Report draft tool

trustcv supports documentation of validation choices. It does not make a model or device compliant with FDA, EU MDR or any other regulation.

This form turns values you enter into a draft validation summary that you then edit. It runs entirely in your browser; nothing is uploaded.

The tool computes nothing. Point estimates and confidence intervals appear exactly as you enter them, and an interval left blank is shown as "not provided". It applies no thresholds and draws no conclusions. Take the numbers from your own analysis, for example result.summary() or the record described in Documenting validation.

Model and intended use

Model or software name (required)

Version (required)

Developer or manufacturer

Intended use, as you would state it

Data

Number of patients

Number of rows, images or samples

Data sources, sites and collection period

Population characteristics

Validation design

Split method

Number of folds

Grouping unit

Random seed

Interval method Not stated corrected_t (corrected t-interval over folds) oof_bootstrap (bootstrap of out-of-fold predictions) Other (describe under notes)

Interval level (%)

trustcv overall status Not reported PASSED NEEDS_REVIEW NOT_FULLY_VERIFIED FAILED

trustcv version

Why this split suits the data

Leakage controls and integrity check notes

For example: preprocessing inside the pipeline, tuning in nested CV, checks that were not run and why.

Performance

Enter values on the scale you report them (for example 0.87 or 87). Leave a row empty to omit it. Interval bounds are optional, but enter both or neither.

Metric	Estimate	Lower bound	Upper bound

Confusion matrix counts (optional)

Shown in the draft only when all four counts are entered.

True positives

False positives

False negatives

True negatives

Decision threshold and how it was chosen

Subgroups, limitations and notes

Subgroup results, one per line, as you will report them

Limitations

Other notes

Generate draft Download Markdown Download HTML