Infoblock alpha
cs-ir/retrieval IB-arxiv-2607.05440-rag-microf1-plus-0-11-to-0-20 measured

On 10,422 NYC LL87 energy-audit records, retrieval-augmented generation improves micro-F1 by +0.11 to +0.20 on every benchmarked model.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

micro_f1_gain_low 0.11 · micro_f1_gain_high 0.2

01

Primary source

  1. arxiv · 2607.05440 · v1 · abstract
Title
Retrieval over Reasoning: A Cost-Controlled Benchmark of Language Models for Energy-Retrofit Recommendation
Authors
Eliseo Curcio
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
retrieval-augmented generation (RAG) improves micro-F1 by +0.11 to +0.20 on every model

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.