By simulating thousands of random drawings, the program can give a very accurate quantitative evaluation of the different plays.
The shallow search is because the Maven author argues[1] that, due to the fast turnover of letters in one's bag, it is typically not useful to look more than 2-ply deep, because if one instead looked deeper, e.g. 4-ply, the variance of rewards will be larger and the simulations will take several times longer, while only helping in a few exotic situations: "We maintain that if it requires an extreme situation like CACIQUE to see the value of a four-ply simulation then they are not worth doing."
In two-player games, this means that the players can now deduce from the initial letter distribution the exact tiles on each other's racks.
For example, A is generally better than I as a tile, but if there are 7 A's and only 2 I's left in the bag, then maybe we should prefer to keep the I. Parameter fitting was accomplished by tuning the values to predict the total of future scores.
Maven has since switched to a completely different architecture, proposed by John O'Laughlin and implemented in Quackle.
This is the "exhaustive" architecture, where the program has a different rack evaluation parameter for each of the 3 million possible combinations of 0 to 7 tiles.
With the advances in computer power over the last decade, it has become possible to tune such large parameter sets.
The downside of using an exhaustive approach is that Maven lost the ability to vary evaluations as a function of the tiles that remained in the bag.
The great human champion Ron Tiekert had studied Scrabble by playing out individual positions dozens of times, and tabulating results.
By the mid-1990s, computers had become fast enough that Maven used simulation to choose moves in competitive games under tournament time controls.
The problem with Alpha Beta is that some Scrabble endgames require 14 moves to play out, and it is not possible to search that deeply.
Since B* is reasonably robust in the presence of a small percentage of error in the bounds, Maven can solve endgames that other approaches cannot.
A further refinement makes Maven's endgame solutions asymptotically optimal even in the presence of errors.
Because there are some situations (e.g., two blanks, stuck-with-Q) that require extra effort, the calculation is performed progressively.
This policy does not produce play that is theoretically perfect, because it is impossible to know what the true initial distribution of unseen tiles should be.
Assuming a uniform distribution does well, and it is possible to calculate inferences about unseen tiles that marginally improves on that assumption.
That is, in theory there are situations where players maximize expectation by randomly choosing moves according to a probability distribution.