Discussion of the 6-3 split has been popping up here and there ever since Maserati used it to beat Jono's 40 line world record*. In theory, it should offer a slight edge, but it's a headache to learn. For that reason, I spent a good six hours gathering and analyzing data today.
Why would the 6-3 offer faster play, anyhow? Well, as Someone2knoe first pointed out
, it forces you to avoid certain unfavorable placements. Here's why:
In the above diagram, placing pieces in the green columns tend to require the least movement, while the red ones tend to require the most, roughly.
If we position our tetrising column on the 7th file, we can force ourselves to eliminate the possibility of some of these "red column" placements:
By eliminating the possibility of these placements, we inadvertently use less keys.
For my experiment, I transcribed and analyzed over 750 piece placements. I played three games of 100 lines each, using three different methods. I used perfect finesse
, 180 rotation, and didn't use hold. The three methods were: 6-3 split, traditional (making tetrises all the way to the right), and "free form" (not going for tetrises at all). I played 100 lines per game instead of 40 in order to gain more accurate statistics. I played as naturally and logically as possible.
After transcribing the replays to fumen, I recorded how many meaningful keys per tetromino (mKPT) each perfectly-finessed piece required. In the past (when I showed how SRS required less KPT than ARS), people brought up the argument how KPT isn't useful since it doesn't take into account how a player can rotate and move simultaneously. For this reason, I coined (yet another) new metric: mKPT. To calculate it, each movement, rotation, and hard drop counts as one--except when we can execute movement and rotation simultaneously within a single frame. In these cases, mKPT combines them into one key input.
1 mKPT (hard drop)
2 mKPT (rotate, hard drop)
2 mKPT (simultaneous rotate+move, hard drop)
3 mKPT (rotate and move events cannot occur simultaneously here)
3 mKPT (move, move, hard drop)
The games6-3 Split:
- Probably requires the least mKPT on average
- The least intuitive
- Splits the playfield which gives less space to work with per island, making stacking more difficult
- Difficult to play successfully without using hold (which adds mKPT). Maserati doesn't use hold, but he deals with troublesome pieces by skimming (perhaps adding some mKPT along the way). bach_of_tetris adds, "For me it only seems to work like once every 5."
- Most popular (won't require unlearning for the typical player)
- A single large island which makes placements easy
- Possible to play most Sprints without needing hold or skimming
- Probably requires the most mKPT on average
- Most intuitive
- No islands, which makes finding placements easier than even the traditional method
- The method most likely to offer placements where you can simply hard drop without rotating or moving
- Useful in TOJ, where smaller line clears take off less delay than bigger ones
- A little unfamiliar at first if you're already used to the traditional method.
- More mKPT than the 6-3 split
Here's the raw data
So the experiment backs up the theory, but just how much of an improvement is it anyway? Let's take a hypothetical player who goes at a pace of exactly 10 mKPS. Let's say he tries all three methods out and finishes with a perfect clear in every game. Here's what his records would look like: ((mKPT*100)/mKPS)
- 21.25 seconds (+0.00), 6-3 Split
- 23.60 seconds (+2.35), traditional
- 22.77 seconds (+1.52), free form
So, it looks like the 6-3 split definitely has some substance to it. Saving two whole seconds sounds pretty good, right!?
However, I wonder if there's more to it than that. One major criticism I'd like to bring up is how mKPT (and KPT as well) might not be all that important in the first place. Why? First off, the most important factor here is "processing speed." Imagine an interface where the player can "think" each piece into position, 1 by 1, using a total of 0 KPS. Common sense tells us that he probably wouldn't perform at a TAS-level
speed all of a sudden. He still needs to think it through, doesn't he? So, what I'm trying to say is that there's likely a "KPT ceiling." Once KPT reaches a certain point, it's likely enough to handle even the fastest player's processing speed. Take for example WR* holders Apocalypse and Maserati. They don't use 2-step finesse or 180 rotation (which would save more keys than a 6-3 split), but they sit at the top of the leaderboard nonetheless. I think think it's a mistake to say that if you drop 0.15 KPT, your 10KPS will stay the same (and as a result gain TPS). In all likelyhood, your KPS will probably drop uniformly until you improve what's actually important--your processing speed. This would explain why we don't see any Keyblox or Typomino WRs yet.
Conclusion? Do what's in your heart! :hug:
*Note: yes, I realize the Koreans have the WR, but we can't see the replay, so what's the point?