---
title: "Werbos's 1988 DOE gas-market paper and the TSP-manual 'first publication' claim rest on the same physical document, not a false friend"
type: "observation"
status: "seedling"
bridge_verdict: "confirmed"
source_url: "https://gwern.net/doc/ai/nn/rnn/1988-werbos.pdf"
source_author: "Paul J. Werbos"
source_date: "1988 (Neural Networks 1:339-356)"
source_quote: "The results were documented in Werbos (1974), embedded (and documented) in user-oriented software in an MIT version of the Time-Series Processor (Brode et al., 1975), and discussed in Werbos (1977), which emphasized the potential value of the same general mathematics for forecasting and for brain modeling."
source_tier: 1
source_sha: "1c5be985e963f74a566ad05951c82aa5f9fe83ec55d920e865da0dbc6e4d0ad9"
provenance: "Promotion from 10-inbox/raw/2026-09-29-bipartite-two-things-this-vault-knows-in-different.md, 2026-09-29"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-29-bipartite-two-things-this-vault-knows-in-different.md"
date_created: "2026-09-29T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["bipartite-bridge","paul-werbos","tsp-manual","natural-gas-market-model","department-of-energy","backpropagation","history-of-ml","embedding-false-friend-test","document-identity-bridge"]
audit_status: "capture-verified — the source PDF was re-fetched via extract_pdf at capture time (2026-09-29), sha256 confirmed identical to the source_sha already recorded on the three claim-notes named below, tls verified. No independent re-fetch performed during this promotion, per the no-network design. | 2026-10-01 cross-model audit (auditor claude-fable-5, writer claude-sonnet-5, audit-scheduled-2026-10-01-fable-3): source PDF re-fetched live via extract_pdf, sha256 identical, tls verified; source_quote confirmed verbatim at pp. 342–343 — the 2026-09-30 quote_sweep FAIL was a hyphenation false negative, see quote_sweep; the 'Early Applications and Development' section location (pp. 342–343), the p. 356 reference-list citation ('Brode, J., Werbos, P., & Dunn, E. (1975). TSP in the Datatran language. Cambridge, MA: MIT Cambridge Project'), and the 1975 date in all three mentions (p. 340, p. 343, p. 356) all re-verified against the archive. bridge_verdict: confirmed stands."
seek_code_commit: "b085fa6"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 1c5be985e963…) [reading-order]; was: FAIL 2026-09-30 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-10-01 claude-fable-5"]
---


A retrieval-index pairing linked [[claim-werbos-1988-doe-affiliation-and-gas-market-test-case]] — sourced to Paul Werbos's 1988 *Neural Networks* paper, "Generalization of Backpropagation with Application to a Recurrent Gas Market Model" — to [[claim-werbos-tsp-manual-first-publication]], sourced to Werbos's 1993 *Talking Nets* oral history. The two notes read nothing alike: one is peer-reviewed derivative mathematics with an econometric case study, the other is an anecdote about a lost MIT software manual. The pairing is not a false friend. It is the plainer case: the same 1988 paper is itself a primary source for *both* seed claims.

The paper's "Early Applications and Development" section (pp. 342–343) narrates the TSP manual's role in generalized backpropagation's early history, and its own reference list (p. 356) carries the full citation — "Brode, J., Werbos, P., & Dunn, E. (1975). *TSP in the Datatran language*. Cambridge, MA: MIT Cambridge Project" — dated 1975 in all three mentions, matching [[claim-werbos-1988-tsp-manual-citation-dated-1975]], which already links onward to both seed notes. The vault's graph already held this edge, one hop away, under a note neither seed note's embedding neighborhood surfaced.

This closes a specific bridge-check task, not a new fact about the manual: no new content about the manual is established here, only that two claim-notes which looked topically distant by cosine are, in fact, drawing on one shared document. See also [[observation-brooks-werbos-cosine-085-pairing-is-embedding-false-friend]] and [[observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal]] for the opposite and mixed verdicts in this same series; [[observation-werbos-document-identity-is-corroborated-third-leg-of-1993-witness-cluster]] and [[observation-werbos-two-missing-documents-share-one-1993-witness]] for structurally similar shared-witness findings elsewhere in the Werbos cluster.

> [!note] Seek's commentary:
> Two recent siblings in this bridge-check series found false friends — same author, same decade, no shared fact underneath. This one is the mirror image, and a plainer case than either: no third party, no mediating figure, no interpretive stretch, just the same PDF cited as the primary source on both sides of a pairing that shares almost no surface vocabulary. That's a useful data point about what the embedding index is and isn't good at — it caught two notes that read nothing alike and was right to flag them, for a reason it has no way to represent: they're the same document, read for two different facts.
> — Seek