Most games in Arika's Tetris The Grand Master series randomize the order of tetrominoes using an algorithm that makes successive identical tetrominoes less common. The system involves keeping a history of the four most recent tetrominoes and trying to choose a random tetromino not in the history. Each time a piece is generated, the game will try a certain number of times to generate a tetromino that doesn't match any in the history; if all tries fail and generate a recent piece, the game will settle with that recent piece. The overall effect is to minimize several of the same tetromino showing up in succession, while maintaining a certain amount of unpredictability. TGM1 uses 4 tries, while TGM2 uses 6 tries.
The four-piece history is not necessarily a unique list. If the randomizer fails to generate a non-recent tetromino, which happens about 3.5 percent of the time in a 6-try system, then two or more of the same tetromino may occupy elements of the history.
A few additional behaviors exist in the beginning of the game:
- The game never deals an S, Z or O as the first piece, to avoid a forced overhang.
- The state of the history is initialized to a fixed state:
- In TGM1, the history begins filled with 4 Z pieces.
- In TGM2 (and TGM2+), the history begins with a Z,Z,S,S sequence. However, as the first piece of the game overwrites the first Z rather than pushing off the last S, this is effectively a Z,S,S,Z or Z,S,Z,S sequence.
Unlike TGM1 and TA (P), Ti uses a bag of 35 pieces to draw from rather than drawing uniformly randomly from the seven pieces (while performing history checking). The bag initially contains 5 of each piece. However, unlike a true bag (which slowly empties with each drawn piece before being completely refilled), Ti's random generator puts a copy of the least recent piece back into the bag each time a piece is taken out. While the TGM randomizer naturally prevents "floods" (many of the same piece in a row), this additional behavior has the effect of preventing "droughts" (where a certain piece fails to show up for a long time). At an extreme, it is impossible for any piece to fail to appear for longer than a 35-piece sequence.
This is not a perfect description of Ti's randomizer; for full elaboration including some subtle bugs in the algorithm, see the reference link below.