Version history — Retrieval Practice¶
Current edition: v3 · 2026-09-28 · domain-specific abstraction
Corrections¶
Checking this entry found 4 errors, now corrected: against its sources, or where it contradicted itself or its arithmetic did not hold. Each is listed here so a reader can see what changed. How references were checked.
- 2026-09-25 · Factual correction. Described the classroom quizzes as generative free recall; they were multiple-choice, and the article now says they still required retrieval before feedback.
- 2026-09-25 · Factual correction. Misdescribed the middle-school clicker study (teachers, ungraded quizzes, material 'only reviewed'); a research assistant gave low-stakes quizzes and the comparison material was unquizzed or re-presented.
- 2026-09-25 · Factual correction. Said students predicted that restudying would help more; what the study measured was confidence ratings in the second experiment, and the sentence now reports that.
- 2026-09-25 · Factual correction. Gave the one-week retention figures of the Roediger and Karpicke study as about 60% versus 40%; the reported figures are 56% versus 42% (Experiment 1) and 61% versus 40% (Experiment 2).
Editions¶
| Version | Date | Change | What changed |
|---|---|---|---|
| v3 | 2026-09-28 | First citations: 6 claims in this entry now cite 5 sources, each footnote saying what that source supports. 4 passages corrected against the sources that establish them, with independent review. | prose — detailed body (prose edited) |
| v2 | 2026-09-19 | Both layers now take their origin domain from one shared 65-domain roster. The domain-specific layer had drifted to 2,744 free-text domain labels, most used once and many just restating the entry's own subdomain, sharing only 18 of them with the closed list the primes used — so the same domain name meant different things in the two layers and their counts could not be read together. All 2,716 off-roster labels were mapped onto the roster, and the finer specialty each entry was authored with was kept as its subdomain, now a list rather than a single value; subdomains that only repeated the domain under an older or broader name were dropped. The prime side no longer reads a hand-maintained registry that had fallen behind the corpus — the Mathematics page listed 66 primes when 154 claimed the domain — and the 61 primes carrying no origin domain at all were classified. Every domain also rolls up into one of eight groups, shown on each entry and browsable in its own right. | structure — origin (fields changed) |
| v1 | 2026-08-24 | First published edition. | first published edition |
Editions are cut at publish time from a content hash of the entity's canonical source, so a version changes only when the content actually changed. Full canonical content for every edition, and the exact per-part diffs between them, are in the downloadable editions archive (editions_archive.jsonl); the current-state index is versions.jsonl.