Welcome to the Proofig AI Review Assistant. When Proofig AI surfaces a suspected finding, the Review Assistant explains what the evidence shows, what it might mean, and what you can do next.

It supports your judgment. It does not replace it, and it does not decide anything for you.

The Review Assistant is a side panel on the Review page, the screen where you step through the suspected issues detected in your manuscript.

The comparison image stays the main evidence. The panel sits alongside it and updates as you move between findings, so the guidance always matches the finding you are looking at.

For each finding, the panel presents:

Each finding carries a priority level:

Priority indicates the recommended review level and order of attention. It does not mean the finding is confirmed.

Signal and Concern are two independent ratings, and keeping them separate matters.

Signal indicates the strength or maturity of the technical evidence detected by Proofig. It is shown as a bar with a label: Strong, Moderate, or Beta (developing).

Concern indicates how serious the issue could be if confirmed and unexplained after review. It is shown as High, Moderate, or Low.

A strong signal means the technical match is clear. It does not mean that anything improper happened. A high concern does not mean the issue is confirmed. It means that if the finding is real and cannot be explained, it could meaningfully affect image integrity, interpretation, or editorial and institutional handling.

Reading Signal as though it were Concern, or reading Concern as though it were a conclusion, is how legitimate figures end up over-flagged.

'What do I see?' describes what is actually on screen for the current finding: which panels are involved, which pages they appear on, how many matching features were found, how much of each image the matched region covers, and any transform such as flip, scale, or rotation.

This description is generated for each finding from the detection data, so it reflects that specific comparison.

The same description is carried into the report. When you generate a report, each finding it contains is accompanied by the same plain-language description of what can be seen, so anyone reading the report has that context.

'What does it mean?' interprets the evidence. It explains what the pattern typically indicates, and it also sets out other explanations that could account for it.

This section is written to help you weigh the finding, not to tell you what to conclude.

'Quick check' gives you a short numbered list of things you can do right away to inspect the evidence yourself. Typical steps include hiding the overlay to compare the two panels directly, checking whether the figure legend states a relationship between them, and ruling out shared background or template elements.

'Suggested filters' names the investigation tools that tend to help for that finding type. For western blots, that is often Histogram Equalization and Emboss. For microscopy, Invert and a color map are often more useful.

One caveat on Histogram Equalization: it is very useful for revealing subtle contrast differences, but it redistributes intensity values and can make compression, scan, or background artefacts more visible. Always interpret it alongside the original image.

Each finding ends with a decision: is the suspected finding legitimate, or a potential issue?

If the finding appears explained, irrelevant, or unsupported after your review, move on to the next one. Legitimate explanations are common: reuse disclosed in the legend, a higher-magnification view of the same region, a shared control panel, duplicated text, or a graphical summary.

If the concern persists, use 'Add to report' so that the finding is documented for clarification or follow-up.

Adding a finding to the report is not a final determination. The report is a review document containing selected findings for clarification or follow-up. It does not conclude that a problem exists.

After you add a finding to the report, the panel suggests next steps appropriate to your role. These are examples of the kind of guidance shown, not a complete or binding list.

When you communicate with authors, keep the language neutral. Ask them to clarify the relationship between the panels or to provide the relevant original files, and follow your journal, publisher, or institutional process.

When a western blot is flagged for suspected manipulation, the Review page can display a heatmap overlay that highlights the region which contributed to the signal. Instead of assessing the whole blot without a starting point, you can see which area to inspect, then apply the investigation tools to examine it.

The highlighted region shows where a signal was detected. It is not proof that manipulation occurred.

This applies to western blots and the manipulation signal only.

A finding is a prompt for careful human review, not a conclusion.

That is an overview of the Proofig AI Review Assistant. To learn more about the page it sits on, see the Proofig Review Page Overview. For the investigation tools it suggests, see Key Detection Tools and Filters.

To learn more about Proofig AI and its AI-powered tools, visit our homepage.