I have embarked on this experimentative pipedream. Please do tell if there is a better issue. I searched for replay and keyframe.
First goal is to rewind a replay and do timeline jumps without desyncs or speedups.
I'm trying to keep desync resolutions documented so its possible to refactor the POC into regression tests and some separate PR's that can be useful even if the main "replay bundle" store might not.
I do believe I'm following approach number 2 in this comment ("perfect sync-safe saves") #1388 (comment)
Discussion of the feature: #2704
Other related issues:
#1179
#350
So outside of this goal (but very interesting) is the later optimization stage to try to shrink the storage and transfer of the keyframes.
WIP branches:
https://github.com/tetrisface/RecoilEngine/tree/keyframes-rewind-catchup
https://github.com/tetrisface/Beyond-All-Reason/tree/replay
2026-06-09 update
even though the agent might succeed it might not even be possible to work towards merging anything meaningful due to the scale of the work. but i will let it stew for a bit more out of curiosity.
2026-06-13
large body of work. but I dont think the POC has to grow much more now.
DISCLAMER: Completely done with AI (GPT-5.5 on max effort)
I have embarked on this experimentative pipedream. Please do tell if there is a better issue. I searched for replay and keyframe.
First goal is to rewind a replay and do timeline jumps without desyncs or speedups.
I'm trying to keep desync resolutions documented so its possible to refactor the POC into regression tests and some separate PR's that can be useful even if the main "replay bundle" store might not.
I do believe I'm following approach number 2 in this comment ("perfect sync-safe saves") #1388 (comment)
Discussion of the feature: #2704
Other related issues:
#1179
#350
So outside of this goal (but very interesting) is the later optimization stage to try to shrink the storage and transfer of the keyframes.
WIP branches:
https://github.com/tetrisface/RecoilEngine/tree/keyframes-rewind-catchup
https://github.com/tetrisface/Beyond-All-Reason/tree/replay
2026-06-09 update
even though the agent might succeed it might not even be possible to work towards merging anything meaningful due to the scale of the work. but i will let it stew for a bit more out of curiosity.
2026-06-13
large body of work. but I dont think the POC has to grow much more now.
DISCLAMER: Completely done with AI (GPT-5.5 on max effort)