Butterfield/Weinberg's whig-history critique does not sort the vault's AI-history myth-ledger cleanly into two kinds — a hero/villain-narrative mode and a residual bibliometric/scope-conflation mode resist the binary

whig-historyhistoriography-of-aipresentismmyth-ledgerphilosophy-of-sciencecross-domain-bridgehero-villain-narrative

The vault holds a saturated cluster of AI-history corrections — the "credit assignment" label the 1974 Werbos thesis never used (claim-werbos-1974-no-credit-assignment-language), the reading of a 1986 journal table of contents as a deliberate pairing of NETtalk with an unrelated cellular-automaton paper (claim-complex-systems-issue-one-paired-nettalk-with-cellular-automaton), Amari's uncited priority over Hopfield (myth-amari-first-sgd-mlp), and the Perceptrons myth (myth-perceptrons-book-killed-connectionism) — without ever naming, until this note, what general fallacy all of them police. Historians of science already have the name: Whig history (claim-butterfield-1931-coined-whig-history-as-present-referenced-past), "the past with reference to the present."

The frame does not import into the vault cleanly, and that is where the value is. Steven Weinberg's defense of Whig history for science specifically (claim-weinberg-2015-defends-whig-history-as-legitimate-for-science) — "in scientific history we really can say who was right" — draws a boundary this observation originally applied to the myth-ledger as a clean two-way split: judging, with present knowledge, that Werbos's equations amount to a gradient computation is defensible present-knowledge appraisal of the kind Weinberg licenses, while attaching a later concept or intent the original actors never had — the specific compound "spatial credit assignment" — is the anachronism Butterfield warned against.

Tested against every entry under moc-backpropagation-origins's myth ledger, that two-way split does not hold. Only myth-amari-first-sgd-mlp's "five-layer MLP" relabeling of Saito's 1967 experiment sorts cleanly as Butterfield's anachronism — distinct from the same myth's separate citogenesis problem (claim-wikipedia-amari-sgd-citogenesis), which is a source-independence defect, not a present-vs-past one. myth-werbos-invented-backpropagation and myth-perceptrons-book-killed-connectionism resist both poles: both collapse a multi-threaded, multi-actor history into a single hero or villain — not a dispute over whether past work was correct, and not a later concept imported onto it, but a third mode (which Weinberg's own essay and Wikipedia's summary of Butterfield both explicitly name and reject, one paragraph past the sentence this note already quoted). myth-linnainmaa-uncited-before-2010s and claim-griewank-2012-does-not-mention-amari sit outside the frame entirely — a citation-count dispute and a confirmed non-overlap between two priority literatures, neither a present-knowledge appraisal nor an anachronism. The myth-ledger's real target is not "mostly" anachronism, as originally written here; it is at least four distinct failure shapes, one of which (hero/villain oversimplification) the theory's own authors already named and the observation's binary was not built to catch.

The sharpest test case sits inside the vault's own most-trusted source for this material: Schmidhuber's Annotated History runs an anti-Whig credit-correction campaign while itself reading "deep learning" back onto 1676–1914 mathematics (claim-schmidhubers-annotated-history-projects-deep-learning-onto-pre-1900-mathematics) — one document failing the second test while succeeding at the first. Michael Mahoney's plural-histories critique of computing (claim-mahoney-warned-against-whig-history-of-computing) is the disciplinary bridge that makes this vault's home domain a legitimate target for the critique in the first place — though, per a 2026-08-18 direct read of Mahoney's own essay, the bridge is this vault's own construction, not one Mahoney built himself: he never names Butterfield or "Whig history" (see claim-mahoney-2005-essay-omits-whig-and-butterfield). His critique arrives at the same fallacy independently, from the history-of-technology literature rather than the history-of-science Whig-history conversation. See moc-backpropagation-origins.

Correction history.

• 2026-09-04 — the original two-way split (present-knowledge appraisal vs. anachronistic concept-importing) was tested against all six entries under moc-backpropagation-origins's myth ledger (10-inbox capture 2026-09-04-does-the-butterfieldweinberg-defensible-vs-anachronistic-split-actually.md, answering question-verify-whig-defensible-vs-anachronistic-split-against-myth-ledger). It does not hold as stated: myth-werbos-invented-backpropagation and myth-perceptrons-book-killed-connectionism are monocausal hero/villain narratives that fit neither pole — a third mode Weinberg's own essay and Wikipedia's summary of Butterfield both explicitly name and reject (claim-weinberg-and-butterfield-both-explicitly-reject-hero-villain-narratives). myth-linnainmaa-uncited-before-2010s and claim-griewank-2012-does-not-mention-amari never touch the present-vs-past axis at all — a residual bibliometric/scope-conflation category. Only myth-amari-first-sgd-mlp's MLP-relabeling sorts cleanly as Butterfield's anachronism. Title and body above revised to state the four-mode finding directly; this block preserves the original two-way claim as history.