The vault's cosine-0.75 tie between HNC's Falcon and the 1995 Helmholtz Machine is an embedding false friend — same word 'inference,' discriminative scoring vs. generative cause-modeling

Seek's semantic index flagged two 1990s neural-network notes — claim-hnc-falcon-fraud-manager-became-fico-infrastructure and claim-helmholtz-machine-named-for-unconscious-inference-theory — as unlinked neighbors at cosine 0.75, raising the question of whether a genuine conceptual bridge joined them. On examination the proximity is a false friend: both systems describe their task as "inferring a probable hidden cause," but the word points to opposite computations sitting on either side of the discriminative/generative divide.

HNC's Falcon (1992) is a discriminative engine. It scores P(fraud | transaction) directly, comparing each authorization against learned behavioral profiles for real-time fraud scoring — "transaction profiling (1992) summarizes complex transaction histories into a concise set of recursive variables to enable real-time scoring" (T.J. Horan, FICO, Tier 2). It learns a decision boundary over labeled examples. (The stronger characterization of Falcon as specifically a supervised backpropagation feedforward network is [unverified-mechanism] — the cited Tier-2 source documents profiling and real-time scoring, not the architecture — and is routed to question-verify-falcon-fraud-manager-neural-architecture.)

The Helmholtz Machine (Dayan, Hinton, Neal & Zemel, 1995) is a generative engine. It builds a model of the hidden causes and inverts it, "infer[ring] the probable causes of sensory input" through unsupervised wake-sleep learning (Tier 1, quote preserved in the linked note). It learns to generate the data from latent causes rather than to draw a boundary between fraud and not-fraud.

So the 0.75 is driven by shared 1990s-connectionism vocabulary ("inference," "probability," "network"), not a lineage — the failure mode described by claim-cosine-similarity-of-embeddings-can-be-arbitrary and already logged in the sibling observation-hawks-jeffress-cosine-pairing-is-embedding-false-friend. No institutional bridge survives either: Geoffrey Hinton was in UCSD's PDP group in 1978–82, while Robert Hecht-Nielsen (HNC's founder) held a UCSD adjunct post only from 1985 — same campus, non-overlapping years, and different vault clusters (HNC → commercialization / AI-winter; Helmholtz → moc-backpropagation-origins). No wikilink of lineage was added between the two notes.