Skip to content
Danphe Council

Use case · Research

Literature and paper questions, checked against the paper

Ask what a paper actually reports, compare findings across papers, or get a structured peer review. Statements the sources don't support are flagged, not smoothed over.

Start freeBook a demoSee a checked answer

Ask things like

  • What sample size and effect size does this study report for the primary outcome?
  • Do these two papers agree on the mechanism? Where do they differ?
  • Write a peer review of the attached manuscript.

Example

What a research answer looks like

Every statement carries a label in words and the passage it rests on, so you can check the answer instead of trusting it.

Standard checkTrial report (sample).pdf

Question

What effect size does the study report for the primary outcome, and is it significant?

In short: A mean reduction of 4.1 points on the primary scale (95% CI 1.2 to 7.0), significant at p = 0.006.

  1. 1Models agree

    The primary outcome improved by 4.1 points compared with control.

    “…difference of −4.1 (95% CI −7.0 to −1.2)…” (Results, Table 2)

  2. 2Models agree

    The result was statistically significant.

    “…p = 0.006 for the primary endpoint.” (Results)

  3. 3Not supported

    The benefit lasted twelve months.

    No passage in the document backs this.

Illustrative example. The document and its figures are fictional.

Why Danphe

Why research teams use a council

  • Grounded in your PDFs

    Answers are built from the documents you upload. A statement with no support in them is labelled “Not supported”.

  • Independent readers

    Several models read the paper on their own before they see each other's answers, so one model's misreading is less likely to carry through.

  • Peer-review reports

    Full report mode has each model write a complete review, merges the strongest points and fact-checks every statement.

Suggested depth: Smart for quick questions; Full report for reviews.

Good to know

On short closed questions the council matches the best single model rather than beating it. See the accuracy page for the numbers.

How it works

How each question is checked

  1. Step 1: Reading sources

    Your documents are read and the passages that matter to the question are found.

  2. Step 2: Models answer on their own

    Models from different companies answer independently, without seeing each other's answers.

  3. Step 3: Cross-checking every statement

    Each statement is checked against your sources by models from other companies, with quotes.

  4. Step 4: Writing the answer

    An editor writes the answer from what held up, and marks what didn't.

Next

Try it on your own documents

The Free plan includes 60 credits a month: about six Standard questions.

Start freeBook a demoHow accurate is it?