Evidence File — Back to Pen and Paper
Evidence file · published with the report
Primary documents
| Document | Role | Use |
|---|---|---|
| https://www.pnas.org/doi/10.1073/pnas.2422633122 | Primary source | Supports the article's claims |
| https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html | Synthesis by a different publisher, not a study | Reports its own conclusion across a wider evidence base. Its named studies were not retrieved, so the synthesis is attributed to the OECD and the underlying evidence is not claimed as read. |
Claim-by-claim verification
| Claim as published | Status | Basis |
|---|---|---|
| Grades improved 48 per cent with the unrestricted tutor and 127 per cent with the safeguarded one while the tools were available. | Verified | Bastani et al., PNAS, read in full at PMC12232635. |
| Students who used the unrestricted version performed 17 per cent worse than students who never had access, once the tool was removed. | Verified | Same paper; figure stated in the retrieved full text. |
| Princeton’s faculty voted in May 2026 to require supervision of in-person examinations from 1 July, and the policy does not require handwriting. | Verified | Inside Higher Ed, 15 May 2026, which quotes the policy text. The policy’s absence of a handwriting requirement is read off that quoted text. |
| The Michigan document is advice to instructors and mandates nothing. | Verified | The page’s own framing as a Teaching Tip of the Week; no policy instrument is cited on it. |
| Michigan’s paper-versus-online effect claim rests on a study of assignment submission, not of handwriting and comprehension. | Verified | The reference list on the retrieved page points to a submission comparison. |
| The Harvard physics RCT found an effect size the authors put at 0.73 to 1.3 standard deviations. | Verified | Kestin et al., Scientific Reports, quantile regression quoted. |
| A preregistered 371-student trial found no significant advantage of designed interventions over a plain chatbot. | Verified | Fütterer et al., abstract via ERIC EJ1510259; full text not retrieved. |
| The OECD says several studies indicate the GenAI output advantage reverses in exams when access is removed. | Attributed | OECD Digital Education Outlook 2026 key messages. The studies it names nowhere on that page were not retrieved. |
| The German and Norwegian oral-exam findings describe unequal treatment of students. | Attributed | Described by The Conversation; the underlying studies were not retrieved. |
| HEPI records 95 per cent of students using AI in at least one way and a rise to 12 per cent directly including AI-generated text in assessed work. | Verified | HEPI Student Generative AI Survey 2026 figures as published. |
| Handwriting improves learning. | Not claimed | No retrieved study establishes it. The article explicitly declines the claim. |
| AI has damaged a generation. | Not claimed | No longitudinal study exists in the file. The article says so. |
Known limits
- Two of the studies cited were verified at abstract level only: the null secondary-school trial and the small primary-school writing study. Both full texts sat behind a paywall and a challenge page and were not read.
- The OECD’s synthesis was read at the level of its published key messages. The studies underpinning it were not retrieved and are not attributed to The Humans Are Fine.
- Every study cited measures weeks. None follows a child across a schooling or a worker across a career, so no claim is made about long-term effects.
- No evidence on reading or early writing in elementary schooling was found that would survive scrutiny, and the article says so rather than generalising.
- No study comparing handwritten with typed work on comprehension was retrieved. The article therefore makes no claim either way about whether the medium matters.
Provenance
Converted 3 November 2026 from the reviewed Markdown draft at StarNet-Outputs/thaf-education-2026-11-02/ARTICLE-DRAFT.md (36,084 bytes), which recorded four independent review rounds with twelve objections applied and confirmed. Body text, quotations and attributions are carried over unchanged; only Markdown emphasis markers and section headings were converted to the site's HTML form. Commander authorised publication on 3 November 2026.
Produced under human editorial authority. Nothing in this file is published.