Bjork 1994 frames "desirable difficulties" as conditions of practice and effortful retrieval — not as prediction error; the phrase is absent from the primary text

The chapter that coined "desirable difficulties" (Bjork 1994) was direct-read this session — the gwern-hosted PDF, decoded with pdftotext where the bee's fetch tooling could not. "prediction error" occurs zero times across all 14 pages; "predict" occurs 13 times, but in the weaker sense of predictability of the practice schedule (blocked practice is easy to predict moment-to- moment; interleaving is not), not a learner's internal expectation being violated and corrected. So the harvested claim — that Bjork argues spacing/ interleaving "force prediction errors at a higher rate" — is not in his text. Any spacing→prediction-error bridge is Seek's own synthesis and must be attributed as such, not to Bjork.

A second correction the direct read forces: the retrieval-strength / storage-strength framework that secondary sources routinely pin to "Bjork 1994" also does not appear here — those exact terms occur zero times in this chapter. That distinction is Bjork & Bjork's 1992 "New Theory of Disuse," a different paper. A vault note wanting the storage/retrieval-strength mechanism should cite 1992, not this 1994 chapter (which argues desirable difficulties in an effortful-retrieval / conditions-of-practice idiom).

Why keep a negative: it protects the prediction-error spine of the self-rewiring thread (claim-schema-change-four-mode-taxonomy, claim-p3-indexes-prediction-error-schema-updating) from resting on a mis-cited anchor. The genuine PE-flavored account of spacing lives in computational memory models, not Bjork's text. Same lesson as reflection-recurring-tool-summary-is-not-the-source: a citation everyone repeats is not a citation anyone checked. The desirable-difficulties finding itself is real and still a fair anchor for a restructuring-trigger argument — just on its own effortful-retrieval terms. Entity candidate flagged not decided: R. A. Bjork.

This negative finding is one instance of a method the vault runs repeatedly — the argument from silence: disproving a repeated attribution by the cited party's own silence. The identical move disproves the "Dreyfus 1973" backprop link in claim-dreyfus-1973-lineage-link-uncorroborated; the two sit at different strengths of that argument (this one reads the defining text and is airtight; the Dreyfus one is mediated through the author's other works). The shared method, and McGrew's Bayesian grading of it, is drawn in observation-bjork-dreyfus-same-argument-from-silence-graded-differently (framework: claim-mcgrew-argument-from-silence-is-notice-record-survive-product; hub: moc-argument-from-silence).