Simulate a two-component mixture whose per-tau fits cross (for O4 validation)
Source:R/noncrossing.R
sim_mixqr_cross.RdOne component is well-behaved; the other has a near-flat slope and strong
heteroscedasticity (error scale growing across the design), so its
independently fitted per-tau quantile lines cross in finite samples – exactly
the within-component crossing Wu & Yao (2016, sec.5) flag as common "when the
sample size is small." Use to demonstrate that noncrossing = "none" crosses
while "rearrange" does not. (A valid location-scale model has non-crossing
true quantiles; crossing here is the finite-sample estimation artefact the
method repairs.)