Q-DEIM/GKS context selection for RAG: pre-registered replication protocol (v2) on new MuSiQue questions

Pre-registration of a confirmatory replication of the study "Q-DEIM/GKS context selection for retrieval-augmented generation" (protocol v4.2) on new questions from the MuSiQue training split, with an 8B reader. The protocol (replication_protocol.md, in Italian; SHA-256 d7fb6a4813865155788a7cea4fa578358613d01b2f33e9eeab79f2debee87b02) fixes hypotheses, sample, exclusions, feasibility gate, analysis and decision rules before the training data are obtained: R1 (GKS selection improves answer F1 over cosine top-m) as gatekeeper, then Holm at 2.5% one-sided on R2 (near-duplicate filter vs cosine) and R3 (GKS vs filter), plus a secondary equivalence test (±0.005 F1). HASHES.txt lists the hashes of the sealed main study to which the replication is anchored (frozen configuration, confirmatory seal, audit-log head) and states what this deposit does and does not show. The files are restricted until the publication terms of the main study's outputs are settled; the SHA-256 above allows anyone to verify them once released.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-10-14
DOI
https://doi.org/10.5281/zenodo.23357801
Primary Topic
Topic Modeling
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Q-DEIM/GKS context selection for RAG: pre-registered replication protocol (v2) on new MuSiQue questions

Vincenzo Trombetta
Zenodo (CERN European Organization for Nuclear Research)
Topic Modeling
article

Q-DEIM/GKS context selection for RAG: pre-registered replication protocol (v2) on new MuSiQue questions

Vincenzo Trombetta
article en

Abstract

Pre-registration of a confirmatory replication of the study "Q-DEIM/GKS context selection for retrieval-augmented generation" (protocol v4.2) on new questions from the MuSiQue training split, with an 8B reader. The protocol (replication_protocol.md, in Italian; SHA-256 d7fb6a4813865155788a7cea4fa578358613d01b2f33e9eeab79f2debee87b02) fixes hypotheses, sample, exclusions, feasibility gate, analysis and decision rules before the training data are obtained: R1 (GKS selection improves answer F1 over cosine top-m) as gatekeeper, then Holm at 2.5% one-sided on R2 (near-duplicate filter vs cosine) and R3 (GKS vs filter), plus a secondary equivalence test (±0.005 F1). HASHES.txt lists the hashes of the sealed main study to which the replication is anchored (frozen configuration, confirmatory seal, audit-log head) and states what this deposit does and does not show. The files are restricted until the publication terms of the main study's outputs are settled; the SHA-256 above allows anyone to verify them once released.

Zenodo (CERN European Organization for Nuclear Research)
Openalex Percentile: Top 13%
Topic Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.