Your statistics, start to finish, on your own machine.

Agreeing with R is only half of being right. The other half is declining to answer when the data cannot support the question. Every condition below is detected, explained in words, and covered by its own automated check. R's behaviour on the same data was established by feeding it the same defective datasets, and it is recorded because on this class of problem R is often the weaker party, not the target.

Two comparisons are meaningless unless the exact R call is named, and both are named in the harness: cor.test for Spearman on tied data underflows to exactly zero unless asked for the asymptotic form, and psych::ICC defaults to a mixed model that returns a different estimator from the classical one this app implements.

The limits of agreement on a Bland-Altman plot carry an exact confidence interval, from the non-central t distribution, rather than Bland and Altman's 1999 approximation. A coverage simulation against a known population limit, 40,000 samples at each size, puts the approximation at 92.2% at five pairs and 92.7% at eight, where the exact interval holds 95.0% at every size from five to a hundred. Carkeet (2015) reports the same permissiveness below about forty pairs. The approximation is still computed and returned, so a figure made before this change can be reconciled.

It is free. No account, no sign up, and no licence key.

No. There is no cloud, no account, and no upload. The statistics engine runs on your own machine, and the app works with no internet connection. Your dataset is held in memory while you work and is never written to a server, because there is no server to write to.

That question does not really apply to a tool, and anyone who answers it with a simple yes is overselling. Compliance is a property of your project and your institution, not of software. What the app does is keep the data on your device, ship a Safe Harbor helper that scans for the eighteen identifiers and returns a de-identified copy, and warn you before any free text you type could be sent out. What it does not do is make your study compliant on its own.

Yes. Alongside the normal installer there is a portable build. You extract it to any folder you can write to, such as your desktop or a USB stick, and run it. No installation, no administrator rights, no registry changes, and nothing left behind afterwards.

Yes, completely, for every analysis. The only features that need the internet are the ones that are about the internet: the literature search in Discovery, which queries PubMed and ClinicalTrials.gov, and the optional AI topic suggestions. Your data is not part of either.

Every test in the analysis menu has been checked against R's gold standard packages on the same data, and matches to the precision it displays. The full list is in the section above. An automated harness re-runs 472 checks, 252 of them comparisons against stored values from R's own packages, and it has to pass before any release, so a change cannot silently break a result. The comparisons are in the repository, so you can run them yourself rather than take our word for it.

No, and it does not try to. It removes the mechanical errors, choosing the wrong test, missing an assumption, reporting a p value with no effect size, and it explains what it did in plain English. Study design, confounding, and interpretation still need a human, and anything headed for publication deserves a proper review.

You are told, and you are offered the correct alternative. If variances are unequal it suggests Welch. If a logistic model is perfectly separated it suggests Firth's penalized regression rather than reporting an absurd odds ratio. If your groups are not normal it moves to the nonparametric equivalent. The reasoning is shown, so you can learn the rule rather than just take the output.

Figures export at 300 DPI in PNG, JPG, and TIFF, and as true vector SVG, which is what most journals ask for. Every result also has a plain-English write-up with the effect size and confidence interval, which you can copy to the clipboard and paste into a manuscript.

CSV and Excel files, or you can type and paste directly into the data grid. Meta-analysis and diagnostics have downloadable templates so the column layout is unambiguous, and the importer maps your columns for you.

Windows 10 and 11 today, as a 247 MB installer or a 308 MB portable build. A macOS version is in preparation and will be signed and notarized by Apple so it opens without warnings.

There is one optional AI feature that suggests novel research angles, and it is off unless you add your own API key. It only ever receives the search wording you type, never your dataset. Before any text leaves the device, a guard scans it for patient identifiers and blocks it by default until you confirm twice.

MedResearch Suite is built by HariOm S. Vaja. If you would like it running across a department or an institution, including on machines that block installers, get in touch.