Candidate claims
Not approved. Each candidate is built from one "This claim dies if …" sentence shared by one or more Kunskapsbank notes. The kill sentence is quoted in Swedish, verbatim; the claim sentence and English kill condition are first drafts. Candidates have no ledger and are not in the brief. Only Jonatan promotes a candidate. K01 and K12 were promoted on 2026-09-25 as C3 and C2.
K02 candidate: not approved
Giving every pupil their own computer or tablet (1:1) does not raise average knowledge results in Swedish schools under realistic use.
Serves: Q2 · Ages: foralder-tidig-barndom, grundskola-hog, grundskola-lag, gymnasium, overgripande
What would kill this: A Swedish/Nordic difference-in-differences study or RCT in which 1:1 devices under realistic use raise closed-book knowledge without more distraction.
Kill sentence in the bank (Swedish): Svensk/nordisk DiD eller RCT där 1:1-enheter under realistisk användning höjer closed-book-kunskap utan ökad distraktion.
Overlap: none noted
8 linked sources (6 supports, 2 limits)
Supports (6)
Limits (2)
Review state: first-pass draft, pending human review.
K03 candidate: not approved
Taking notes by hand gives better conceptual, closed-book learning than typing, when the note-taking instructions are the same.
Serves: Q1+Q2 · Ages: foralder-tidig-barndom, grundskola-hog, grundskola-lag, gymnasium, overgripande, vuxen-livslangt
What would kill this: A preregistered replication in which keyboard is at least as good as pen on conceptual/closed-book outcomes in the same population with note-taking instructions held constant.
Kill sentence in the bank (Swedish): Preregistrerad replikering där tangentbord ≥ penna på conceptual/closed-book-utfall i samma population när anteckningsinstruktion hålls konstant.
Overlap: C1 (the bank frames this as "conceptual encoding"); K24
7 linked sources (2 supports, 4 limits, 1 challenges)
Supports (2)
Limits (4)
Challenges (1)
Review state: first-pass draft, pending human review.
K04 candidate: not approved
More evening screen use worsens sleep and next-day retention, even after controlling for confounders.
Serves: Q1+Q2 · Ages: foralder-tidig-barndom, gymnasium, overgripande
What would kill this: Longitudinal or RCT evidence in which more evening screen use does not worsen sleep or next-day retention after confounders are controlled.
Kill sentence in the bank (Swedish): Longitudinell/RCT-evidens där ökad kvällsskärm inte försämrar sömn eller nästa-dag-retention efter confound-kontroll.
Overlap: none noted
7 linked sources (7 supports)
Supports (7)
Review state: first-pass draft, pending human review.
K05 candidate: not approved
For beginning readers at risk, systematic phonics gives better decoding and comprehension than "balanced" or non-systematic approaches.
Serves: Q1 · Ages: foralder-tidig-barndom, grundskola-hog, grundskola-lag, overgripande
What would kill this: Large-scale evidence that non-systematic/"balanced" decoding instruction without phonics matches systematic phonics for at-risk early readers on decoding and comprehension.
Kill sentence in the bank (Swedish): Storskalig evidens att icke-systematisk/‘balanced’ avkodning utan phonics matchar systematisk phonics för at-risk early readers på decoding+comprehension.
Overlap: none noted
7 linked sources (7 supports)
Supports (7)
Review state: first-pass draft, pending human review.
K06 candidate: not approved
Retrieval practice, spacing and interleaving beat rereading or massed practice on delayed closed-book tests in real classroom time.
Serves: Q1 · Ages: overgripande
What would kill this: A meta-analysis in which retrieval/spacing/interleaving lose to rereading/massed practice in realistic classroom time with a delayed closed-book test.
Kill sentence in the bank (Swedish): Meta där retrieval/spacing/interleaving förlorar mot reread/massed under ekologisk klassrumstid med delayed closed-book.
Overlap: C0 (retrieval part). The draft recommends folding the retrieval part into C0 and keeping spacing and interleaving as separate candidates if wanted
6 linked sources (5 supports, 1 limits)
Supports (5)
Limits (1)
Review state: first-pass draft, pending human review.
K07 candidate: not approved
Working-memory training does not produce robust transfer to school subjects that survives placebo control and delayed testing.
Serves: Q1 · Ages: foralder-tidig-barndom, grundskola-hog, grundskola-lag, gymnasium, overgripande, vuxen-livslangt
What would kill this: A preregistered meta-analysis in which working-memory training gives robust far transfer to school subjects that survives placebo control and a delayed test.
Kill sentence in the bank (Swedish): Preregistrerad meta där WM-träning ger robust far-transfer till skolämnen som överlever placebo och delayed test.
Overlap: none noted
6 linked sources (6 supports)
Supports (6)
Review state: first-pass draft, pending human review.
K08 candidate: not approved
Feedback about the task and the process helps delayed learning in school more than generic or ego-focused feedback.
Serves: Q1 · Ages: grundskola-hog, grundskola-lag, gymnasium, overgripande, vuxen-livslangt
What would kill this: A meta-analysis in which generic/ego feedback systematically beats task/process feedback on delayed learning in K–12.
Kill sentence in the bank (Swedish): Meta där generisk/ego-feedback systematiskt slår uppgifts-/processfeedback på delayed learning i K–12.
Overlap: none noted
6 linked sources (6 supports)
Supports (6)
Review state: first-pass draft, pending human review.
K09 candidate: not approved
Swedish national measurements (PISA, PIRLS, inspection and inquiry reports) show falling levels and widening gaps, and the practice they recommend is linked to closed-book results.
Serves: Q2 · Ages: grundskola-hog, grundskola-lag, overgripande
What would kill this: A new representative Swedish measurement that reverses the reports' picture of levels and gaps and shows that the recommended practice has no link to closed-book outcomes.
Kill sentence in the bank (Swedish): Ny representativ svensk mätning som vänder rapportens nivå-/gapbild och visar att rekommenderad praktik saknar samband med closed-book-utfall.
Overlap: This is a bundle of Swedish situation reports; probably not one testable claim
6 linked sources (6 supports)
Supports (6)
Review state: first-pass draft, pending human review.
K10 candidate: not approved
In the early school years, writing instruction that includes explicit handwriting/fluency training gives better composition than instruction without it.
Serves: Q1 · Ages: foralder-tidig-barndom, grundskola-lag, overgripande
What would kill this: A meta-analysis in which early writing instruction without a handwriting/fluency component matches explicit handwriting instruction on composition outcomes.
Kill sentence in the bank (Swedish): Meta där tidig skrivundervisning utan handstil/fluency-komponent matchar explicit handskriftsundervisning på composition outcomes.
Overlap: none noted
5 linked sources (5 supports)
Supports (5)
Review state: first-pass draft, pending human review.
K11 candidate: not approved
Free access to phones or laptops during lessons lowers learning for the pupil (and for neighbours) compared with a ban or a phone box.
Serves: Q2 · Ages: grundskola-hog, gymnasium, overgripande
What would kill this: A field RCT or difference-in-differences study in which free phone/laptop access during lessons does not lower learning for the pupil (or neighbours) compared with a ban/box.
Kill sentence in the bank (Swedish): Fält-RCT/DiD där fri telefon-/laptopaccess under lektion inte sänker lärande för eleven (och grannar) relativt förbud/låda.
Overlap: K14, K18, K27, K29
3 linked sources (1 supports, 1 limits, 1 challenges)
Supports (1)
Limits (1)
Challenges (1)
Review state: first-pass draft, pending human review.
K13 candidate: not approved
Seductive details and extraneous multimedia lower transfer, for the same instruction time.
Serves: Q1+Q2 · Ages: overgripande
What would kill this: A replication in which seductive details/extraneous multimedia raise transfer for the same instruction time.
Kill sentence in the bank (Swedish): Replikering där seductive details/extraneous multimedia höjer transfer under samma instruktionstid.
Overlap: none noted
3 linked sources (3 supports)
Supports (3)
Review state: first-pass draft, pending human review.
K14 candidate: not approved
School phone bans that are actually enforced (phones handed in) raise girls' grades and maths results and reduce bullying.
Serves: Q2 · Ages: grundskola-hog, overgripande
What would kill this: A Swedish register difference-in-differences study with the same gender/SES split showing no effect on girls' GPA/maths and no reduction in bullying after an actual hand-in policy (not just a ban on paper).
Kill sentence in the bank (Swedish): Svensk register-DiD med samma köns-/SES-split visar nolleffekt på flickors GPA/matte och ingen mobbningsminskning efter faktisk inlämningspolicy (inte bara pappersförbud).
Overlap: K11
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K15 candidate: not approved
Unprotected answer-giving GPT harms upper-secondary closed-book learning, while a tutor version that withholds answers is pedagogically different and mitigates the harm.
Serves: Q2 · Ages: grundskola-hog, gymnasium
What would kill this: A replicated upper-secondary RCT in which unprotected GPT Base does not harm closed-book results compared with control and the Tutor arm no longer differs pedagogically from Base.
Kill sentence in the bank (Swedish): Replikerad gymnasie-RCT där oskyddad GPT Base inte skadar closed-book relativt kontroll och Tutor-armen inte längre skiljer sig i pedagogik från Base.
Overlap: C2 (this single-study sentence is now part of C2's kill condition and bounds)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K16 candidate: not approved
ChatGPT use improves student learning (Deng et al. 2025 meta-analysis).
Serves: Q2 · Ages: gymnasium, overgripande, vuxen-livslangt
What would kill this: A bias-corrected analysis using only delayed closed-book outcomes gives zero or negative effect.
Kill sentence in the bank (Swedish): Bias-korrigerad analys med endast delayed closed-book ger noll/negativ effekt.
Overlap: C2 (opposite direction; linked to C2 as a challenge)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K17 candidate: not approved
ChatGPT raises learning achievement (Doo & Park 2026 meta-analysis).
Serves: Q2 · Ages: overgripande
What would kill this: The positive effect disappears when analyses are restricted to closed-book retention.
Kill sentence in the bank (Swedish): När analyser begränsas till closed-book retention försvinner positiv g.
Overlap: C2 (opposite direction; linked to C2 as a challenge)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K18 candidate: not approved
Sweden's proposed national school phone collection will reduce actual phone use in lessons and improve calm and safety.
Serves: Q2 · Ages: grundskola-hog, grundskola-lag, overgripande
What would kill this: A national follow-up after entry into force shows unchanged actual phone use in lessons and no improvement in calm/safety indicators.
Kill sentence in the bank (Swedish): Nationell uppföljning efter ikraftträdande visar oförändrad faktisk mobilanvändning i lektion och ingen förbättring i studiero/trygghetsindikatorer.
Overlap: K11
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K19 candidate: not approved
GenAI improves learning outcomes (Fan et al. 2026 meta-analysis, higher education).
Serves: Q2 · Ages: vuxen-livslangt
What would kill this: K–12 meta-analyses with mandatory closed-book follow-up show about zero or negative retention despite positive practice performance.
Kill sentence in the bank (Swedish): K–12-metas med obligatorisk closed-book follow-up visar g≈0 eller negativ retention trots positiv practice-performance.
Overlap: C2 (opposite direction; linked to C2 as a challenge)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K20 candidate: not approved
For children aged 1–8, plain digitisation of a story lowers comprehension, while story-congruent digital enhancements can help.
Serves: Q1+Q2 · Ages: foralder-tidig-barndom, grundskola-lag
What would kill this: A preregistered meta-analysis of ages 1–8 showing no effect of plain digitisation on comprehension and no harm from incongruent hotspots.
Kill sentence in the bank (Swedish): Preregistrerad meta på 1–8 år visar null för plain digitization på comprehension och att incongruent hotspots inte skadar.
Overlap: C3 (used as a bound on C3)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K21 candidate: not approved
Undesigned, free GenAI use does not systematically raise closed-book learning in K–12 without AI-literacy requirements (position of Kasneci et al. 2023).
Serves: Q2 · Ages: grundskola-hog, grundskola-lag, gymnasium, overgripande
What would kill this: Large-scale evidence that undesigned free GenAI systematically raises closed-book K–12 learning without literacy requirements.
Kill sentence in the bank (Swedish): … (position falsifieras inte med en p-värde): om storskalig evidens visar att odesignad fri GenAI systematiskt höjer closed-book K–12 utan litteracitetskrav, faller paperets riskvikt — empirin avgörs av AI-003/009/metas med closed-book-flagga.
Overlap: C2 (position paper, not an effect estimate)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K22 candidate: not approved
Intelligent tutoring systems raise K–12 learning compared with business as usual, but effects on standardised tests are much smaller than on local tests.
Serves: Q2 · Ages: overgripande
What would kill this: A replicated meta-analysis with strict standardised outcomes and preregistration showing a median ITS effect of about zero against business as usual in K–12, regardless of alignment.
Kill sentence in the bank (Swedish): Replikerad meta med strikta standardiserade outcome och preregistrering visar median ITS-effekt ≈0 mot BAU i K–12, oberoende av alignment.
Overlap: K23, K25, K26, K28 (tutoring-software cluster)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K23 candidate: not approved
Intelligent tutoring systems outperform non-ITS computer-based instruction on standardised K–12 measures when content coverage is equal.
Serves: Q2 · Ages: overgripande
What would kill this: A new comprehensive meta-analysis (post-2020, including GenAI-as-ITS claims) showing ITS no better than non-ITS computer-based instruction on standardised K–12 measures when both have the same content coverage.
Kill sentence in the bank (Swedish): Ny comprehensive meta (post-2020, inkl. GenAI-as-ITS-claim) visar ITS ≤ non-ITS CBI på standardiserade K–12-mått när båda har samma content coverage.
Overlap: K22, K25, K26, K28
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K24 candidate: not approved
Paraphrasing notes by hand gives an encoding advantage over typed verbatim notes.
Serves: Q1+Q2 · Ages: gymnasium, vuxen-livslangt
What would kill this: A large preregistered K–12 replication showing no encoding advantage for handwriting/paraphrase, and the early-years letter-circuit findings (Longcamp/James) rejected.
Kill sentence in the bank (Swedish): Storskalig preregistrerad K–12-replikation visar ingen encodingfördel för handskrift/parafras och tidiga-års-motor/letter-circuit-fynd (Longcamp/James) underkänns — då faller även “penna för begrepp”-tumregeln från just denna linje.
Overlap: K03; C1
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K25 candidate: not approved
A guard-railed AI tutor inside a mastery requirement gives better delayed closed-book learning than the same platform without AI.
Serves: Q2 · Ages: grundskola-hog, overgripande
What would kill this: A replicated K–12 RCT in which a guard-railed LLM without mastery beats computer-assisted learning on delayed closed-book, or in which mastery+AI no longer differs from computer-assisted learning on practised retention.
Kill sentence in the bank (Swedish): Replikerad K–12-RCT där guard-railed LLM utan mastery slår CAL på delayed closed-book eller där mastery+AI inte längre skiljer sig från CAL på practiced retention.
Overlap: C2 (a limit on C2); K22, K23, K26, K28
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K26 candidate: not approved
A coaching AI (Khanmigo) adds to standardised maths only when pupils actually engage in substantive dialogue with it.
Serves: Q2 · Ages: grundskola-hog, overgripande
What would kill this: A K–12 RCT in which coach-AI with a high substantive dialogue rate (at least the median pupil) still has no effect compared with the same platform without AI on standardised maths.
Kill sentence in the bank (Swedish): K–12-RCT där coach-AI med hög substantiv dialog-rate (≥ median elev) fortfarande ger nolleffekt vs samma platform utan AI på standardiserad math.
Overlap: K22, K23, K25, K28
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K27 candidate: not approved
Laptop multitasking in class lowers comprehension for the user and for peers who can see the screen.
Serves: Q2 · Ages: gymnasium, overgripande, vuxen-livslangt
What would kill this: A Swedish 1:1 study with realistic multitasking showing no peer spillover when screens are visible during instruction, and no effect on the user's closed-book result.
Kill sentence in the bank (Swedish): Svensk 1:1-studie med ekologisk multitasking visar ingen peer-spillover när skärmar är synliga under genomgång och användareffekten uteblir på closed-book.
Overlap: K11
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K28 candidate: not approved
Well-designed step-based tutoring systems come close to human tutoring, and answer-giving GenAI without guardrails does not match them on closed-book outcomes.
Serves: Q2 · Ages: overgripande
What would kill this: Large K–12 meta-analyses showing that designed step-based ITS systematically underperform human tutoring by more than 0.3 d, and that answer-giving GenAI matches ITS on closed-book outcomes without guardrails.
Kill sentence in the bank (Swedish): Storskaliga K–12-metas visar att designad step-based ITS systematiskt underpresterar human tutoring med >0,3 d och att answer-giving GenAI matchar ITS på closed-book utan guardrails.
Overlap: K22, K23, K25, K26; C2
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K29 candidate: not approved
The mere presence of one's own smartphone, even silent in a pocket, lowers available working memory, so physical separation matters.
Serves: Q2 · Ages: gymnasium, overgripande, vuxen-livslangt
What would kill this: A preregistered Swedish lower/upper-secondary replication with 1:1 showing no effect of physical separation versus "silent in the pocket" on working memory and closed-book encoding during lessons.
Kill sentence in the bank (Swedish): Preregistrerad svensk högstadie-/gymnasiereplikation med 1:1 visar nolleffekt av fysisk separation vs “tyst i fickan” på WM och closed-book encoding under lektion.
Overlap: K11; C1 (conditions for encoding)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.
K30 candidate: not approved
AI chatbots improve students' learning outcomes (Wu & Yu 2024 meta-analysis).
Serves: Q2 · Ages: grundskola-hog, gymnasium, overgripande, vuxen-livslangt
What would kill this: The effect disappears in K–12 closed-book testing after the novelty window.
Kill sentence in the bank (Swedish): Effekten försvinner i K–12 closed-book efter novelty-fönstret.
Overlap: C2 (opposite direction; linked to C2 as a challenge)
1 linked source (1 supports)
Supports (1)
Review state: first-pass draft, pending human review.