This is my first submission to SDA. I never expected to beat the former record by such a gap ! Initially my plan was to reach around 0:06:15 - 0:06:20, but by dint of practice, I realized under 0:06:00 could be possible. I'm extremely happy with this run ; of course it's improvable (by finding new strats, or with more luck against the 2 last bosses or with Lizard King), but I'd be surprised if someone reaches under 0:05:30 someday... I'd like to thank Claire who gave me the idea to do this speedrun, and guys like Kirkq, Wizphix and M_88 who helped me a lot with their videos !

[The Drawbridge]
You need 10 daggers to kill the dragon. My strat is 7 then 3. I use the 2 daggers when I kneel down on purpose ; I noticed that if you stand up too fast after the dragon shoots his fireball, the dragon has time enough to launch a second fireball before dying, and you'll lose time to kneel down another time. These 2 daggers are just here to give me a timing to stand up at the right moment. Thanks to that, I can kill the dragon before he launches another fireball, and I can directly walk to the door.

[The Entrance Hall]
You can think I lose time by keeping my Gold Bonus (as Lizard King didn't appear). That's right, it means I don't skip the Gold Bonus screen. But it helps me to go a little faster in next level, you'll understand why.

[The Gold Mines]
In this level, I usually take 8 Gold Bonus with my strat. As I kept the first one taken in previous level, the trick is to take one more Gold Bonus (the one over the second moving platform) to have precisely 10 Gold Bonus at the end of the level. Thanks to that, the Gold Bonus screen will be passed faster than with 8 Gold Bonus. The way I pass the last moving platform is very tricky, but with practice, I can reach a solid successrate here. Surprisingly a Gold Bonus is added to my total, even if I don't jump directly on it (and you can notice it doesn't disappear) ! I don't know if it's a bug or something else, but you can try, it works every time ! I couldn't beat the trolls faster than that : it's hard to kill the one to the left at his first appearance, but I managed to do it here. Otherwise you lose around 2 seconds to wait his second appearance and kill him.

[The Hall Of The Grim Reaper]
The first hidden energy isn't really necessary, but as I don't lose time by taking it, it's safer. The Reaper can be beaten faster with more luck (for example if he shoots only low fireballs). A more aggressive strategy is also possible, but I think it requires some luck too.

[Singe's Cavern]
Until the very last moment, I planned to jump over the third baby dragon, but it's very tricky. And even if it's faster, I realized I lost this benefit because of the second flying dragon (who drops something like a rock), as I have to wait him just after I killed the sixth baby dragon, to move safely to the next island. So I switched to a safer strat. Against Singe, I counted the number of hits, so I knew what I was doing at the end, don't worry. ;-)

Disclaimer
I did this run on a Powerpak. In case it matters for future reference, I used a ROM with checksum 7bb417bc023f48cf6184d9bc6b7e87f5.

SDA history
02-Feb-2018 - 0:07:21 by ktwo (https://ia903109.us.archive.org/19/items/2018-04-29-SDA_Speed_Runs/)

Speedrun information
For information about speedrunning Dragon's Lair 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/Dragon%27s_Lair_(NES)

Background
After my 7:21 run, 'SBDWolf' lowered the time to 7:16 and found some baby dragon skips and faster fights in the last level. I was interested to pick up the game again at some point and try to lower my pb by implenting the new findings.

As a first step, I decided to TAS this version of the game towards the end of 2022. With the limited control scheme and straight-forward gameplay, there wasn't a whole lot of new findings until reaching Singe, the last boss. The "standard" method of fighting Singe is to stay on the left side. With this method, getting a 15-puff fight, with 145 frames between each puff, was considered very good (I had 16 in my previous pb and SBDWolf had 18 in his). But in a TAS setting, it's possible to manipulate the smoke puffs and I ended up getting it down to a 6-puff fight. All the puffs then needed to be of the right kind (out of high or low), each one depending on the RNG-function. 2^6 = 64... Could it be done in real-time? 1/64 wasn't completely out of the question... It would be a big grind, especially considering that the execution wasn't trivial, but my interest had been caught...

Fast-forward to summer-23 and I decided it was time to finally get started. I expected it to occupy me for a long time, possibly requiring over 100 Singe fights (given that my execution was not 100%). The rest of the game is honestly pretty easy, so it didn't take long before my Singe fight counter started ticking up. A little over a week after having started the grind, I got my 11th attempt at Singe. Most of the previous attempts had ended already on the 1st puff, so I didn't get into the fight with high expectations. But puff after puff went my way until Singe was eventually defeated! This was the run that I'm now submitting.

Note the about random factor in this speedrun
The strats I was going for in this speedrun were very RNG-dependent, with a low probability of success. The first factor, already described above, was attempting to get a Singe fight with the TAS-pattern. When I attempted this, I thought the probability was 1/64. A deeper analysis of the RNG-algorithm that was done after this speedrun was completed showed that the probability was more like 1/75 (see sdakb for more details). The RNG-analysis actually revealed a method to improve the chances of a successful fight (to ~1/41), but that was after completing this run and would anyways not have helped me during this grind. The other big factor was requiring a Lizard King appearance in level 1 (saves a little over 2s), with roughly a 50% chance.

Before starting the grind, I decided to save all attempts reaching Singe. I expected my twitch channel to eventually be filled up with failed Singe attempts, but as already mentioned, I surprisingly only needed 11 Singe fights to complete a run. For whatever it's worth, the first 9 attempts are still highlighted on my twitch channel (https://www.twitch.tv/ktwo123), as well as the full stream of the session containing Singe fight 10 and 11.

Run comments

Stage 0
It's possible to attack the dragon 8 times in the first cycle, saving 20 frames in the second cycle. It's not that hard, but was not consistent enough for me, so I always went for a 7-3 fight instead of creating a reset point here.

Stage 1
* I was at a stage in the grind where I was still learning the game (getting familiar with every pixel of the run and dying in every imaginable way). I hadn't yet gotten the consistency I felt was needed to start resetting over not getting the Lizard King (LK) in this level. So it's just a nice bonus that he appeared in this attempt. When I did attempts back in 2018, I reset if the LK didn't appear, so he was in my previous pb as well.
* Had the LK appeared slightly earlier, I might have reacted in time and not jumped, which would have saved half a second (if the LK doesn't appear here, a jump is needed).
* The snake boss fight at the end has been slightly improved since my last pb. I now jump-attack into the fight and I also jump-attack against the last snake. The time save is around a quarter of a second.

Stage 2
* When creating the TAS for this game, I took a closer look at how the "Dragon's breath" in this level worked (see sdakb for the details). It resulted in some useful observations also for real-time attempts. Before starting doing attempts, I thought I would be able to save 0.5-1s. Looking at the splits while making attempts, it's more like ~0.3s on average. No risk involved, so still worth doing of course.
* It's possible to get the LK at the end of the level, but I'd say it's a 1/10 chance or worse. With the RNG already required to complete a run, it was out of the question for me to reset over the LK not appearing here. Same decision as in my previous pb.
* It's possible to tighten up the mine troll fight and get 4 hits in the first cycle on the second troll, resulting in a 20 frames time save in the second cycle. When my positioning ended being right, I attempted the quicker kill, but I didn't reset if I didn't get it. Unfortunately didn't get it in this attempt (same as my previous pb).

Stage 3
This whole level runs on a global timer, so most completions will look very similar.
* In my previous pb I made an ugly mistake against the boss, which I managed to correct this time around for a time save of a little over one and a half second.
* In the elevator section to stage 4, it's possible to use a small jump instead of a big jump. This saves 5 frames. The visual cue is pretty good and it's not particularly hard, but I lost attempts here and there from doing short jumps. Given the small time save, I decided to trade time for safety here in my attempts.

Stage 4
There are plenty of optimizations in this level, all found by SBDWolf'.
* Use jump-attacks against the 1st and 2nd baby dragons (BD)
* Manipulated the 3rd BD to get it to take a low take-off angle, so it can be jumped over (see sdakb)
* Looking back at the run, it might have been possible to make more out of the LK appearing. However, it also looks like it messed up the bubble cycle before the 5th BD and made that fight slower. I'm not sure if I could have done something different. Regardless of the reason, it looks like there is a little time to save in this section (half a second or so?).
* I jumped over the 7th baby dragon also in my previous pb, but the manipulation is now faster
* I lost several attempts by one of the axes being blocked at the 8th baby dragon (= instant reset). I still don't understand why that happens, but moving a bit more to the left seemed to remove this problem. The time loss is a little over half a second, compared to the ideal case, but I decided to trade time for safety in my attempts.
* The jump after the 8th baby dragon can be done with a small jump, saving 5 frames over doing a big jump, but it's frame-perfect (= 5 actual frames) and I again decided to go for additional safety in my attempts. The same thing is true for the jump to the platform with the D-box.
* As described in the sdakb, there are two methods described for fighting Singe, while attempting a 6-puff fight. The first is the TAS-method. It's not too difficult to execute, but there are some parts that are a bit tricky. The second method is definitely easier than the TAS-method, but also 45 frames slower. For the sake of consistency, I decided to go with the second method in my attempts, possibly switching over later on if my consistency went up.
* The Singe fight started badly when I jumped too early into the fight. Jumping early means the first axe will miss, resulting in a 7-puff fight. There is a bit of execution around the 2nd BD, but the rest of the fight is pretty much on rails, while crossing the fingers for the smoke puffs to all be of the right kind. On top of the 7th puff meaning another random puff included in the fight, it also meant a time loss of around three quarters of a second.

Final thoughts
I'm still almost in disbelief that I sit here with a completed run. I expected this to be a never-ending project that would haunt me for a long time. At the point I was in the grind, I accepted mistakes here and there and did not require the LK to appear in the first level. So getting the LK and no major execution mistakes in the completed run feels satisfying. That being said, there are still many ways to save time by including faster strats and/or trying for even more RNG-dependent events. I guess you should never say never, but I have a hard time seeing myself getting back to this game again though.

Credits
* 'SBDWolf' - For having found several route improvements, especially in stage 4. He also contributed to the game knowledge through discussions.