Installing it from the CD (the same one for Windows or Macintosh) is very easy. You just tell it what operating system you have, and it installs itself. Once it is installed onto your hard disk, you do not need the CD again.
The graphics are very nicely done. The appearance of the board is the best I have seen in any program. But on a Windows system, it does require 256 colours: with only 16, it runs but essential information is invisible.
It has options of 19-by-19, 13-by-13, and 9-by-9 boards; all four combinations of human and program playing Black and White; up to nine handicap stones; and three skill levels. It allows you to take back moves.
It can record partly-played games for re-loading later, and does this using popular "Ishi" format. I am particularly pleased about the last point: most authors of Go programs, for reasons which I cannot guess at, devise their own new and incompatible formats for game records. Mark Boon has shown that there is no need for this, and I hope that other Go programers will follow his example.
A consequence of this is that you can use it to play through Ishi-format game records, on a more attractive board than other game-recording programs. You can also use it to record games in Ishi format, by setting it to a "Human versus Human" game (however it ignores all comments and variations). And if you play a game against it, you can then play through the game with another program such as GoScribe or Yago, and add a commentary.
I found it easy to beat on nine stones, whereas I have not yet managed to beat HandTalk, the current world computer Go champion, on nine stones (I am 1-kyu). So when I played them against each other, I expected HandTalk to win. I set each to its maximum strength, to play on a full board with Japanese rules, and Goliath as black giving 5½ points komi. After a rather chaotic game, Goliath won by 16½ points.
There must be something about the different styles of the two programs which allows me to beat one of them much more easily than the other. Their styles certainly differ greatly. Goliath seems to make good sensible standard shapes, while HandTalk has way of starting complicated and unreasonable fights, and then reasoning out, sometimes correctly, how to win them.
As you play, you have an option of four soundtracks of background music, or none. You need a soundcard to hear this. I have none, so I cannot comment on the music.
If you use it, you may find that it appears to be very slow.
It is not: it can complete its part in a full-board game in ten
minutes, on my 66mhz 486. But (on my Windows 95 system) it has
a problem with detecting that it is its turn. I found that I
can overcome this by playing my stone, and then twitching the
mouse just to wake it up.
Tsume Go Goliath
Tsume Go Goliath is another program by Mark Boon. It
displays Tsume Go problems, and knows the answers to them.
Indeed, using new techniques in artificial intelligence,
it not only knows the right answers, it is able to produce
refutations for all the wrong answers.
System requirements, and installation, are the same as for Goliath 3.5, above.
It has a repertoire of almost 1,000 tsume problems, varying in difficulty from 20-kyu up to shodan. (That is what the manual says. The program rates me as 11-kyu, so I would like to think that the problems are actually more difficult than this.)
It operates in two modes, "browse" and "test". In "browse" mode, you select one of its 979 problems, and try to solve it. In "test" mode, the program uses its assessment of your strength, saved from a previous session, and selects a problem for you. Either way, when you think you have found the solution, you click on the point where you want to play. The program answers (without revealing whether you are right or wrong) and it is your move again. Eventually, if you get the entire sequence right, you see an animation of a bobbin-doll bowing to you and congratulating you. If you get it wrong, it continues playing until your failure is obvious, and then the bobbin-doll appears, bows, and tells you to try again.
At any point while trying to solve a problem, you can back up one move, back up all the way to the start, or give up and ask it for the answer.
In "test" mode, there is also a timer for each problem, set to two minutes. The program continuously adjusts its estimate of your strength, and is able to show you a graph of how this has been changing.
The problems are all Black to play: some to kill, some to live. For a few of them, the objective can only be achieved in ko - this is not stated, and you are meant to read it out for yourself. For others, it is possible to win outright, and if you only achieve a ko the bobbin-doll then tells you so.
I found this program much more rewarding than a book of tsume-go problems. This was because of its ability to refute all possible mistakes sensibly. If I try tsume-go problems from a book, and get one wrong, the book can generally not show me why my answer is wrong. This not only spoils that problem for me, it leaves an unpleasant feeling which distracts me from subsequent problems.
The problems are "real" ones, from Japanese composers. I found these more rewarding to try to solve than the computer-generated problems, with sealed-off edges, that are offered by Thomas Wolf's  GoTools program. However, Tsume Goliath Go, unlike GoTools, cannot solve problems entered by the user. Moreover it has only a finite stock of problems (GoTools is supplied with 12,000 and can generate more). But there are plans to supply further problems for it, for customers who have worked their way through all 979.
For anyone wanting to buy a good program with which to improve their reading ability, I would recommend Tsume Go Goliath.
review by Nick Wedd, January 1997