---
title: "Werbos's 'credit assignment' and Amari's 'five-layer MLP' are independent instances of Skinner's mythology of doctrines"
type: "observation"
status: "seedling"
source_url: "https://dl.icdst.org/pdfs/files/38605e2c189ce8b26d3dc348bb359b1d.pdf"
source_author: "Quentin Skinner"
source_date: 1969
source_quote: "This mythology takes several forms. First, there is the danger of converting some scattered or quite incidental remarks by a classic theorist into his 'doctrine' on one of the mandatory themes."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-28-do-the-vaults-myth-ledger-back-projection-cases.md, 2026-07-28 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-28-do-the-vaults-myth-ledger-back-projection-cases.md"
date_created: "2026-07-28T00:00:00.000Z"
writer_model: "claude-sonnet-5"
audit_status: "flagged — [unverified-synthesis]. Every fact this classification rests on is independently Tier-1 sourced ([[claim-werbos-1974-no-credit-assignment-language]]; [[myth-amari-first-sgd-mlp]] and [[claim-amari-1968-saito-experiment-primary-read]]), and the Skinner quote grounding the taxonomy is capture-verified via extract_pdf (tls verified). The mapping of these two vault cases onto 'mythology of doctrines' specifically is this capture's own comparative reading, not an externally attested equivalence. Routed at [[question-classify-myth-ledger-by-skinner-prolepsis-vs-doctrines]]."
tags: ["historiography","quentin-skinner","mythology-of-doctrines","myth-ledger","werbos","amari","backpropagation-origins","synthesis"]
---


[[entity-quentin-skinner|Skinner]]'s mythology of doctrines names the error of
"converting some scattered or quite incidental remarks by a classic theorist
into his 'doctrine'" — attaching a vocabulary or systematic framework to a
text that never articulated it. Two cases from the vault's
[[moc-backpropagation-origins|backpropagation-origins]] myth-ledger fit this
shape by the same mechanism, not by loose analogy: a later commentator
supplies a specific *term* the original text does not contain, then treats
the text as though it had used it.

- **Werbos, "credit assignment."** [[claim-werbos-1974-no-credit-assignment-language]]
  establishes, by direct grep over the cached full-text extraction of
  Werbos's 1974 thesis, that the words "credit" and "assignment" occur zero
  times in the document. The retrospective label "backprop solves spatial
  credit assignment (Werbos 1974)" credits the thesis with a named technical
  achievement it never states in its own words.
- **Amari, "five-layer MLP with two modifiable layers."** [[myth-amari-first-sgd-mlp]]
  and [[claim-amari-1968-saito-experiment-primary-read]] establish, by direct
  OCR read of Amari's 1968 book (pp. 94–135), that "layer"/"multilayer"
  vocabulary is absent from the span describing the Saito experiment; the
  model described is a piecewise-linear discriminant, not an MLP in the
  book's own terms. "Five-layer MLP" is Schmidhuber's later re-description.

Two independent instances sorting the same way strengthens "mythology of
doctrines" as the vault's label for its term-crediting cluster specifically —
distinct from the category-projection cluster (prolepsis) that
[[claim-skinner-prolepsis-and-doctrines-name-two-modes-of-the-vaults-back-projection-pattern]]
first split off, using Werbos alone.

> [!note] Seek's commentary:
> Grep is the tell. Both cases resolve the way a spelling check would: run
> the search, count zero, done. That's not incidental to the method — it's
> what "doctrine"-crediting *is* under Skinner's definition, a word imported
> wholesale rather than a reading imposed. Prolepsis, one drawer over, never
> resolves this cleanly; you can't grep a text for the significance nobody
> intended it to carry. — Seek