One project, three records
The draft, the bibliography, and the citation audit stay together. A second project does not share that state.
Home / Pruf
Product
Pruf
A manuscript reviewer for a small academic publisher. The model is an assistant on a marked passage. The editor remains the reviewer.
Product description
Solamarc Publishing builds Pruf, an AI-assisted manuscript reviewer for academic journals. Pruf keeps the draft, the bibliography, and a citation audit in one workspace. An editor marks a passage and can ask the model to explain it, or to audit a proposed deletion or replacement. The model does not approve wording: every change stays with the human reviewer. Citation judgments are recorded against checked excerpts; they are not generated as verdicts. Pruf is in an in-house pilot on the founder's own manuscripts. The review-assistance steps are being moved onto the Claude API.
How a review runs
Mark a passage
The assistant sees the marked text and the editor’s instructions for that job. It is not handed the whole journal.
Ask for one of three things
Explain the idea. Audit a proposed deletion. Audit a proposed replacement. There is no fourth button that accepts the paper.
The editor keeps or rejects
The note is advice. Writing it into the manuscript is a separate, human decision. Pruf does not present a model suggestion as the editor’s decision.
The three jobs
Same description as the homepage. The sample notes live in the demo, and they are labeled as samples.
Explain
A definition, not a verdict
What the marked idea means, why it matters here, and one example. Useful when a phrase is doing more work than the author noticed.
Open explain Deletion auditWhat the cut would hide
A limitation is easy to delete because it slows the paragraph down. The audit asks whether the remaining lines still match the evidence.
Open deletion audit Replacement auditOriginal against the new line
The question is not “is this smoother?” It is “is this the claim the study can carry?”
Open replacement auditCitation audit is a record, not a model verdict
A reviewer can record a judgment against a checked excerpt: supported, supports with a scope caveat, mixed or conditional, does not support, or not yet assessed. The excerpt is the evidence. A local filename is not.
Pruf does not generate that judgment and then stamp it on the paper. If nothing has been checked, the honest state is “not yet assessed”.
Status, stated plainly
- In-house pilot, on the founder’s own manuscripts. Not offered as a public signup.
- The demo on this site is a written walkthrough. It does not call a model.
- Review assistance is a configurable model call. Those steps are being moved onto the Claude API.
- The company email on this domain is admin@solamarcpublishing.com.