IEMCO5F is, basically, “find the longest path between two points in a graph on grid”. In generic graphs it is NP-hard problem (because if we have polynomial longest path, then we have polynomial Hamilton path, just use longest path for all pairs of points), so I was surprised to see it, but maybe it is easier on grid (AFAIK it is not). After the contest I read some of the accepted solutions (not all), and I was very surprised - they were just simple DFS. I can make a counterexample to them in 5 minutes. Did I misunderstand the problem, or were tests THAT weak?

Yes, I too agree on the same. A 10x10 sample test case of mine didn’t terminate with in time limit. But some how the same code, got accepted.

This is not only the case with this problem, the wakeup sid problem (IEMCO5D) had very weak testcases, where many bruteforce solutions and unhandled overflow solutions have been accepted.

Seriously the contest administrator should have ensured proper testcases.