How to judge a paper

A paper on EntryLib shows what it claims and what each claim rests on. Its scores check that record; they do not check the science, and nothing on EntryLib is peer review. Before relying on a claim, look at the things below. Each takes a click or two on the paper's page.

1. Read the claim exactly

A claim holds under its conditions: a data set, a horizon, a population, a range of parameters. Read its full statement, its scope and its conditions before quoting it. A finding stated without them is a different, broader claim.

2. Open its evidence

Click the claim's evidence (E1, E2…). Ask: is it the paper's own result, or another paper's? Does the table or figure actually show what the claim says, for the same measure and the same cases? Is the material, data and code, there to obtain, or only described?

3. Check what it assumes

Assumptions (A1, A2…) say what must hold for the claim to follow. Some are checked in the paper, some are only stated, and some cannot be tested. A claim is no stronger than its unchecked assumptions.

4. Read its limits and open questions

The Limitations and Open questions sections say where the authors think the findings may not hold. A limitation that touches the claim you want to use matters more than any score.

5. Look at the Support score, and why it is not higher

Support starts at 100 for each claim and loses points for named problems in its record: weak support, a test not planned beforehand, material a reader cannot obtain, a missing scope or link to earlier work, open limitations and unchecked assumptions. Click the score to see which. A high score says the record is complete and consistent; it does not say the claim is true. A low score tells you what to check first.

6. Know how the paper was made

• Made with the lab The research was recorded while it was done, and the record is linked. You can see which hypotheses were tested, which failed and what was changed on the way.
• Written as a claimset The authors wrote it directly, usually with an AI agent. You rely on them for how the research was done.
• From a published paper Made from a paper published elsewhere. It is as good as its source, and its conversion guarantor answers only for its faithfulness to that source.

7. Check novelty and earlier work

Novelty is the authors' own account: new, extends, replicates or refutes earlier work. EntryLib compares it with the claims already published here and flags a close match the paper does not acknowledge. It cannot compare with work published elsewhere: the references are where the authors place their work.

8. Check the version and any notice

Cite a version, such as entrylib:2026.000001@v2#C4. A newer version or a notice at the top of the page (updated, expression of concern, retracted) tells you whether the claim still stands as published.

9. Ask your AI, but check what it says

Ask your AI on every paper page copies a message for your assistant with the paper's address. Useful questions: "What does C3 rest on, and is that evidence the paper's own?", "Under which conditions does C3 hold?", "Which assumptions behind C3 are untested?" Ask it to cite claims by their identifiers, and open them to check.

The responsible researcher named at the top of each paper answers for it. To raise a problem with a paper, see Who runs EntryLib.