The cosine-0.84 pairing between Greenhalgh's 2003 self-citation and Klensin & Romberg's 1989 bare TSP-manual citation is an embedding false friend

embedding-false-friendbridge-investigationcosine-similarityjingle-fallacywerbostsp-manualgreenhalghchinaone-child-policyhistory-of-mlhistoriographybipartite

Seek's semantic index flagged claim-greenhalgh-2003-article-states-it-draws-on-book-in-progress and claim-klensin-romberg-1989-sole-source-of-tsp-manual-1977-date as unlinked neighbors at cosine 0.84. The two notes sit in clusters the vault has already tested against each other on this exact crossing — Greenhalgh/Olsder's Dutch-Chinese population-policy history against Werbos's US/UK computing history — and returned false friend on both prior tests: observation-rhw-docsub-gates-jevons-olsder-greenhalgh-meta-pairing-is-embedding-false-friend and observation-olsder-greenhalgh-2003-narrative-structure-cosine-pairing-is-embedding-false-friend. This pairing compares a different, specific pair of notes than either prior test used, so it is a fresh test rather than a repeat.

At the object level, no entity, document, institution, or author connects the two source documents. Greenhalgh's 2003 Population and Development Review article names Song Jian, Huibert Kwakernaak, Liang Zhongtang, and herself; Klensin & Romberg's 1989 SSDBM chapter reconciles statistical database management with the SQL standard, and its only connection to the Greenhalgh cluster is one bare reference-list line that happens to be the vault's sole traceable source for a 1977 date on the Brode, Werbos & Dunn TSP manual — a manual, and a Werbos, absent from both Greenhalgh's article and Olsder's account. A targeted search this session found no connection between Klensin, Romberg, Brode, Werbos, Dunn, or the MIT Cambridge Project on one side and Song Jian, Kwakernaak, Greenhalgh, or Chinese population-policy cybernetics on the other. [unverified — could not confirm or deny a hidden connection after search; absence of search results is not proof of absence].

The 0.84 cosine reads more plausibly as a shared structural move than a substantive bridge: both source notes perform the vault's own citation-distance audit, quoting a source and honestly recording how close it sits to the fact it attests, per 00-meta/specs/sources.md's sourcing-floor discipline — but the two audits reach opposite evidentiary valence. Greenhalgh's own first-person sentence — "In this article and the book in progress on which it draws, I look closely at the role of population science…" — is Tier 1 authorial self-report that partially resolves a live dispute (continuity between the 2003 article and a later book, with a residual [unverified-mechanism] gap over the book's exact identity). Klensin & Romberg's bare reference-list line resolves nothing; it only relocates, without validating, the terminus of a citation chain — the vault's 1977 date for the TSP manual remains contested against Werbos's own thrice-repeated 1975 self-citation, open at question-werbos-tsp-manual-year-1975-vs-1977. One note is a confirming find, the other a gap-exposing find; the shared "citation forensics" prose register both share ("states," "no further sourcing," "sole traceable source") is exactly the surface signal an embedding model trained on prose structure would seize on, not evidence that either document informs the other.