at
Restamps two solutions onto a common evaluation number, the later of the two, and returns them in the order supplied.
A dynamic penalty scales a violation by a factor that grows with the evaluation number, so that a search may roam infeasible early and is pushed toward feasibility later. That factor belongs to the moment of comparison, not to the moment either solution happened to be evaluated. Judging an incumbent at the clock it was born with against a challenger at the current one penalizes the challenger harder for the same violation, by the ratio of the two clocks, and the effect compounds: while the incumbent stands its clock does not advance, so the bar rises against every later candidate and more effort makes matters worse rather than better. A search started outside the feasible region can be frozen on its starting point.
Using the later of the two clocks needs no solver state, is right when two archived solutions are compared, and keeps the intended semantics: late comparisons are stricter about feasibility than early ones, but both sides of any one comparison meet the same standard. Restamping is free for a solution already at that number.