screeningReplications

The replications per candidate used by the screening stage. Null (the default) disables screening; it must be supplied exactly when screenK is.

Size this against the constraint, not by habit. Feasibility is tested with a one-sided interval, so a candidate is declared feasible only when its upper limit falls below zero, and at small replication counts that is unreachable for any candidate near the limit. On a P(event) <= 0.05 constraint at 99% overall confidence, measured against ResponseConstraint.testFeasibility itself:

n=30   only 0 violations out of 30 passes; 1 in 30 (p = 0.033) already fails
n=100  passes up to p = 0.01;  p = 0.02 fails
n=200  passes up to p = 0.02

So screening at 100 replications against a 0.05 limit still rejects every candidate sitting at a realistic 0.02 to 0.04, and the screening stage inherits the degeneracy it was added to cure. Roughly 200 is where discrimination arrives for that constraint. A tighter limit needs more.