Rebel 9 Review by     Claudio Bollini 



When a hyper-top program like Rebel reappears with a new version, it 

inevitably produces a lot of waves on the computer chess sea. The 

questions that always arise are: Is the engine noticeably stronger than 

previous versions? Does this release have more and better features? In a 

few words: Was the upgrade justified?



Many of us think that Rebel is so strong that only one other program 

(Hiarcs 6) can compete with it for the crown of the "strongest program 

available", as well as for the "most human like style". The proper Ed

Schroder's philosophy was (is it yet?) to keep this approach as the main

goal, and that is why he stays in the DOS platform. So, I believe that

here the chess engine itself has to be considered as the most important 

feature, contrarily to some windows bells and whistles programs.



However, to be fair to those who care strongly on the GUI side of Rebel,

I will also analyze some new database and playing options.





Graphical User interface and general features



Rebel 9 is certainly a full-featured program, even for today's high

standards. There are a lot of options to modify the chess engine 

behavior and capabilities, but these remain almost unchanged from 

Rebel 8.



An interesting new possibility is the autoplay option, that allows you

to run matches to test different settings against each other, or to 

confront Rebel 9 with previous engines (Rebel 8, Rebel 7, Rebel Decade).

However, the latter is not so interesting as for Fritz 5, which includes

modules with a truly different approach like Hiarcs 6 or Doctor 2. It 

would be great to see the next Rebel release including some "foreign" 

engines, for better testing of its real strength.



The main screen appearance of the Rebel family was always for my taste 

the clearest and nicest I have ever seen of all the chess programs, and 

Rebel 9 fortunately keeps the tradition with some minor improvements. I

really love the big board of great aspect, that doesn't seize any vital 

space for each time more informative windows.



Nevertheless, there is a problem which becomes more serious with each 

new release: the options distribution in the menu. The layout seems to

me here a bit untidy and confusing. Only a few examples: there are

searching database functions both into and outside the database screen; 

in the "options" item you can find a file saving entry, and the "PGN" 

option gathers PGN, EPD and BookUp management. Definitely, I/O 

operations are quite unclear. In my opinion, a reorganization and

reassignment of the different functions are urgently needed.



The new on-line manual is very welcome, as it has been a missing feature 

in strong DOS programs. The bad news is that a paper manual is no longer 

available. Obviously, Rebel team wanted to cut their costs, but a really

portable manual, to carry it anywhere to study it thoroughly, or just to 

take a glance, is almost indispensable, especially for those who are not

familiarized with chess programs.





Database and book



No doubt, the greatest effort for this Rebel 9 version (putting aside 

the engine issue) has been invested in new book and data management 

capabilities. 



The book size developed by Jeroen Noomen has been increased from 870,000

to 1,250,000 positions, upgrading it with new lines from latest theory. 

Besides, it has a wide repertoire including a balanced selection of 

frequently played lines. A remarkable feature is its new "genius book 

learning system", that automatically adjusts the frequency scores of 

Rebel's book according its historical performances. Therefore, it tends

to avoid losing lines and turns to frequently successful lines.



The information about the learning method was very scanty, but if things 

are like it seems to be, I believe that there is something not so 

accurate here. It is rather logical to learn from bad experiences with

a given variation and adjust its frequency down. But the symmetrical 

method is not necessarily adequate! If after winning, Rebel increases 

the score of the line used, regardless the opponent Elo or its own 

setting, the modified frequencies will become, at least, arbitrary.



There is a very useful option that lets Rebel 9 engine analyze any 

opening book in Rebel format (.mvs), and save the evaluated scores in

it, not only to have an idea about which lines are more promising 

according Rebel's opinion, but also to find out which ones fit better 

with its own style. The standard Rebel book comes with pre-analyzed 

scores from its few first moves. You can analyze also your own books,

but you have to convert them first into Rebel format. For do this there

is an external program which is included in Bonus CD.



Now Rebel 9 has a database environment with local options to perform 

databases export/import, creation and analyze, PGN conversion, doubles 

removing, searching by names and text, etc. There are some new and very 

nice features: When you select a game from the database list, you see 

two at the right small boards: the first one is the terminal position

from the ECO line, and the second, the final game position. An 

evaluation button can be clicked to view Rebel's evaluation. Besides,

the overview can be extended to 15 boards, which show the entire 

evolution of the game. (Surprisingly, you can find the same option in

the main screen, inside "Extended" option, with the difference that

there Rebel will put its evaluation below of each board).



The searching options have been extended, but (again, the options layout

problem), the most important of them must be located outside the 

database environment, in the main screen. They are: "on moves", for 

listing the games at the same move number of present game; "on 

position", to scan for identical boards; "pattern", that searches in

"exclusive" mode for same pieces at same positions (not AND/OR options);

and "on material", similar to pattern option, searches for the exact 

number of pieces entered.



As an interesting addition, the bonus CD has a lot of useful data: i.e., 

a selected database of 107,000 top master games; 56 Mb of thematic

editable books; many test sets with 1258 endgames, 300 positions from 

Win in Chess, 1001 tactical sacrifices, among others.





Chess Engine



To evaluate the announced strength improvement of 30 Elo due to the new 

engine (plus another 20-50 for the book learning function), was the 

hardest task of all. 



Of course, the gain from the book side can be only probed playing 

hundred of games with and without the function, and even in that case it

will be too fuzzy to measure, since the book autocorrection is a slow

process, not a unique on/off state. Besides, after how many games could

the 20-50 Elo gain be verified, and under which conditions?



So I concentrated in the engine, and ran a lot of positions; but they 

failed to find an appreciable difference between Rebel 8 and 9. 

According these tests the strength margin is too narrow. For instance, 

I've processed 310 positions from a set of endgame positions included 

in the Bonus CD, with an MMX 166, 32 Mb RAM, at 1 min of fixed time.

The hits were 228 for Rebel 9 vs. 226 for Rebel 8 (with higher total 

solving time for Rebel 9). Something similar happened with middlegame

(tactical and positional) tests.



Another fact I noticed is that Rebel 9 is slower than R8 in many 

positions I tested (from 10% to 30%). I supposed that this was a clear

symptom of the addition of a considerable amount of knowledge to the 

evaluation algorithm, that incidentally was not been reflecting in test 

positions. Ed Schroder confirmed me later that there is a 20% slow down

in nodes per second, and that effectively it is due to the presence of 

much more knowledge. "Result, better (overall) positional play", 

according Ed's own words. However, he doubted that I could arrive at a

conclusion; even more, he affirmed that nobody can clearly determine 

engine improvements these days.



He also called into question how representative a set of tests could be,

telling me that at least 300 or 400 games are needed to get a good

estimation. I partially agree with him: there is no doubt that i.e. a

match of several games against a strong GM would be the best way to find

out program improvements, but as there are so many fixes to test and 

ideas to implement that it is impossible to subdue each change to such

matches. So, tests are essential for programmers at development time.

And once the release is out, they become simply a reasonable indicator 

for users and testers.



Rebel can now deepen twice its search, up to 60 plies. There were some

positions I sent Ed that Rebel 8 could not solve, and now Rebel 9 could,

due to this new capacity. But this is not only a matter of some rare

positions. When Rebel 9 arrives at the endgame, it is not unusual to see

its new maximum depth indicator going beyond 40 or 50 plies.



As for the endgame, there were some rook and pawn endgames that now 

Rebel seemed to be able to solve. Ed told me that he has tuned up the

endgame evaluations, and now the scores are more reliable.



The famous Rebel style remains untouched. Terrifically well balanced in

all the stages of game, able to understand and react accordingly in a

wide type of positions. Other programs could specialize better in 

certain themes or phases, but no one covers so fairly all the aspects 

of game. Rebel is a master at harmonizing the evolution and cooperation

between pieces, and moving in pursuit of a specific goal. It really 

deserves its popularity for its enjoyable genuine human style.



As I said, I couldn't determine yet any change in Rebel 9 playing style,

so personalities' screen will remain the same. For the first time since

I started my reviews, a new version doesn't imply a displacement from

previous one. (Naturally, this is not necessarily bad!):



  Tricker                     Hunter      

    * - - - - - - + - - - - - - *         R9 = Rebel 9

 ^^ |             .       TK    |         F5 = Fritz 5

 AD |             . F5      M6H6|         H6 = Hiarcs 6

    |             .  WC  NM     |         WC = WChess

    |             .             |         TK = The King/CM5000     

    |             .          R9 |         M6 = MChessPro 6

    | . . . . . . . . . . .G5. .|         NM = NIMZO 3.5

    |             .             |         G5 = Genius 5               

    |             .             |                                   

    |             .             |         

    |             .             |         AD = Aggressiveness

Degree      

    |             .             |         SD = Strategical Degree     

    * - - - - - - + - - - - - - *    SD>

  Hesitant                    Timid





Conclusion



If you are a strong player and you do not yet have any master class 

chess program, Rebel 9 will be probably your best choice. It has many

assorted features to support a large set of book, database and user 

interface operations. And, most important, you have its extraordinary

chess engine, capable of passing the Turing Test in front of a 

Grandmaster.



We believe that Rebel 9 suffers an overgrowth of a somewhat disorderly 

set of features. But we sincerely hope that Ed Schroder will keep 

prioritizing the develop of his valuable engine, and continue investing

his best efforts in heightening even more the strength and style of 

that which is perhaps the best chess program available.