For information about speedrunning Adventure Island and to get a better understanding of what's going on in this run, go to the guide in the SDA knowledge base (referred to as sdakb in my comments below):
https://kb.speeddemosarchive.com/Hudson%27s_adventure_island

SDA history
2013-Jun-28 - 0:37:21 by ktwo (http://archive.org/download/2013-10-27-SDA_Speed_Runs/

Background
This game and me go back a long time. According to my first post in the forum, I started speedrunning it some time during the summer of 2011. At that time, I was playing on PAL. It took me pretty much a year and a great deal of effort before I got a run I was happy with.
The following year, I finally took the step over to NTSC (better late than never). One of the first games I then tried my hands on was Adventure Island. My goal was simply to put up a time for the NTSC-version. After two weeks or so, I got a deathless run that was the fastest reported run (although there wasn't really anyone speedrunning this game properly then). The run itself was pretty terrible though and the recording even worse. I was experimenting with recording with my Dazzle, which ended up with a lot of inserted frames. I still decided it achieved the goal I had.
As time passed on, I felt more and more that my NTSC-run was a mistake that I had to correct at some point in time. Eventually, I decided that it would be my next game after I was done with Ikari Warriors. I therefore began working on this game again late summer 2017 to finally be able to lay this game to rest. The attempt that ended up in this submission was quite unexpected. After a break of around 2 months, I was slowly picking up from where I left off. This run basically came without being completely up-to-speed. Unfortunately, I feel this can be noticed in some areas, where I made mistakes that I usually manage to avoid.

Run history
While speedrunning this game, I posted short updates in the bio section of my twitch page. To preserve some of the history this run has gone through, I'll just copy-paste those notes here:
Edit (2017-08-19): I should have most of the levels memorized enough for starting to make full-game attempts. I'll probably still continue playing on good attempts instead of resetting so I can get more practice of later levels.
Edit (2017-09-08): My first completed attempt since I picked this game up again ended with a 37:29. It included a death that cost ~21 seconds. Based on that attempt, a 37:05 would probably look very good, while over 37:10 would probably start looking questionable.
Edit (2017-09-10): A good attempt (~37:07 pace) that ended in 8-3 is my pb so far in terms of proper attempts go.
Edit (2017-09-26): I managed to get my first completed attempt without deaths, 37:08 (offline). The execution was a bit rough and I lost the skateboard in 6-1, so I hope that I can bring the time down by another 5 seconds or so. (I've found several new time savers, so my estimate is a bit lower than what I wrote in one of my previous edits)
Edit (2017-09-28): My best online attempt was completed in 37:25. It had a death in 8-1, but was otherwise good and probably on pace for around 37:02.
Edit (2017-11-12): The first stream in a while. I've continued doing offstream attempts though. I've found several new time savers, but all of them are small, so there is no change to the estimate (I would probably be content with a 37:03, pretty happy with a 37:02, happy with a 37:01, and so on). I'm now up to 5 proper attempts reaching world 8.
Edit (2018-02-10): I'm making a new attempt to complete a run I'm happy with of this game. There are only a few very minor changes to the route since my last update.

Run comments
I'll comment here on noteworthy events of this particular attempt. For more detailed level comments, refer to the guide in the sdakb linked above. It's not possible to go into detail of what has changed since my previous run. The changes are simply too numerous to list. I'm guessing it's in the vicinity of 70 changes or so that I've made to the routes. While the new run is better executed, the vast majority of the time save comes from the improved routes. Maybe 13 seconds or so (give or take)?
1-2 It's possible to save half a second by using riskier strats (see sdakb). Even though it's early, the tricks needed were too precise for me to include.
1-4 I collected a carrot before the first rolling boulder. The jump in between the fruits is somewhat precise and I miss it fairly frequently. The same jump has to be done three more times in the run. I got the other jumps. 3 out of 4 is ok-ish.
2-1 I always attempt a speed manipulation in the last section. There is no consistent way to speed manipulate on foot, so it's random if you get it or not, but on average it's definitely worth going for it. I didn't get the speed manipulation in this attempt, which would have saved around a quarter of a second.
2-2 I always attempt a speed manipulation in this stage as well. It's difficult to say for sure if I got it, but I believe I didn't get it in this attempt. Only a handful of frames lost though.
2-4 A 5-jump boss fight is always a let-down, but sadly something that happens most of the time. To do it in 4 jumps, you can only miss one fireball. Around half a second lost.
3-1 I didn't land properly on the second to last platform in the moving platform section and had to make a "brake jump" to avoid falling down. A little over half a second lost.
3-3 It's possible to save a little less than half a second by using riskier strats (see sdakb). The tricks required were a bit too tight for me to use in attempts though.
3-4 5 jumps on the boss is the best I can do with the setup I use.
4-2 One unintentional apple collected at the end of section 3 (8 frames lost). It's somewhat tight to avoid it, but I still rarely pick it up, so a bit disappointed here. The carrot at the end can be avoided, but appears to depend on the sub-pixel. I haven't been able to figure out a method to manipulate it so it's basically a coin toss. Or actually, I probably skip it only like one out of three or four, so it's not too surprising that I collected it here.
4-3 A few days after this run was completed, I figured out a virtually risk-free way of avoiding one more fruit for a cost of 1 frame. See in the sdakb.
4-4 Disaster... I unintentionally collected a fruit during the jump before the line of frogs. This distracted me and I didn't speed manipulate, which resulted in me riding at 80 speed for too long. I also lost my board right before the very end, which is painful to look back on. Almost one second lost in total. At least the boss fight was decent. 6 jumps is to be expected (5 jumps is very tough and requires no missed attacks).
5-1 By jumping from the last cloud over the fruit, you can avoid the last small jump, which would save around 8 frames. That jump isn't too hard, but failing it will either lead to collecting the fruit or dropping into the sea. It's late enough in the run that I don't want to add any risk of that kind for such a small time save.
5-3 16 frames lost due to two unintentional fruits collected (the last banana in the moving platform section and the banana in the last platform section). It's also possible to save half a second in the moving platform section (see sdakb). I feel that it should be possible with enough work to include it in a speedrun attempt, but I never got consistent enough to really go for it.
5-4 A mistake half-way through at the first racoon. I missed shooting a rock, which almost always results in me crashing into it instead (= reset). I managed to save it this time, but for a bit of lost time and an additional fruit collected. The boss fight was awkward. I started attacking too early and got misaligned with the first boss attack. Normally, you can just keep attacking and the boss's fireball will pass below Higgins. Anyway, I had to steer out of the way and ended up with a 7-jump fight. I would have liked a 6-jump fight. A little over a second lost on the fight alone.
6-2 Unnecessary mistake near the end. Which fruits to collect have been carefully planned out. I messed up and collected a fruit too many (at the very end of the ice section). Because of the eggplant, I needed to brake near the top to fix the situation or the energy would have run out. Half a second lost.
6-3 The beginning is very risky and after a lot of trying back and forth, I settled with a compromise where I traded half a second for safety. The faster route is shown in the sdakb. I also had some trouble scaling the steps in section 2, which cost me around half a second. This stage is one of the tougher and something usually goes wrong, so I'm not too upset.
6-4 Terrible boss fight. 10 jumps is far too much. I use a slightly different approach for this fight by starting to attack as early and far away as possible, so it's normal to have a few extra jumps, but 10 that's at least one too many.
7-3 I mistimed the jump on the moving platform before the three birds and had to brake or I would have jumped into the fire. Small time loss though. Half a second could be saved when entering the bonus room (see sdakb). I go for the faster strat in 2-3, but it involves a certain failure rate, which I've decided was not worth having as late in the game as 7-3.
7-4 A 10-jump boss fight is on the high end. I'm also using a fairly conservative strat with only two jumps between each of the boss's fireballs. With the right spacing, it's possible to squeeze in another jump. I often get hit with that strat, so I'm playing it safe at that point.
8-1 In my opinion the hardest stage in the run. Especially when considering how late it comes in the game. However, the stage went pretty much perfect this time. I even got the speed manipulation in the second half of section 2. It only saves a quarter of a second, but still nice.
8-2 I collected a bit too much fruit and the very end can be done faster by not shooting the octopus and bird (to avoid lag). Still, it's a lot about survival at this point, so I'm not complaining.
8-3 Very good stage. I finished with the minimum amount of bonus with the route I'm using. There is another half a second to save in the moving platform section (same as 5-3, see sdakb).
8-4 The last boss fight might look a bit conservative, but it was a good fight for my standards (even though I admit that it can be done faster by being more aggressive and getting 3 attacks to connect per jump).

Final thoughts
Once the routes started to settle, I felt that a time I would be happy with would be around 37:03. My sum of best is currently 36:47, so a sub-37 time is very possible to achieve. Very hard though and I'm not sure I've been on a pace, where I could clearly say that it would have been a sub-37 had I only finished. The (very low) 37:05 of this run is therefore partly disappointing. As mentioned above, this run came a bit out of nowhere, while getting back into the game. So there are mistakes I think would have been avoided if I had had more practice. I'm almost certain that another completion without any costly mistakes would beat this time. However, lately this speedrun has sadly been more about "getting the run done" than enjoying a good game to speedrun. So even though I'm not completely satisfied with this run, this is likely my last pb for quite some time.

Useless knowledge
The amount of hidden eggs appearing in this run is not purely a coincidence. I tried to use every opportunity to show off hidden eggs, where it didn't cost any time. Out of the 58 hidden eggs in the game, I had a chance to reveal 35. I ended up with 31 though (missed ones in 4-4, 6-2 and 2x8-1).

If you want to discuss or share information related to speedrunning this game (tricks, strategies, ...), use this thread:

https://forum.speeddemosarchive.com/post/adventure_island.html

If you have questions or comments about the run, use the existing verification thread:
https://forum.speeddemosarchive.com/post/adventure_island__may_6th_2013.html

Techniques and general speedrun strategies

If there ever were a competition for the most linear game, Adventure Island would certainly be a top candidate. Most of what's seen in this video should therefore be straight forward to understand, but I'll use this section to fill in the gaps and add some more detail for those interested in that.

 

The fastest way to clear a level is to use the skateboard at "run speed", while shooting as few enemies as possible. Shooting and running use the same button (B), so shooting requires a momentary decrease of the speed. Playing on console, one can roughly say that one shot costs one frame. Another important aspect of each level is to finish with the life meter as low as possible as that reduces the countdown after each level (each 20 on the life bar costs a total of 2 frames of countdown). Every pellet in the life bar is worth 160 life, and lasts 5 seconds during the actual gameplay. The life bar can be increased by picking up "fruit" (I know some of them are not fruits, but I always call them fruits anyways):

- Bananas, apples - 80

- Tomatoes, carrots - 160

- Pineapples (gold bags?) - 240

 

In practice, running out of life is only an issue in a few stages, so most of the time it's just about avoiding fruit as much as possible. There are two ways to make the life bar decrease faster. The first is to pick up an eggplant (hidden in eggs in certain levels). The eggplant makes the timer run down five times faster than normal. It stays with Master Higgins until two pellets of the life bar remain. The other way is to trip on stones, which reduces the life bar by 320 (=32 frames of countdown). It also means a slower speed costing 16 frames, but the overall gain is still 16 frames.

 

Let's talk about Master Higgins' speed a bit. If we fire up this game on the emu of preference and put a watcher on RAM-address 05A4, we'll get to observe the horizontal speed. The acceleration/deceleration is 3 (subpixels/frame) when on the ground or when dropping from a ledge, 1 when jumping and 1 when subject to ice physics (whether on the ground or not). Interestingly enough, there are several stable max speeds (=no acceleration possible when at these or higher speeds):

- Skateboard (B) - 80, 81 and 82

- Skateboard (no B) - 73

- Foot (B) - 73, 74 and 75

- Foot (no B) - 49

 

If exceeding the max speed, such as getting a speed of 74 on skateboard (no B), the speed will begin to oscillate between 74 and 71 every other frame.

It goes without saying that one wants to go at max speed in a speedrun. When on the skateboard and shooting during a jump (a common situation), the speed will invariably end up at 80 (first deceleration to 7X something and then reacceleration by 1/frame until 80 is reached). This is a decrease of 2.5% of the speed compared to 82, which over the course of the run quickly adds up. However, if one lands without holding the B-button, the speed will always be 73, right? Pressing the B-button after landing will then accelerate to 76, 79 and... 82. This is very easy to perform, but the side effect is that part of the jump has to be performed without B-speed, thus being slow. I count that if I can go 82 for the length of roughly one screen, then it pays off to set it up. Especially in the later levels, the layout and enemy positions don't allow for much speed manipulation though. It should also be added that the jumps (length and height) are affected by the speed, which is also a factor when determining which is the most appropriate speed in each area.

Ok, now we have talked about the skateboard. Is it possible to manipulate the max speed on foot as well? Let's see... So we jump and shoot without repressing the B-button, which will set the speed to 49. Upon landing, press the B-button, which will change the speed to 52, 55, 58, 61, 64, 67, 70 and... 73. So here we always end up with the slower of the three valid max speeds... Unfortunately, it's therefore not possible to reliably manipulate the speed on foot to 75. However, it is possible to release the B-button just before landing and repress it right after. If done correctly, this will first slow down Higgins to a random speed between 49 and 73 and then reaccelerate by 3 until 73, 74 or 75 has been reached. While this is interesting to study and could possibly be used in one or two places to save a couple of frames, I chose not to use it because I prefer to know my speed at all time since it affects the jumps for example. Before moving on from this slightly technical section of the comments, we should keep in mind that since the start speed is 0, it's always possible to achieve 75 on foot at the start of a level, which will then be kept until shooting during a jump.

Level comments

1-1 - A little known fact is that throwing an axe at an egg, stone or rolling boulder, will make the axe bounce off and that will trigger the next snake to spit fire at you. I hope that can help a casual player or two out there. :-)

1-2 - It's possible to jump over the bonus pot at full speed, which would save half a second to a second. It's risky though. (picking up a bonus pot makes the bonus countdown be done twice + some waiting -> instant run killer)

1-4 - I jump too early on the first jump and pick up an apple. At the first raccoon, I jump too late and pick up a tomato. This jump is admittedly pretty difficult though as jumping too early makes you land on the rock -> lose skateboard -> run over. The boss always takes 8 hits to kill, no matter whether it's axes or fireballs. Every subsequent boss takes an additional 2 hits to kill. I shoot three shots every jump. I've seen that the Japanese players shoot 4 per jump. I've tried to get that to work, but with no success. It's not about mashing (you can only have two shots on screen at the same time), but more about learning a good rhythm. While learning to do 4 shots per jump that consistently connect is possibly the biggest potential time saver over this run, it should be mentioned that the Japanese players miss more shots than I do. Still, the potential is there. Another thing worth mentioning is that it's possible to run after the boss once it flies away to exit at early as possible. However, this ends up wasting time (maybe 2-3 sec over the course of a run). The time it takes for the screen transition to the bonus counting depends on when the boss leaves the screen and not when Master Higgins leaves the screen. By running after the boss, the screen will scroll to the right making it longer for the boss to fly off the screen.

2-2 - I jump too late and pick up an apple in section three, at the rolling boulder.

2-3 - Interesting skip here. I lose the skateboard to reveal a hidden egg (not possible while riding the skateboard) to grab the key. This activates two of the platforms in the level that will now take you to a bonus area. When leaving a bonus area, the game will send you to the section after the one you were in when entering the bonus (this prevents an infinite loop). There are a couple of bonus areas in the game, but their location don't let you skip enough of the respective levels compared to the time wasted going there. What makes it special here (and in 7-3) is that the second platform (I jump over the first) is located so that Master Higgins can just barely trigger checkpoint four of the level before entering the bonus area (a checkpoint doesn't have to be passed, it just needs to be loaded to count). Upon exiting, Master Higgins is sent to the goal, skipping the whole section four.

3-4 - Missed a shot on the boss. One second lost.

4-2 - I miss a jump, which cost one second + picking up extra unintentional fruit costing half a second in extra countdown.

5-1 - The jump to the egg with the skateboard is possibly the hardest jump in the game.

5-2 - In section 1, I should have shot the bat after the bat+boulder instead of jumping over it and picking up an extra apple. The jumps over the moving platforms near the end are maybe a notch easier than the jumps in 5-1, but still major run killers.

5-3 - Could have avoided the banana in the last section, but I usually end up picking up one of the fruits there, so it was maybe to be expected.

6-1 - The first egg is an eggplant, but I haven't found a way to make use of it (I always run out of life). Neither the TAS nor the Japanese players use it, so I guess it's all good. I'd like to throw out a tip for the casual players. In the first section with all the octopus, just run at B-speed and they're all gonna be lined up perfectly without interfering with Higgins' path. Trying to dodge is just asking for trouble...

6-3 - The beginning is one of the hardest sections for me in a speedrun. I lost a second from a mistimed jump.

7-1 - I pick up an extra apple near the end, which turned out to be unnecessary (you can exit the level with 0 on the life bar). Since having failed an attempt or two from running out of life just before crossing the finish line, I tend to play it a bit safe.

7-2 - The TAS manages to get through the end on skateboard, which saves a couple of seconds in 7-3. However, I have never even been close doing it on console or emu. Not sure how much manipulation would be required, but it looks to me like a TAS-only trick.

7-3 - See the 2-3 description for how the skip works. Interesting to note is that section 4 of 7-3 was by far the section I had the most trouble with before this skip was found. Sometimes you're lucky, I guess. :-)

7-4 - A few missed shots on the boss, but I only had to make one extra jump (8 iso 7 to get the 20 hits), which is acceptable by my standards.

8-3 - My jumps get a bit misaligned in the beginning, but not much time is lost, I believe.

8-4 - I miss several shots on the last boss, but the time loss is fairly small as long as the last hit connects before the boss has reached the back wall (which it just did in this run). Landing the last hit earlier still requires him to moving the same distance back to the wall before falling down the pit.

Competition (as of writing - early 2013)

When I started speedrunning this game, I was (to my knowledge) the only western player doing so. Since then, the game has seen some race activity, but no serious effort has (to my knowledge) been done to get a good time. There has however been a bit of speedrun activity for this game in Japan. The record has been pushed down several times and now stands at 37:40 (37:35 sda-timing). The Japanese runners don't use the speed manipulation (at least not yet), but I know they are familiar with it. This means that their times will always be ~15-20 seconds from an optimal run by default. In my eyes, it also looks like they don't use very effective routes (which fruits to collect, which enemies to kill etc). For comparison, my time would be 36:59 (36:54 sda-timing) if the same gameplay had been performed on an ntsc-machine. Take that comparison as you will, but at least it shows that there should be lots of room for improvement in the ntsc-category. It should also be mentioned that my comparison does not take into account any differences caused by lag, which is bound have a negative impact on ntsc.

Final thoughts

Despite a few imprecisions in the run, I'm quite happy with the end result. I don't intend to come back and improve this time. As a whole, there were some interesting discoveries, which made playing the game fun. However, I personally think it's too linear to be fun to speedrun from this point on. A good proportion of the attempts (possibly as many as 50%, but don't quote me on that) were actually ruined by finger fatigue from constantly holding right+B. Of all the challenges a speedrun presents you with, that's obviously not one of the more exciting ones. I guess the ntsc-version is better in that regard though since the levels go by slightly faster.

Credits

- SDA for publishing this run and for providing the community with a non-commercial platform for hosting speedrun videos.

- Anyone who has contributed to the development of Anri-chan. If a tool is so easy that even I can encode videos, then it must good!

- Macco and Takenoko (Japanese runners) for giving me a few new ideas that helped me improve my time (special thanks to Takenoko for having videos with the 2-3 and 7-3 skips).

- Randil for explaining about the max speeds.