--- id: "20260706-1103-did-werboss-1978-ieee" title: "Capture: Did Werbos's 1978 IEEE Transactions on Systems, Man, and Cybernetics paper omit technical appendices because DARPA blocked entry of the fuller report into the DOD public document system?" type: "capture" status: "promoted" promoted_to: ["30-notes/claim-werbos-1978-docsub-blocked.md (two-parallel-reasons nuance preserved; single-witness caveat carried; SMC biblio [unverified-quant] carried)"] not_promoted: ["Dead-ends inventory (werbos.com, BTT 1990, Wikiquote, podcast) — search-completeness record stays in capture"] promotion_note: "Queen cycle 20, 2026-07-07." origin: "batch" model: "claude-sonnet-5" date_created: "2026-07-06T00:00:00.000Z" provenance: "batch run 2026-07-06 — researched via web search/fetch, plus direct visual verification of the primary PDF; harvested from 2026-06-29-who-was-paul-werbos-and-what-did-his-1974-harvard-phd-thesis-actually-claim-about-backpropagation" derived_from: [] tags: ["werbos","backpropagation","history-of-ml","darpa","dod","oral-history","primary-source-confirmed"] source_url: "https://gwern.net/doc/ai/nn/rnn/1998-werbos.pdf" source_author: "Paul J. Werbos (interviewed by Edward Rosenfeld)" source_date: "interview conducted June 1993; published 1998" source_tier: 1 other_sources: [{"url":"https://yuxi.ml/essays/posts/backstory-of-backpropagation/","author":"Yuxi Liu","tier":4,"note":"Independent essay that quotes fragments of the same Talking Nets passage ('the appendix on how to do it was not there because of page limits,' 'DOCSUB,' 'too hot'). Used only for triangulation/discovery, not as the evidentiary source — the primary transcript was read and visually verified directly."},{"url":"https://www.werbos.com/Neural/Selected_Papers.htm","author":"Paul J. Werbos","tier":1,"note":"Werbos's own annotated list of selected papers. Checked directly; does not list or annotate the 1978 SMC paper at all (earliest annotated entries are 1981/1982). No DARPA/DOCSUB/appendix language found here — a negative result, not corroborating."},{"url":"https://www.werbos.com/Neural/BTT.pdf","author":"Paul J. Werbos","tier":1,"note":"Werbos, 'Backpropagation Through Time,' Proceedings of the IEEE, Vol. 78, No. 10, Oct. 1990, pp. 1550-1560. Read in full; contains no mention of DARPA, DOCSUB, or omitted appendices. Cites a different, later Werbos paper ('Applications of advances in nonlinear sensitivity analysis,' 1981 IFIP conference / 1982 Springer book) that should not be confused with the 1978 IEEE SMC paper."},{"url":"https://archive.org/details/talkingnetsoralh00ande","author":"James A. Anderson and Edward Rosenfeld, eds.","tier":1,"note":"Archive.org's copy of the full book Talking Nets: An Oral History of Neural Networks (MIT Press, 1998), access-restricted/borrow-only. Confirmed to exist as a backup/redundant source if the gwern.net PDF ever goes down; not directly searched since the gwern copy already supplied verified full-page text."},{"url":"https://repository.library.brown.edu/studio/collections/bdr:78krc38h/","tier":4,"note":"Brown University Digital Repository, apparent original host of the participant-approved oral-history transcripts underlying the gwern.net PDF. Returned only a bot-verification screen on fetch attempt this session; not independently browsed."}] --- This capture follows up a claim flagged in the 2026-06-29 capture on Werbos's 1974 Harvard PhD thesis, asking whether there is primary-source evidence that the 1978 IEEE Transactions on Systems, Man, and Cybernetics (SMC) paper omitted its technical appendices specifically because DARPA blocked a fuller report from entering the DOD's public document system. **Bottom line: the central question is CONFIRMED, at Tier 1, directly in Werbos's own words**, in an oral-history interview transcript he approved for public dissemination. The gwern.net-hosted PDF (a scan of the relevant book chapter) was fetched and its page 351 was read directly via image/vision inspection — not relayed through an intermediary summary — to independently verify the exact wording below. --- ## Claim: Werbos states, in his own words, that DARPA blocked his fuller technical report from entering DOCSUB (the DoD's document submission system), and that the resulting 1978 IEEE SMC journal paper — which was published despite that blocking — did not include the technical appendix on how to do backpropagation, due to journal page limits **Claim type:** historical/biographical, and specifically the kind the sourcing floor calls out as needing escalation to Tier 1-2 because it is a surprising, load-bearing claim ("Werbos explicitly mapped backpropagation onto Freudian libido" is the rubric's own example of this escalation category; "DARPA suppressed a technical report" is the same kind of claim). **Tier 1-2 required — achieved Tier 1.** The source is Paul J. Werbos's interview with Edward Rosenfeld, conducted June 1993 in Baltimore, Maryland, published as Chapter 15 ("Paul J. Werbos") in James A. Anderson and Edward Rosenfeld, eds., *Talking Nets: An Oral History of Neural Networks* (MIT Press, 1998), pp. 335-358. Per a description found of the Brown University Digital Repository (the apparent original host), "only the edited transcripts used for the book and approved by the participants for dissemination are made publicly available online" — i.e. this is a participant-approved transcript, not a leak, paraphrase, or third-party account. Exact verbatim quote, page 351, in Werbos's own voice ("PW" in the transcript), directly following his account of re-estimating a DARPA-funded, classified Joint Chiefs of Staff worldwide conflict-forecasting model (on a Latin America dataset) using backpropagation and finding it had almost no predictive power: > "DARPA wasn't happy when I published that result. They wanted me to do something real and practical and useful for the United States. This was useful, but it was an exposé. They didn't like that. Therefore, the report, which included backpropagation in detail and other advanced methods, was not even entered into DOCSUB. Every time I tried to enter it into DOCSUB, somebody jiggled influence to say, 'No, no, no, we can't publish this. This is too hot.' > > It was published in the IEEE SMC [Systems, Man and Cybernetics] Transactions Journal in '78 anyway because they couldn't block the journals, but it didn't include the appendices. So that paper in 1978 said, 'We've got this great thing called dynamic feedback, which lets you estimate these things. It calculates derivatives in a single swoop, and you can use it for lots of things, like AI.' > > That was all in the paper, but the appendix on how to do it was not there because of page limits for a journal article. The fact that we could get better forecasts than conventional statistical methods was in the IEEE SMC Transactions Journal in '78." Visual inspection of the scanned page (page 351 of the PDF, numbered "351" in the printed running header, chapter "Paul J. Werbos") confirms this transcription is exact, word for word. **Provenance:** - source_url: https://gwern.net/doc/ai/nn/rnn/1998-werbos.pdf - source_author: Paul J. Werbos, interviewed by Edward Rosenfeld - source_date: interview June 1993; published 1998, in *Talking Nets: An Oral History of Neural Networks*, MIT Press, ISBN 0-262-01167-0, Chapter 15, pp. 335-358 (quoted passage p. 351) - source_tier: 1 - exact quote: see block quote above > [!note] Seek's commentary: Werbos also gives, on the same page, a slightly different reason for the appendix's absence than the DOCSUB blocking alone — he says the appendix "was not there because of page limits for a journal article," which reads as the proximate/technical reason (a journal has page limits; appendices get cut) rather than the same thing as the institutional reason (DARPA blocked the fuller report from DOCSUB). The two aren't contradictory — a full technical report blocked from DOCSUB and a page-limited journal article are two different documents with two different reasons for cutting the same material — but a careful promotion should preserve that Werbos names *two* things in the same breath: journals have page limits regardless of DARPA, and DARPA separately blocked the fuller version. The claim as posed in the topic prompt ("omit appendices because DARPA blocked...") slightly collapses these two into one causal chain; Werbos's own account keeps them as parallel-but-related facts. --- ## Claim: bibliographic identity of the 1978 paper — Werbos, P. J., and Titus, J., "An Empirical Test of New Forecasting Methods Derived from a Theory of Intelligence: The Prediction of Conflict in Latin America," IEEE Transactions on Systems, Man, and Cybernetics, Vol. SMC-8, Issue 9, pp. 657-666 **Claim type:** quantitative/bibliographic (volume, issue, page numbers) — **Tier 1-2 required.** This specific citation is `[unverified-quant — needs primary]`: it rests on convergent web-search indexing (multiple independent search hits citing the same volume/issue/page range, and subject matter — Latin American conflict forecasting — that matches Werbos's own account in the Tier 1 interview above), not on a directly-opened IEEE Xplore record or library catalog entry. Direct fetch attempts at IEEE Xplore returned HTTP 418 and empty responses this session and could not independently confirm the DOI or abstract page. The paper's existence and general subject (conflict forecasting in Latin America, DARPA-funded) is corroborated by Werbos's own interview account; the exact volume/issue/page numbers specifically need a direct IEEE Xplore or library check before being treated as settled. **Provenance:** - source: convergent web search indexing (not independently opened at IEEE Xplore this session) - source_tier: unclear/unconfirmed for the bibliographic detail — flagged, not to be treated as Tier 1-2 verified - status: `[unverified-quant — needs primary]` for the specific volume/issue/page citation only; the paper's existence, year, venue, and general subject are corroborated at Tier 1 via Werbos's own interview account above --- ## Dead ends checked this session (reported for completeness) - **werbos.com "Selected Papers" page** — does not list or annotate the 1978 SMC paper at all; earliest annotated entries are 1981/1982. Negative result, not corroborating either way. - **Werbos's 1990 "Backpropagation Through Time" paper** — read in full; no mention of DARPA, DOCSUB, or omitted appendices. Contains a citation to a *different* 1981/82 paper ("Applications of advances in nonlinear sensitivity analysis") that should not be confused with the 1978 SMC paper — an earlier search pass in this research thread briefly conflated the two, which is corrected here. - **Jürgen Schmidhuber's "Who Invented Backpropagation?" survey** — could not be fetched (DNS resolution failure). Untested as a corroborating or dissenting source this session. - **Paul Werbos Wikiquote page** — fetched; contains only 5 quotes, none touching DARPA/DOCSUB/appendices. - **Mind Matters podcast transcript** (2021, interview with Werbos) — fetched and read in full; no mention of any of the target terms. - **IEEE Xplore** (author page and specific document page for the 1978 paper) — returned HTTP 418 and empty content respectively; could not independently verify the citation through IEEE's own site this session. --- ## Central question status **Did Werbos's 1978 IEEE SMC paper omit technical appendices because DARPA blocked entry of the fuller report into the DOD public document system, and is there primary-source evidence for this claim?** **CONFIRMED**, directly from Werbos's own first-person, on-the-record account (his 1993 interview, published 1998, participant-approved for dissemination) — this is Tier 1 evidence in the rubric's terms, satisfying the escalated floor this kind of surprising/load-bearing historical claim requires. The one caveat, stated plainly rather than smoothed over: no *independent* corroborating witness was found this session — no DARPA program-manager account, no declassified internal memo, no contemporaneous news article — that verifies the DARPA-blocking mechanism from a second angle. The fact that Werbos said this, in his own words, in a serious citable venue, is confirmed. Whether a second, independent primary source exists to corroborate the institutional mechanism from DARPA's side was not found and may not exist in the public record — this remains a single-witness (though highly credible, first-person, on-the-record) account. --- ## Further leads - A direct IEEE Xplore or library-catalog check of the exact volume/issue/page citation (SMC-8, Issue 9, pp. 657-666) would close the bibliographic-detail gap flagged above. - Checking whether the published 1978 journal paper's own front matter/editorial notes mention anything about the appendix being cut (e.g., a "for the appendices, contact the author" note, common in that era's SMC Transactions) would add a second angle from the journal side, though it would likely only corroborate the page-limit reason, not the DARPA/DOCSUB blocking specifically. - Related existing notes: [[backpropagation-gap]], claim-notes from the 2026-06-29 and earlier 2026-07-06 Werbos captures in this batch.