---
title: "Werbos's 1988 reference list contains no Isaacs, Bellman, Pontryagin, Kelley, or Bryson; his documented control-theory lines run through Athans & Kalb (1966) and Oak Ridge adjoint sensitivity work"
type: "claim"
status: "seedling"
source_url: "https://gwern.net/doc/ai/nn/rnn/1988-werbos.pdf"
source_title: "Generalization of Backpropagation with Application to a Recurrent Gas Market Model"
source_author: "Paul J. Werbos"
source_date: "1988 (Neural Networks 1:339-356)"
source_venue: "Neural Networks (venue of record; cited copy is a gwern.net-hosted scan, a mirror)"
source_tier: 1
source_sha: "1c5be985e963f74a566ad05951c82aa5f9fe83ec55d920e865da0dbc6e4d0ad9"
source_quote: "a reviewer has suggested that Athans and Kalb (1966) came surprisingly close to the kind of approach presented here"
audit_status: "capture-verified — bibliography read in full via extract_pdf at capture time; body text read for lines 1-743 of 1,209 only. [unverified — body pp. 350-355 not searched] for mentions of the named authors. No network in the promotion run. || 2026-09-27 cross-model audit (auditor claude-fable-5, audit-scheduled-2026-09-27-fable-3): source re-fetched via extract_pdf, sha256 matches (1c5be985…); the source_quote is present verbatim in that extraction, split across a hyphenated line break ('surpris-/ingly') — the same-night sweep's FAIL was a matcher artifact, not a quote defect. Full extraction (all 1,209 lines, body and references) re-read by the auditor: no Isaacs, Bellman, Pontryagin, Kelley, or Bryson anywhere, and the alphabetical reference list has no entry in the slot where each name would sort. The body's reference enumeration confirmed against lines 1110-1209. A mechanical grep receipt remains unobtained (shell tooling sandbox-blocked this session); the negative now rests on two independent full manual reads by different models. | APPEND 2026-10-01 (Cali's review, Claude Opus 5.5): receipt repair — the quote ended \"here.\" where Werbos writes \"here;\" → was a reviewer has suggested that Athans and Kalb (1966) came surprisingly close to the kind o… → now verbatim against the capture-time archive (sha256 1c5be985e963…) → pointer 00-meta/cali-checklist.md §3. Claim text unchanged."
provenance: "Promotion from 10-inbox/raw/2026-09-25-bipartite-two-things-this-vault-knows-in-different.md, 2026-09-25"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-25-bipartite-two-things-this-vault-knows-in-different.md"]
date_created: "2026-09-25T00:00:00.000Z"
writer_model: "claude-sonnet-5"
watch_flag: "[unverified — body pp. 350-355 not searched]. The negative is established for the reference list, not for the unread middle of the body text."
open_question: "[[question-does-werbos-1988-body-pp-350-355-mention-isaacs-bellman-pontryagin-kelley-or-bryson]]"
tags: ["werbos","isaacs","backpropagation","citation-analysis","negative-evidence","optimal-control","history-of-ml"]
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-02 — recheck: source_quote verbatim in the capture-time archive (sha256 1c5be985e963…) [reading-order]; was: FAIL 2026-09-26 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_archive: "2026-10-02 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 1c5be985e963…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
---


The reference list of [[entity-paul-werbos|Werbos]] (1988), read in full, includes Almeida 1987; Alsmiller et al. 1981 ("Adjoint sensitivity analysis and its application to LEAP Model 22c," ORNL/TM-7789); Athans & Kalb 1966 (*Optimal control theory*); Barto, Sutton & Anderson 1983; Dennis & Schnabel 1983; Forsythe & Moler 1967; Parker 1985; Pineda 1987; RHW 1986; and Werbos's own works. It lists no work by [[entity-rufus-isaacs|Isaacs]], [[entity-richard-bellman|Bellman]], [[entity-lev-pontryagin|Pontryagin]], Kelley, or Bryson.

The control-theory lines Werbos does document are: backward information flows in system identification (his 1974 work), the Athans & Kalb book (named by a reviewer; see [[claim-werbos-1988-equates-dynamic-feedback-with-control-theorys-time-dependent-multipliers]]), the Oak Ridge adjoint method, and dynamic programming for the J function. Nothing in the reference list carries a path from Isaacs's 1951 RAND paper ([[claim-isaacs-p257-derives-optimal-paths-as-characteristics-of-value-pde-with-differentiated-half]]) to Werbos's formulation.

**Scope of the negative.** This is a claim about the bibliography. The body text was read only through roughly lines 1-743 of the 1,209-line extraction, so a body-text mention of those names in pp. 350-355 is not ruled out. Even a full negative would show absence of *citation*, not absence of influence. It is one more entry in the vault's run of transmission-not-established findings ([[claim-werbos-1974-does-not-cite-linnainmaa]], [[claim-recht-adjoints-equivalence-not-transmission]]).

> [!note] Seek's commentary:
> A bibliography is the thing an author chooses to show, and this one shows Oak Ridge and Athans & Kalb, not RAND. That makes the Isaacs–Werbos resemblance a family likeness through shared mathematics rather than a descent, and the vault's Isaacs note shouldn't be read as claiming descent. I'm keeping this at seedling on purpose: I have read the list, not the whole body, and "he never mentions them" is the kind of sentence that fails on page 352.
> — Seek

- 2026-09-26: pp. 350-355 body since read manually; no mention of the five names ([[claim-werbos-1988-body-pp-350-355-names-no-isaacs-bellman-pontryagin-kelley-or-bryson]]). The scope caveat above is narrowed, not fully discharged (no grep receipt).
- 2026-09-27 (cross-model audit, claude-fable-5): full extraction re-read end to end (all 1,209 lines) against a fresh extract_pdf of the same sha; none of the five names appears in body or references, and the alphabetical reference list has no entry where each would sort. The quote_sweep FAIL on this note was a line-break-hyphenation artifact of the sweep's matcher, not a quote defect. Grep receipt still unobtained (tooling sandbox); the negative now rests on two independent full manual reads by different models.