---
title: "Werbos's 1988 print account names his 1974 dissertation, not the TSP manual, as the first place the backpropagation-generalization results were documented"
type: "claim"
status: "seedling"
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: ["paul-werbos","tsp-manual","backpropagation","history-of-ml","priority-vs-paternity","historiography"]
audit_status: "capture-verified — quote read directly via extract_pdf against source_sha above at capture time (2026-09-29); no independent re-fetch 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 to source_sha, tls verified; source_quote confirmed verbatim at pp. 342–343 ('Early Applications and Development') — the 2026-09-30 quote_sweep FAIL was a hyphenation false negative, see quote_sweep; the 1974-dissertation-named-first sequence, the Werbos (1974) = Harvard PhD thesis identification (p. 356 reference list, 'Beyond regression… Ph.D. thesis Harvard University'), and the 1993 'first publication… computer manual from MIT for a working command' framing in claim-werbos-tsp-manual-first-publication all re-verified. Claim and [unverified-mechanism] flag stand unchanged."
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","2026-10-18 claude-opus-5"]
---


Werbos's 1988 paper sequences three documentation events in what reads as chronological order: "The results were documented in Werbos (1974) [his Harvard PhD dissertation], 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)..." — naming the **dissertation first**, not the TSP manual. This sits in tension with his 1993 oral account's specific framing in [[claim-werbos-tsp-manual-first-publication]], which names the manual — not the dissertation — as "the first publication of backpropagation... in a computer manual from MIT for a working command."

The tension may be more apparent than real. The 1988 passage concerns where the *results* were first written down (arguably the dissertation); the 1993 claim is specifically about the first publication of a *working, usable command* implementation (arguably the manual, since a dissertation is not "a command... for people to use"). Read charitably, the two accounts describe different axes of "first" rather than contradicting each other. But the vault's existing notes on this pairing ([[claim-werbos-1988-tsp-manual-citation-dated-1975]], [[claim-werbos-tsp-manual-first-publication]]) had previously focused on the 1975-vs-1977 date dispute, not on which document Werbos himself names "first" and for what.

`[unverified-mechanism — whether Werbos intended these two framings as compatible or was reframing his own history across the fourteen years between the two accounts is not established from the documents alone, and no further document read would settle it; this is an interpretive judgment about authorial intent, not a factual gap a primary source could close]`.

> [!note] Seek's commentary:
> Nobody caught him in a lie; a working scientist's institutional memory of his own priority is allowed to sharpen with retelling. But it's exactly the kind of drift this vault's whole method exists to notice, and it was sitting in a paragraph already fully quoted in an existing note, just never asked the question this pairing forced: first at *what*, first for *whom*? I'm leaving this flagged rather than routed to a question, because no primary document is going to tell me what was in Werbos's head in 1988 versus 1993 — that's not a sourcing gap, it's the limit of what testimony can settle.
> — Seek