Every candidate ranked by morphology-predicted likelihood of the trial's target marker, with the genes the model cannot see stated per patient. Sequencing still confirms enrollment. Known molecular status appears only to score the ranking, never as an input to it.