Tile (hex) laying maze/race game concept

Much like many other people here, I have what I think could be a solid core mechanic for a game. What I don't have are all the (important!) details to turn a concept mechanic into a functional game. After hearing about this forum elsewhere on the internet and reading some of the fascinating threads here, I thought I'd post here asking for input.

I'm basically looking at a hex tile-laying maze game in two stages. The tiles are hexes, where each side is a wall. Each tile has 2, 3, 4, or 5 doors on it in various combinations (so there are 3 different types of 4-door tiles, 4 different types of 3-door tiles, etc). The distribution of tiles (i.e., how many of each type are in the deck) is still indeterminate.

There are two phases to a complete game. In phase one, the players build the board and traverse it. In phase two, the players move back across the board and can modify it. First past the post wins.

Phase 1
Each player must take three actions on each turn, and each action can be one of three things: draw a tile, lay a tile, move his pawn. Any combination/count is allowed, so a complete turn might be drawing three tiles, or drawing two tiles and laying one tile, or moving the pawn across three tiles, etc.

There are a number of starting tiles equal to the number of players, featuring two doors on adjacent sides. To begin the game, each player lays down his starting tile such that they form a line of adjacent hexes, touching along the sides, with all the doors facing the same direction, so for three players the start line would look like this bad ASCII:

/-\+
\-/+
/-\+
\-/+
/-\+
\-/+

The pluses indicating the sides which have doors.

Tiles are drawn from a communal deck, and kept secret from other players. There will probably need to be a maximum "hand" size, but I have no idea what it should be. I'm also considering a semi-open system, where some number of tiles are kept face up on the table, and can be drawn at the cost of losing one action, but my inclination is to keep it simple/secret.

One of the open questions is whether I have an actual board on which tiles can be laid, or if the tiles themselves make up the board. The advantage to the former is it removes the problem of defining the far end of the playing surface; it would just be the opposite edge of the board. It would also allow me to introduce a special space in the center of the board: a pre-laid tile with all six sides having doors. Each player's eventual path to the finish line would have to pass through this space. I think this would help ensure player interaction in tile laying (as opposed to each person laying completely independent, non-intersecting tile paths). The advantage of not having a board, of course, is simplicity - I don't have to worry about board size from a balancing perspective. And it would potentially scale better to smaller/larger numbers of players. I keep waffling back and forth on this; input would be much appreciated.

In any event, tiles are laid such that they adjoin at least one other tile along an edge (though if I do go with an actual board, this rule could be removed or made optional).

Pawns are moved one tile per each action dedicated to movement by the player, so 0 to 3 tiles per turn. Obviously, they can only move through door-door pairs. Once a pawn reaches the far end of the board (however that ends up being defined), Phase 2 begins.

Phase 2
This runs in a very similar fashion to Phase 1, with the following differences:

Any player who has reached the far end of the board may no longer draw or lay tiles, but may now rotate tiles that have already been laid at a rate of 60 degrees (one facing change) per action spent.

Any player who has not reached the far end of the board may no longer draw tiles, but may still lay down any tiles still in his hand. He may also rotate tiles already laid at a rate of 60 degrees per action spent.

The first player to return to the starting line wins.

Problems/Questions
1. Initial tile counts. Most of my open questions regarding how this works out can only be determined by some initial playtesting. But to even get to the solo playtesting stage, I need some idea of how many tiles of each variety should exist. Problem is, I don't have any idea how to approach rough-answering that question.

2. Whether or not to have a fixed playing surface, as I outlined above. There's an elegance to having the tiles be the whole game, of course. If I could find a way to ensure interaction between different players' paths (the problem I'd solve with a mandatory, fixed central tile) and an elegant way to define the end of the board, I'd remove any real need for a fixed play area.

3. Maintaining balance between phases. My original conception was to have only people who had reached the far end be able to rotate tiles. But the problem occurred to me that the first person to get there would have no particular reason to rotate anything; he'd just retrace his route at 3 tiles a turn. I'm hoping that granting everyone that ability as soon as anyone crosses the line addresses that (and adds a level of cost/benefit as to timing that move). I'm not sure this is the best way, though.

I've got some other spitball-type modifications (like adding directionality to the hexes, so doors can only be gone through in one direction - and making the ability to modify a hex's directionality an action players can take) to the game I'm tossing around, but they'll have to wait until I've tried out what I've already got.

Anyway, feedback/input much appreciated!