Your vote redirection idea is interesting, but a simple 1 to 1 mapping may make it just as difficult to find a surrogate voter as it is to research a valid candidate. I've tossed around the idea of a learning system where you could log your preference for multiple issues, then based on those preferences, the system could predict your preference on future issues based on the logged preferences of others. I think a system like that would be a great aid for representatives to use to visualize the current thoughts of their constituents, and could be an intermediate step towards the system you propose here.