This movie is 31 frames /
00:00.52 seconds faster than the
current movie. The rerecord count is not accurate.
Background
In March 2020, I submitted an improvement of 1504 frames /
00:25.03 seconds to Randil’s very old Heroes of the Lance movie. After the submission was published, ktwo posted some ideas that might shave off a few more frames, and some of those ideas proved to be valid. The biggest one was a trick to improve screen 2-8, the Flaming Gauntlet.
I started incorporating ktwo’s ideas into an improved run while also testing some of my own ideas. I found some improvements early in the run and was 24 frames ahead after screen 1-5. I maintained the 24-frame improvement through screen 2-6. RNG was a little worse on 2-7, but I still was 19 frames ahead.
Then I reached 2-8, the Flaming Gauntlet. I was astonished by how bad the RNG was. All of my gains and then some rapidly vanished. I tried variations to change the RNG, but it remained awful. It appears to be timing-based. I had not realized that the RNG in my previous run was fantastic. Put simply, any gains made before the first flame of the gauntlet are meaningless if the flame has bad RNG. Thus, I abandoned all the work put into improving the beginning part of the run and instead started working from the good 2-8 of the previous run. Here is the file of the abandoned attempt:
userfiles/info/63462110983891496
Changes
The route has 31 screens total. The new content starts on screen 2-8.
| # | Screen | Former HP | Current HP | Total Frames Ahead | Notes |
|---|
| 1 | 1-1 | 19 | 19 | 1 | Converted from FCEUX to Bizhawk. |
| 14 | 2-8 | 7 | 7 | 39 | ktwo found that there is a 'teleportation' glitch caused by activating the menu at the right time. I confirmed that the game divides the window position into 16-pixel increments, and the game places the character at the left edge when opening the menu. The Flaming Gauntlet screen starts at window position 144. But when Goldmoon runs left through the first flame, the number eventually changes to 142. At that time, if the menu is opened, Goldmoon's position will shift to the next 16-pixel mark, which is 128. Likewise, opening the menu at 126 would shift Goldmoon to 112. Additional time saved on this screen is from getting a clean run through the flames and arrows. |
| 15 | 2-9 | 7 | 7 | 28 | The first problem on this screen is that the baby dragon has worse RNG than the previous run. The second problem is that Goldmoon's movement has to be perfectly manipulated in order to make the exit spawn in the good location (062). If the exit spawns 16 pixels further left (048), it takes an extra 21 frames to reach it. It's very difficult to find the input combination that loses only the 11 frames. |
| 16 | 2-10 | 7 | 7 | 29 | There's a half-frame difference that requires adjusting on almost every screen. |
| 17 | 3-1 | 7 | 7 | 28 | |
| 18 | 3-2 | 7 | 7 | 29 | |
| 19 | 3-3 | 7 | 7 | 28 | |
| 20 | 3-4 | 7 | 7 | 28 | |
| 21 | 3-5 | 7 | 7 | 29 | |
| 22 | 3-6 | 6 | 6 | 21 | Of course at the next opportunity for worse RNG, the game gives worse RNG. But, I managed to manipulate it to maintain the same HP. I will need it later. |
| 23 | 3-7 | 6 | 6 | 20 | |
| 24 | 3-8 | 6 | 6 | 21 | |
| 25 | 3-9 | 6 | 6 | 25 | I unexpectedly improved this room by minimizing the time Goldmoon has to wait after firing Hold Person at the baby dragon. |
| 26 | 3-10 | 6 | 6 | 23 | A small delay is needed to maintain the good RNG from the warrior on the next screen. I could have created the delay on 3-11 too. The timing is the same either way. |
| 27 | 3-11 | 5 | 5 | 26 | ktwo noted that it's possible to low attack the spider rather than use Hold Person. A regular attack is 3 frames faster than using Hold Person. |
| 28 | 3-12 | 5 | 5 | 26 | |
| 29 | 3-13 | 5 | 5 | 26 | |
|
| 30 | 3-14 | 2 | 1 | 29 | Another unexpected improvement happens after I jumped clean through the first troll. That discovery was also fortunate because I used up every spare point of Goldmoon’s HP on the second troll. |
| 31 | 4-1 | 2 | 1 | 31 | ktwo pointed out a better way to do the final menu. |
| End | End | 2 | 1 | 31 | |
