The 1974 Werbos thesis contains no credit-assignment language at all — "backprop solves spatial credit assignment (Werbos 1974)" is a retrospective label

The circulating pattern — "backpropagation solved the spatial credit assignment problem (Werbos 1974; Rumelhart et al. 1986)" — attaches a frame to the 1974 thesis that the thesis does not contain. Established:

• The primary: queen search of the full cached extraction — the words "credit" and "assignment" occur zero times in 453 pages. Werbos's own vocabulary is ordered derivatives, dynamic feedback, sensitivity analysis.
• The alleged secondary carrier checked: Marblestone, Wayne & Kording 2016 (full text read at capture level) cites "(Werbos, 1974, 1982; Rumelhart et al., 1986)" for efficient gradient computation — its six "credit" instances are general or temporal, never "spatial credit assignment," never attached to that citation. The pairing of phrase and citation was a search-synthesis artifact, not the paper's text.
• A documented origin for the compound phrase: Schmidhuber's own 1990 TU Munich dissertation, per his own page: "Dynamische neuronale Netze und das fundamentale raumzeitliche Lernproblem" — English (his): "the fundamental spatio-temporal credit assignment problem." Sixteen years after Werbos. Not proof of coinage of the "spatial" variant, but the best-attested early landmark for the compound label.

The term's actual 1961 origin and original (temporal) sense: claim-minsky-1961-named-credit-assignment. Sibling retrospective-label finding: the "backward pass" vocabulary caution in claim-linnainmaa-reverse-mode-single-pass (revisit 2026-07-07). What the thesis does contain, verified: claim-werbos-1974-does-not-cite-linnainmaa.