An analysis of Microblizz WR 40-line run

[belzebub]

Time for an analysis of Microblizz' amazing world record. First the (correct), general stats:

time   : 20.55
ppm   : 300.72992
pps   : 5.0121655
pieces   : 103
holds   : 5

The youtube video version still has a bug for the ppm/pps (corrected in my mod). This is what it should look like:

[!--ImageUrlBegin--][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/general.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


And here is the fumen of the full sprint:

[fumen]110@eessnbuswbC7UAAEcqUmboUydYimbaitbnSAEcdDmb?dDndlzmblz4bGTAEc6Gob4GxdIwobKwqbjSAEcNRnbNRpd1?lnb1l0bcSA6bwNnbxNnbwNddQpnbRpnbQp6byRA5boUmbqU?qdaimbYiAcnRACcdDmbdDWdlzmblzLccSA6bwNnbxNnbwNb?dQpnbRpnbQp8bNRA3bUKobVKnbUKIdssobtsnbssQcDRABc?NRnbNROd1lnb1lSc5QA2bEYobEYobEYobEY3c8eob8eob8e?ob8eOc+QA2b4Gob4Gnb5GDdIwobIwnbJwWcMQA2bwNnbxNn?bwN9cQpnbRpnbQpecaQA0bpUoboUoboU4cZiobYiobYihcf?QA0bcDobdDobcD2ckzoblzobkzjcJPAzbEYobEYobEYobEY?ic8eob8eob8eob8emcOQA1b5Gnb4Gob4G7cJwnbIwobIwfc?VPAzbUKnbWK3cusnbss3cDPA9bNRnbNRpc1lnb1l7cJNAvb?EYobEYobEYobEY8b8eob8eob8eob8eQdiOA7bqUmboUecYi?mbaiKdVPAzbUKnbWK6cusnbss0cDOA7bNRnbNRRc1lnb1lV?dJSA5bEYobEYobEYobEYZd8eob8eob8eob8epbeQAdc4Gob?4Gnb5GTcIwobIwnbJwfccQA2bwNnbxNnbwN6cQpnbRpnbQp?hcnPA+bdDmbdDwclzmblz1cvOAxbcDobdDobcDfckzoblzo?bkz9cDQA/bNRnbNR8c1lnb1lmcqOAyboUoboUobpUacYiob?YiobZiBd1OAybUKnbWKvcusnbssAdEOA6bxNobxNRcRpobR?pUd+OAyb4Gob4Gnb5GjcIwobIwnbJw6cJSA5bEYobEYobEY?obEYZd8eob8eob8eob8epbqOAacoUoboUobpUybYiobYiob?ZiBd8PA1bwNnbxNnbwNxcQpnbRpnbQprcPQA0bcDobdDobc?D8ckzoblzobkzdcNPAzbUKobVKnbUKmcssobtsnbss2cjPA?+bNRnbNRzc1lnb1lwcuPA0b5Gnb4Gob4GycJwnbIwobIwpc?mOA7b6Gob4GgcIwobKwEdaOAwbpUoboUoboUScZiobYiobY?iLdlOA7bWKnbUKfcssnbusHdDOA7bNRnbNRYc1lnb1lOdsN?AxbwNnbxNnbwNPcQpnbRpnbQpRdHOA7bdDmbdDVclzmblzT?dJSA5bEYobEYobEYobEYZd8eob8eob8eob8epbDPAlcNRnb?NRCc1lnb1l6cWPAyb4Gob6G3cKwobIw2cJSA5bEYobEYobE?YobEYZd8eob8eob8eob8epbHRApcdDmbdDlclzmblzVciQA?/bqUmboUCdYimbaiic8QA3bwNnbxNnbwNDdQpnbRpnbQpXc?FQA+bWKnbUK9cssnbusmc1PA0bUKnbWKEdusnbsspcDQA/b?NRnbNR1c1lnb1ltcWPAyb4Gob6G9cKwobIwwcsOAzbwNnbx?NnbwNhcQpnbRpnbQp9cqOAyboUoboUobpUacYiobYiobZiB?dJSA5bEYobEYobEYobEYZd8eob8eob8eob8epb/RAfccDob?dDobcDrckzoblzobkzDcaQA0bpUoboUoboU2cZiobYiobYi?jcERAAcxNobxNJdRpobRpWcjQAAcNRnbNRFd1lnb1lccWQA?0b4Gob6GLdKwobIwgc1PA0bUKnbWKEdusnbsspcnPA+bdDm?bdDwclzmblz1cJSA5bEYobEYobEYobEYZd8eob8eob8eob8?epbFRAocWKnbUKhcssnbusYcKRA3boUoboUobpUOdYiobYi?obZiIccQA2bwNnbxNnbwN3cQpnbRpnbQpkcvQA1bcDobdDo?bcDDdkzoblzobkzVcjQAAcNRnbNRAd1lnb1lhcuQA2b5Gnb?4Gob4GFdJwnbIwobIwUcJSA5bEYobEYobEYobEYZd8eob8e?ob8eob8epbyRAhcoUmbqUBdaimbYiBcNRA3bUKobVKnbUKI?dssobtsnbssQcJSA5bEYobEYobEYobEYYd8eob8eob8eob8?eqbjRACcNRnbNRSd1lnb1lNcfQA0bcDobdDobcD2ckzoblz?obkzjc+QA2b4Gob4Gnb5GHdIwobIwnbJwScMRA4bwNnbxNn?bwNNdQpnbRpnbQpMcmQA/b6Gob4GAdIwobKwgcfPAybcDob?dDobcDkckzoblzobkz3ctQA2bUKobVKnbUKFdssobtsnbss?Uc8PA1bwNnbxNnbwN1cQpnbRpnbQpncDQA/bNRnbNR4c1ln?b1lqcJSA5bEYobEYobEYobEYZd8eob8eob8eob8epbSRAgc?oUmbqU1caimbYiOciQA/bqUmboUDdYimbaihctRA4bUKobV?KnbUKWdssobtsnbssBcHRABcdDmbdDPdlzmblzTcMQA2bwN?nbxNnbwN9cQpnbRpnbQpecjQAAcNRnbNRCd1lnb1lfcJSA5?bEYobEYobEYobEYZd8eob8eob8eob8epbGSAqc6Gob4GxcI?wobKwEcJRA3bEYobEYobEYobEYGd8eob8eob8eob8e+bmRA?Bc6Gob4GUdIwobKwKcjRACcNRnbNRRd1lnb1lOcaRA2bpUo?boUoboUKdZiobYiobYiNcvQA1bcDobdDobcDDdkzoblzobk?zVcJSA5bEYobEYobEYobEYZd8eob8eob8eob8epbAAAXecD?mbpUdDhbEYhbNRoU4GcDhbEYqbAAA[/fumen]

So almost sub20, an amazing 300+ ppm and 5+ pps! He used 103 pieces and a few holds, so there is still room for a very little improvement here.

You can see the (new) numerical reports with the graphs here:
[a href=\\\"http://tetris.vanderhoven.be/MicroBlizz%20-%20WR%20-%20Analysis%20Part%202.html]http://tetris.vanderhoven.be/MicroBlizz%20...d%20Record.html

And you have a report with a fumen that you can drag next to the (interactive) graphs here:
http://tetris.vanderhoven.be/Microblizz_WR...y%20Report.html

General

In the graph we see that his speed is rather consistent, a few parts where he went a bit slow and a slow recovery overall. In the Piece speed graphs we see that placements are all very consistent, with not so much peaks.

Global speed graph (in ppm, higher is better):
[!--ImageUrlBegin--][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/speed.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Piece speed graph (in frames, more frames is bad):
[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/piecespeed.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/piecespeed.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Finesse

Now for the finesse, Microblizz only once did a keypress more then the 2-step finesse scheme on the wiki, making his kpt roughly the same as is ekpt (ignoring the hold bug for now):

kpt   2.8737862
ekpt   2.8640773

It's not really a fault if we look where the difference comes from, in frame 131, he placed an I in column 3 by rotating and tapping it twice. You can see it in this picture:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/finesseI.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/finesseI.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


The 2-step finesse scheme says, das to wall, rotate right, drop. Because the das to wall needs perfect timing and Micro is very fast in tapping twice, his scheme might be optimal for him. Also the rotation and the first tap left can go together. So no time wasted by Finesse other then the uses of hold. Also remark this position is not so common and the fact that it's for most players quite difficult to get perfect I-finesse at high speed.

Use of Hold

But then, why the hold, well Microblizz makes smart use of the hold, that is, he tries to only keep I's. He seems to always keep his first I (as also seen in previous records of Blizz). Then at frame 283 he changes his I, because he can't play the Z. Then he gets a J and an S, and instead of just dropping it, he decides to rotate it and use his hold again to put the Z in place. It's unclear why and it looks like to me this was confusion due to using hold and the nature of the S/Z pieces. Then another I comes and Blizz uses hold again to get the I into hold. Due to the nature of the finesse detection and the 'hold' bug, this is not signaled and we had to use the replays for this. Amazing he could do this at this speed. We see just a little speed drop under the global speed at this time.

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/283.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/283.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]
[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/284.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/284.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/307.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/307.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/315.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/315.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/323.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/323.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Then at frame 1161 he uses hold again to change a T for an I:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/1161.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/1161.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


I don't really understand why, also the I is placed not on the side gap, but on the 9th column:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/1175.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/1175.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Maybe that was a missdrop, as Micro needs now 4 pieces before his next I (getting to 103 pieces) to finish the game. However I don't think it was a missdrop and MicroBlizz strategy was just to keep on stacking tetrises. After all his sprint were only tetrises, wich is a noble thing on it's own. I wonder what the second best time in the world is for doing sprints with only tetrises.

If the I was on the side, he could probably finish with one piece less (I-10 -> J-9, O-7, L-3, S-2). For example, an alternative ending from piece 97:

[fumen]110@eessnbuswbCLHEcqUmboUydYimbaitbnSAEcdDmbdD?ndlzmblz4bGTAEc6Gob4GxdIwobKwqbjSAEcNRnbNRpd1ln?b1l0bcSA6bwNnbxNnbwNddQpnbRpnbQp6byRA5boUmbqUqd?aimbYiAcnRACcdDmbdDWdlzmblzLccSA6bwNnbxNnbwNbdQ?pnbRpnbQp8bNRA3bUKobVKnbUKIdssobtsnbssQcDRABcNR?nbNROd1lnb1lSc5QA2bEYobEYobEYobEY3c8eob8eob8eob?8eOc+QA2b4Gob4Gnb5GDdIwobIwnbJwWcMQA2bwNnbxNnbw?N9cQpnbRpnbQpecaQA0bpUoboUoboU4cZiobYiobYihcfQA?0bcDobdDobcD2ckzoblzobkzjcJPAzbEYobEYobEYobEYic?8eob8eob8eob8emcOQA1b5Gnb4Gob4G7cJwnbIwobIwfcVP?AzbUKnbWK3cusnbss3cDPA9bNRnbNRpc1lnb1l7cJNAvbEY?obEYobEYobEY8b8eob8eob8eob8eQdiOA7bqUmboUecYimb?aiKdVPAzbUKnbWK6cusnbss0cDOA7bNRnbNRRc1lnb1lVdJ?SA5bEYobEYobEYobEYZd8eob8eob8eob8epbeQAdc4Gob4G?nb5GTcIwobIwnbJwfccQA2bwNnbxNnbwN6cQpnbRpnbQphc?nPA+bdDmbdDwclzmblz1cvOAxbcDobdDobcDfckzoblzobk?z9cDQA/bNRnbNR8c1lnb1lmcqOAyboUoboUobpUacYiobYi?obZiBd1OAybUKnbWKvcusnbssAdEOA6bxNobxNRcRpobRpU?d+OAyb4Gob4Gnb5GjcIwobIwnbJw6cJSA5bEYobEYobEYob?EYZd8eob8eob8eob8epbqOAacoUoboUobpUybYiobYiobZi?Bd8PA1bwNnbxNnbwNxcQpnbRpnbQprcPQA0bcDobdDobcD8?ckzoblzobkzdcNPAzbUKobVKnbUKmcssobtsnbss2cjPA+b?NRnbNRzc1lnb1lwcuPA0b5Gnb4Gob4GycJwnbIwobIwpcmO?A7b6Gob4GgcIwobKwEdaOAwbpUoboUoboUScZiobYiobYiL?dlOA7bWKnbUKfcssnbusHdDOA7bNRnbNRYc1lnb1lOdsNAx?bwNnbxNnbwNPcQpnbRpnbQpRdHOA7bdDmbdDVclzmblzTdJ?SA5bEYobEYobEYobEYZd8eob8eob8eob8epbDPAlcNRnbNR?Cc1lnb1l6cWPAyb4Gob6G3cKwobIw2cJSA5bEYobEYobEYo?bEYZd8eob8eob8eob8epbHRApcdDmbdDlclzmblzVciQA/b?qUmboUCdYimbaiic8QA3bwNnbxNnbwNDdQpnbRpnbQpXcFQ?A+bWKnbUK9cssnbusmc1PA0bUKnbWKEdusnbsspcDQA/bNR?nbNR1c1lnb1ltcWPAyb4Gob6G9cKwobIwwcsOAzbwNnbxNn?bwNhcQpnbRpnbQp9cqOAyboUoboUobpUacYiobYiobZiBdJ?SA5bEYobEYobEYobEYZd8eob8eob8eob8epb/RAfccDobdD?obcDrckzoblzobkzDcaQA0bpUoboUoboU2cZiobYiobYijc?ERAAcxNobxNJdRpobRpWcjQAAcNRnbNRFd1lnb1lccWQA0b?4Gob6GLdKwobIwgc1PA0bUKnbWKEdusnbsspcnPA+bdDmbd?Dwclzmblz1cJSA5bEYobEYobEYobEYZd8eob8eob8eob8ep?bFRAocWKnbUKhcssnbusYcKRA3boUoboUobpUOdYiobYiob?ZiIccQA2bwNnbxNnbwN3cQpnbRpnbQpkcvQA1bcDobdDobc?DDdkzoblzobkzVcjQAAcNRnbNRAd1lnb1lhcuQA2b5Gnb4G?ob4GFdJwnbIwobIwUcJSA5bEYobEYobEYobEYZd8eob8eob?8eob8epbyRAhcoUmbqUBdaimbYiBcNRA3bUKobVKnbUKIds?sobtsnbssQcJSA5bEYobEYobEYobEYYd8eob8eob8eob8eq?bjRACcNRnbNRSd1lnb1lNcfQA0bcDobdDobcD2ckzoblzob?kzjc+QA2b4Gob4Gnb5GHdIwobIwnbJwScMRA4bwNnbxNnbw?NNdQpnbRpnbQpMcmQA/b6Gob4GAdIwobKwgcfPAybcDobdD?obcDkckzoblzobkz3ctQA2bUKobVKnbUKFdssobtsnbssUc?8PA1bwNnbxNnbwN1cQpnbRpnbQpncDQA/bNRnbNR4c1lnb1?lqcJSA5bEYobEYobEYobEYZd8eob8eob8eob8epbSRAgcoU?mbqU1caimbYiOciQA/bqUmboUDdYimbaihctRA4bUKobVKn?bUKWdssobtsnbssBcHRABcdDmbdDPdlzmblzTcMQA2bwNnb?xNnbwN9cQpnbRpnbQpecjQAAcNRnbNRCd1lnb1lfcJSA5bE?YobEYobEYobEYZd8eob8eob8eob8epbGSAqc6Gob4GxcIwo?bKwEcpeB7eEOjBjhBylBvfBAAA[/fumen]

Manipulations

Due to the fact Micro has such a good finesse scheme, he minimized the needed manipulations:

total rotations 77
total moves   262
total moves (das=1) 116

If we compare this with Lapsilap's record (the previous world record):

total rotations 77
total moves   276
total moves (das=1) 112

Exact the same amount of rotations (kinda low also) and ten (total) moves less. At such high keypress speed this makes a difference. Now with 100 pieces so many moves was strange (100 pieces -> around 2,6 moves per piece with 2-step finesse?) and the total moves seemed not to take das into calculation, so i included an (experimental) feature that does calculation more right and uses the move key presses that are also used in the finesse. This gives us 4 more key presses for Micro <-> LapsiLap, so even more impressive.

If we compare with a 23.9s 40L sprint of Maserati - he uses another stacking, 6-3, for the I:

total rotations 62
total moves   248
total moves (das=1)   110

(this seems still in line with the findings of Caffeine's mKpt, but we are getting too far away here and the difference is not that huge any more, might be interesting to investigate more in here).

Speed

The next statistics we look at is the speed. Where Lapsilap used +/-13 (=209.9 ms) frames per piece, MicroBlizz only uses +/-12 frames on average and with 199.5 ms he is the first to go under 200ms per piece.

Also Microblizz seems more consistent, where LapsiLap's piece speed varies with 12 frames, Microblizz varies only with 9.7. But this also has to do with the piece placements and positions (some pieces need more complex placements).

Now the interesting part are the three slowest pieces of Microblizz (aka outliers):

J   at 0.68 s   took 0.32 s (=19 frames) - (x=6,y=18,direction=0)
T   at 6.7 s   took 0.37 s (=22 frames) - (x=1,y=9,direction=2)
L   at 15 s   took 0.32 s (=19 frames) - (x=5,y=15,direction=2)

We can clearly see them in the speed graphs too.

The first J is the 4th piece he placed and I have really no clue why it was so slow. My guess is that the LS combination went really fast and then he needed a little bit of thinking time to get back to the J, but again, that's really a guess and 19 frames is still not slow.

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/400.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/400.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Then the T at frame 400 (6.7s) is just a quite heavy placement. It goes with it tip down in the 3th column, therefore has to rotate 180 degrees. Microblizz used double rotation for this: rotate left, move left, move left, rotate left, verify if correct, harddrop. That's just a lot of manipulations for a piece to place it. What I see happening or want to believe is that really pro players tend to have less of such difficult placements. They are like "bad" placements, like making a gap but more of a soft fault and their brain tries to avoid it. Such T placement was also a point of interest in Lapsilap's sprint, where Lapsilap used 3 taps for placing such a T on the right.

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/919.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/919.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Same story for the L at frame 919, needs 180 rotation, Microblizz uses two rotations and needs two moves. Also it's a bit a difficult stacking position as the gap is two wide there and it's a bit more uncomfortable (well for me)

What we see is that this corresponds to the local speed drops. For example the piece 77 (the "slow" L at 15s, frame 919) corresponds with a speed drop in local speed (see graph local vs global speed). Also the T piece at 6,7s corresponds with a local speed drop below global speed. This happens not so long after the 'hold' speed drop, resulting in a bit speed loss there. So in fact Microblizz didn't make (finesse) mistakes, but this slowest pieces just costed him time. It's amazing that his global speed recovers from those initial speed drops and grows a little bit until the end. This was also visible in Lapsilap's sprint.

The fact that a "slow" piece takes more time then an equal "fast" piece, has also the effect that it counts harder then a piece that has an equal "fast" time. Eg, if you drive 100km from A to B at 50km/h and you drive back at 100km/h, your average speed will not be 75km/hour but lower (3 hours for 200km is 66km/h). So what you want is consistency rather then spikes, or better, no low spikes.

Now if MicroBlizz needs a faster time, he needs a sprint without those slower piece placements, so I'd guess he has to avoid pieces that need 4 inputs (two rotations and two moves). I don't know if pro players naturally avoid those situations as if they are avoiding gaps. The fact that Microblizz uses 2 rotations and not 180 rotation is extremely impressive.

Extra - Speed
Ok now some new stuff. We know some placements take more time, because they are more complex. For example das tap back double rotate needs more keys then a simple drop. If the drop takes 3/4 of the time of the complex move, the drop is the bottleneck and needs improvement. So let's look at the average frames per key that are used for every piece:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/framesperkey.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/framesperkey.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Interesting, a new graph, a new view. In the beginning we see that for piece 24 and 29, 16 frames are used. Yet they are really not difficult placements (I checked this manually by going frame by frame because it's so strange). Let's first look at piece 24:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/24.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/24.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


The reason why this one is slow is probably because MicroBlizz sees what's gonna happen, if he places the O there, there is no good place for the Z. Probably a slight hesitation there as Blizz uses normally around 12 frames for a das, what happens to be also the average of his placement speed overall.

Now let's look at piece 29:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/29.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/29.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


To be honest I can't really explain this one. It is a piece with only a rotation, so every frame counts here. Single rotations costs normally only around 9 frames, so there is a little bit thinking time here. He just recovered from his hold adventure and might now need to buffer again looking at the previews or he might have been looking at the gap on the right and to see if there was an opportunity for the S to skim. Only he can tell.

If we look at the global speed graph, we see that those two frames are also in the zones of the slower parts of the sprint with the holds and confusion, confirming (or, not contradicting) what is said before.

Now the slowest one... piece 84 using 17 frames per key:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/84.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/84.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


This is a move with only one tap, normally costs on average 10 frames, not 17. It's guessing here again what is happening. Maybe Blizz had the natural feeling to put it on the other blue J to make the block form and realized that his stack would grow a bit strange and had to supress that feeling to change his mind and just drop it, not loosing too much time on this piece because it's a quick placement, but filling mid space with a slow piece also slows down in general.

We see in the general speed graph that it's in a slower zone, but not taking down the global speed too much because it was not so much out of proportion for a placement. Indeed, 17 frames (because it was only one move, the frames per key press is the same as the total time it took) is not so far away from the average of 12 and is still under the 95% certainty interval of the total of the pieces, where the ideal slowest number of frames of a piece is 18 frames.

Extra - Manipulations

Now let's look at the speeds of the different manipulations. I added a new feature to get a proper look at this. First look at the manipulations that MicroBlizz used (this is still 'beta'). He uses a lot of those:

DAS_ROTATE: 27 times
DAS_TAPBACK: 13 times
TAP_ROTATE: 11 times
TAP: 9 times
ROTATE: 8 times

I'm not going into too much detail here, but we can clearly see that MicroBlizz is heavily making use of DAS. If we sum up everything, almost 1/2 of his sprint is rely'ing on DAS.

Now let's take this a bit further and see what are Microblizz slowest movements in general. Those manipulation to get a piece in place will cost most of the time:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/manipulations.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/manipulations.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


We can quickly see that double tap and double rotation is the most costly one. This happened twice:

DOUBLE_TAP_DOUBLE_ROTATE   2 times - used frames [22, 19]

Once using 22 frames and once using 19 frames. If you look at the speed graph, for piece speed in frames, you can quickly recognize that the one using 22 frames is piece 32:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/speedgraph32.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/speedgraph32.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


if we look at this piece, it's our old friend T:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/400.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/400.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


If we look at the other one, we see it's piece 77, yet another old friend:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/919.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/919.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


This way we can see that by looking from different positions at the sprint, we get additional and complementing results.

What's interesting is that those 2 placements count for 2/3 of all the outliers. There are only 3 outliers that get outside the 95% certainty interval and those two are part of it. So if MicroBlizz could somehow manage to do a sprints without double movekeypress double rotation, this will significantly bring his time down.

The next one is the row is just one double rotation that was actually just in the 95% save time interval with 28 frames, it is piece 23:

[!--ImageUrlBegin--][a href=\\\"http://tetris.vanderhoven.be/images/23.png\\\" target=\\\"_new\\\"][!--ImageUrlEBegin--][img width=\\\"400\\\" class=\\\"attach\\\" src=\\\"http://tetris.vanderhoven.be/images/23.png\\\" border=\\\'0\\\' alt=\\\"IPB Image\\\" /][!--ImageUrlEnd--][/a][!--ImageUrlEEnd--]


Somehow this piece went a bit slow, you can imagine what happened if Micro used 180 in his sprint. This double rotations makes it all so more impressive.

When we go further there are no real suprises, of course complex moves cost more. Das tapback with double rotate came in twice with an average of 15 frames and then we have a big class of manipulations that are well around the average timing, 12-13 frames with the normal DAS the same as the average (12 frames). As last the really fast ones come, like tap, rotate, drop, all 10 frames and lower.

Ideal time

Now if we take away those "slower" piece placements that were outside the 95% interval of all the other piece placements, and we recalculate the sprint with 103 pieces, we get at an ideal time of 20.137 s. What this means is that if MicroBlizz manages to get a sprint without those slow 4-manipulation moves (or uses 180 rotation) he would get at 20.137s.

If he manages to do it without the hold, the slight hesitations, with less the 103 pieces, and fortunate stacking, he will be sub20.

QED.

--
References used:
- Microblizz record stats: [a href=\\\"http://tetris.vanderhoven.be/MicroBlizz%20-%20WR%20-%20Analysis%20Part%202.html]http://tetris.vanderhoven.be/MicroBlizz%20...20Part%202.html
- Microblizz WR sprint fumen report: http://tetris.vanderhoven.be/Microblizz_WR...y%20Report.html
- Previous world record stats: http://tetris.vanderhoven.be/Lapsilap%20-%...d%20Record.html
- Previous Microblizz record stats: http://tetris.vanderhoven.be/MicroBlizz%20...-%2022.78s.html
- Maserati Sprint: http://tetris.vanderhoven.be/Maserati%20-%...0-%2023.98.html

[larrytetris]

Holy hell.

[farter]

i guess it took hundreds of times longer than 20.55s to make this analysis..
awesome

btw 60 frames full fumen seems dead
Request-URI Too Large
The requested URL's length exceeds the capacity limit for this server.
Apache/2.2.14 (Ubuntu) Server at maggie.priorweb.be Port 80

[CaptainPaul]

After reading this... I'm not sure what i'm more impressed by. The run or the analysis. Holy hell. Well done.

[belzebub]

i guess it took hundreds of times longer than 20.55s to make this analysis..
awesome

btw 60 frames full fumen seems dead
Request-URI Too Large
The requested URL's length exceeds the capacity limit for this server.
Apache/2.2.14 (Ubuntu) Server at maggie.priorweb.be Port 80

It's actually the server that's the problem here (and the way fumens work). If you download the report to the report directory of my nullpomino fumen mod you can watch the replay on your local computer without a problem. The problem is that the generated fumen (with 60 frames per second) is really huge -_-

[farter]

It's actually the server that's the problem here (and the way fumens work). If you download the report to the report directory of my nullpomino fumen mod you can watch the replay on your local computer without a problem. The problem is that the generated fumen (with 60 frames per second) is really huge -_-

yeah i see that
but... i remember that if not much is changed between frames, the data will be a lot shorter (especially when only the active piece is moved but the field is not changed) than one whose field blocks are modified manually... in other word there might be some compression for tetris (but i haven't read the code carefully)... \
i guess.. with full compression (and instant das which makes changes even fewer), it shouldn't be so long..?
also.. in the frame-per-piece fumen sometimes the active piece is partially out of the field..? bug?
EDIT: frame 18, 22 30, etc
EDIT again:
[fumen]110@7e/1cISmBXcB2nBzdBMbBiYB3SBMaBNQBzOBJMBeNB?MJB6HBvGBZABuIBl8AT6AZrAy3AF+ATtA5eB+IB8HB39A/5?AzEBK3AF4A0tAe7A5eBK3AsCB/JBN9Az/AuDB26A6yAV5Az?wA8vA3uA5eBz6Am8A5eB3NByLBsMB1JBlDBTBBm/A85AK3A?5eBvWB6GB0MB7enzJBmHBlDB39A5eB1QBqTBMGBvNBTHBOO?B5eBCYBNQBZeBTRBvGBePBMSB2MBv8ANOB8EBzCB5eBiRBS?MBtXBnOBMJBTIB5eB2aBZUB2XBzQB6RB/NB5eBAAA[/fumen]
manual copy... it's really a lot shorter...

Code Select Expand


110@7e/1cISmBXcB2nBzdBMbBiYB3SBMaBNQBzOBJMBeNB?MJB6HBvGBZABuIBl8AT6AZrAy3AF+ATtA5eB+IB8HB39A/5?AzEBK3AF4A0tAe7A5eBK3AsCB/JBN9Az/AuDB26A6yAV5Az?wA8vA3uA5eBz6Am8A5eB3NByLBsMB1JBlDBTBBm/A85AK3A?5eBvWB6GB0MB7enzJBmHBlDB39A5eB1QBqTBMGBvNBTHBOO?B5eBCYBNQBZeBTRBvGBePBMSB2MBv8ANOB8EBzCB5eBiRBS?MBtXBnOBMJBTIB5eB2aBZUB2XBzQB6RB/NB5eBAAA


[belzebub]

yeah i see that
but... i remember that if not much is changed between frames, the data will be a lot shorter (especially when only the active piece is moved but the field is not changed) than one whose field blocks are modified manually... in other word there might be some compression for tetris (but i haven't read the code carefully)... \
i guess.. with full compression (and instant das which makes changes even fewer), it shouldn't be so long..?
also.. in the frame-per-piece fumen sometimes the active piece is partially out of the field..? bug?

To be able to see a live replay I fumenized every frame (60 per second). There is compression and encoding, but it's not enough to compensate, the query of url's is just limited on the server. The data is not so huge but i can't change the server limit's because i don't own the server.

The part where the active piece is partially out is a known bug, I just didn't take time to place it exactly for every piece, also it's always at row 2. If it's a problem for somebody i'll fix it if I have time (i'm glad people actually look at this thx!)

[farter]

To be able to see a live replay I fumenized every frame (60 per second). There is compression and encoding, but it's not enough to compensate, the query of url's is just limited on the server. The data is not so huge but i can't change the server limit's because i don't own the server.

The part where the active piece is partially out is a known bug, I just didn't take time to place it exactly for every piece, also it's always at row 2. If it's a problem for somebody i'll fix it if I have time (i'm glad people actually look at this thx!)

er.. i made a manual copy below..

and seems it's 7 lines vs. your.. eh i didn't count that..

also i think put the current piece where it locks is a good idea, just like my copy

[belzebub]

er.. i made a manual copy below..

and seems it's 7 lines vs. your.. eh i didn't count that..

also i think put the current piece where it locks is a good idea, just like my copy

The fumen you posted is in the report too, it's in the fumen tab and in the dialog that pops up when loading the page. That fumen just gives the piece sequence. The replay tab contains a replay of the full game with every frame in it and is much bigger. It has correct timing and stuff and can be used to count frames between pieces etc..

edit: i see now that you mention the same fumen, however i'd prefer not to discuss it further here.

Can we move this dicussion to the fumen mod thread? It's getting off topic for the analysis i'm afraid.

[FelipeMayrink]

Sorry to spam this question everywhere but, where could I get Micro's replay code of this run?

[Panda]

DO NOT WASTE YOUR SCIENTIFIC ANALYSIS SKILLS HERE, GO TO MEDICINE CANCER NEEDS YOU