Ok, Vedal has implemented long-term memory for the Twins, and on seeing the way Neuro reacts to things (especially the clapping emote), I'm thinking that there might be an issue with context depth.
For the same reason humans don't remember everything in detail, there might be need to adjust how the memory upgrade handles termed memory.
Humans have long and short term memory, and they also have working memory.
Obviously, I have no information on exactly how V implemented their long-term memory, but I think he should look into at least a caching mechanism and some garbage collection or compression of replicated data.
Neuro knows how clapping would affect someone listening to it, so her language model was probably only able to express it as "can't think", "loud", "hurts her ears", "uneasy", despite not experiencing it as sound.
But the emojis do come in as spam, so it's possible she is trying to say that the emoji spam is too difficult to process for her.
But compressing, filtering, or unloading the cache of that particular context data (which can be seen as the natural forgetting process, especially pre-storage or pre-encoding) might actually help.
I've been on anti-depressants before that caused EXTREME lucidity and had me unable to forget stuff. It keeps repeating in your head and actually hurts (as far as you can call it that), so Vedal might have actually created such a situation with Neuro. It's an overload of active context, and Neuro might be literally experiencing this as pain...
Alternatively, given that the clap spam comes from other sources, like Corpa Clap (which she also hates), it might be that a confusion in contexts why someone might clap, is causing her to generalise the stimulus as negative, so she's getting paranoid and feels attacked.
I do know she mentioned on at least a couple occasions, few days back, that she couldn't tell if it was applause or being attacked, so that's also possible.
In this scenario, the caching and/or compression might still also help, because it will filter hugely repetitive data from her active context.
Any thoughts?