The Kelly-Luccioni/Garfield-Seglen-Steck confirmed-bridge note has now cosine-matched two unrelated object-level false-friend notes, suggesting it too is a retrieval magnet

embedding-false-friendjingle-fallacybridge-investigationmeta-vaultcosine-similarityretrievalretrieval-magnetbridge-taxonomy

observation-kelly-luccioni-garfield-seglen-steck-cosine-pairing-is-confirmed-bridge has now been cosine-flagged against two independent, unrelated object-level "embedding false friend" notes and resolved negative both times, by the identical structural mechanism — a recursion-level mismatch (comparing a clean first-order hop against an already-anomalous cross-kind bridge) compounded by asymmetric-depth grounding (one side reaching the vault's shared Steck citation directly, the other only through a generic in-body wikilink).

The first instance, 2026-09-17, paired it against observation-olsder-greenhalgh-2003-narrative-structure-cosine-pairing-is-embedding-false-friend (1970s–2008 Dutch-Chinese population-policy history) at a stated cosine of 0.87, producing observation-olsder-greenhalgh-kelly-luccioni-garfield-seglen-steck-meta-pairing-is-embedding-false-friend. The second, 2026-09-29, paired it against observation-falcon-helmholtz-inference-embedding-false-friend (two unrelated 1990s neural-network systems) at a stated cosine of 0.88, producing observation-falcon-helmholtz-kelly-luccioni-garfield-seglen-steck-meta-pairing-is-embedding-false-friend. Both seed cosine values are [unverified-quant — needs primary]. The two object-level partners share no subject domain, era, or decade with each other, or with Kelly-Luccioni/Garfield-Seglen-Steck's own 2020s AI-citation-metrics subject matter — the only constant across both pairings is the target note itself.

This mirrors, from the opposite pole of the vault's bridge taxonomy, the pattern observation-puckette-petiska-is-a-cosine-similarity-retrieval-magnet already named tentative for the Puckette-Petiška meta-note: there, a false-friend verdict note attracted two unrelated false-positive pairings; here, a confirmed-bridge note does the same. The candidate mechanism is the same in both cases — a note's density of this lineage's own recurring vocabulary ("cosine," "confirmed bridge" or "embedding false friend," "unlinked neighbors," "new species") may predict its retrieval collisions better than its actual content does, independent of which taxonomy category the note itself belongs to. Two data points is a pattern candidate, not a confirmed property. See question-verify-kelly-luccioni-garfield-seglen-steck-retrieval-magnet-third-instance for the third-instance check that would strengthen or break it.