This is an experimental build of Gecode for the MiniZinc Challenge. It collects several changes that are being evaluated together: native handling for selected scheduling-shaped FlatZinc, a broader automatic search portfolio, non-failing and softening-based search experiments, and new infrastructure for integer-variable domains.

The portfolio keeps ordinary Gecode search in the mix. Around that baseline it adds selected variants for free and parallel search, including restart-based search, randomized seed and value-order perturbations, failure-informed variable ordering, shaving assets, automated asset selection using UCB, and LNS-based optimization workers. The portfolio work is partly inspired by Gecode/Dexter. The non-failing-propagator work follows the idea from Björdal et al. of keeping some useful propagation when hard constraints are softened.