We are not able to settle the (non-)existence of a polynomial kernel for a parameterized combinatorial NP-complete problem (we also tried to apply some recent lower bound techniques to prove the non-existence of a polynomial kernel under reasonable complexity-theoretic assumptions). So we are searching for major open problems that could be used in a parameter-preserving reduction to "underline its hardness".
What are major parameterized NP-complete problems for which it is unknown if they have a polynomial kernel ? Is there a survey/technical report on the subject?
An example could be ODD CYCLE TRANSVERSAL (OCT), the task of making an undirected graph bipartite by deleting as few vertices as possible, parameterized by the number of allowed vertex deletions (though Stefan Kratsch and Magnus Wahlström recently showed a randomized polynomial kernel for OCT)