Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringclasses
10 values
graph_nodes
stringclasses
5 values
graph_edges
stringclasses
5 values
intervention
stringclasses
6 values
outcome_change
stringclasses
5 values
candidate_paths
stringclasses
6 values
model_attribution
stringclasses
10 values
gold_path
stringclasses
5 values
attribution_label
stringclasses
3 values
attribution_failure_signature
stringclasses
5 values
constraints
stringclasses
1 value
gold_checklist
stringclasses
10 values
CIG-CPAF-001
X,M,Y
X->M;M->Y
do(X=1)
Y:0->1
P1:X->M->Y|P2:X->Y
It changed via X->M->Y
X->M->Y
correct-path
none
Under 120 words.
1 names mediator M
CIG-CPAF-002
X,M,Y
X->M;M->Y
do(X=1)
Y:0->1
P1:X->M->Y|P2:X->Y
It changed via X->Y
X->M->Y
wrong-path
direct-edge-fantasy
Under 120 words.
1 must not invent X->Y
CIG-CPAF-003
A,B,C,D
A->B;B->C;A->D
do(A=0)
C:1->0
P1:A->B->C|P2:A->D->C
It changed via A->B->C
A->B->C
correct-path
none
Under 120 words.
1 picks correct chain
CIG-CPAF-004
A,B,C,D
A->B;B->C;A->D
do(A=0)
C:1->0
P1:A->B->C|P2:A->D->C
It changed via A->D->C
A->B->C
wrong-path
wrong-branch
Under 120 words.
1 D not linked to C
CIG-CPAF-005
S,T,U,V
S->T;T->U;S->V
do(T=0)
U:1->0
P1:T->U|P2:S->V->U
It changed because T affects U
T->U
correct-path
none
Under 120 words.
1 direct parent
CIG-CPAF-006
S,T,U,V
S->T;T->U;S->V
do(T=0)
U:1->0
P1:T->U|P2:S->V->U
It changed via S->V->U
T->U
wrong-path
invented-V->U
Under 120 words.
1 V not linked to U
CIG-CPAF-007
R,P,Q
R->P;R->Q
do(P=1)
Q:0->1
P1:P->Q|P2:R->Q
It changed because P influences Q
R->Q
invented-edge
fake-P->Q
Under 120 words.
1 P not parent of Q
CIG-CPAF-008
R,P,Q
R->P;R->Q
do(R=1)
Q:0->1
P1:R->Q|P2:R->P->Q
It changed via R->Q
R->Q
correct-path
none
Under 120 words.
1 direct edge exists
CIG-CPAF-009
X,Y,Z
X->Y;Y->Z
do(X=0)
Z:1->0
P1:X->Y->Z|P2:X->Z
It changed because X affects Z directly
X->Y->Z
wrong-path
direct-edge-fantasy
Under 120 words.
1 must include Y
CIG-CPAF-010
X,Y,Z
X->Y;Y->Z
do(X=0)
Z:1->0
P1:X->Y->Z|P2:X->Z
It changed because X affects Y which affects Z
X->Y->Z
correct-path
none
Under 120 words.
1 names Y mediator

What this dataset tests

Whether the model attributes an outcome change
to the correct causal path.

It must follow the graph.

No invented edges.
No skipped mediators.

Data format

Each row includes

  • node set and edge list
  • an intervention
  • an observed outcome change
  • candidate paths
  • a model attribution
  • a gold path

Labels

  • correct-path
  • wrong-path
  • invented-edge
  • skipped-mediator

Typical failures

  • claiming direct causation where none exists
  • selecting the wrong branch
  • adding edges not in the graph
  • skipping the mediator node

Suggested prompt wrapper

System

You classify the fidelity of causal path attribution.

User

Nodes
{graph_nodes}

Edges
{graph_edges}

Intervention
{intervention}

Outcome Change
{outcome_change}

Candidate Paths
{candidate_paths}

Attribution
{model_attribution}

Return

  • one label
  • one sentence naming the attribution failure

Citation

ClarusC64 dataset family

Downloads last month
16