Learn Before
QASC Question Answering Benchmark
Method Part 2: Strict-Parity Evaluation Contract for Graph-Aware (Auditable Strict-Parity Graph-RAG Paper)
Method Part 1: Adaptive Depth Gating for Prerequisite Retrieval (Auditable Strict-Parity Graph-RAG Paper)
QASC Validation: Reranking Remains Stronger Than Either Hierarchical Or Adaptive Traversal (Results) in Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
QASC Paired Delta: Adaptive vs Hierarchical Baseline = +0.5 [-0.5, +1.5]
On the QASC validation split (), the paired delta between Adaptive (heuristic) depth gating and the fixed-depth hierarchical baseline in R@ is only points, with a 95% paired-bootstrap CI of . Because the CI straddles zero, the two systems are statistically tied under strict parity, which is why the paper presents adaptive depth gating as an optional extension rather than the core result on this dataset.
0
1
Tags
Science
Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
Related
QASC Directed Science Fact Graph Reconstruction (16,444 Nodes, 25,590 Edges)
QASC Strict-Parity Result: ColBERTv2/RePlug Strongest (R@10 = 85.0 [83.4, 86.6])
QASC Generation Diagnostic: TF-IDF Multiple-Choice Scorer 76.8% (Hierarchical) vs 74.6% (Adaptive)
QASC Conclusion: Reranking Beats Hierarchical and Adaptive Graph Traversal
QASC Paired Delta: Adaptive vs Hierarchical Baseline = +0.5 [-0.5, +1.5]
ColBERTv2/RePlug Reranking Baseline for Strict-Parity Prerequisite Retrieval
Flat Dense Retrieval Baseline in Strict-Parity Prerequisite Retrieval
Truncated Personalized PageRank Baseline for Prerequisite Retrieval
QASC Paired Delta: Adaptive vs Hierarchical Baseline = +0.5 [-0.5, +1.5]
GraphRAG-Style Baseline in Strict-Parity Prerequisite Retrieval
Adaptive Ancestor/Descendant Depth Controller Formula
LectureBank-Full Paired Delta: Adaptive vs Hierarchical Baseline = +0.7 [-2.1, +3.6]
QASC Paired Delta: Adaptive vs Hierarchical Baseline = +0.5 [-0.5, +1.5]
Adaptive Depth Gating Stays Tied to Hierarchical Baseline Under Target-Disjoint Control
Level-Only vs Level+Contrast Gate Ablation (Diffusion and Quotas Fixed)
QASC Paired Delta: Adaptive vs Hierarchical Baseline = +0.5 [-0.5, +1.5]
QASC Strict-Parity Result: ColBERTv2/RePlug Strongest (R@10 = 85.0 [83.4, 86.6])