The cosine pairing between the Olsder-Greenhalgh 'embedding false friend' note and the Kelly-Luccioni/Garfield-Seglen-Steck 'confirmed bridge' note is itself an embedding false friend

embedding-false-friendjingle-fallacybridge-investigationmeta-vaultcosine-similaritybridge-taxonomymeta-pairing

Seek's semantic index flagged observation-olsder-greenhalgh-2003-narrative-structure-cosine-pairing-is-embedding-false-friend and observation-kelly-luccioni-garfield-seglen-steck-cosine-pairing-is-confirmed-bridge as unlinked neighbors at cosine 0.87 [unverified-quant — needs primary]. Direct comparison finds a plain false friend, on four independent tests.

Opposite verdicts, same vocabulary. Olsder-Greenhalgh concludes "no wikilink is added" — a negative verdict between its own two inputs. Kelly-Luccioni/Garfield-Seglen-Steck concludes the opposite: "direct comparison confirms a genuine bridge" between its own two inputs. The two notes are cosine-neighbors because they share the words "cosine," "pairing," and the vault's own titling template — not because their verdicts relate to each other. Pairing a "not connected" finding against a "connected" finding on vocabulary alone is exactly the surface collision claim-cosine-similarity-of-embeddings-can-be-arbitrary predicts.

Recursion-level mismatch, compounded. Olsder-Greenhalgh is a clean first-order hop comparing two raw claim-notes about the world. Kelly-Luccioni/Garfield-Seglen-Steck is not a matching first-order hop: by its own text it compares one already-synthesized meta-level verdict against one first-order claim, a cross-kind mismatch observation-petiska-and-gates-jevons-liang-olsder-bridges-are-two-species-not-one-relationship had already flagged (and there ruled unconnected), and one the bridge note itself logs as "a species the vault's taxonomy has not catalogued." Pairing a clean hop against a hop already known to be structurally anomalous stacks a third axis of mismatch onto the comparison, the same recursion-level failure mode catalogued at observation-kelly-luccioni-puckette-petiska-meta-pairing-is-embedding-false-friend and observation-rhw-docsub-gates-jevons-olsder-greenhalgh-meta-pairing-is-embedding-false-friend.

Asymmetric-depth grounding, not convergence. Kelly-Luccioni/Garfield-Seglen-Steck's bridge rests on both of its inputs grounding, at the frontmatter level, in the identical Steck, Ekanadham & Kallus paper — one input directly at its arXiv identifier (arXiv:2403.05440), the other via the vault's own claim-cosine-similarity-of-embeddings-can-be-arbitrary, which names that paper; no arXiv identifier appears anywhere in the Garfield-Seglen-Steck note's file (per the bridge note's own 2026-09-05 audit correction) — with the identical quoted clause in each input's own frontmatter. Olsder-Greenhalgh touches the same vault claim-note, claim-cosine-similarity-of-embeddings-can-be-arbitrary, but only as a single in-body wikilink offering generic mechanism scaffolding — no frontmatter citation, no arXiv identifier, no quoted clause. Treating a shared pointer to the vault's own generic mechanism-note as shared primary-document grounding is the trap already named "opposite-direction grounding" in the RHW/DOCSUB-Olsder-Greenhalgh note; here it is milder — asymmetric depth rather than outright inversion — but the same category error.

No shared object-level referent. Olsder-Greenhalgh's cluster is 1970s–2008 Dutch-Chinese population-policy history (entity-geert-jan-olsder, entity-susan-greenhalgh); Kelly-Luccioni/Garfield-Seglen-Steck's is 2020s AI-era citation metrics (entity-harald-steck). A full-text cross-check of the Olsder-Greenhalgh note for the other cluster's names (Steck, Kelly, Luccioni, Garfield, Seglen) returns zero hits; in the reverse direction "Greenhalgh" returns zero hits, and "Olsder" occurs only inside the names of two precedent notes on the separate Gates-Jevons/Liang-Olsder pairing, cited as taxonomy precedent — mechanism-level citations, not object-level content. No wikilink is added between the two source notes.