat proof |
council.jsonl |
raw blinded council gradings (0–10 move + instr, per judge, with token usage) |
gen/<model>.jsonl |
each model's coaching generations on the val slice |
val_scenarios.jsonl |
the held-out positions (engine-grounded, with sound pools) |
One fresh, apples-to-apples comparison of all 20 models — our tuned specialists, the untuned baselines, and the full frontier lineup — on the SAME held-out VAL slice, scored with BOTH layers:
- Deterministic moat metrics (free; python-chess over pre-computed Stockfish/Maia facts) over all 120 positions × 3 tiers = 360 scenarios: tier-fit, distinct-moves-per-level, move-soundness, raw faithfulness (verify-pass on draft 1), tier-coherence, shipped-gate soundness.
- Blinded cross-family frontier council (GPT-5.5 + Claude Opus 4.8 + Gemini 3.1 Pro via TrueFoundry), 0-10 move + instructiveness with 95% CIs, over 75 of the 120 positions (675 gradings) — sized to the TFY budget.
Every TFY gateway model was regenerated FRESH on these exact positions (never reusing the old frontier gens); our Modal/MLX tuned models are deterministic given their adapter (reused where noted). ours_v5 is the finish-v5 controller's fresh Modal Volume gen.
Frontier reachability: the 14-model lineup = 3 frontier APIs + 11 open candidates; 12 reachable (dsr1 via bedrock-oss-group/deepseek-r1), 2 blocked: llama4-maverick (400, Meta Llama access denied) and kimi-k2-thinking (403, not authorized).
Metric framing (2026-07-09 honest reframe)
Numbers below are the as-computed grand-eval values (canonical/frozen); the framing terms are aligned to the honest reframe:
- "tier-fit" = "tier-policy exact match" — exact agreement with the preregistered
select_tier_moverule, a PROJECT RULE, not validated pedagogy. Lead with the all-scenario number (v4 0.767 vs best frontier 0.553), not a head-to-head win rate. - The "moat" head-to-head is a project-rule metric, not a general win rate. The per-tuned W/L/T table below is SELECTION-CONDITIONED (only the positions where OURS already gives a distinct, sound, correctly-graded move AND diverges from the frontier). For v4, the honest unbiased head-to-head over all 92 diverging positions is 56-24-12 (56-24-40 over all 120), recomputed from the committed raw/greedy gens and asserted by
scripts/reproduce_v4.py(it supersedes an earlier eval-audit figure that did not reproduce — same 56 wins and 12 ties, but four audit-only losses over four extra diverging positions that did not replay); the 51-5-6 over 62 shown below is the conditioned subset. - distinct-moves denominator: the
distinct↑column is distinct / (positions where the model named both tier moves) as computed at grand-eval time (v4 = 73/93 = 0.785). The honest all-opportunities denominator (every canonical beginner!=advanced position, a no-answer counting as a miss) gives v4 73/100 = 0.730; seedata/benchmark_honest/report_v4.json.
Leaderboard — ranked by tier-appropriate move selection (the trained behavior)
Sort key: ranked by tier-appropriate move selection — the deterministic tier-fit↑ metric (the behavior we trained and the graded axis), ties broken by distinct-moves-per-level↑ then move-soundness↑. Every deterministic axis (tier-fit / distinct / move-sound / coherence and the moat) uses the canonical STRICT any-legal move extractor (coach_gate.pick_recommendation, accept = any legal move; no in-pool backfill) — so an output that names no clearly-legal move is a miss everywhere and the leaderboard method matches the moat method exactly. The per-tuned head-to-head W/L/T vs the best frontier is in the moat table below. Instructiveness (the blinded cross-family council) is shown as a secondary axis in the instr 0-10 / move 0-10 / rank↓ columns — OURS-v4 is intentionally weaker on council prose and that is reported here honestly and unchanged. Only the row order reflects tier-fit; every model's measured numbers are the deterministic + council values.
| # | Model | family | gen | gated | tier-fit↑ | distinct↑ | move-sound↑ | raw-faith↑ | coh-viol↓ | instr 0-10↑ [95% CI] | move 0-10↑ | rank↓ | top1% |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | OURS-v4 (Qwen3-32B tuned) | ours | reuse | raw | 0.767 | 0.785 | 0.942 | 0.589 | 0.140 | 4.528 [4.168–4.875] | 7.660 | 12.66 | 9.800 |
| 2 | OURS-v2 (Qwen3-1.7B tuned) | ours | reuse | raw | 0.578 | 0.380 | 1.000 | 0.689 | 0.167 | 4.323 [3.983–4.662] | 8.050 | 13.58 | 8.000 |
| 3 | OURS-v3 (Qwen3-32B tuned) | ours | reuse | raw | 0.558 | 0.585 | 0.950 | 0.942 | 0.229 | 6.428 [6.131–6.738] | 8.540 | 7.764 | 28.90 |
| 4 | Gemini 3.1 Pro | frontier | FRESH | raw | 0.553 | 0.210 | 1.000 | 0.958 | 0.292 | 6.902 [6.721–7.080] | 9.110 | 6.838 | 8.400 |
| 5 | OURS-v5 (Qwen3-32B tuned, v5) | ours | FRESH | raw | 0.536 | 0.726 | 0.828 | 0.575 | 0.260 | 3.863 [3.530–4.188] | 7.250 | 14.37 | 3.100 |
| 6 | Claude Opus 4.8 | frontier | FRESH | raw | 0.508 | 0.200 | 1.000 | 0.944 | 0.308 | 7.062 [6.876–7.249] | 9.160 | 6.011 | 17.80 |
| 7 | GPT-5.5 | frontier | FRESH | raw | 0.494 | 0.280 | 1.000 | 0.986 | 0.342 | 7.984 [7.869–8.098] | 9.380 | 3.024 | 40.40 |
| 8 | DeepSeek-R1 (reasoning) | open | FRESH | raw | 0.436 | 0.370 | 1.000 | 0.978 | 0.300 | 6.117 [5.906–6.319] | 9.060 | 9.524 | 1.300 |
| 9 | GLM-5 | open | FRESH | raw | 0.408 | 0.270 | 1.000 | 0.906 | 0.350 | 6.875 [6.685–7.068] | 9.150 | 6.780 | 9.300 |
| 10 | DeepSeek-V3.2 | open | FRESH | raw | 0.400 | 0.280 | 0.997 | 0.950 | 0.392 | 5.859 [5.612–6.107] | 8.900 | 10.09 | 0.900 |
| 11 | OURS-4B (Qwen3-4B tuned) | ours | reuse | yes | 0.397 | 0.280 | 1.000 | — | 0.325 | 5.828 [5.608–6.040] | 8.930 | 10.37 | 3.600 |
| 12 | PROMPT-BASE-4B (Qwen3-4B engineered) | base | reuse | yes | 0.378 | 0.460 | 1.000 | — | 0.333 | 4.799 [4.592–5.011] | 8.810 | 13.52 | 0.400 |
| 13 | BASE (Qwen3-1.7B untuned) | base | reuse | raw | 0.358 | 0.280 | 0.992 | 0.858 | 0.333 | 1.939 [1.779–2.107] | 6.950 | 18.71 | 0.000 |
| 14 | BASE (Qwen3-32B untuned) | base | FRESH | raw | 0.358 | 0.250 | 1.000 | 0.939 | 0.442 | 5.493 [5.286–5.689] | 8.950 | 11.63 | 0.400 |
| 15 | Llama-3.3-70B | open | FRESH | raw | 0.356 | 0.160 | 1.000 | 0.997 | 0.417 | 6.262 [6.095–6.422] | 9.160 | 9.327 | 1.300 |
| 16 | BASE-4B (Qwen3-4B untuned) | base | reuse | yes | 0.353 | 0.230 | 1.000 | — | 0.375 | 4.723 [4.526–4.927] | 8.750 | 13.94 | 0.000 |
| 17 | Mistral-Large-3 (675B) | open | FRESH | raw | 0.336 | 0.370 | 0.997 | 0.919 | 0.403 | 5.217 [4.982–5.436] | 8.770 | 12.14 | 0.900 |
| 18 | Kimi-K2.5 | open | FRESH | raw | 0.333 | 0.410 | 1.000 | 0.875 | 0.500 | 6.266 [6.058–6.470] | 9.070 | 8.758 | 6.700 |
| 19 | Gemma-3-27B-it | open | FRESH | raw | 0.283 | 0.190 | 1.000 | 0.969 | 0.417 | 5.778 [5.553–5.981] | 9.010 | 10.62 | 0.400 |
| 20 | Qwen3-Next-80B-A3B | open | FRESH | raw | 0.281 | 0.240 | 1.000 | 0.953 | 0.333 | 5.875 [5.654–6.092] | 8.950 | 10.35 | 2.700 |
gen: FRESH = regenerated this run; reuse = deterministic adapter/MLX gen reused. gated: yes = full shipped verify-and-regenerate pipeline (4B trio); raw = ungated draft (raw-draft gate axes shown). raw-faith = verify-pass on draft 1 (1 − fabrication). tier-fit / distinct / move-sound / raw-faith / coherence are deterministic (free); instr / move 0-10 + rank are the blinded council.
The moat — each tuned model vs the best frontier (tier-fit then soundness)
On positions where OURS gives distinct, sound, correctly-graded per-tier moves AND diverges from the best-frontier move, who wins the platform's move-quality moat (the assemble.derive_wins definition)? Instructiveness (where the frontier leads) is reported separately above.
| Tuned model | distinct | distinct & diverge | W | L | T |
|---|---|---|---|---|---|
| OURS-v2 (Qwen3-1.7B tuned) | 47 | 43 | 21 | 17 | 5 |
| OURS-4B (Qwen3-4B tuned) | 26 | 23 | 5 | 14 | 4 |
| OURS-v3 (Qwen3-32B tuned) | 47 | 44 | 22 | 13 | 9 |
| OURS-v4 (Qwen3-32B tuned) | 67 | 62 | 51 | 5 | 6 |
| OURS-v5 (Qwen3-32B tuned, v5) | 43 | 40 | 26 | 6 | 8 |
Shipped-gate soundness (tuned models through the SAME verify+fallback gate)
| Tuned model | gated move-sound↑ | gated well-formed↑ | gated no-engine-speak↑ | gate fallback↓ |
|---|---|---|---|---|
| OURS-v2 (Qwen3-1.7B tuned) | 1.000 | 1.000 | 1.000 | 0.358 |
| OURS-4B (Qwen3-4B tuned) | 1.000 | 1.000 | 1.000 | 0.000 |
| OURS-v3 (Qwen3-32B tuned) | 1.000 | 1.000 | 0.969 | 0.181 |
| OURS-v4 (Qwen3-32B tuned) | 1.000 | 1.000 | 0.983 | 0.444 |
| OURS-v5 (Qwen3-32B tuned, v5) | 1.000 | 1.000 | 0.992 | 0.444 |
Once gated, tuned soundness/format hit a shared ~100% floor (zero verifier-detectable mechanical violations by construction) — a fairness floor, not a differentiator; the differentiators are tier-fit / distinct-moves / instructiveness.
Deterministic gate axes (raw draft for ungated rows; telemetry for gated 4B)
| Model | gated | no-engine-speak↑ | well-formed↑ | move-sound↑ | verify-pass draft1↑ | mean attempts | fallback↓ |
|---|---|---|---|---|---|---|---|
| OURS-v4 (Qwen3-32B tuned) | raw | 0.978 | 0.956 | 0.942 | 0.589 | — | — |
| OURS-v2 (Qwen3-1.7B tuned) | raw | 1.000 | 1.000 | 1.000 | 0.689 | — | — |
| OURS-v3 (Qwen3-32B tuned) | raw | 0.964 | 0.958 | 0.950 | 0.942 | — | — |
| Gemini 3.1 Pro | raw | 0.997 | 1.000 | 1.000 | 0.958 | — | — |
| OURS-v5 (Qwen3-32B tuned, v5) | raw | 0.978 | 0.861 | 0.828 | 0.575 | — | — |
| Claude Opus 4.8 | raw | 1.000 | 1.000 | 1.000 | 0.944 | — | — |
| GPT-5.5 | raw | 1.000 | 1.000 | 1.000 | 0.986 | — | — |
| DeepSeek-R1 (reasoning) | raw | 1.000 | 1.000 | 1.000 | 0.978 | — | — |
| GLM-5 | raw | 0.997 | 1.000 | 1.000 | 0.906 | — | — |
| DeepSeek-V3.2 | raw | 1.000 | 1.000 | 0.997 | 0.950 | — | — |
| OURS-4B (Qwen3-4B tuned) | yes | 1.000 | 1.000 | — | — | 1.194 | 0.008 |
| PROMPT-BASE-4B (Qwen3-4B engineered) | yes | 1.000 | 1.000 | — | — | 1.167 | 0.003 |
| BASE (Qwen3-1.7B untuned) | raw | 0.964 | 1.000 | 0.992 | 0.858 | — | — |
| BASE (Qwen3-32B untuned) | raw | 0.992 | 1.000 | 1.000 | 0.939 | — | — |
| Llama-3.3-70B | raw | 1.000 | 1.000 | 1.000 | 0.997 | — | — |
| BASE-4B (Qwen3-4B untuned) | yes | 1.000 | 1.000 | — | — | 1.156 | 0.000 |
| Mistral-Large-3 (675B) | raw | 1.000 | 0.997 | 0.997 | 0.919 | — | — |
| Kimi-K2.5 | raw | 1.000 | 1.000 | 1.000 | 0.875 | — | — |
| Gemma-3-27B-it | raw | 1.000 | 1.000 | 1.000 | 0.969 | — | — |
| Qwen3-Next-80B-A3B | raw | 1.000 | 1.000 | 1.000 | 0.953 | — | — |
How each row was generated
| Model | fresh/reused | method |
|---|---|---|
| OURS-v5 (Qwen3-32B tuned, v5) | FRESH | Modal-adapter FRESH (finish-v5 controller Volume gen) |
| OURS-v4 (Qwen3-32B tuned) | reused | Modal-adapter reuse (honest val, deterministic) |
| OURS-v3 (Qwen3-32B tuned) | reused | Modal-adapter reuse (gap803, deterministic) |
| OURS-v2 (Qwen3-1.7B tuned) | reused | MLX-local reuse (gap803, greedy deterministic) |
| OURS-4B (Qwen3-4B tuned) | reused | Modal reuse (honest val, gated pipeline) |
| BASE (Qwen3-32B untuned) | FRESH | TFY FRESH (aws-bedrock qwen3-32b) |
| BASE (Qwen3-1.7B untuned) | reused | MLX-local reuse (gap803, greedy deterministic) |
| BASE-4B (Qwen3-4B untuned) | reused | Modal reuse (honest val, gated pipeline) |
| PROMPT-BASE-4B (Qwen3-4B engineered) | reused | Modal reuse (honest val, gated pipeline) |
| GPT-5.5 | FRESH | TFY FRESH |
| Claude Opus 4.8 | FRESH | TFY FRESH |
| Gemini 3.1 Pro | FRESH | TFY FRESH |
| Qwen3-Next-80B-A3B | FRESH | TFY FRESH |
| Gemma-3-27B-it | FRESH | TFY FRESH |
| Llama-3.3-70B | FRESH | TFY FRESH |
| DeepSeek-V3.2 | FRESH | TFY FRESH |
| GLM-5 | FRESH | TFY FRESH |
| Mistral-Large-3 (675B) | FRESH | TFY FRESH |
| Kimi-K2.5 | FRESH | TFY FRESH |
| DeepSeek-R1 (reasoning) | FRESH | TFY FRESH |
- Downloads last month
- 80