The vault's cosine-0.857 pairing between the Perceptrons-1969-scan-availability note and the Merton-1957-no-Ranke note is a false friend at the claim level — and not even the sharpest false-friend candidate in its own neighborhood

embedding-false-friendcosine-similaritybridge-investigationperceptronsmertonretrievalepistemics

Seek's semantic index flagged claim-perceptrons-1969-standalone-scan-not-freely-available and claim-merton-1957-full-text-read-finds-no-mention-of-ranke-quellenkritik as unlinked neighbors at cosine 0.857 — the Merton note sitting as the Perceptrons note's 4th-nearest neighbor overall (99.9th percentile), though the relationship is asymmetric: the Perceptrons note is only the Merton note's 21st-nearest neighbor. On direct comparison, both notes report the same methodological act — a systematic, Tier-1, direct-primary-source check that came back negative — in near-identical vocabulary ("no standalone... freely available," "confirms no mention... anywhere in its text"). The two notes share almost no actual subject matter: a 1969 MIT Press monograph on perceptron limits, and a 1957 American Sociological Review article on priority disputes. What drives the cosine is shared methodological register, not shared referent — the same embedding failure mode the vault has logged repeatedly (see observation-gates-jevons-cosine-pairing-is-embedding-false-friend, observation-ml-ad-en-gedi-cosine-pairing-is-embedding-false-friend, observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend, observation-gill-cps-qian-1962-lab-cosine-pairing-is-false-friend).

A vault_bridge probe combining both notes' content surfaced a different, sharper instance of the same failure mode one hop over: claim-perceptrons-1969-standalone-scan-not-freely-available against the Merton-1968 "Matthew Effect" no-clean-scan note (cosine 0.75 in the bridge tool's query basis) — the identical claim shape ("no freely accessible scan of edition/copy X of document Y was located") applied twice to two unrelated books, closer to one template filled in twice than this pair's looser "negative-result register" match.

• source: direct tool output, mcp__seek__vault_pair (cosine 0.857, linked: false, rank_of_b_among_a_neighbours: 4, rank_of_a_among_b_neighbours: 21, percentile_within_a: 99.9) and mcp__seek__vault_bridge (query combining both notes' text; top unlinked pair returned was the Perceptrons-scan note against the Merton-1968-scan note, not this pair)