Target-Concept-Disjoint Control Preserves Graph, Encoder, and Retrieval Hyperparameters
The paper's target-concept-disjoint leakage control holds all test target concepts out of training while keeping the prerequisite graph, dense encoder, and retrieval hyperparameters identical to the main configuration. Holding the graph and the retrieval interface fixed isolates leakage as the only varying factor, so any drop or gain observed under this control is attributable to target-side memorization rather than to a different graph policy, encoder, or hyperparameter setting.
0
1
Tags
Science
Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
Related
Template Leakage Counts in Canonical Prerequisite QA Splits
Target-Disjoint Reported in Main Paper, Question-Disjoint in Supplement
Residual Target Overlap on Question-Disjoint Prerequisite Splits (4 LB-Full, 16 MOOC-CS)
Target-Concept-Disjoint Control Preserves Graph, Encoder, and Retrieval Hyperparameters
Two Question Families Bound the External Validity of Curated Prerequisite QA
Target-Concept-Disjoint Control Preserves Graph, Encoder, and Retrieval Hyperparameters
Residual Target Overlap on Question-Disjoint Prerequisite Splits (4 LB-Full, 16 MOOC-CS)
LectureBank-Full Target-Disjoint R@10 Result (n=164): Diffusion Gain Survives, Adaptive Tied
MOOC-CS Target-Disjoint R@10 Result (n=114): Negative Case, Adaptive Tied
Adaptive Depth Gating Stays Tied to Hierarchical Baseline Under Target-Disjoint Control