A conformal prediction interval is a confidence band with a finite-sample, distribution-free coverage target. If the engine reports a 90% conformal interval of [56, 82], the aim is that about 90% of similar future predictions contain the realised value — assuming exchangeability. Wider intervals reflect higher per-ticker uncertainty, not lower expected return.
Vovk-Gammerman-Shafer 2005 conformal prediction provides finite-sample, distribution-free coverage targets: if the model says [56, 82] is a 90% interval, about 90% of similar future predictions should contain the realised value. The measured live rate can run below target while calibration data matures. We use Mondrian conformal — different intervals per tail-alignment × regime bin. Wider interval = more uncertainty about this name.