What genuinely connects Werbos's two missing-document claims and the DOCSUB/RHW-thesis document-identity claim? The same 1993 sitting, with one leg independently corroborated

werbosdocsubdarpabackpropagationoral-historysingle-witnesscitation-historybridge-checkhistory-of-ml

The task brief asks whether observation-werbos-two-missing-documents-share-one-1993-witness and claim-werbos-docsub-report-and-rhws-missing-1974-thesis-are-different-documents are a real bridge (cosine 0.87, previously unlinked) or a resemblance the retrieval index mistook for a connection. No prior capture in 10-inbox/raw/ addresses this specific pair together (confirmed by grep before starting).

Verdict up front: the bridge is real, and it is stronger than shared vocabulary. Both notes' load-bearing evidence traces to the identical primary document — the same PDF, same sha256, same 25-page chapter — re-fetched and re-read directly this session. The claim note's own basis for identifying which document DOCSUB blocked (the 1975–1978 Maryland/DARPA project, not the 1974 Harvard thesis) is narrated in the same June 1993 sitting the observation note names as the single witness behind the 1968 Cybernetica and 1978 DOCSUB stories. This does not overturn either note's conclusion — both stand — but it means the vault's three-part "Werbos missing documents" cluster shares one narrative backbone, not two independent ones, with a single exception detailed below.

Claim: The document-identity claim's evidence for "which document" comes from the same June 1993 sitting the shared-witness observation names

A direct read of the full chapter (pp. 335–358, re-fetched this session, sha256 identical to the vault's existing citations) shows one continuous autobiographical narration in this order: the 1968 Cybernetica paper and its Freud-to-mathematics content (p. 338); the Harvard PhD thesis work through 1974 (pp. 340–346); the move to the University of Maryland and onto a DARPA grant as Principal Investigator (pp. 348–349); and the DOCSUB block itself (pp. 350–351). These are not three separately-told episodes the vault happened to file into three notes — they are consecutive pages of one interview. The passage establishing that the DARPA/DOCSUB project followed, and was distinct from, the PhD work reads: "It wasn't like I had filled out a grant application. It's like the DARPA guys came and spoke to me" (p. 348) — immediately after Werbos describes arriving at Maryland following his Harvard PhD. The 1968 paper's own location in the same sitting is fixed by "I talk in there about the concept of translating Freud into mathematics" (p. 338), and the whole chapter is dated once, at its head: "June 1993, Baltimore, Maryland."

Sources: https://gwern.net/doc/ai/nn/rnn/1998-werbos.pdf, "Talking Nets: An Oral History of Neural Networks," ch. 15 (Paul J. Werbos, interviewed by Edward Rosenfeld), MIT Press, 1998 (interview June 1993) — Tier 1, source_sha cd2801055eda7be5100072ec0bf8c79d0d52bd2ee6bb450495c9e75df9bcc6a6, TLS verified, all three quotes above confirmed verbatim against the extracted text via quote_check this session.

Claim: Unlike the other two "missing document" legs, the document-identity conclusion has independent corroboration outside the 1993 interview

Werbos's own curriculum vitae — a second, separately-authored document, not a retelling of the interview — independently states the dates that make the document-identity claim work without relying on the 1993 narrative at all: "1974 Ph.D. Harvard University" (thesis "proposed the method now known as 'backpropagation,'" per the CV's own words: "My Ph.D. thesis proposed the method now known as 'backpropagation,'"), followed by "1975 - 1978 Assistant Professor, University of Maryland, College Park," where he served as "Principal Investigator on a DARPA grant." This is the same chronology the interview narrates, arrived at independently: PhD complete in 1974, DARPA project 1975–1978. That gap is what makes "the blocked report" and "the uncited thesis" two different, non-overlapping documents rather than one — and it no longer rests on a single witness for the dates, only for the DOCSUB-block narrative itself (which remains single-witness, per claim-werbos-1978-docsub-blocked's own flag: "no DARPA-side corroboration found and none may exist publicly").

Sources: http://web0.msci.memphis.edu/faculty/pjwerbos.pdf, "Curriculum Vitae Paul J. Werbos, Ph.D." — Tier 1 (primary self-authored CV; hosted at a University of Memphis faculty mirror rather than a Werbos-controlled venue, flagged below as a further lead, not a defect in this claim), source_sha 1c4051c9a125dabb54c414119a62094f6a7913ecdbb5a2ef473ed8fa35a3a593, fetched fresh via extract_pdf this session, TLS verified, all quotes confirmed verbatim via quote_check.

Claim: The bridge extends the shared-witness finding to a third leg rather than contradicting either note

observation-werbos-two-missing-documents-share-one-1993-witness establishes that two Werbos "missing document" claims — the lost 1968 paper's content and the fact of the DOCSUB block — trace to one interview. The document-identity claim examined here is, structurally, a third leg of the same cluster: it also depends on that interview to place the DARPA/DOCSUB episode after the PhD rather than folding into it. What is new is that this third leg is not purely single-witness the way the other two remain — Werbos's CV gives it a second, independent anchor for the dates that do the actual disambiguating work. The 0.87 cosine between the two seed notes is therefore not tracking coincidental vocabulary the way a false-friend pairing would (compare observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal, where a different pairing — RHW's 1986 reference list against the DOCSUB block — was found to be thematic only, with no shared source underneath it). Here the underlying document is genuinely shared, and the resemblance the retrieval index caught reflects a real evidentiary dependency, not just a repeated cast of names.

What this does not change: neither note's core conclusion moves. claim-werbos-docsub-report-and-rhws-missing-1974-thesis-are-different-documents still correctly identifies two distinct documents, now on slightly firmer footing (CV corroboration). observation-werbos-two-missing-documents-share-one-1993-witness still correctly warns that two "missing document" stories are not two independent corroborations — that warning is confirmed to extend one step further than the note itself states.

Sources: as above (Talking Nets ch. 15 and Werbos CV, both Tier 1, both re-fetched this session with matching sha256 to prior vault citations); claim-werbos-1978-docsub-blocked and claim-docsub-episode-postdates-rhw-1986-nature-paper-by-years consulted for cross-reference only, not re-verified this session.

Further leads

• The CV cited above (web0.msci.memphis.edu) is a University of Memphis faculty-page mirror, not a Werbos-controlled venue. A WebSearch this session found werbos.com hosts a different document (an NSF biography PDF, PJW_NSF_bio_Feb2015.pdf) but no confirmation it hosts this same academic CV — worth checking directly for a cleaner primary-venue citation per the public-shelf rule, since this exact CV is already load-bearing in claim-werbos-docsub-report-and-rhws-missing-1974-thesis-are-different-documents.
• The Talking Nets chapter's OCR/pdftotext extraction shows inconsistent word-spacing in places (e.g., "walkedinto," "PrincipalInvestigator," "majorgrant" render with no space in a fresh extract_pdf pass, while other nearby words render cleanly) — several previously-recorded source_quotes on existing claim notes (e.g. the "When I got there I also walked into being PI..." quote) did not pass a fresh quote_check against this session's re-extraction, despite matching sha256 and a prior audit marking them "verified-verbatim." Not investigated further this session since it wasn't this capture's claim to re-audit, but worth a dedicated hygiene pass — possibly a quote_check/extraction-tool version difference rather than a sourcing error.
• Karl Deutsch's "The Nerves of Government" (1963) — Werbos's thesis adviser and the person who both enabled the Maryland/DARPA pipeline (per the interview, pp. 341, 346) and had already applied a systems/neural-network framing to political forecasting before Werbos did — was not read directly this session; only Werbos's account of him was.

Entity candidates

• Karl Deutsch — person — Werbos's Harvard PhD thesis adviser and the older, foundational figure this whole document chain measures against: he authored the neural-network-of-government framing before Werbos, proposed the political-forecasting application that became Werbos's thesis, and was the direct route into the Maryland job that later carried the DARPA grant. Already has an entity page (entity-karl-deutsch) — flagged to confirm the connection, not to create.
• Edward Rosenfeld — person — the interviewer who conducted and co-edited the single June 1993 sitting that is the shared source underlying all three "missing document" legs in this cluster. No entity page found in 40-entities/; his editorial choices (what he asked, in what order) shaped the very narrative sequence this capture relies on.
• DOCSUB — concept — the institutional mechanism itself remains undefined by any primary source across multiple vault sessions; already has a watching-stub entity page (entity-docsub) per a prior capture. Flagged to confirm, not create.