A preprint is a scientific paper that the author has posted before a journal has reviewed it. That is the whole idea, and it is the whole risk. Teaching a class to read one is teaching a class how science actually moves.
Frame the object before you open it
Before anyone opens the file, put the definition on the board. A preprint is a research article that has been made public by the author without having gone through the review process a journal would run (Class E, standard usage across arXiv, bioRxiv, and Zenodo). That word, without, is the whole lesson. The paper may be excellent. It may be a draft. The class does not know yet, and neither does the author's field.
Tell the students what changes because of that word. The claims are the author's claims. The math has not been checked by a stranger. The figures may still be being revised. Nothing in the file has the weight that a journal stamp would add. This is not a reason to dismiss the paper. It is a reason to read it with a specific posture.
The four questions a class asks together
Give the students four questions and keep them on the board for the whole read.
- Who wrote it, and what would they gain from being right or wrong? Authorship is a claim, and it belongs to a person or a group with a history the students can look up (Class C, the author list and affiliations are in the header of the file).
- What is the specific claim? Not the topic. The claim. A topic is "active inference in classrooms." A claim is "students who kept an evidence log revised their essays more often than students who did not." Make the class rewrite the claim in one sentence.
- What would prove it wrong? This is the falsifier question, and it is the one students almost never see modeled. If the author has already written what would sink their own claim, that is a strong signal of intellectual honesty (Class F). If they have not, the class writes one for them, out loud, together.
- What has not yet happened to this paper? A named reviewer has not read it. A second lab has not reproduced it. An editor has not decided it belongs in a specific conversation. Make that list visible so the students know what a peer review would still need to add.
Tag the claims as you read
Read a passage, then stop. Ask a student to tag the last claim with one of the evidence classes the class already uses.
- A for an empirical result the author ran and reports.
- B for a code artifact or inspectable procedure the author links to.
- C for a configuration or integration claim, the sort of "here is how the pieces are wired" statement that shows up in methods sections.
- E for a citation to another author's published work.
- F for a falsifier the author has stated.
- U for a claim the author asserts without a receipt yet.
Two things happen when a class does this. Students stop reading a paper as one uniform block of authority, and they start seeing it as a tissue of claims with different weights. That is what a careful reader does, and it is a habit that transfers to news articles, textbooks, and their own writing.
What a preprint is not
A preprint is not a journal-published article, and the class must say the words out loud. It is not a press release. It is not a settled result. It is a serious offer from an author to the field, made in public, with the receipt attached. Treat it as an offer, not a verdict.
This is the posture that lets a class talk honestly about work in progress. UNI publishes its own working notes as preprints for exactly this reason, on an attainable path toward General Natural Intelligence, natural not artificial. The whole point of putting a claim in public before review is that anyone, including a classroom, can start looking for where it fails.
The write-up
End the lesson with a one-page write-up per student. Two paragraphs on what the paper claims, one paragraph on what the class thinks the strongest falsifier is, and one paragraph on what a peer review would still need to check. Grade it on the tags, not on whether the student agreed with the author.
Where to go next
- The Stratified Palimpsest benchmark, for teachers
- Evidence classes as classroom vocabulary
- A rubric that rewards falsifiability
- The workshop for educators
AI-authorship fence: this post was drafted with LLM assistance, reviewed and edited by a human author, and published under human editorial responsibility. No claim is made that an LLM authored the reasoning. The framing is a working hypothesis on the attainable path toward General Natural Intelligence, natural not artificial, tested in the open.