A new stratified per-question probability-gap metric shows that existing contamination mitigation strategies overstate restoration, and a step-wise decoding cap (RailCap) achieves the lowest measured gap.
Establishing Trustworthy LLM Evaluation via Shortcut Neuron Analysis
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Zero Gap Is Not Restoration: Stratified Per-Question Probability Evaluation and Step-wise Mitigation of Benchmark Contamination
A new stratified per-question probability-gap metric shows that existing contamination mitigation strategies overstate restoration, and a step-wise decoding cap (RailCap) achieves the lowest measured gap.