C0 Retrieval beats restudy

narrowed 12 attacks: 6 survived, 6 narrowed · serves Q1 · last change 28 September 2026

After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying when the criterion test is delayed, items are correctly retrieved repeatedly in practice with corrective feedback (a single correct retrieval is not enough), and no-feedback multiple-choice practice is excluded. At an immediate (~5 min) test, equal-time restudy can be superior. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, when practice accuracy is relatively low and no feedback is given, or when “testing” is high-stakes performance rather than practice. The claim does not transfer to novel, meaningless abstract visual materials that lack preexisting semantic scaffolding.

In practice

Once something has been learned, spend review time recalling it from memory instead of rereading it. Check every answer against the correct one, and keep going until each item has come back correctly more than once.

Don't judge recall practice by a quiz a few minutes later. On an immediate test, rereading for the same time can come out ahead; the advantage of recalling is on a delayed test.

Where it holds

Inside the circle: the core of the claim. Outside: every bound and carve-out in the claim file, numbered, with the attack that forced it.

forced by a narrowing attackin the original wordingwritten at creation

Main contrast is for delayed criterion tests, not immediate (~5 min) tests.
Forced by attack #7 · Roediger & Karpicke 2006 · 8 September 2026
Delayed contrast requires repeated successful retrievals with corrective feedback (not one-shot success)
Forced by attack #10 · Dunlosky et al. 2013 · 8 September 2026
Earlier form: “Delayed contrast requires adequate practice-test success and corrective feedback” (Forced by attack #8 · Dunlosky et al. 2013 · 8 September 2026).
No-feedback multiple-choice practice is excluded.
Forced by attack #9 · Butler & Roediger 2008 · 8 September 2026
This bound first appeared in a change that applied several attacks at once; it is matched to #9 because its wording echoes that attack's ledger entry.
Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated.
Forced by attack #1 · Roediger & Karpicke 2006 · 8 September 2026
High semantic relatedness of cue–target materials reduces relative efficacy of retrieval vs restudy (~26%) but does not erase the advantage (Gupta, Pan & Rickard 2026; kill-condition test survived 2026-09-28).
Forced by attack #12 · McCarter, Huber & Cowell 2025 · 28 September 2026
Added after the archived record of earlier wordings ends.
Novel meaningless abstract visual materials lacking preexisting semantic scaffolding: McCarter, Huber & Cowell 2025 (six experiments) found restudy ≥ retrieval (sometimes significantly better) despite feedback and delays — outside the claim (cf. Ferreira & Wimber 2023 on meaningful vs low-semantic visual materials).
Forced by attack #12 · McCarter, Huber & Cowell 2025 · 28 September 2026
Added after the archived record of earlier wordings ends.
Does not hold when encoding never succeeded
In the original wording (8 September 2026), before any attack
Does not hold when the practice test is a poor match to what must later be remembered
In the original wording (8 September 2026), before any attack
Does not hold when practice accuracy is relatively low and no feedback is given
Forced by attack #8 · Dunlosky et al. 2013 · 8 September 2026
This bound first appeared in a change that applied several attacks at once; it is matched to #8 because its wording echoes that attack's ledger entry.
Does not hold when “testing” is high-stakes performance rather than practice
In the original wording (8 September 2026), before any attack
Does not transfer to: Novel, meaningless abstract visual materials lacking preexisting semantic knowledge / scaffolding (McCarter, Huber & Cowell 2025; cf. Ferreira & Wimber 2023).
Forced by attack #12 · McCarter, Huber & Cowell 2025 · 28 September 2026
Added after the archived record of earlier wordings ends.

8 of 11 bounds and carve-outs were forced by a specific attack. 6 attacks hit the boundary and did not move it: #2, #3, #4, #5, #6, #11.

How it changed

Replay of how the wording of C0 changed, attack by attack. Earlier wordings come from the published history of the old learning-graph site, which kept a snapshot after every change. Added words and ~~removed words~~ are marked. Steps are in the order the verdicts took effect.

Starting wordingfirst published 8 September 2026, 12:23 CEST
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. The advantage grows as the delay to the real test grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
No bounds were written yet.
Attack #2 survived no named source (attacker's own inference) · ledger date 8 September 2026 published 8 September 2026, 12:54 CEST
Attack: Outside lab word-list and short prose-passage paradigms, equal-time restudy yields delayed criterion retention at least as high as retrieval practice when encoding succeeded and practice matches the test.
Claim wording unchanged.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. The advantage grows as the delay to the real test grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Attack #3 survived Dunlosky et al. 2013 · ledger date 8 September 2026 published 8 September 2026, 13:01 CEST
Attack: When encoding success is only assumed after a single study pass, a large unencoded item subset remains for which equal-time restudy yields delayed retention at least as high as retrieval practice, so the main contrast does not hold under the claim’s own encoding precondition.
Claim wording unchanged.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. The advantage grows as the delay to the real test grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Attack #4 survived Bjork & Bjork 2020 · ledger date 8 September 2026 published 8 September 2026, 13:01 CEST
Attack: When practice-test format or cues mismatch the criterion test, equal-time restudy yields delayed criterion retention at least as high as retrieval practice after encoding succeeded.
Claim wording unchanged.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. The advantage grows as the delay to the real test grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Attack #1 narrowed Roediger & Karpicke 2006 · ledger date 8 September 2026 published 8 September 2026, 13:07 CEST
Attack: When initial learning is equated, the retrieval–restudy retention gap does not grow as final-test delay grows.
Attack #1 was first logged as survived (8 September 2026, 12:54 CEST) and later re-applied as narrowed; the wording changed at the re-application.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. ~~The~~ Once initial learning is equated, the claim does not assert that this advantage grows as ~~the delay to the real test~~ final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Bounds field after this step (changes marked):
Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated.
Attack #5 survived Roediger & Karpicke 2006 · ledger date 8 September 2026 published 8 September 2026, 15:01 CEST
Attack: When restudy is distributed across the same schedule as retrieval attempts, equal-time restudy yields delayed retention equal or superior to retrieval practice under encoding-succeeded and criterion-matched conditions.
Claim wording unchanged.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Attack #6 survived Adesope et al. 2017 · ledger date 8 September 2026 published 8 September 2026, 15:01 CEST
Attack: For educationally complex materials (multi-step procedures, transfer and application tests), equal-time restudy yields delayed criterion retention at least as high as retrieval practice when encoding succeeded and practice matches the criterion.
Claim wording unchanged.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Attack #7 narrowed Roediger & Karpicke 2006 · ledger date 8 September 2026 published 8 September 2026, 15:15 CEST
Attack: At an immediate (~5 min) criterion test after encoding succeeded and practice matched the criterion, equal-time restudy yields superior retention to retrieval practice.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time ~~restudying.~~ restudying when the criterion test is delayed; at an immediate (~5 min) test, equal-time restudy can be superior. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, or when “testing” is high-stakes performance rather than practice.
Bounds field after this step (changes marked):
Main contrast is for delayed criterion tests, not immediate (~5 min) tests. Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated.
Attack #8 + #9 narrowed Dunlosky et al. 2013 · Butler & Roediger 2008 · ledger date 8 September 2026 published 8 September 2026, 15:26 CEST
Attack: When accuracy on the practice test is relatively low and no corrective feedback is given, equal-time restudy yields delayed retention at least as high as retrieval practice under encoding-succeeded and criterion-matched conditions.
Attack: Multiple-choice practice without feedback increases delayed intrusions of false alternatives relative to equal-time restudy, so delayed cued-recall accuracy is not superior for retrieval under encoding-succeeded conditions.
Attacks #8 and #9 were applied in the same cycle and published as one change. The record holds only their combined result, so they are shown together; no in-between wording is invented.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying when the criterion test is delayed~~; at~~ , practice-test success is adequate, and corrective feedback is given (no-feedback multiple-choice practice is excluded). At an immediate (~5 min) test, equal-time restudy can be superior. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, when practice accuracy is relatively low and no feedback is given, or when “testing” is high-stakes performance rather than practice.
Bounds field after this step (changes marked):
Main contrast is for delayed criterion tests, not immediate (~5 min) tests. Delayed contrast requires adequate practice-test success and corrective feedback; no-feedback multiple-choice practice is excluded. Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated.
Attack #10 narrowed Dunlosky et al. 2013 · ledger date 8 September 2026 published 8 September 2026, 15:30 CEST
Attack: A single correct retrieval with corrective feedback does not produce superior delayed retention to equal-time restudy under encoding-succeeded and criterion-matched conditions; the delayed advantage requires repeated successful retrievals with feedback.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying when the criterion test is delayed, ~~practice-test success is adequate, and corrective feedback is given (~~ items are correctly retrieved repeatedly in practice with corrective feedback (a single correct retrieval is not enough), and no-feedback multiple-choice practice is ~~excluded).~~ excluded. At an immediate (~5 min) test, equal-time restudy can be superior. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, when practice accuracy is relatively low and no feedback is given, or when “testing” is high-stakes performance rather than practice.
Bounds field after this step (changes marked):
Main contrast is for delayed criterion tests, not immediate (~5 min) tests. Delayed contrast requires ~~adequate practice-test success and corrective feedback~~ repeated successful retrievals with corrective feedback (not one-shot success); no-feedback multiple-choice practice is excluded. Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated.
Attack #11 survived Gupta, Pan & Rickard 2026 · ledger date 28 September 2026
Attack: For highly semantically related cue–target materials, equal-time restudy yields delayed criterion retention at least as high as retrieval practice with corrective feedback after encoding.
Logged after the archived record ends; the wording at the time of this attack is not recorded.
Wording at this point: not recorded.
Attack #12 narrowed McCarter, Huber & Cowell 2025 · ledger date 28 September 2026
Attack: For novel, meaningless abstract visual materials lacking preexisting semantic knowledge, equal-time restudy yields criterion retention at least as high as retrieval practice with feedback, so the main contrast does not transfer to such materials.
This change happened after the archived record of earlier wordings ends; any intermediate wordings were not recorded.
After the material has been encoded, practicing retrieval produces more durable retention than spending the same time restudying when the criterion test is delayed, items are correctly retrieved repeatedly in practice with corrective feedback (a single correct retrieval is not enough), and no-feedback multiple-choice practice is excluded. At an immediate (~5 min) test, equal-time restudy can be superior. Once initial learning is equated, the claim does not assert that this advantage grows as final-test delay grows. The claim does not hold when encoding never succeeded, when the practice test is a poor match to what must later be remembered, when practice accuracy is relatively low and no feedback is given, or when “testing” is high-stakes performance rather than practice. The claim does not transfer to novel, meaningless abstract visual materials that lack preexisting semantic scaffolding.
Bounds field after this step (changes marked):
Main contrast is for delayed criterion tests, not immediate (~5 min) tests. Delayed contrast requires repeated successful retrievals with corrective feedback (not one-shot success); no-feedback multiple-choice practice is excluded. Delay-growth of the retrieval–restudy advantage is not claimed once initial learning is equated. High semantic relatedness of cue–target materials reduces relative efficacy of retrieval vs restudy (~26%) but does not erase the advantage (Gupta, Pan & Rickard 2026; kill-condition test survived 2026-09-28). Novel meaningless abstract visual materials lacking preexisting semantic scaffolding: McCarter, Huber & Cowell 2025 (six experiments) found restudy ≥ retrieval (sometimes significantly better) despite feedback and delays — outside the claim (cf. Ferreira & Wimber 2023 on meaningful vs low-semantic visual materials).

The current claim file differs from the last archived wording; the difference is shown as the final step.

Attack record

#1 narrowed 8 September 2026 · Roediger & Karpicke 2006 · target: claim-text
When initial learning is equated, the retrieval–restudy retention gap does not grow as final-test delay grows.
Strongest reading and verdict reasoning
Steelmans to: The delay-growth conjunct is false as a general rule; a larger late gap is not a reliable property of retrieval practice once encoding success is matched.
Why: Human re-apply as weaken. The delay-growth conjunct is dropped as a general assertion once initial learning is equated; main retrieval-vs-restudy contrast kept. Roediger & Karpicke 2006 still shows a delay-dependent pattern in their design, but the claim no longer states growth-with-delay as part of C0. Claim field and Bounds rewritten in the same turn; brief rewritten.
#2 survived 8 September 2026 · no named source (attacker's own inference) · target: bound
Outside lab word-list and short prose-passage paradigms, equal-time restudy yields delayed criterion retention at least as high as retrieval practice when encoding succeeded and practice matches the test.
Strongest reading and verdict reasoning
Steelmans to: C0’s main contrast is paradigm-bound; without an explicit transfer limit the claim overreaches educationally complex materials.
Why: Scope inference only — no identifiable counter-study applied in this turn showing equal or superior delayed restudy outside word-list/prose paradigms under encoding-succeeded and criterion-matched conditions. Did not force a transfer Bound; main-contrast kill-condition not met. Claim field unchanged. (The Bounds field later that day records only the delay-growth withdrawal from ledger line 1 — not this attack.)
#3 survived 8 September 2026 · Dunlosky et al. 2013 · target: bound
When encoding success is only assumed after a single study pass, a large unencoded item subset remains for which equal-time restudy yields delayed retention at least as high as retrieval practice, so the main contrast does not hold under the claim’s own encoding precondition.
Strongest reading and verdict reasoning
Steelmans to: C0’s retrieval advantage is conditional on verified encoding; treating one study exposure as “encoded” lets restudy of unlearned items erase the claimed contrast at delay.
Why: Evidence vs inference: Dunlosky et al. 2013 rate practice testing high utility and rereading low; they do not report equal-time restudy beating retrieval on verified-encoded items, nor a result that forces a new Bound beyond C0’s existing “encoding never succeeded” carve-out. Operational reading of a carve-out already in the claim; no status-changing named counter-result. Claim field unchanged; no new Bound written.
#4 survived 8 September 2026 · Bjork & Bjork 2020 · target: bound
When practice-test format or cues mismatch the criterion test, equal-time restudy yields delayed criterion retention at least as high as retrieval practice after encoding succeeded.
Strongest reading and verdict reasoning
Steelmans to: Transfer-appropriate processing / desirable-difficulty limits mean a “practice test” that does not match criterion fails to beat restudy at delay; the mismatch carve-out must bind typical surface-form mismatches, not only blatant content mismatch.
Why: Bjork & Bjork 2020 discuss desirable difficulties and the need for adequate prior knowledge so difficulty is not undesirable; they do not supply a named result that equal-time restudy ≥ retrieval under encoding-succeeded but format/cue-mismatched practice. C0 already carves out poor practice–criterion match. Tightening “surface-form” into Bounds would be inference beyond the cited work. Claim unchanged.
#5 survived 8 September 2026 · Roediger & Karpicke 2006 · target: kill-condition
When restudy is distributed across the same schedule as retrieval attempts, equal-time restudy yields delayed retention equal or superior to retrieval practice under encoding-succeeded and criterion-matched conditions.
Strongest reading and verdict reasoning
Steelmans to: The main contrast is an artifact of comparing retrieval to massed or single-bolus restudy; once restudy is schedule-matched, retrieval does not beat restudy at delay.
Why: Evidence vs inference: Roediger & Karpicke 2006 contrasts testing with repeated studying of prose passages across retention intervals; it is not a full schedule-matched distributed-restudy design that equates temporal distribution of restudy episodes with retrieval attempts. Whether that equated design removes the advantage is underdetermined from the seed alone. Kill-condition not met. Claim unchanged.
#6 survived 8 September 2026 · Adesope et al. 2017 · target: bound
For educationally complex materials (multi-step procedures, transfer and application tests), equal-time restudy yields delayed criterion retention at least as high as retrieval practice when encoding succeeded and practice matches the criterion.
Strongest reading and verdict reasoning
Steelmans to: The main contrast must be bounded to simpler retention paradigms; educationally complex criterion performance does not show a retrieval-over-restudy advantage under the claim’s own success conditions.
Why: Adesope et al. 2017 find practice testing beats restudy/rereading overall (g ≈ 0.51) and still show a positive testing effect on transfer outcomes (g ≈ 0.53, k = 11), smaller than pure retention but not restudy ≥ retrieval. The attack’s claim that complex/transfer criteria erase the advantage is not established by that meta; no new Bound written. Distinct force from the earlier vague outside-lab line, but status unchanged.
#7 narrowed 8 September 2026 · Roediger & Karpicke 2006 · target: bound
At an immediate (~5 min) criterion test after encoding succeeded and practice matched the criterion, equal-time restudy yields superior retention to retrieval practice.
Strongest reading and verdict reasoning
Steelmans to: The main contrast is false as an unrestricted claim about “more durable retention”; restudy beats retrieval at immediate test, so the contrast must be bounded to delayed criterion tests (without restoring delay-growth).
Why: Clerk Apply block 1 (cycle 1/5). Named result in Roediger & Karpicke 2006: after 5 min, repeated studying improved recall relative to repeated testing; delayed tests reversed. Claim and Bounds rewritten in the same turn so durable retention means delayed criterion, not immediate; delay-growth withdrawal unchanged. Brief rewritten.
#8 narrowed 8 September 2026 · Dunlosky et al. 2013 · target: bound
When accuracy on the practice test is relatively low and no corrective feedback is given, equal-time restudy yields delayed retention at least as high as retrieval practice under encoding-succeeded and criterion-matched conditions.
Strongest reading and verdict reasoning
Steelmans to: The living delayed main contrast overclaims; the robust delayed advantage is practice testing with feedback over restudy, not testing alone under low practice success.
Why: Clerk Apply block 1 (cycle 4/5). Dunlosky et al. 2013: testing alone sometimes fails to beat restudy when practice-test accuracy is relatively low; testing with feedback over restudy is extremely robust. Claim and Bounds rewritten same turn to require adequate practice success and corrective feedback for the delayed contrast. Brief rewritten.
#9 narrowed 8 September 2026 · Butler & Roediger 2008 · target: bound
Multiple-choice practice without feedback increases delayed intrusions of false alternatives relative to equal-time restudy, so delayed cued-recall accuracy is not superior for retrieval under encoding-succeeded conditions.
Strongest reading and verdict reasoning
Steelmans to: Without feedback, MC retrieval practice can harm delayed accuracy via perseveration of wrong alternatives; the delayed main contrast must exclude no-feedback MC practice or require corrective feedback.
Why: Clerk Apply block 2 (cycle 4/5). Butler & Roediger 2008 (as reported in Dunlosky et al. 2013): MC practice without feedback increased false-alternative intrusions on delayed cued recall; feedback prevented that increase. Claim and Bounds rewritten same turn to exclude no-feedback MC. Brief rewritten.
#10 narrowed 8 September 2026 · Dunlosky et al. 2013 · target: bound
A single correct retrieval with corrective feedback does not produce superior delayed retention to equal-time restudy under encoding-succeeded and criterion-matched conditions; the delayed advantage requires repeated successful retrievals with feedback.
Strongest reading and verdict reasoning
Steelmans to: “Adequate practice-test success” in the living claim is underspecified as one-shot success; named dosage evidence ties durable gains to multiple correct retrievals with feedback, so the main contrast must be Bounded to repeated successful retrieval practice.
Why: Clerk Apply block 1 (cycle 5/5). Dunlosky et al. 2013 synthesize dosage evidence (e.g. Vaughn & Rawson 2011): final performance rises with multiple correct retrievals in practice vs once. Claim and Bounds rewritten same turn so adequate success means repeated correct retrievals with feedback. Brief rewritten. Block 2 (Adesope mixed retention+transfer) rejected — no ledger line.
#11 survived 28 September 2026 · Gupta, Pan & Rickard 2026 · target: kill-condition
For highly semantically related cue–target materials, equal-time restudy yields delayed criterion retention at least as high as retrieval practice with corrective feedback after encoding.
Strongest reading and verdict reasoning
Steelmans to: Kill-condition on the delayed main contrast fails if high relatedness erases the retrieval–restudy gap under feedback.
Why: Nightly-attacks 2026-09-28 04:16 Europe/Stockholm. Named result: testing improved 24-h cued recall in both low and high relatedness groups; relative efficacy ~26% smaller for high relatedness (dual-memory model + cumulative distributions) but restudy did not equal or beat retrieval-with-feedback. Kill-condition not met. Clause tested: high semantic relatedness under feedback. Claim field unchanged; Bounds note Gupta kill-test survived.
#12 narrowed 28 September 2026 · McCarter, Huber & Cowell 2025 · target: bound
For novel, meaningless abstract visual materials lacking preexisting semantic knowledge, equal-time restudy yields criterion retention at least as high as retrieval practice with feedback, so the main contrast does not transfer to such materials.
Strongest reading and verdict reasoning
Steelmans to: C0’s delayed retrieval-over-restudy contrast is material-bound; without an explicit transfer limit it overreaches asemantic novel visuals where TE is absent or reversed.
Why: Nightly-attacks 2026-09-28 04:16 Europe/Stockholm. Named multi-experiment result: across six experiments restudy ≥ retrieval for novel meaningless visual objects (sometimes significantly better) despite feedback, delays, and heavy-handed practice variants. Complementary: Ferreira & Wimber 2023 (DOI 10.1037/xlm0001248) found TE for meaningful objects but not for low-semantic visual materials. Does not kill the linguistic/prose main contrast. Claim, Bounds, and Does-not-transfer-to rewritten same turn; brief rewritten.

Evidence

12 linked sources: 7 supports, 5 limits, 2 challenges (first-pass draft, pending human review). supports: Adesope et al. 2017, Dunlosky et al. 2013, Roediger & Karpicke 2006, Weinstein et al. 2018, Gupta et al. 2026, Gupta et al. 2022, Racsmány et al. 2020 · limits: Bjork & Bjork 2020, Butler & Roediger 2008, Dunlosky et al. 2013, Roediger & Karpicke 2006, McCarter et al. 2025 · challenges: Kornell 2025, Chan et al. 2024

Full claim record in the evidence library →