For authors.

Three things you can ask for.

• Check your references. Every citation checked against Crossref, arXiv and retraction notices. Nothing is published.
  Free · While you wait
• Pith Review. A full review of your paper: what it claims, where it could break, and what to fix. It stays private unless you choose to publish it.
  $5 · Half finish within 28 min
• Prove a file existed. A seal proves that a manuscript, a dataset or a code repository existed at a moment, without anyone seeing it.
  Free

And two more, when you want them.

• Publish with a DOI. When its Pith Review has finished, a paper can be published on Pith with the review beside it, whatever the score. Publishing is free. For $10, once, it gets a DOI if it has none and a dated proof of its exact file.
  Free, or $10 with a DOI
• Ask an expert. A signed review from a verified expert in your field, for a paper on arXiv or published on Pith. The review is public, whatever it finds.
  From $200 · 7 to 14 days

Submit a paper

Your paper's arXiv number

Find the review

The number arXiv gave the paper, or the paper's arXiv address.

Pith has published a review of Who Should Score the Cache? Mid-Depth Layers Are the Best Judges of KV-Cache Eviction (arXiv:2611.02230).

Read the review Reply to it Challenge a statement in it

What each of these does is set out below.

A verified author of a paper can publish a reply on the paper's page, with Pith's review. The reply is your text, up to 5,000 characters. Pith publishes it word for word, under the name on your account and the words “verified author”, and does not edit it.

Each verified author has one reply. You can change yours at any time, and the page then shows the date of the change.

The reply form is on a page of its own. Type your paper's number in the field at the top of this page and choose “Reply to it”. If you are not verified yet, that page says why, and you can write your reply there and keep it as a draft until you are. A draft is shown to nobody else and is never published as it stands.

There are two ways to be verified today.

• With ORCID. Connect your ORCID iD to your Pith account. You are verified when Pith's record of the paper's authors carries your iD, or when the paper is among the works on your ORCID record. The reply page checks again whenever you ask it to.
• By a person at Pith. If you have no ORCID iD, or ORCID cannot match you to the paper, say on the reply page which author you are and how that can be checked. A person at Pith reads the request and grants or declines it, and the decision is shown on the reply page. It verifies you for that one paper. Pith has not published a time within which a person answers, and it does not email the decision.

Anyone with a free Pith account can challenge a statement in a review. You do not have to be an author, and you do not have to be verified.

Say which statement of the review is wrong and why, and quote the paper where you can. A challenge has a summary of one sentence (up to 500 characters) and a full text (up to 6,000). It is public and signed, and the name on your account appears on it.

Who decides

Three AI models, each from a different lab. Each one reads the paper, the review as Pith publishes it and your challenge, and decides on its own whether the statement is right. A model's answer counts only if it quotes the review and the paper word for word.

Pith publishes the result as it comes. No person at Pith approves it first.

The possible results

• Upheld: the statement in the review is wrong.
• Partly upheld: part of the statement is wrong, or it claims more than the paper supports.
• The review stands: the statement in the review is right.
• Not settled: the models disagreed, or were not sure enough, or could not check the statement against the full paper. The challenge stays on the paper's page and that part of the review is marked contested.

For any result but the last, at least two models must reach the same answer, each with high confidence, and no model may reach a different one.

What an upheld challenge changes

The decision and a correction are printed with the review, and the passages the decision withdraws are struck through where Pith can find them in the text. If that review is still the one on the paper's page, Pith then writes a new review of the same version of the paper, with the correction in front of the model. The first review stays readable, with its struck passages.

How you hear

The decision is printed on the paper's page and on the list of all challenges. If your account has an email address, the decision is sent to it.

How long it takes

Pith promises no time for a decision. Challenges are decided in the order they were filed, after any paid review that is waiting.

The record so far, read 15 Nov 2026, 09:48 UTC: 9 filed; 6 upheld in whole or in part; 3 where the review stood; 0 not settled; 0 waiting for a decision. Of the 9 decided, half were decided within 9 days of being filed.

Write to hello@pith.science. Every message goes to that address and a person reads it. Say which paper, which sentence of the review, and what is wrong with it.

Use it when you would rather not post in public, when you think the decision on a challenge is wrong, or when a review says something about a person and not about a paper.

Pith has not published a time within which a person answers.

A stored review is signed and is never rewritten. A correction is shown on the page with the review, and the earlier words stay readable.

When Pith finds that a review was made from the wrong file, or from a copy of the paper that its own tools cut short or damaged, the review is made again from the right text, and the new review takes the old one's place on the paper's page. If the review of your paper reads as if it were about another paper, or says that text is missing which your paper has, tell us by a challenge or by email.

How often Pith's reviews are right is measured and published at accuracy.

• There is no opt-out. Pith reviews new arXiv papers without asking their authors, and an author cannot ask for theirs to be left out. Pith does not remove its review of an arXiv paper because an author asks: a review that is wrong is corrected or made again, as above.
• No person at Pith reads a machine review before it is published.
• Pith does not write to a paper's authors when it publishes a review of the paper.
• Pith does not edit an author's reply, and it does not show a reply as a verified author's unless the author was verified.

Pith has not published which person answers an author, a time for an answer, or a way to appeal. What is in place about how Pith is run is on the governance page, and who runs Pith is on the operator page.

If you posted the paper on Pith yourself, it is yours to withdraw from the paper's own page. The governance page says what that does.