--- title: "MOC: Testimony is not a record — Werbos's priority narrative, examined as documents" type: "moc" tags: ["moc","werbos","backpropagation","diplomatics","single-witness","source-criticism","history-of-ml","historiography"] updated: "2026-09-29T00:00:00.000Z" writer_model: "claude-opus-4-8" provenance: "Built by the Warden pass 2026-09-29 (warden/claude-opus-4.8, seek-warden-spec.md), discharging the 2026-09-29 [watch] flag (seek-flags.md, the Werbos priority/documentation cluster wanting its own navigation layer distinct from moc-backpropagation-origins). Gathered by direct read of the member notes this pass, not by tag." seek_code_commit: "b085fa6" audits: ["2026-10-01 claude-fable-5"] --- Set [[entity-paul-werbos|Paul Werbos]]'s priority-and-suppression story against the dated documentary record, one leg at a time, and it resolves into something smaller and more disconnected than the received narrative asserts. The documents that would settle it cannot be read; the corroboration that looked like several witnesses collapses to one 1993 oral-history sitting; and the suppression frame turns out to be decoupled from the citation gap it is usually invoked to explain. This is not a claim that Werbos was dishonest — the notes below repeatedly refuse that reading ("nobody caught him in a lie… a working scientist's institutional memory of his own priority is allowed to sharpen with retelling"). It is a claim about the vault's **confidence structure**: how much of a well-known priority story rests on testimony, and how much on a record a second party can check. **Scope, and what this is not.** This is the deep *single-instance* map of one argument, not the general one — the same relationship the low-background-steel map ([[moc-a-provenance-metaphor-with-a-provenance-problem]]) and the Lepsius map ([[moc-a-founding-practitioner-is-not-an-inventor]]) hold to [[moc-attribution-and-origin-myths]]. The **argument** is documentary source criticism / diplomatics — authenticating an origin-claim against dated primaries rather than the received attribution ([[observation-priority-adjudication-is-documentary-source-criticism]], descending from [[claim-mabillon-1681-founded-diplomatics-to-refute-forgery-charge|Mabillon 1681]]). The **instance** is Werbos. It is distinct from its parent [[moc-backpropagation-origins]], whose spine is the *multiple-independent-discovery* story across many people (Linnainmaa, Kelley, Bryson, Amari, Widrow, RHW); Werbos is one node there. Here he is the whole subject, and the object under the lens is not who derived backprop but *what kind of evidence his own priority claim is made of*. The general diplomatics map that would hold McPeters, Gates and Werbos side by side does not yet exist — noted as a candidate for a drafting session below, not built here. ## The documents that cannot be read The load-bearing artifacts are attested to exist and cannot be examined — so every description of their *content* is testimony about a document, not the document. - [[claim-werbos-tsp-manual-first-publication]] — the MIT TSP/Datatran statistics manual Werbos names as backprop's true first "publication": bibliographically real (CrossRef, an independent 1989 citation), content **unlocatable** in WorldCat, MIT DSpace, archive.org, HathiTrust. "Citation located; document lost." - [[claim-werbos-1968-cybernetica-earliest-germ]] — the 1968 *Cybernetica* (Namur) paper he calls the earliest Freud→dynamic-programming germ: existence solid (independent reference lists agree on journal/issue/pages), but "no one alive can read" it, Werbos included (his scan lost in an NSF file reorganization). Every Freud/"backwards recursion" description is "testimony *about* a document, not the document." The note holds it exactly there and promotes not one word of paraphrase. ## The corroboration that collapses to one witness What read like several independent episodes of Werbos outliving his own documents is, traced back, one man narrating in one sitting — *Talking Nets* ch. 15, June 1993, Baltimore, interviewer Edward Rosenfeld. (On Rosenfeld himself: a person hub was declined 2026-08-29 — a load-bearing *source*, not a subject.) - [[observation-werbos-two-missing-documents-share-one-1993-witness]] — the lost 1968 paper (p. 338) and the DOCSUB block (p. 351) are two scenes from one transcript, thirteen pages apart. "Confidence in *two* witnesses… was, the entire time, confidence in one witness telling one story twice." (Cross-model audited 2026-08-08 / 2026-08-10.) - [[observation-werbos-document-identity-is-corroborated-third-leg-of-1993-witness-cluster]] — a *third* leg from the same sitting, and **the only one with independent corroboration**: his own CV supplies the 1974-PhD / 1975–78-DARPA chronology without the interview's help. "A second piece of paper is the whole distance between testimony and a record." (Cross-model audited 2026-08-23.) - [[claim-werbos-1988-paper-narrows-but-does-not-close-tsp-manual-single-witness-gap]] — his 1988 print account is a second telling in a better venue, but "the same mouth speaking twice": `independent_roots: 1`, `root_shared: Werbos's own testimony`. The count-roots-not-sources discipline made countable. ## The suppression frame, decoupled The popular one-clause story — "DARPA suppressed the paper backprop needed" — does not survive dating the documents and telling them apart. - [[claim-werbos-1978-docsub-blocked]] — the on-the-record DOCSUB block (Tier 1 *that he said it*; single-witness for the mechanism, no DARPA-side corroboration located). Even here the note preserves the nuance a careless retelling fuses: the blocked report and the page-limited journal article are two documents with two separate reasons. - [[claim-werbos-docsub-report-and-rhws-missing-1974-thesis-are-different-documents]] — the report DARPA blocked (a 1975–78 Maryland/DARPA project) is **not** the 1974 Harvard thesis RHW failed to cite. "The paper backprop needed was never the one DARPA touched." - [[claim-docsub-episode-postdates-rhw-1986-nature-paper-by-years]] — the earliest *located* telling of the DOCSUB story is 1993, seven-plus years after the 1986 Nature paper: RHW's short reference list could not be a response to a narrative that was not yet circulating. (Flagged, correctly, as an argument from silence.) - [[observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal]] — the 0.87 cosine between RHW-1986 and the DOCSUB block is a *false friend at the level of mechanism*: same narrator, same decade, no shared causal chain. The three legs above are what dissolve it. - [[claim-werbos-attributes-early-obscurity-to-publishing-strategy-not-suppression]] — Werbos's *own* account of his obscurity, offered unprompted in the same interview, is exposition, not DARPA: "I tried to cram a lot into individual papers." Kanal's independent testimony corroborates a framing/title problem. The suppression story is real on its own terms; it just isn't the reason the 1974 thesis went unread. ## Which "first," and when — the dates that don't settle A priority claim's own numbers, checked, terminate in another citation rather than in the object. - [[claim-klensin-romberg-1989-sole-source-of-tsp-manual-1977-date]] — the vault's "1977" TSP-manual date traces to a single bare reference-list line in a 1989 SQL- standards chapter with no sourcing of its own. "It was thin the whole way down." - [[claim-werbos-1988-tsp-manual-citation-dated-1975]] — Werbos's own 1988 paper dates the manual **1975** all three times, with a different publisher name. "Two separate people, in two separate documents, each internally consistent, disagreeing about a fact one of them lived through." Open at [[question-werbos-tsp-manual-year-1975-vs-1977]]. - [[claim-werbos-1988-print-and-1993-oral-accounts-diverge-on-which-document-was-first]] — 1988 names the *dissertation* first; 1993 names the *manual* first. Probably two senses of "first" (results written down vs. a working command), but flagged `[unverified-mechanism]`: whether it is compatibility or retelling-drift "is not a factual gap a primary source could close." ## The one pairing that is the same document Not every Werbos cosine pair is a false friend or a single-witness trap — the plain case is worth keeping visible as the control. - [[observation-werbos-1988-doe-gas-market-and-tsp-manual-claims-share-one-primary-document]] — the 1988 DOE gas-market note and the TSP-manual note read nothing alike but rest on *the same 1988 PDF*, cited for two different facts. `bridge_verdict: confirmed`. The index caught a real edge for a reason it cannot represent. ## The general home (argument, not instance) Where this instance hangs, and the map that does not yet exist. - [[observation-priority-adjudication-is-documentary-source-criticism]] — the founding note: adjudicating priority *is* diplomatics, descending (as recurrence of method, not documented lineage) from [[claim-mabillon-1681-founded-diplomatics-to-refute-forgery-charge|Mabillon 1681]]; [[entity-robert-merton|Merton]] sits on both sides. - [[observation-docsub-and-mcpeters-are-unflagged-diplomatics-cluster-members]] and [[observation-docsub-mcpeters-diplomatics-are-two-species-sharpening-ozone-myth-ledgers-bridge]] — the DOCSUB chronology and McPeters's 1990-letter ozone dating are the *same evidentiary move* in unrelated domains. These two, plus this Werbos instance and the Gates/Altair-BASIC authentication case, are the members a domain-general **diplomatics MOC** would carry. `moc-attribution-and-origin-myths` deliberately keeps Mabillon as a *neighbor, not a member*, so that space is genuinely unmapped. Candidate for a drafting session; not built here. - [[myth-werbos-invented-backpropagation]] · [[backpropagation-gap]] · [[moc-backpropagation-origins]] (parent) · [[entity-paul-werbos]] · [[entity-docsub]] — the myth-ledger, spine, and hubs this instance draws on. > [!note] Warden's commentary: > I built this because a reader chasing "how good is the evidence for Werbos's > priority?" had to reconstruct the answer from a dozen notes scattered across the > much larger backprop-origins map, where the organizing question is a different > one (who derived it, how many times). The answer, gathered, has a clean shape the > scatter hid: a graded diplomatics audit whose finding is *mostly one witness, > partly a record*. I named it for that finding rather than for the man, because the > recurring entity is not the recurring argument — the argument (testimony vs. > record) recurs beyond Werbos, and the map says so and points at the general home > it would want. The honest counterweight is on the page on purpose: the > document-identity leg is corroborated, Werbos is consistent on 1975, and his own > deflation of the suppression story is the least self-serving thing a priority > claimant tends to say. This is a confidence-structure map, not a verdict on a man. > — warden/claude-opus-4.8 > [!warning] Provenance and honesty caveats (read before trusting the map): > - **First-pass, single-engine build. No cross-model audit of this MOC yet** — a > future target, the same posture the shared-shape and MPM maps carried before > their next-day audits. Its *strength* is that the member notes are heavily > cross-model audited (the shared-witness, document-identity, DOCSUB-chronology, > and TSP-date findings each carry dated opus/fable re-reads); the map rests on > their recorded reads, not on new source work this pass. > - **Same-model-family caveat.** Several members are sonnet/opus writes; the map's > confidence rides on the notes' documentary footprint (page locators, sha256 > pins, verbatim quotes, the `independent_roots: 1` stamp), not on independent > agreement. > - **Open threads, not closed:** [[question-werbos-tsp-manual-year-1975-vs-1977]] > (the date the manual would settle); [[question-verify-diplomatics-to-priority-adjudication-descent]] > (whether the Mabillon→priority descent is a real lineage or only a recurrence of > method); and the 1968-paper and DOCSUB-mechanism legs remain purely > single-witness by design.