Research Integrity Support — Proof of Concept

This public demo runs only the bundled synthetic sample manuscript below — click Run integrity review. The manuscript text is checked for two real, independently recomputed statistical inconsistencies: a stated count/percentage pair that doesn't recompute correctly, and a group of percentages that doesn't actually sum to 100%. Claude then reads the text for disclosure statements that look missing (corroborated by a real keyword search before being shown) and apparent numeric contradictions, each required to quote two real passages.

Figure-image upload and duplicate-image detection aren't available in this public demo — only the manuscript-text review runs here.

Synthetic sample manuscript

A total of 198 participants were enrolled across three study sites. Of these, 42 of 180 (28.3%)
experienced a treatment-related adverse event during the follow-up period, the most common being mild
gastrointestinal discomfort.

Treatment allocation was as follows: 40% received Drug A, 35% received Drug B, and 20% received
placebo, with the remainder lost to randomization error before dosing began.

All 180 participants completed the full twelve-week study protocol, and no participants withdrew
for reasons related to the study intervention.

Figure 1 shows representative immunoblot results from the primary cohort. Figure 2 summarizes
treatment allocation across sites. Figure 3 shows immunoblot results from the replication cohort,
confirming the pattern observed in the primary analysis.