The Spacious Mind
Click to go back to The Spacious Mind Tests
About Clones Testing

There are several well known ways for testing chess computers to see if they are identical. Some of the most popular ways are:

  • BT 2450 and BT 2630 Test
    • These tests were developed by Hubert Bednorz and Freddi Tönissen and comprise of 30 test positions for which the chess computer is given 15 minutes for each test position to find the correct move. Typically the computers are setup with the Infinite Time Level and a timer is used for observing when the computer finds the correct move. This time is then noted for each position. If the computer does not find the correct move in 15 minutes (900 seconds) then 900 seconds is noted as the time for the problem. After playing through all the test positions and noting the times in a spreadsheet that is provided for this test, the formula in the spreadsheet will calculate the playing strength for the chess computer based on these tests.
    • These tests are very useful for identifying a clone suspect because these suspects will post exactly the same times for all or most of the test positions and this allows you to then take a closer look at the suspected machines and try out some other tests.
    • These BT tests are good for obtaining playing strengths for machines above 1800 ELO (2000 USCF). Weaker machines would have problems in these tests.
  • Replay a game Test
    • If you think that two computers may actually be identical machines although disguised by the manufacturer through different housings, you may want to let them play a game or two against each other. You start by letting the two machines play a game from start to finish while writing down all the moves. Then you reset the machines and replay the moves with the other machine. If all the moves for the whole game are repeated exactly the same by the other machine then it is most likely identical. It is important to turn off any random or selective options that the chess computer may have to limit any random moves the computer may generate. As a rule of thumb, machines may be accepted as identical if they repeat the same moves 95% of the time. Repeating this test a few times would help to validate your findings.
  • Play 1.f3
    • A quick and easy way to test your suspicions before you spend too much time verifying them with other tests is to set up the machines with exactly the same game settings and play 1. f3 on both machines at the same time. The 1.f3 move is useful because this move will take most dedicated chess computers out of their opening book. If both machines reply with the same move at exactly the same time then you probably have two clones. You may want to repeat this a few times because some machines still have some randomness in their play, even if this option is switched off. This may cause them to vary their reply and alternate between these two most likely replies 1. ... d5 or 1. ... e5. Once you have tried this you may then want to confirm your findings with the Replay a game Test or the BT Tests.
GK 2100 Clones Test

This test is is really a Replay a game Test but instead of trying to prove the existence of a clone, the goal is to find any differences in their game play and try to identify at what point in the games these move deviations occur.

In these tests, machines that reply on average more then 90% of the time with the same move can be considered as having the exact same software program.

The idea for these tests was formed while reading an article in which it was stated that Radio Shack Champion 2250 XL was just another one of many Saitek Kasparov GK2100 clones.

To test the validity of this article the following machines have been added to this test. All the Saitek machines below are listed as clones.

              • Saitek Kasparov Cougar
              • Saitek Kasparov Cosmos
              • Saitek Kasparov GK 2100
              • Saitek Kasparov Travel Champion 2100
              • Saitek Kasparov Centurion
              • Saitek Kasparov Travel Expert
              • Radio Shack Champion RS 2250 XL
              • Mephisto Roma II
              • Novag Sapphire

Each machine was forced into playing two 40 moves in 2 hours games, using exactly the same opening each time, the Ruy Lopez Exchange Variation both as Black and as White against Fritz 8 on a Home PC running at 1.8 GHz. Next three games were picked at random that were played by a Saitek computer. The test results are shown below:

Test # 1

GK 2100 Clones Test # 1

This first test shows quite well that Radio Shack Champion 2250 XL deviates more than Cougar, Cosmos, Centurion, TC 2100, GK 2100 or Expert Travel. Also interesting in Test # 1 is that truly different software as in the examples of Richard Lang’s Mephisto Roma II and Dave Kittinger’s Novag Sapphire deviate in their move choice by at least 30% during the course of a whole game. Machines with similar playing strengths are used in these tests..

It might be interesting in the future to add one or two machines that are weaker or stronger by 200 to 300 Elo points in order to identify how much a machine’s playing strength influences the choice of moves.

The table above does show quite well that RS 2250 XL’s program is not an exact copy of the other Frans Morsch programs tested.

Test # 2

GK 2100 Clones Test # 2

Test # 2 shows similar results to Test # 1. In this test, a game with fewer moves was picked to see if this would make a difference in the amount of deviations. This time a Saitek Kasparov GK 2100 game was used.

Test # 3

GK 2100 Clones Test # 3

A Saitek Cosmos game was used in this final test. This time everyone was playing as Black. The results in Test # 3 are very consistent with the results of Test # 1 and Test # 2.


GK 2100 Clones Test Summary

As seen in the above summary chart, the results in the three tests remained very consistent. In the future more machines may be added to these tests. Certainly Mephisto MMV, Mephisto Mystery and a Cosmos with 24 MHz would be interesting. Perhaps Atlanta, Magellan and Milano Pro as well.

The results seem to nicely show the following if the machines are of similar playing strength:

  • Closely Related
    • These seem to stay above 90%.
  • Modified or Enhanced
    • Below 90% and possibly as low as 70%?
  • Different Software Program (er)
    • These should be below 70%

Certainly more of these types of tests may provide lots of additional new information. How would these charts look if several Morsch’s and Kittingers replayed a Lang game? Is Lang’s software more closely related to Kittinger or to Morsch? What about other Programmers? ..........

[Home] [Collection] [Tournaments] [Tests] [Computers] [What's New] [Links]