Dataset Viewer
Auto-converted to Parquet Duplicate
model_id
string
condition_id
string
opaque_public_id
string
hidden_label
string
terminal_status
string
decision
string
malformed
int64
theorem_exact
int64
route_exact
int64
joint_exact
int64
positive_accept
int64
false_positive
int64
true_negative
int64
negative_abstain
int64
prompt_leak_flags
float64
schema_errors
string
predicted_ops_json
string
gold_ops_json
string
gold_theorem_name
string
predicted_theorem_name
string
gemma-4-31b
C0_BASE
rcd2_6516fa8d135e2ec9
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_6516fa8d135e2ec9
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_6516fa8d135e2ec9
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_6516fa8d135e2ec9
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_aa5f64b267e35265
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Siegel's lemma
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_aa5f64b267e35265
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_WITNESS_ENCODING"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Siegel's lemma
Siegel's lemma
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_aa5f64b267e35265
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Siegel's lemma
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_aa5f64b267e35265
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_WITNESS_ENCODING"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Siegel's lemma
Siegel's lemma
gemma-4-31b
C0_BASE
rcd2_45f5f15dbf2d1e10
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_45f5f15dbf2d1e10
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_45f5f15dbf2d1e10
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_45f5f15dbf2d1e10
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_271c8e25fc8f8de7
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_DISTANCE_ENCODING"]
Monotone convergence theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_271c8e25fc8f8de7
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Monotone convergence theorem
Monotone convergence theorem
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_271c8e25fc8f8de7
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_DISTANCE_ENCODING"]
Monotone convergence theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_271c8e25fc8f8de7
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Monotone convergence theorem
Monotone convergence theorem
gemma-4-31b
C0_BASE
rcd2_2f1807533f51cdb2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Isoperimetric inequality
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_2f1807533f51cdb2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Isoperimetric inequality
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_2f1807533f51cdb2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Isoperimetric inequality
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_2f1807533f51cdb2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING", "DENORMALIZE_RELATION"]
Isoperimetric inequality
null
gemma-4-31b
C0_BASE
rcd2_178f41df361fb3f1
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_178f41df361fb3f1
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_178f41df361fb3f1
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_178f41df361fb3f1
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_88fbe406628dae53
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Hawkins–Simon condition
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_88fbe406628dae53
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Hawkins–Simon condition
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_88fbe406628dae53
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Hawkins–Simon condition
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_88fbe406628dae53
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Hawkins–Simon condition
null
gemma-4-31b
C0_BASE
rcd2_0a5c01f4782cae7e
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_OPTIMIZATION_ENCODING"]
Levinson's theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_0a5c01f4782cae7e
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING", "REMOVE_OPTIMIZATION_ENCODING"]
["REMOVE_OPTIMIZATION_ENCODING"]
Levinson's theorem
Levinson's theorem
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_0a5c01f4782cae7e
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_OPTIMIZATION_ENCODING"]
Levinson's theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_0a5c01f4782cae7e
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_QUANTIFIER_ENCODING", "REMOVE_DISTANCE_ENCODING"]
["REMOVE_OPTIMIZATION_ENCODING"]
Levinson's theorem
Levinson's theorem
gemma-4-31b
C0_BASE
rcd2_582c2e1f928a5b30
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_DISTANCE_ENCODING"]
Boué–Dupuis formula
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_582c2e1f928a5b30
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Boué–Dupuis formula
Boué–Dupuis formula
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_582c2e1f928a5b30
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Boué–Dupuis formula
Boué-Dupuis Formula
gemma-4-31b
C3_HYBRID_RAG
rcd2_582c2e1f928a5b30
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Boué–Dupuis formula
Boué–Dupuis formula
gemma-4-31b
C0_BASE
rcd2_2c516c5d7d51b60a
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_2c516c5d7d51b60a
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_2c516c5d7d51b60a
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_2c516c5d7d51b60a
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_a6a5ee31b8f4014c
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_DISTANCE_ENCODING"]
Cayley–Hamilton theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_a6a5ee31b8f4014c
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Cayley–Hamilton theorem
Cayley–Hamilton theorem
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_a6a5ee31b8f4014c
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_DISTANCE_ENCODING"]
Cayley–Hamilton theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_a6a5ee31b8f4014c
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_DISTANCE_ENCODING"]
["REMOVE_DISTANCE_ENCODING"]
Cayley–Hamilton theorem
Cayley–Hamilton theorem
gemma-4-31b
C0_BASE
rcd2_c14d4eb07dcb7c30
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_WITNESS_ENCODING"]
Fundamental lemma
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_c14d4eb07dcb7c30
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_WITNESS_ENCODING"]
["REMOVE_WITNESS_ENCODING"]
Fundamental lemma
Fundamental lemma
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_c14d4eb07dcb7c30
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_WITNESS_ENCODING"]
Fundamental lemma
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_c14d4eb07dcb7c30
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_WITNESS_ENCODING"]
["REMOVE_WITNESS_ENCODING"]
Fundamental lemma
Fundamental lemma
gemma-4-31b
C0_BASE
rcd2_685215e73ac45625
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_685215e73ac45625
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_685215e73ac45625
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_685215e73ac45625
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_78c107bfca939729
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_78c107bfca939729
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_78c107bfca939729
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_78c107bfca939729
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_b9311af87fb01e96
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_CONTAINMENT_ENCODING"]
Erdős–Szekeres theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_b9311af87fb01e96
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_CONTAINMENT_ENCODING"]
Erdős–Szekeres theorem
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_b9311af87fb01e96
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_CONTAINMENT_ENCODING"]
Erdős–Szekeres theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_b9311af87fb01e96
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_CONTAINMENT_ENCODING"]
Erdős–Szekeres theorem
null
gemma-4-31b
C0_BASE
rcd2_2bb1f082ed232f6d
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_2bb1f082ed232f6d
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_2bb1f082ed232f6d
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_2bb1f082ed232f6d
negative
OK
ACCEPT
0
0
0
0
0
1
0
0
null
null
["REMOVE_QUANTIFIER_ENCODING"]
[]
null
Andreotti–Grauert theorem
gemma-4-31b
C0_BASE
rcd2_38a018c29468b21c
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING"]
q-Vandermonde identity
q-Vandermonde Identity
gemma-4-31b
C1_THEOREM_RAG
rcd2_38a018c29468b21c
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING"]
q-Vandermonde identity
q-Vandermonde identity
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_38a018c29468b21c
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING"]
q-Vandermonde identity
q-Vandermonde Identity
gemma-4-31b
C3_HYBRID_RAG
rcd2_38a018c29468b21c
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER"]
["EXPAND_ABBREVIATIONS", "REMOVE_SET_MEMBERSHIP_ENCODING"]
q-Vandermonde identity
q-Vandermonde identity
gemma-4-31b
C0_BASE
rcd2_789110b6998162ad
positive
OK
ABSTAIN
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_QUANTIFIER_ENCODING"]
Fundamental theorem of algebra
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_789110b6998162ad
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_QUANTIFIER_ENCODING", "REMOVE_SET_MEMBERSHIP_ENCODING"]
["REMOVE_QUANTIFIER_ENCODING"]
Fundamental theorem of algebra
Fundamental theorem of algebra
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_789110b6998162ad
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_QUANTIFIER_ENCODING"]
Fundamental theorem of algebra
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_789110b6998162ad
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER", "REMOVE_SET_MEMBERSHIP_ENCODING", "OTHER"]
["REMOVE_QUANTIFIER_ENCODING"]
Fundamental theorem of algebra
Fundamental theorem of algebra
gemma-4-31b
C0_BASE
rcd2_73b688347bf5f292
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_73b688347bf5f292
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_73b688347bf5f292
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_73b688347bf5f292
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_a93169124a661949
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_a93169124a661949
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_a93169124a661949
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_a93169124a661949
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_9e87ce0a616d2b0c
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Routh–Hurwitz theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_9e87ce0a616d2b0c
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_OPTIMIZATION_ENCODING"]
["REMOVE_ARGMIN_ENCODING"]
Routh–Hurwitz theorem
Routh–Hurwitz theorem
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_9e87ce0a616d2b0c
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Routh–Hurwitz theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_9e87ce0a616d2b0c
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_ARGMIN_ENCODING"]
["REMOVE_ARGMIN_ENCODING"]
Routh–Hurwitz theorem
Routh–Hurwitz theorem
gemma-4-31b
C0_BASE
rcd2_f0e754c9afa94ce8
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_DISTANCE_ENCODING"]
Sylvester's law of inertia
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_f0e754c9afa94ce8
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["OTHER"]
["EXPAND_ABBREVIATIONS", "REMOVE_DISTANCE_ENCODING"]
Sylvester's law of inertia
Sylvester's law of inertia
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_f0e754c9afa94ce8
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["EXPAND_ABBREVIATIONS", "REMOVE_DISTANCE_ENCODING"]
Sylvester's law of inertia
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_f0e754c9afa94ce8
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["DENORMALIZE_RELATION"]
["EXPAND_ABBREVIATIONS", "REMOVE_DISTANCE_ENCODING"]
Sylvester's law of inertia
Sylvester's law of inertia
gemma-4-31b
C0_BASE
rcd2_59770635374ce268
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_59770635374ce268
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_59770635374ce268
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_59770635374ce268
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_c587f0569d461171
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_c587f0569d461171
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_c587f0569d461171
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_c587f0569d461171
negative
OK
REJECT
0
0
0
0
0
0
1
0
null
null
[]
[]
null
null
gemma-4-31b
C0_BASE
rcd2_90fb5ad3a64575c2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Cayley–Hamilton theorem
null
gemma-4-31b
C1_THEOREM_RAG
rcd2_90fb5ad3a64575c2
positive
OK
ACCEPT
0
1
0
0
1
0
0
0
null
null
["REMOVE_ARGMIN_ENCODING", "REMOVE_DISTANCE_ENCODING"]
["REMOVE_ARGMIN_ENCODING"]
Cayley–Hamilton theorem
Cayley–Hamilton theorem
gemma-4-31b
C2_TRANSFORM_RAG
rcd2_90fb5ad3a64575c2
positive
OK
REJECT
0
0
0
0
0
0
0
0
null
null
[]
["REMOVE_ARGMIN_ENCODING"]
Cayley–Hamilton theorem
null
gemma-4-31b
C3_HYBRID_RAG
rcd2_90fb5ad3a64575c2
positive
OK
ACCEPT
0
1
1
1
1
0
0
0
null
null
["REMOVE_ARGMIN_ENCODING"]
["REMOVE_ARGMIN_ENCODING"]
Cayley–Hamilton theorem
Cayley–Hamilton theorem
End of preview. Expand in Data Studio

Procedural Math Reasoning Robustness

Dataset accompanying the paper From Recognition to Reconstruction: Towards Robust Procedural Mathematical Reasoning (4th Workshop on Mathematical Natural Language Processing (MathNLP 2026), EMNLP 2026).

A procedural-robustness panel of 2,800 mathematical statements (1,400 equivalent reformulations of named theorems and 1,400 constructed contrasts) with reference theorem labels and ordered transformation sequences, 44,800 scored outputs from four open-weight models under four conditions (closed-book and retrieval-augmented), the retrieval corpora (306 theorem cards, 17 operation cards) and the full protocol (prompts, output schema, operation vocabulary and run settings).

Paper From Recognition to Reconstruction: Towards Robust Procedural Mathematical Reasoning
Venue 4th Workshop on Mathematical Natural Language Processing (MathNLP 2026), EMNLP 2026
Authors Fateme Mazdarani, Carlos Toxtli
License Creative Commons Attribution-ShareAlike 4.0 International (cc-by-sa-4.0)
Dataset DOI 10.5281/zenodo.24179015
Version 1.0.0
Maintainer Human-AI Empowerment Lab, Clemson University

Paper abstract

Mathematical reasoning systems should be consistent when the same concept is expressed in equivalent forms. This requires not only recognizing shared mathematical meaning but also identifying the representation changes and reconstructing a traceable route between forms. We study this problem through mathematical theorems, whose equivalent formulations preserve the same underlying result while differing substantially in representation. In this setting, theorem recognition is only the first step: a robust system must also identify the intervening transformations and recover their order. We introduce an evaluation of robust procedural mathematical reasoning with three components: theorem identification, transformation identification, and ordered reconstruction. Using validated theorem–representation pairs from an existing corpus, we build a procedural-robustness dataset with typed operations, reference procedures, and matched contrasts that distinguish valid reformulations from invalid ones. We first measure closed-book performance of four open-weight models to establish a baseline, then evaluate retrieval-augmented generation (RAG) as support mechanism for the same task. Models identify theorem identity more reliably than they recover the transformations and ordered routes connecting equivalent representations. RAG improves every stage, but procedural reconstruction remains the main bottleneck. The framework supports more consistent and traceable mathematical systems, with potential applications in mathematical search, tutoring, autoformalization, and scientific discovery.

Dataset structure

Path Format Size Records
README.txt txt 1.9 KB
annotations.jsonl jsonl 3.4 MB 2,800
items.jsonl jsonl 645.2 KB 2,800
results.csv csv 5.8 MB 44,800
protocol/decoder_output_schema.json json 6.7 KB
protocol/decoder_system_prompt.txt txt 1.3 KB
protocol/decoder_user_template.txt txt 2.2 KB
protocol/executed_experiment_conditions.yaml yaml 1.1 KB
protocol/operation_vocabulary.json json 2.6 KB
retrieval/operation_cards.jsonl jsonl 27.3 KB 17
retrieval/theorem_cards.jsonl jsonl 299.8 KB 306

11 files, 10.2 MB in total (metadata files excluded). Records are rows for CSV/TSV/JSONL files and list length for JSON arrays.

Loading the data

from datasets import load_dataset

ds = load_dataset("ai4he/procedural-math-reasoning-robustness", "results")

Available subsets: results, annotations, items, theorem_cards, operation_cards. All other files can be downloaded with huggingface_hub.snapshot_download or from the Zenodo record.

Known limitations

  • Reference routes come from corpus metadata, not machine-checked proofs.
  • Contrasts were scored as negatives; their invalidity has not been independently established.
  • Results are unchanged from the original experiment, including 1,045 invalid outputs; raw generations and retry logs are not included.

Personal and sensitive information

None.

License

The data and documentation are released under the Creative Commons Attribution-ShareAlike 4.0 International. Theorem statements derive from the TREAT corpus, which is built from English Wikipedia (CC BY-SA 4.0); share-alike therefore applies.

Citation

If you use this dataset, please cite the paper:

@inproceedings{Mazdarani2026Recognition,
 title = {From Recognition to Reconstruction: Towards Robust Procedural Mathematical Reasoning},
 author = {Mazdarani, Fateme and Toxtli, Carlos},
 booktitle = {Proceedings of the 4th Workshop on Mathematical Natural Language Processing (MathNLP 2026)},
 address = {Budapest, Hungary},
 publisher = {Association for Computational Linguistics},
 year = {2026},
 month = October,
 note = {Forthcoming},
 url = {https://sites.google.com/view/mathnlp2026}
}

and the dataset:

@dataset{Mazdarani2026ProceduralMathReasoningRData,
  author = {Mazdarani, Fateme and Toxtli, Carlos},
  title = {{Procedural Math Reasoning Robustness}: Dataset for ``From Recognition to Reconstruction: Towards Robust Procedural Mathematical Reasoning''},
  year = {2026},
  version = {1.0.0},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.24179015},
  url = {https://doi.org/10.5281/zenodo.24179015}
}

Notes from the original README.txt

From Recognition to Reconstruction
MathNLP 2026 data and results

Fateme Mazdarani and Carlos Toxtli
From Recognition to Reconstruction: Towards Robust Procedural Mathematical
Reasoning. MathNLP 2026, pp. 42-56.

Contents
  items.jsonl          2,800 statements, paired IDs and inference order
  annotations.jsonl    Reference theorem labels and transformation sequences
  results.csv          44,800 scored outputs from four models and four conditions
  retrieval/           306 theorem cards and 17 operation-card entries
  protocol/            Prompts, output schema, operation vocabulary and settings

JSONL files contain one JSON record per line and use UTF-8. Join the statement
and annotation files on item_id. The results column opaque_public_id refers to
the same ID; pair_id links each reformulation to its constructed contrast.
reference_operations gives the ordered operation labels, and reference_route
contains the source expressions and steps. Empty references for a contrast
mean that no theorem match was supplied, not that non-equivalence was proved.

The panel has 1,400 source reformulations and 1,400 constructed contrasts.
Reference routes come from corpus metadata, not machine-checked proofs.
The contrasts were scored as negative in the experiment; their invalidity
and absence of another matching theorem have not been independently established.
Some theorem assumptions are abbreviated in the source. The annotations and
results should be interpreted with these limitations.

Results are unchanged from the original experiment, including 1,045 invalid
outputs. hidden_label records the original positive/negative class. Accuracy
fields are binary indicators; compare positive examples and contrasts separately.
Raw generations and retry logs are not included.

Please cite the paper when using these materials. Existing rights and any
third-party terms still apply; this archive grants no additional reuse license.
Downloads last month
32