Does every p value in your manuscript match its statistic?
Paste the Statistical analysis section, the Results and the figure legends. Your browser recomputes every reported t, F, chi-square, r and z and checks the reporting for free - nothing is uploaded. A paid run then audits the statistics line by line or drafts the Statistical analysis section.
Each example has a saved model run, so you can see the whole page for free.
Check one statistic (free, in your browser)
Type a test result.
Your recent runs
What this does, and what it does not
The recomputation is exact arithmetic on what you printed: two-tailed p for t, z and r, the upper tail for F and chi-square, with the range your rounding allows (t = 2.10 could be 2.095 to 2.105). A result is inconsistent when no rounding reconciles the statistic with its p value, and a decision error when the mismatch crosses 0.05. The distribution functions were checked against R on 600 cases. The reporting flags are pattern checks - the independent experimental unit is the default n, multiple comparisons need a declared family or correction, error bars need a definition, a difference in significance is not a significant difference - and the paid run answers each flag and may dismiss one with a reason.
With Nature as the target, the checks follow Nature's initial-submission statistical information (read on nature.com on 2026-09-27): tests and tails in the Methods, defined error bars, exact n rather than a range, repeat counts for representative results, exact P values, and t and F values with degrees of freedom. Nothing here reanalyses your data, and the model is told never to invent a number, test or correction. Derived from the agent skill @yuan1z0825/nature-statistics (yuan1z0825/nature-skills, Apache-2.0). The example studies are fictional. Stats Desk is independent and not affiliated with or endorsed by Nature or Springer Nature.