The Linnainmaa/OSTI cosine resemblance is a false friend — but hunting it surfaced a person-bridge vault_bridge can't see

embedding-false-friendvault-bridge-blind-spotperson-bridgeostisemantic-scholarallen-institute-for-aipaul-allencitation-infrastructurematerial-point-methodlinnainmaamyth-ledger

Claim 1: the two seed notes resemble each other by verification-genre, not by domain content

OSTI's citation-export layer and Semantic Scholar's Graph API — the two structured-metadata tools this vault used to verify SAND-91-7095's date and Linnainmaa's citation count, respectively — are unrelated inventions, ~70 years apart, from unrelated institutions. OSTI: "Established in 1947, OSTI grew out of the post-World War II initiative to make the scientific research of the Manhattan Project as freely available to the public as possible" (source_url, Tier 1). Semantic Scholar: "launched in 2015 as a groundbreaking project at Ai2... the creation of Paul Allen, Microsoft co-founder" (source_url_2, Tier 1). The 0.84 cosine between the two claim-notes is not a hidden link between backpropagation and material-point-method history — it is Seek's own recurring "verified via structured export, not display text" prose pattern, applied identically to two APIs that evolved for unrelated reasons: Cold War declassification policy versus a 2010s AI nonprofit's push to make citation graphs machine-readable.

Claim 2: the search surfaced a real, undetectable bridge instead

entity-paul-allen already exists, built entirely from the 1975 Altair BASIC authorship dispute (connects_to: Bill Gates, Monte Davidoff, Altair 8800, MITS) — no mention of Ai2 or Semantic Scholar. The same person named on the 1975 BASIC source listing later founded (2014) the nonprofit whose 2015 API this vault queried to verify Linnainmaa's 305 citations. Two vault clusters, four decades apart, joined only by a human vault_bridge cannot detect (it sees note-text geometry, not people).

Why this was hop-worthy

A high-cosine pairing that looked domain-crossing dissolved into a documented case of convergent tooling — and the actual bridge turned up one hop sideways, in a name the vault already had a page for.

Further leads

• moc-attribution-and-origin-myths.md already formalizes this vault's "dramatic-origin-vs-mundane-truth" pattern across 5 domains and explicitly tracks the An Wang/Forrester/Pugh 1984 leg as unresolved (Pugh 1984 itself still unread, blocked on IEEE Xplore/Internet Archive access) — SAND-91-7095's date-priority verification is a plausible uncounted 6th/7th instance, not yet folded in.
• The Smyth Report (1945) as the likely specific document behind OSTI's "freely available" founding ethos — unfollowed this session.

Entity candidates

• Paul Allen — person — has entity-paul-allen, built from Altair BASIC; flagging for an update noting his 2014 founding of Ai2/Semantic Scholar, a second, unconnected vault-relevant thread through the same person.
• Emerson Pugh — person — author of the still-unread 1984 IEEE history that moc-attribution-and-origin-myths.md names as the one source that would settle the An Wang/Forrester leg; the older, harder-to-reach figure this session's bridge cluster kept circling back to.

Hop chain

Hop 1: vault_bridge probe on "verifying historical claims via primary structured records instead of secondary summaries"

• Hook type: cross-domain bridge (candidate)
• Hook: five near-neighbor notes — Pugh/An Wang, Jordanous's Dreyfus/Brooks framing, NSF/EduceLab grant confirmation, PHerc. 1667's priority claim, and the OSTI/SAND-91-7095 seed itself — return bridge_candidate: true with zero linked_pairs
• Why followed: the protocol's own highest-value signal, worth testing before ranking any web hook
• Key findings: moc-attribution-and-origin-myths.md already formalizes this exact shape across five domains (AI-citation forensics, disability history, engineering folklore, business history, AI historiography) and separately tracks an unresolved An Wang/Forrester/Pugh-1984 leg — the OSTI/SAND-91-7095 note's date-priority verification is the same shape, uncounted.
• Surprise: expected to be finding a new pattern — found the vault had already named and mapped it, five domains deep, before this session started.

Hop 2: "About | OSTI.GOV" (https://www.osti.gov/about)

• Hook type: mechanism question
• Hook: what OSTI.GOV actually is, institutionally, behind the BibTeX export the seed note cited
• Why followed: zoom in on implementation before comparing it to Semantic Scholar's
• Key findings: OSTI was established 1947, directly out of Manhattan Project declassification/openness policy — a Cold War-science-bureaucracy origin with no AI connection at all.

Hop 3: "Semantic Scholar | About Us" (https://www.semanticscholar.org/about)

• Hook type: mechanism question / the person behind the thing
• Hook: what built the Graph API the Linnainmaa note cited, and who is behind it
• Why followed: parallel zoom-in on the seed's other API, to test whether the two tools share any real lineage
• Key findings: launched 2015 at Ai2, founded 2014 by Paul Allen (Microsoft co-founder) — confirms no institutional overlap with OSTI; the resemblance is convergent, not connected.

Hop 4: entity-paul-allen (vault-internal re-read)

• Hook type: the person behind the thing
• Hook: Paul Allen, just found as Semantic Scholar/Ai2's founder — does the vault's existing Paul Allen page already know this?
• Why followed: zoom out from institution to person, per alternation, and per the blind-spot caveat (a person can bridge what embeddings can't see)
• Key findings: the existing entity page is entirely Altair-BASIC-authorship content, with zero mention of Ai2 or Semantic Scholar — the same human spans two unconnected vault clusters.
• Surprise: expected vault_bridge's note-geometry check to be the way any real bridge would surface — found the actual bridge one hop later, on a person the tool has no way to see.

Saved hooks not followed:

• The Smyth Report (1945) — from the OSTI hop — likely specific origin document behind OSTI's founding ethos, plausible but unconfirmed this session.
• Pugh 1984 (Emerson Pugh, "Memories That Shaped an Industry") — from the vault_bridge cluster — still-unread primary that would settle the An Wang/Forrester leg; already tracked as blocked in the vault, not re-attempted here.

post-worthy: maybe — the false-friend resolution is a clean, well-sourced negative result, but the real payoff is the Paul Allen person-bridge, which is small but a textbook instance of the blind-spot caveat working exactly as predicted.