The idea with roads burrowing through hills was really retarded, albeit a way to reduce tileset needs. Now instead I implemented roads going over hills, which required drawing incline cobblestone tiles and doing some tricky road interpolation code. Anyway, hills should be reserved for placing caves/mounds inside them, with some neutral monster and a treasure. I think these could be optional: if player decides to explore full map, instead of going straight to the goal, could be reward with some minor stuff. Otherwise players units should be placed right near the dungeon entrance (castle entrance, city tavern, etc...), because wandering through empty map is not that fun. Although I though about making that a challenge in itself - i.e. player should have some engineering units to build a bridge. But that obviously makes gameplay less focused and turns game into a turnbased Minecraft clone.
There also should be some placement mechanics, so defending side would be allowed to move its units before the battle has started. There are several way to implement that. But I though about allowing defender to pick initial center of placement and then place the units inside of it. Or at any indoor place. Afterwards attacker places his/her units, being limited only to outdoor areas, outside of defender's placement circle.
Unfortunately that means visitor would see whole map from the beginning. So I think still limiting visitor placement only to the camp area, and maybe to map borders. Initial placement is actually very important, especially if I later implement multiplayer maps.
Now there is also a retreat mechanic, where defender can retreat at any time, losing any unit which is indoor or in enemy sight. So for maps where defender got attacked in the open field (i.e. no castle or dungeon on that hex), that defender must get immediately surrounded and prevented from retreating without heavy loses. I still think about giving defender there the first turn, he/she will have any chance of breaking out.
As usual, game design is hard, because there are many possibilities and it is not immediately obvious which one is the best. I sometimes fail to see one of the better possibilities, noticing it after some time, and then backtracking to it.
Here is the same seed map, but with road generated going over hill, instead of digging though it. Guess in some cases this approach could make use of suspended bridges.