This is the other experiment in this hub — not the regular solver run, but a stress test built on top of it. For each of 11 cell-type families, we deliberately hide some cells' identity from the solver and check whether it can figure them back out. Click through the stages below (or hit Play) to see exactly what "before", "after", "gate", and "verdict" mean everywhere else in this hub.
Pick a family, then deliberately blind the solver to some of its known cells.
Every family is solved twice — once honestly limited, once with a conditional boost.
Turn raw predictions into a number that means something against random chance.
One label per family, and the table that feeds the other two experiments in this hub.