Skip to content

Latest commit

 

History

History
527 lines (410 loc) · 45.1 KB

AIED-BACKLOG.md

File metadata and controls

527 lines (410 loc) · 45.1 KB

AI-in-Education Article Backlog

Relevant open-access AI-in-education journal articles identified from RSS feeds that have not yet been ingested into the wiki. The maintainer downloads the PDFs from the linked article pages and sends them for full-text ingestion. Sorted by journal.

Policy (2026-08-23): an article is only added to this backlog (never ingested as an abstract-only wiki page) when its full text cannot be retrieved — e.g. the publisher blocks scraping with CAPTCHA, or the fetch times out. The maintainer downloads the PDF and sends it for full-text ingestion.

Scope of this file: it lists work in progress only — articles whose full text is still being sought, and pages waiting on a source before their practice and limitations sections can be written. Candidates appear here only while they are being pursued, and no outcome is recorded; items judged unsuitable for the knowledge base are tracked outside this repository.

Last updated: 2026-11-19 Total backlog: 29 articles awaiting ingest · 52 truncated source texts awaiting PDFs · 19 pages awaiting a full text before their practice and limitations sections can be written

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-19)

Ingested seven papers from full text: five Frontiers journal articles, one Discover Education case study, and one arXiv preprint. Three records cleared the significance screen but their full text could not be retrieved — the publishers answer automated requests with a bot wall — so they await a supplied PDF:

  • Motivational gain trumps cognitive offloading: a MASEM study on the net positive effect of AI use on critical thinking (Smart Learning Environments, 2026) — DOI: 10.1186/s40561-026-00477-7 — Springer Nature Link served the abstract and reference list only.
  • Reconstruction and practice of an AI-enhanced multifactorial teaching model for virology laboratory testing (BMC Medical Education, 2026) — DOI: 10.1186/s12909-026-10835-1 — abstract page only, PDF endpoint blocked.
  • AI-Powered Conversational Interfaces in Higher Education: An Experimental Comparison of Chatbots and Traditional Websites for Information Seeking (Education Sciences, 2026) — DOI: 10.3390/educsci16111966 — MDPI returned Access Denied.

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-18)

Ingested nine papers (arXiv, EdArXiv, and five Frontiers journal articles). Four candidates carried forward from the same window were all resolved on 2026-11-18: two ingested from PDFs the maintainer supplied, two declined.

Ingested from full text once the PDFs arrived (both publisher-blocked on the scan day):

  • Artificial intelligence–based clinical simulation in undergraduate medical education — scoping review of 30 studies — ingested as hernandez-ai-clinical-simulation-scoping-2026.
  • Artificial Intelligence in Middle and Secondary Education — perspective with policy recommendations — ingested as stefanile-middle-secondary-ai-guidance-2026.

A few sources considered for ingestion were not accepted by the maintainer; the decision is recorded privately rather than in this file.

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-16)

Five BMC Medical Education early-access records needed publisher PDFs. Resolved on 2026-11-16: three were supplied and ingested as ago-prompt-literacy-allied-health-2026, tischkau-ai-simulated-patient-formats-2026 and islam-human-vs-ai-sba-dental-2026; the other two were not accepted; the decision is recorded privately rather than in this file.

OpenAlex weekly topic sweep (2026-11-15)

The topic sweep returned 395 candidates for the 7 days to 2026-11-15 (25 dropped as already known or out of scope). Five were selected as worth ingesting — a mapping review of randomized evidence, two systematic reviews in disciplines the corpus covers thinly, a scoping review of adaptive and co-regulation designs, and a controlled study of LLM scaffolds and delayed cost — and none was ingested this week, because the run's full-text budget went to the eight RSS articles above. All five are open access at their landing pages.

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-12)

The prefetch returned 93 candidates in the 2026-11-09..12 window and dropped 28 already-known items. Six had been ingested in the preceding three days (the PISA 2025 scientific-literacy analysis, the Quebec post-secondary AI-literacy survey, the career decision-making self-efficacy SEM, the Jordanian EFL writing trial, the Egyptian extended-TAM study and the generative-operationalism paper), so they were left alone and the prefetch's title-based dedupe missed them only because their slugs differed. Eight were ingested from full text by this run: four Frontiers articles, two MDPI Education Sciences papers and two Discover Education papers.

The remainder of the arXiv CS listing was model training, retrieval, decoding, benchmarking and medical-imaging work with no educational setting, and the physics-education feed returned nothing in the window. The mini review on AI-mediated interpersonal communication was left out as out of scope rather than backlogged: its experiments are about relational authenticity in human communication, not about learning.

Queued for a later run (full text already retrieved or open access)

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-11)

The prefetch returned 88 candidates and dropped 21 already-known items. Seventeen had already been ingested earlier today (the Stanford K-12 update batch, the sustainable higher-ed GenAI review, the medical-education governance op-ed, six EdArXiv/arXiv preprints, and the GPT-5 kinematics study). Seven were ingested from full text by this run: one MDPI randomized trial in EFL writing, three arXiv preprints and three Frontiers articles. The first two entries below were listed on 2026-11-10 and were ingested today (career self-efficacy, Silinx), so they are removed from that section rather than carried again.

The rest of the arXiv CS listing was model-training, retrieval, benchmarking and medical-imaging work with no educational setting, and the physics education feed returned one study with no AI intervention (active learning at a graduate summer school) - out of scope rather than backlog.

None outstanding from this scan. The three items listed here on 2026-11-11 were closed the same day: the early-childhood storytelling study was ingested from the maintainer's PDF, and the other two were closed without ingestion.

Adjacent research awaiting a tier decision

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-10)

The prefetch returned 76 candidates and dropped 21 already-known items. Ten were ingested from full text this run (3 arXiv preprints, 2 EdArXiv preprints, 4 Frontiers articles, 1 MDPI article). Eight more cleared relevance but were already in the wiki under an earlier slug (GenAI use-and-caution trajectories, the UK StudentXGenAI survey, the pham tutor-or-shortcut review, AI4Design, executor-to-partner, school-community ecosystems, the AI integrity crisis, the certification-and-consultation paradigm, the Scandinavian TALIS analysis and the AI-enabled VR STEM bibliometric). HAIML (10.35542/osf.io/qhjgs) surfaced again and already has two pages; nothing was written. The three below cleared relevance but were not reached inside the cap, or carry no intervention whose full text is reachable.

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-09)

The prefetch returned 75 candidates and dropped 21 already-known items. Twelve were ingested from full text this run (2 EdArXiv-backed arXiv preprints, 5 EdArXiv preprints, 5 Frontiers open-access articles). Eleven more cleared relevance but were not reached inside the 20-article cap; nine of those were subsequently ingested from full text, leaving the two below. Both stay listed because neither can be ingested as an article: one is blocked at the publisher, and one carries no AI intervention at all.

Not an AI-in-education study (flagged, no page)

Weekly journal RSS scan (2026-11-08)

The RSS sweep checked 180 feed items: 118 already had pages, 62 were new, and 17 of those cleared the AI-in-education relevance filter. Four were ingested this run from full text (two journal articles, two arXiv papers the daily scan had not reached); the 6 below are in scope but their full text could not be retrieved — ScienceDirect, Wiley and Springer all serve automated requests a bot check or an HTML shell instead of the article.

Computers and Education: Artificial Intelligence (CAEAI)

Computers and Education Open (CEAO)

British Journal of Educational Technology (BJET)

OpenAlex weekly topic sweep (2026-11-08)

The topic sweep returned 359 candidates for the 7 days to 2026-11-08 (21 dropped as already known). Six in-scope reviews and meta-analyses were selected; three are listed here rather than ingested: the OSF preprints serve a 245-byte error page to automated downloads, and the two paywalled items have no open full text. All six are open access at their landing pages, so a browser session or a maintainer-supplied PDF would unblock any of them.

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-06)

Twelve articles were ingested this run (five arXiv, two PsyArXiv, five OpenAlex journal-watch), each with full text saved under raw/papers/. Five more in-scope candidates cleared the relevance screen but fell outside the 20-article cap; full text was not saved for them, so a later run can pick them up:

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-03)

Twelve articles were ingested this run (three arXiv, nine OpenAlex journal-watch), each with full text saved under raw/papers/. Two in-scope candidates cleared the relevance screen but could not be ingested because Taylor & Francis serves them to automated requests as an empty landing page (435 characters of navigation chrome, no body). They are open access, so a browser session or a maintainer-supplied PDF would unblock either:

Daily arXiv + EdArXiv + OpenAlex scan (2026-11-02)

Ten articles were ingested this run (three arXiv, five OpenAlex journal-watch, one PsyArXiv-adjacent Frontiers and one Springer), each with its full text saved under raw/papers/. Two in-scope candidates could not be ingested because the publisher serves them only as HTML metadata or behind a bot check.

Both were resolved on 2026-11-02 when the maintainer cleared this queue: the residency-curriculum interviews were ingested from a maintainer-supplied PDF (bursic-ai-residency-education-curriculum-2026), and the endodontic-item classification study was not ingested and is recorded in the private exclusion list.

Weekly journal RSS scan (2026-11-01)

Five articles were ingested this run with full text saved under raw/papers/. Two in-scope candidates were held back for a significance screen, and both have since been ingested: hooshyar-responsible-trustworthy-educational-data-mining-2026 (the methods comparison) and xian-process-visualization-feedback-news-translation-2026 (the news-translation study).

OpenAlex harvest (2026-11-01)

The weekly topic sweep returned 529 candidates across 13 queries. Six were judged worth pursuing rather than ingested this run, preferring randomized trials, meta-analyses and reviews over single-site perception surveys. Two of those six were set aside on 2026-11-01 and now sit in the private exclusion list: an OSF record that is a study registration rather than a completed analysis, and a Cambridge repository report held under access restriction until 2029-10-30. Four remained before this queue was cleared, and all four have since been ingested: yahya-generative-ai-inquiry-based-learning-review-2026, pan-generative-ai-student-motivation-sdt-review-2026, kharbach-k12-ai-literacy-frameworks-scoping-review-2026 and kim-overtrust-ai-literacy-2026.

Daily arXiv + EdArXiv + OpenAlex scan (2026-10-30)

Nine articles were ingested this run (eight arXiv, one PsyArXiv), each with its full text saved under raw/papers/. All four candidates that fell outside the run's working budget of nine have since been ingested: jingru-fan-agentic-university-collaborative-agents-2026, taufer-human-ai-interaction-haidl-2026, lei-fan-genai-prompting-strategies-2026 and gunay-ozden-process-genre-ai-writing-2026. Each has its full text and PDF archived under raw/papers/ and pdf-sources/. Nothing from this window is outstanding.

Daily arXiv + OpenAlex scan (2026-10-27)

In-scope AI-in-education candidates from this run that were not ingested: the run's cap was eight articles, chosen for measured outcomes over self-report. All are open access; the arXiv items have retrievable full text, so any can be ingested on request.

Daily arXiv + EdArXiv + OpenAlex scan (2026-10-29)

Four articles were ingested this run — three from open-access publisher full text and one from an EdArXiv docx — and every full text is saved under raw/papers/. No in-scope candidate from this window is awaiting a full text. The remaining candidates were either already ingested by the 2026-10-27 and 2026-10-28 runs or outside AI-in-education scope (arXiv machine-learning systems, agent reinforcement learning, robotics and rendering papers, and general-attitudes surveys with no educational setting).

Daily arXiv + OpenAlex scan (2026-10-28)

Twelve articles were ingested this run (9 arXiv, 2 EdArXiv, 1 Frontiers). One in-scope candidate is here only because its full text is not public yet — the publisher serves an early-access abstract page, not the article.

Weekly journal RSS scan (2026-10-11)

Every other candidate from this scan was set aside by the maintainer on 2026-10-27. Those decisions live in the local private exclusion list, which the scan reads so the same papers are not proposed again.

Weekly journal RSS scan (2026-10-25)

Every other candidate from this scan was set aside by the maintainer on 2026-10-27. Those decisions live in the local private exclusion list, which the scan reads so the same papers are not proposed again.

OpenAlex harvest (2026-09-25)

Every other candidate from this scan was set aside by the maintainer on 2026-10-27. Those decisions live in the local private exclusion list, which the scan reads so the same papers are not proposed again.

OpenAlex harvest (2026-09-29)

Every other candidate from this scan was set aside by the maintainer on 2026-10-27. Those decisions live in the local private exclusion list, which the scan reads so the same papers are not proposed again.

Ingested pages whose source text is truncated (PDF needed)

These 52 pages are already in the knowledge base, but their saved full text stops at the old 50,000-character extraction cap, so the results, discussion and limitations sections are missing locally. Publisher bot-protection blocks automated downloads (MDPI answers 403, Springer serves a robot page), so these need the PDF provided directly; then re-extract with pdftotext at the current 250,000-character cap and enrich the page from the recovered text.

Elsevier / ScienceDirect

DOI (publisher paywall or unresolved)

ASEE conference proceedings

  • Artificial Intelligence (AI) and the Future of the Engineering and Computing Workforce: A Systematic Review of Grey Literature and Document Analysis of U.S. Reports (2020–2025) — ai-engineering-computing-workforce-grey-literature-2026 (not attempted)
  • It's Like "X": How Engineering Faculty Metaphors Construct (and Constrain) AI Understanding in Engineering Education — engineering-faculty-metaphors-ai-understanding-2026 (not attempted)
  • Ethical Use of Artificial Intelligence in Engineering Education: A Systematic Review — ethical-use-ai-engineering-education-review-2026 (not attempted)

NBER working papers

  • Making AI Tutoring Productive: Evidence from a Mastery-Based Math Practice Experiment — making-ai-tutoring-productive-mastery-math-2026 (not attempted)
  • One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment — one-click-away-khanmigo-two-year-school-experiment-2026 (not attempted)
  • Virtual Tutoring with Computer-Assisted Learning: An Experiment in Take-Up and Learning — virtual-tutoring-computer-assisted-learning-takeup-2026 (not attempted)

SAGE

onlinelearningconsortium.org

  • Utilizing Generative AI to Counter Learner Groupthink by Introducing Controversy in Collaborative Problem Based Learning Settings — genai-counter-learner-groupthink-2025 (not attempted)

papers.ssrn.com

unresolved

  • Fostering collaborative futures: Multidisciplinary approaches to AI integration in educational ecosystems — fostering-collaborative-futures-ai-ecosystems-2026 (html_failed)
    • 10.1016/j.caeo.2026.100405

Pages awaiting a full text (practice and limitations sections blocked)

These pages are already in the knowledge base, but their saved source text is missing, so their What this means for practice and Limitations sections cannot be written from evidence. Retrieval was attempted on 2026-09-19: each page's title was matched against OpenAlex and an open-access PDF requested via the recorded DOI and Unpaywall. Add the PDF here (or re-run the fetch once access changes), extract with pdftotext, save under the path the page's sources: field names, and the sections can then be written.

Open-access PDF could not be downloaded

No confident match found for the page title

  • Creating an AI-powered platform for generating modelling problems: A case study on direct variation in secondary school — ai-modeling-problem-generation-platform-2026 (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Causal Modelling of Support Interventions for Student Competency Assessment — causal-modeling-competency-assessment-2026 (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Hybrid E-Assessment in Higher Education: Semi-Automated Grading of Paper-Based Written Examinations — hybrid-e-assessment-semi-automated-grading (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Exploring value judgements in grading: will teachers mark down student work assisted by GenAI, and should they? — luo-dawson-value-judgments-grading-2026 (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Persistent AI Agents in Academic Research: A Single-Investigator Implementation Case Study — persistent-ai-agents-academic-research (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Preferred Scaffolding Does Not Lead to Better Learning Performance: Empirical Evidence from AI-Supported Mathematical Modelling — preferred-scaffolding-ai-mathematical-modeling (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • Reshaping Undergraduate Computer Science Education in the Generative AI Era — reshaping-cs-education-genai (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • How Students (Mis)understand Conditionals and Loops -- A Taxonomy — student-misconceptions-conditionals-loops-taxonomy (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page
  • How university students work on assessment tasks with generative AI: matters of judgement — walton-bearman-assessment-judgment-2025 (no OpenAlex match at the 0.85 title-similarity bar)
    • no source URL recorded on the page

Source file contained a different paper - resolved

  • ai-divide-ses-personality-primary-education-2026 - Decoding divides: The role of socioeconomic status and personality traits in AI divides and educational inequality (Wang et al.), doi 10.1016/j.caeai.2026.100566. The saved source carried this paper's frontmatter but another article's body, so nothing on the page could be grounded. The correct PDF was supplied directly, its text is now the page's saved source, and the page's limitations section was written from the paper's own stated limits.

    The mismatch turned out to be one of ten in the same ingestion batch: the saved bodies had been rotated among the files, so nine further pages were reading a neighbouring paper as their source. Each was matched to its own paper by title and the bodies were put back where they belong.

Figures that cannot be checked against the saved source

Responses to Multiple-Choice Assessments*, doi 10.1016/j.caeai.2026.100668. The saved text stops at the old extraction cap, before the results, and the only open-access copy Unpaywall lists is the authors' arXiv preprint (2412.02713), which is an earlier draft that does not report the generational comparison. Two figures stated on the page as findings - the newer models' accuracy against humans (80.8% versus 71.5%) and the chemistry person-fit correlation falling from r = 0.389 to 0.060 - therefore cannot be verified from anything on disk. The practice section now makes the same point qualitatively, which the saved abstract and research questions do support; the key-findings figures need the published PDF before they can be trusted or corrected.

Sources that are summaries rather than papers, and figures that live only in images

  • wang-safety-gap-productive-struggle-2026 - the saved source is a 3.3 KB abstract-level overview with no method, results or limitations text, so the page's limitations describe what a perspective paper can and cannot support rather than study statistics. Needs the full article.
  • zhang-races-consistent-essay-scoring-llms-2026 - the saved source stopped at the old extraction cap; the open-access article page was re-extracted through a browser and the source is now complete (70 KB). Its QWK values are presented only inside figures, which text extraction does not capture, so the page no longer states a QWK percentage: it points to the figures and keeps the numbers that the text does state (semantic similarity 81.3% and 79.3%, pairwise accuracy 73.9% and 63.1%). Read the figure if the QWK value is needed.

Figures stranded by extraction caps

Pages in this section now have a full text supplied by hand where one was missing; the entries below record what could not be checked before it arrived.

structrag-diagram-reasoning-ai-tutoring was the last page in this position: it stated edge-level F1 gains of 26.4 points on irregular, 23.1 on bridge/mesh and 21.6 on ring diagrams, plus 85.8% and 81.4% diagram accuracy, while its saved text stopped at the 50,000-character cap before the results tables. All five figures were confirmed against the published article once a copy was supplied: the gains sentence appears in the results, and the accuracy figures are the reported performance on those two topologies, which the paper itself calls the weaker cases. The page now passes the audit with no unverified numbers.

  • genai-chatgpt-adoption-ethics-students-2026 and idea-framework-metacognitive-genai-2026 both state figures that their saved sources do not contain (185 excluded responses; adjusted prompt-quality differences of +11.77 and +29.19 points). Both sources stop at the old extraction cap mid-sentence, before the results. Both articles are open access, but the publisher serves a robot check to automated requests for the article and the PDF, and the only other location Unpaywall lists is the publisher or a directory entry rather than a repository copy. The figures are recorded here rather than deleted, because a truncated source is a likelier explanation than an invented number, and both pages' claims are internally consistent with what the saved text does show.

Sources lost and needing the paper itself

A re-fetch across capped saved sources was pointed at the first arXiv link it found in each file, which in most files is a citation rather than the paper, so thirty saved sources were briefly overwritten with unrelated articles. Every one has been checked against the source its own frontmatter names, and all thirty are now correct: fifteen restored automatically, four more from arXiv PDFs on a second attempt, and the last eleven supplied by hand as PDFs. The supplied copies are kept under pdf-sources/ with their extracted text beside them, and no marker file remains in raw/papers/. (pdf-sources/ is the ONE store for source PDFs — tooling/scripts/check-pdf-store.py fails the gate if a PDF appears anywhere else in the repository.)

Two more pages state figures their saved source cannot confirm, because the source stops at the old extraction cap before the results and no open-access copy exists: microbit-robotics-machine-learning-teacher-training-2026 (a 1.54-point coding-concepts gap) and edurev-100741-tpack-genai-review (effect sizes of 0.828, 0.831, 0.729 and 0.752). The first is a ScienceDirect article behind a robot check, and Unpaywall reports the second as not open access at all. Their sections use those figures because the page bodies already carried them; both need the PDF.

Checked — no further text available

arXiv copies of these were fetched and extracted; the retrieved text is the same length or shorter than what is stored, so nothing is missing beyond the tail of the reference list. The last three were supplied directly as PDFs and are now stored in full.

Screened in by the sweep, then resolved

The 2026-09-27 OpenAlex sweep (window 2026-09-20 to 09-27) surfaced three papers that cleared the significance screen but whose recorded open-access pdf_url returns an HTML landing page or a bot check instead of a file. Two were then supplied as PDFs and are being ingested:

  • 10.1080/02602938.2026.2734350 — Trade-offs in assessment and feedback design: a trilemma-based model (Assessment & Evaluation in Higher Education, 2026-09-22) — supplied as a PDF, INGESTED
  • 10.1007/s00146-026-03380-4 — "Already outdated and still under review": mapping the landscape of research on adolescent development and AI chatbot use (AI & Society, 2026-09-25) — supplied as a PDF, INGESTED