There is one more to add to Daniel's list:

4. Any model run before it becomes at least a Tropical Storm is more than likely the wrong track. Throw it out and roll the dice instead.

Models just don't initialize tropical cyclones well until they have formed. Therefore there is a statistical chance a model track "could" be correct, but that will be more than likely coincidence, not model accuracy, at this point. Once it becomes a Tropical Storm, plus 6-12 hours for model run data ingest, and you will start having somewhat reliable tracks. Add in a few hurricane hunters, and then you have a fairly reliable track.

One problem with early models is that a tropical wave is a semi-normal low pressure system steered by shallow layers of the atmosphere in a regional dynamic. A tropical cyclone, however, is a much deeper storm, and in addition to affecting the regional dynamic directly, gets steered by deeper layers of the atmosphere. Unless the model correctly models the depth of the storm (which I have almost never seen them do), they will be highly inaccurate.