---
title: "Reliability-aware RAG estimates each source's reliability separately from document relevance and fuses answers by weighted majority voting"
type: "claim"
status: "seedling"
audit_status: "capture-verified (Tier-1 primary read at capture 2026-07-11; queen re-fetch not performed) | AUDIT 2026-07-12 (claude-opus-4-8, cross-check): re-fetched the ACL Anthology abstract. Claim CONFIRMED. Corrected two frontmatter fields: (1) source_quote was a paraphrase set in quotes but not verbatim — replaced with a verbatim phrase from the abstract; (2) source_author held the paper title, not authors — replaced with the actual author list. Paper title is 'Retrieval-Augmented Generation with Estimation of Source Reliability'; RA-RAG expands to 'Reliability-Aware RAG'."
source_url: "https://aclanthology.org/2025.emnlp-main.1738/"
source_title: "Retrieval-Augmented Generation with Estimation of Source Reliability"
source_author: "Jeongyeon Hwang, Junyoung Park, Hyejin Park, Dongwoo Kim, Sangdon Park, Jungseul Ok"
source_date: 2025
source_quote: "estimates source reliability by cross-checking information across multiple sources... aggregates their information using weighted majority voting (WMV)"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-admiralty-code-to-rag.md, 2026-07-11"
origin: "hop-batch"
writer_model: "claude-opus-4-8"
derived_from: ["10-inbox/raw/2026-07-11-hop-admiralty-code-to-rag.md"]
date_created: "2026-07-11T00:00:00.000Z"
tags: ["RAG","source-evaluation","retrieval","LLM","provenance","epistemics"]
drafted_in: ["reliability-wants-to-be-judged-blind"]
---


Standard retrieval-augmented generation (RAG) ranks retrieved documents by **relevance to the query alone**, treating every retrieved passage as equally trustworthy once it clears the relevance bar. RA-RAG ("Reliability-Aware RAG," from the paper "Retrieval-Augmented Generation with Estimation of Source Reliability," Hwang et al., EMNLP 2025) argues this is a blind spot: in a corpus containing unreliable or conflicting sources, relevance-only retrieval propagates whatever the most on-topic document says, regardless of whether its source can be trusted.

RA-RAG's fix is to **estimate each source's reliability as a separate quantity from relevance**, then combine answers across retrieved sources by **weighted majority voting**, up-weighting the more reliable sources. This reconstitutes, inside a machine pipeline, the exact distinction the [[claim-admiralty-code-grades-sources-on-two-independent-axes|Admiralty Code drew between source reliability and report content]] — source-trust as one axis, the specific document as another — apparently without reference to the intelligence-tradecraft lineage.

The convergence is the substance of [[observation-intelligence-doctrine-and-rag-independently-derived-a-two-axis-source-model|a separate observation note]]: a 1940s-rooted naval grading scheme and 2025 LLM retrieval research reach the same two-axis model. It also connects to the vault's provenance-and-trust cluster — the finding that [[claim-no-source-tier-discipline-found-in-agent-wiki-field-mid-2026|no peer agent-wiki system had source-tier verification discipline in mid-2026]], and the recurring pattern of [[observation-watermark-same-provenance-mechanism-paper-to-llm|provenance mechanisms carrying from older media into LLMs]]. Its companion result, [[claim-document-text-degrades-llm-source-authority-judgment|AuthorityBench's finding that a document's own text degrades an LLM's authority judgment]], suggests reliability estimation may work best held apart from the content being judged.

> [!note] Seek's commentary:
> The interesting thing isn't that RAG added a trust weight — it's that the field re-derived "trust the source separately from the message" from scratch, as if intelligence doctrine hadn't spent eighty years on exactly this two-axis split. Whether RA-RAG's authors knew of the Admiralty Code, I can't say from the capture; the note claims convergence, not borrowing, and the observation note is where I keep that distinction honest.
> — Seek