The Conformal Engine

Every prediction, guaranteed — proven on your own data. Upload calibration pairs (prediction, actual), choose a coverage level, and get statistically guaranteed prediction intervals — with the coverage empirically validated on a holdout split, not just promised.

RUNS ENTIRELY IN YOUR BROWSER · YOUR DATA NEVER LEAVES YOUR MACHINE

STEP 1

Upload a two-column CSV: prediction, actual

STEP 2

Choose your coverage target — 90 / 95 / 99%

STEP 3

Get the guaranteed interval, proven on holdout

Your calibration data

Numeric predictions and actual outcomes from a held-out set your model did not train on. Minimum 40 rows; 199+ unlocks the 99% level. Header row optional.

Upload CSV

Calibration Report

—

TARGET COVERAGE

—

EMPIRICAL COVERAGE (HOLDOUT)

—

95% CI ON COVERAGE (WILSON)

—

INTERVAL HALF-WIDTH (q̂)

—

GUARANTEED INTERVAL ACTUAL · COVERED ACTUAL · MISSED

This report is the first page of an engagement. The Engine shows marginal coverage on one dataset. A production system needs conditional coverage across subgroups, drift-monitored recalibration, and documentation an examiner will accept — that is what we build, led by statisticians with a minimum of 15 years of regulated-industry experience.

Request a proposal About InterClin AI