It’s great to see that CP is so much used at google. Even more than MIP! I guess it’s because CP is one of the most flexible technique you can find in combinatorial optimization. When it is mixed with LNS (CP + LS) it scales very well.
I can’t agree more with that one. I would also add an item to build the right graphical user interface for your application is also important and very time consuming. Unless you are in a very big team with specialists, this is also something you have to do.
This slides is a bit arsh but probably so true. I guess the community should take it as a challenge and try to publish things that may be eventually used at google.
I find it very sad that it seems so difficult to publish implementation details.
It would be so nice to have for instance a paper of Laurent describing the architecture and design choices he made for or-tool solver. I hope the reviewers of CP this year (probably busy reading the papers now) will take this remark of Laurent into account ;-))