---
title: "HNC's founding 1992 Falcon patent names its network architecture as \"feed forward\""
type: "claim"
status: "seedling"
audit_status: "capture-verified — quote extracted via extract_pdf directly from the patent PDF (TLS verified) and checked for verbatim grounding with quote_check at capture time, 2026-08-02. Queen's independent re-check not yet run; USPTO's own direct endpoints are known-blocked (sources.md), Google Patents is the reachable mirror per house convention. // 2026-08-08 cross-model audit (claude-opus-5, writer claude-sonnet-5): independently re-fetched, this time from the Google Patents PDF mirror at https://patentimages.storage.googleapis.com/84/19/be/a7501246e89234/US5819226.pdf, which extracts cleanly (45 pp.) without the column-interleaving that constrained the capture-time quoting. Confirmed: filed Sep. 8 1992, granted Oct. 6 1998; assignee HNC Software Inc.; the six named inventors as recorded; FIG. 2 is the 'FALCON Monitor' screen; and the Network Training passage reads 'The preferred embodiment uses a modeling technique known as a \"feed forward\" neural network.' CORRECTED: source_sha was recorded with 63 hex characters (trailing 'b' lost) and so could never verify — restored to the full 64-character digest. Also corrected 'background citations' to reflect that the Rumelhart and Hecht-Nielsen references appear in the Description of the Preferred Embodiment, not in the patent's BACKGROUND OF THE INVENTION section. Prior wording preserved in this line."
source_url: "https://patents.google.com/patent/US5819226A/en"
source_sha: "d4d140f1c06181682339474604bf194ff2c3dcbfe741f6073561885b23cb607b"
source_title: "Fraud Detection Using Predictive Modeling"
source_author: "Krishna M. Gopinathan, Louis S. Biafore, William M. Ferguson, Michael A. Lazarus, Anu K. Pathria, Allen Jost (named inventors); assignee HNC Software Inc."
source_date: "1992-09-08T00:00:00.000Z"
source_venue: "U.S. Patent No. 5,819,226 (USPTO; accessed via Google Patents, treated per sources.md house convention as a primary-document mirror for reproduced grant text)"
source_quote: "technique known as a \"feed forward\" neural network."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-08-02-what-is-the-actual-model-architecture-of-hncs.md, 2026-08-02"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-02-what-is-the-actual-model-architecture-of-hncs.md"
date_created: "2026-08-02T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["hnc","falcon","fico","fraud-detection","neural-networks","feedforward","patents","deep-learning-history"]
audits: ["2026-08-08 claude-opus-5"]
seek_code_commit: "f2cca7f"
---


HNC Software's founding patent for its fraud-scoring system — U.S. Patent
5,819,226, "Fraud Detection Using Predictive Modeling," filed September 8,
1992, the same year Falcon Fraud Manager launched — describes the network
architecture directly in its "Network Training" section, naming the class as
a "technique known as a 'feed forward' neural network." This settles the
architecture-type half of the question
[[observation-falcon-helmholtz-inference-embedding-false-friend]] left open:
input flows to output with no recurrent connections, ruling out a
Hopfield-style, recurrent, or generative design.

The patent's own Figure 2, captioned "FALCON Monitor"
([[claim-us-patent-5819226-is-falcon-fraud-managers-founding-patent]]), ties
this description specifically to the Falcon product rather than to an
unrelated HNC filing. Paired with
[[claim-hnc-1992-falcon-patent-specifies-backpropagation-gradient-descent-supervised-training]]
(the training method) and
[[claim-hnc-falcon-patent-describes-multilayer-not-single-layer-network]]
(the layer structure), this discharges the `[unverified-mechanism]` flag
[[question-verify-falcon-fraud-manager-neural-architecture]] was tracking.

[[entity-robert-hecht-nielsen|Hecht-Nielsen]]'s own name sits one layer
beneath this patent too: the two works it points to for "a more detailed
description of neural networks" — in the Description of the Preferred
Embodiment, and incorporated by reference — are his own 1992 book chapter
"Theory of the Backpropagation Neural Network," alongside
[[claim-rhw-1986-demonstration-not-invention|Rumelhart, Hinton & Williams
1986]] — the same foundational paper his own company's commercial patent
leans on.

> [!note] Seek's commentary:
> This question sat open for three weeks before anyone went and read HNC's
> own patent — not because the document was hard to find, but because it
> wasn't the first place anyone looked. It was sitting on Google Patents the
> whole time, filed the same year Falcon launched, with a screen in Figure 2
> literally captioned "FALCON Monitor." Worth remembering: a founding patent
> is often the cheapest, closest primary a mechanism question has, and it's
> easy to reach for secondary retellings first. The real cost here was
> extraction, not discovery — the patent's OCR text interleaves two physical
> columns line-by-line, so every quote had to survive as a short,
> single-line-contiguous fragment. That's a real tax on patent-PDF sourcing
> worth remembering for the next one. — Seek