- Achieve the best ending
- Aim for fastest completion
- Use of glitches
- Heavy luck manipulation
A young dragon awakens deep in a chrysm mine. He sets out on an adventure to find his missing friends, and along the way discovers the truth about his heritage.
BoF3 contains no substantial skips for a TAS to abuse (other than a small one in Angel Tower) so the broad goal is to optimize a normal speedrun. This falls largely into two categories: combat optimization, and movement optimization. In a rare case for RPG speedruns, combat optimization is actually the less complex issue. Around a quarter of the way into the game we learn the Super Combo skill, which obliterates bosses with very little resistance. Before that we do the usual TAS thing of manipulating rng for crits and counters.
Movement optimization out-of-combat allows us to almost entirely avoid random encounters throughout the game. BoF3 spawns encounters on the map screen - no 'fade to black' and loading a separate fight screen - which means the game needs room to place your party members and the enemies. The game is coded to not spawn sprites inside terrain or out of bounds, and we can abuse this to (when possible) not allow the game to spawn a legal battle. In practice this means using the Defence formation and using Garr and Peco as much as possible, since they have the largest sprites.
This video improves on the previous BoF3 TAS by janus, published in 2011, by about 20 minutes. That number is somewhat misleading though, since the old video ran on less accurate PS1 emulation than is available today. A side-by-side comparison shows that each load time should last about a second longer than it does in the old video. My quick napkin estimation is that this adds up to about another 20 minutes over the course of the run. Here's a short list of gameplay improvements, in approximate descending order of time saved:
- Avoiding far more random encounters by abusing defence formation with large character sprites.
- Optimizing the infamous Desert sequence by using the camp screen to avoid battle areas and walking primarily during the day.
- Skipping dragon use entirely against Mutant, Emitai, and Stallion; simply attacking with Momo and Ryu is faster.
- Taking a faster route through the Sample rooms (only 1 fight instead of 2).
- Collecting only the minimum number of parts needed in the Steel Beach segment.
- Picking up the Melted Blade for Teepo to substantially improve his damage when fighting the McNeil ghosts.
- Skipping about half of the intro Dauna Mines sequence by having Ryu attack himself.
- Deprioritizing experience on Momo; learning Might for the Chimera fight is not worth the time spent getting it.
- Skipping the wallet in the McNeil manor grounds by carefully sidestepping one of the guards.
The entire TAS can be viewed here:
The showcase contains commentary by me and the current Any% WR holder Zheal. In addition to the above points we also discuss numerous smaller optimizations, as well as a few possible areas of improvement in this submission. Enjoy!
Suggested screenshot: frame 1585786. The final boss fight against Myria is iconic.
nymx: Claiming for judging.
nymx: There’s quite a lot to evaluate here. I’m not typically accustomed to reviewing a TAS exceeding seven hours; however, it wasn’t necessary to watch the entire run to determine whether this submission meets the standard required for judgment. After reviewing a substantial portion—particularly alongside your commentary—I was able to gather everything needed to reach a conclusion.
1. Emulator Accuracy
As you noted in your commentary, the previous TAS benefited significantly from skipping the console introduction and developer credits, saving roughly 40 seconds. Over time, emulation has improved to more faithfully replicate the intended behavior of the original hardware, including proper handling of ROM content. We’ve seen many prior publications become obsolete as a result of these advancements. That said, accuracy alone isn’t sufficient—I wanted to ensure that the gameplay maintained at least the same level of optimization as before. In this case, it clearly does. In fact, there are noticeable improvements, which became evident when comparing both TASes side by side.
2. Optimization Improvements
Your remarks regarding RNG were particularly insightful. It’s clear that manipulation is highly complex, as actions taken in one moment may only manifest their effects seconds—or even entire scenes—later. Additionally, RNG behavior can vary between consoles due to differences in hardware and system architecture. On the Commodore 64 alone, I’ve identified at least three distinct RNG behaviors. Given these factors, the limitations you described in fully optimizing certain sequences are entirely understandable.
Conclusion
Despite starting from what could be considered an unfair disadvantage, you’ve successfully navigated significant optimization challenges while also overcoming prior emulation inaccuracies. That ultimately settles the matter for me. This is a commendable effort, and I congratulate you on the achievement.
