Statistics and reports

The number comes from a sealed dataset.

Every result names its method, its snapshot, its parameters, and the software versions. A run that is not confirmatory is labelled exploratory.

Statistical methods built in

Diagnostic accuracy

Sensitivity, specificity, predictive values, and likelihood ratios with Wilson score or Clopper–Pearson intervals. AUC with a DeLong interval.

Analytical performance

Precision from replicate runs, linearity, limits of blank, detection, and quantitation, and method comparison with Bland–Altman, Deming, and Passing–Bablok.

Workbench methods

43 methods built on SciPy and statsmodels, including t-tests, regression, survival, mixed models, and sample size. 21 are checked against a reference, 6 in part, and 16 have no reference check.

Exploratory only until checked

A method without a reference check runs for exploratory analyses only. The Studio shows which methods are checked, checked in part, or not checked.

Sample size

Sample-size calculations are stored with their inputs and method, so a reviewer can see how the number was reached.

Not included

There is no double programming, no independent R implementation, and no interim or adaptive design workflow.

Pre-specification and reproducibility

Reference checks

Checked methods are compared with published worked examples in the automated test suite. This is not a signed method validation report.

Analysis plan versions

The analysis plan is versioned and approved with a signature, and the dataset is frozen with a checksum. The system records the time of each.

Exploratory label

A workbench run is confirmatory only when a statistician cites a signed analysis plan and a frozen dataset. Every other run is labelled exploratory in its output.

Recorded runs

Each run records the snapshot checksum, the parameters, the analysis code hash, the library versions, and the study definition version. A rerun is compared with the original.

Figures

Kaplan–Meier, forest, ROC, waterfall, swimmer, and CONSORT-style participant flow figures among others, exported as SVG, PDF, or the plotted data as CSV.

No validation claims

The report states the results. ValidTrials never writes "this product is clinically validated." That judgment remains with the manufacturer and the authority.

Bring us the protocol.

We read the study, configure ValidTrials, and the sites work in one record. The first conversation is about your protocol, not a software tour.

Further reading