Question 1

How can I trust it not to hallucinate extracted data?

Accepted Answer

Jev never writes text here: it answers typed questions with probabilities to select line ids, and code copies the quote verbatim with file, page or paragraph, and highlight position. Nothing is paraphrased, so nothing can be invented by the model.

Question 2

Where do my PDFs and answers actually live?

Accepted Answer

Everything stays in the browser with no uploads or server storage, and improved readers re-find saved quotes by exact text, graying rather than dropping non-matches. Any saved answer deletes with its own control.

Question 3

Why does it need both Choice screening and Noul checks?

Accepted Answer

The screening Choice is relative across a chunk while verifying Nouls are absolute per line, which is what multi-row answers like Mean (SD) under Age need. Adjacent lines merge and table labels attach; below-threshold questions return Not found or Unclear.

Question 4

How does this scale beyond one trial report?

Accepted Answer

Import from EndNote, Zotero, Mendeley, PubMed, Scopus, Web of Science, Covidence, or Rayyan, then ask across every study at once or only the missing answers. The extraction table exports one row per study or one row per quote plus bias, retraction, and PRISMA tabs.

Question 5

What do the two passes over a file actually do?

Accepted Answer

Screening is one request per chunk of two or three pages of one file, carrying every question at once: a Choice asking which line answers the question plus none, and a Noul asking whether the passage answers it. Verification is then per question, taking the best lines and their neighbours with one Noul each, asking whether line B129 itself answers the question. Code then takes lines with a Noul of 0.5 or more, merges adjacent lines and attaches the table label. Line ids are prefixed by file letter, A001, B001, C001, so every quote says which file it came from.

Question 6

What does a template actually cost across a study?

Accepted Answer

The eighteen-question template costs about a cent and five seconds over three files, because each screening request is speculative and carries every question against the same text. Asking every study in a project only asks what each still lacks: questions it never answered, questions reworded since, and every question when a file was added after the answer, about a cent a study for eighteen questions. An answer you already checked or annotated is never overwritten by a reworded question; it stays under a new id beside the new one.

Question 7

What does the extraction table produce for a manuscript?

Accepted Answer

Two exports: the table, one row per study with its reference, how many answers are checked and each question's value and quotes, and the quotes, one row per quote. Under each question it counts the studies that report it, which is how you spot what a literature rarely reports. Tabs hold Risk of bias, Retractions and PMC, Second reviewer and Report, where Write the methods paragraph drafts the text PRISMA 2020 items 9 and 10 ask for from the project's own numbers. The Report tab also draws the PRISMA 2020 flow diagram and downloads it as SVG.

Jev Reviewer

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community