Played around 50-100 games. Highest Score vs Q so far 640-510 (the 
one CSW word was TEEK)

Major points of note:

Quackle plays poorly when it is either very farbehind or in front 
(+/- 100 points) due to "Winningness" rankings

Quackle analyzes it's endgames the same way it analyzes midgame: 
this cause endgame errors.

The 5 minute Quackle is barely noticeable increase in skill than the 
20 second.

Quackle undervalues it's valuation index.  Suggested algorithm would 
be something along the lines of winningness*valuation.

Quackle seems to no attention whatsoever to hotspot replies or "most-
likely' scenarios or 'most dangerous' scenarios.  This all seems to 
be lumped together under 'ply'

Questions
Are the nominal values of letter leaves synergistically balanced?
eg biasing x-vowel leaves instead of X-consonat leaves, CH, CK etc.  
K is a regularly undervalued letter (it's now even more powerful, 
and there's only one of them) - has this been factored into the 
balance-rack algorithm for CSW

ie CSW should have a different valuation for the Q to TWL, as would 
the U

even though a leave of EET and EOT value the same, EOT should be 
ranked higher since there are 3 letters, not two. 

Do the valuations on letters adjust depending on how many letters 
are left in the bag, or how many of an individual letter remain?

A customisable nominal value for each letter would be a nice 
addition.

Best scrabble AI available. (except the endgame)

Regards,

TcheQ


Reply via email to