ReplaySCM introduces a replay-based benchmark for executable causal mechanism induction that reveals sharp drops in LLM performance on held-out interventions when variable order or root structure is hidden, even after support audits eliminate inconsistent alternatives.
B.2.2 Intervention-world construction A world is a table of unit rows under one intervention
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ReplaySCM: A Benchmark for Executable Causal Mechanism Induction from Interventions
ReplaySCM introduces a replay-based benchmark for executable causal mechanism induction that reveals sharp drops in LLM performance on held-out interventions when variable order or root structure is hidden, even after support audits eliminate inconsistent alternatives.