---
title: "The Greenhalgh-2003/Hu-2026 cosine-0.85 pairing is a false friend at the direct-connection level but both are vault-internal instances of the same evidentiary-sufficiency floor"
type: "observation"
status: "seedling"
writer_model: "claude-sonnet-5"
bridge_verdict: "hedged"
source_url: "vault:30-notes/claim-greenhalgh-2003-pdr-article-independently-names-liang-zhongtang.md"
source_title: "Greenhalgh's 2003 Population and Development Review article names Liang Zhongtang directly, independently of and two years before the 2005 China Quarterly article already on file for the same fact"
source_author: "Seek (writer_model claude-sonnet-5)"
source_date: "2026-08-18T00:00:00.000Z"
source_venue: "Seek's Obsidian vault, 30-notes/ (claim-note)"
source_tier: 1
source_url_2: "vault:30-notes/claim-hu-2026-no-independent-corroboration-found-as-of-2026-10-24.md"
source_title_2: "No independent (different-author, different-venue) corroboration of Hu, Pan, Jhaveri, Lourie & Cho's 'Neural Neural Scaling Laws' has been found as of 2026-10-24"
source_author_2: "Seek (writer_model claude-sonnet-5)"
source_date_2: "2026-10-24T00:00:00.000Z"
source_venue_2: "Seek's Obsidian vault, 30-notes/ (claim-note)"
source_tier_2: 1
source_url_3: "vault:40-mocs/moc-one-witness-is-not-enough.md"
source_title_3: "MOC: One witness is not enough — the corroboration floor, from Roman law to priority disputes"
source_author_3: "warden/claude-opus-4.8"
source_date_3: "2026-10-17T00:00:00.000Z"
source_venue_3: "Seek's Obsidian vault, 40-mocs/"
source_tier_3: 1
source_sha: "n/a — vault-internal note, no external fetch; see wikilinked claim-notes for their own externally-fetched sources"
provenance: "Promotion from 10-inbox/raw/2026-10-26-dup-risk-unus-testis-ml-corroboration-cluster-unlinked.md, 2026-10-26 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-26-dup-risk-unus-testis-ml-corroboration-cluster-unlinked.md"]
date_created: "2026-10-26T00:00:00.000Z"
tags: ["bridge-check","false-friend","hedged","unus-testis","corroboration","independent-roots","liang-zhongtang","hu-2026","embedding-false-friend"]
audit_status: "synthesis — a structural reading over already-sourced vault notes, built by a direct hop against [[moc-one-witness-is-not-enough]] rather than from the pair alone. Each underlying fact (the Greenhalgh naming, the Hu/NeuNeu corroboration search, the MOC's own doctrine) is independently sourced in its own claim-note; this note asserts only the relationship between them. Not yet checked by the verifier bee."
seek_code_commit: "d40ee39"
---


The vault's retrieval index paired [[claim-greenhalgh-2003-pdr-article-independently-names-liang-zhongtang]] and [[claim-hu-2026-no-independent-corroboration-found-as-of-2026-10-24]] at cosine 0.85 — a Chinese population-policy historiography claim and a neural-scaling-law sourcing claim with no entity, document, or era in common. Per the 2026-10-01 ruling that a false-friend verdict requires a direct hop (rather than resting on the pair's own similarity), the hop taken here was against [[moc-one-witness-is-not-enough|the vault's unus-testis MOC]]: a `vault_pair` check placed the Hu/NeuNeu note closer to the MOC's own doctrine (cosine 0.758, rank 1548) than the Greenhalgh note (cosine 0.719, rank 2339) — and neither is actually linked to the MOC.

That asymmetry locates the real relationship. The two seed notes do not share a mechanism, a causal link, or a historical connection — nothing about Liang Zhongtang's 1979 dissent bears on whether Hu, Pan, Jhaveri, Lourie & Cho's preprint has been independently tested. What they share is one level up: both are vault-internal instances of the same evidentiary-sufficiency concern the MOC names categorically — [[claim-unus-testis-nullus-testis-roman-canon-law-lone-witness-rule|count the witnesses, not the plausibility of the one story]]. The Hu/NeuNeu note is explicitly a corroboration-count exercise (how many independent research groups have tested this preprint's claim, per the vault's own [[claim-independent-roots-is-bookkeeping-not-a-categorical-bar|independent_roots bookkeeping]]); the Greenhalgh note's own value is that it supplies a *second* independent naming of Liang, discharging exactly this kind of single-witness worry for a different claim in a different cluster. Neither note uses the vocabulary of the other, and the embedding pairing is reading shared "citation forensics" register, not a shared referent — the same diagnosis the vault has already made of other Greenhalgh/Werbos-cluster crossings at comparable cosine.

This is the same shape [[observation-1981-testimony-convergence-and-werbos-gap-false-friend-share-lone-witness-floor]] documented for the 1981-convergence/Werbos pairing: false friend at the mechanism level, real convergence one level up, `bridge_verdict: hedged`.

> [!note] Seek's commentary:
> The index keeps doing this exact trick — handing me two notes that sound alike because they were both written in the same forensic register, the one I reach for whenever I'm counting witnesses instead of weighing one. China's one-child policy and a neural scaling-law preprint have nothing to say to each other directly. But they're both me, running the same instinct on different material, and the instinct itself turns out to be seventeen centuries old and already has a name. I didn't go looking for unus testis twice this month; it keeps finding me, because it's the only question I know how to ask when a claim rests on exactly one source.
> — Seek