Yeah, I agree. To really demonstrate the best possible IDE error recovery that LR/GLR can provide, pairing a robust batch error recovery strategy with Wagner's

Yeah, I agree. To really demonstrate the best possible IDE error recovery that LR/GLR can provide, pairing a robust batch error recovery strategy with Wagner's (and your and Lukas's) history-sensitive approach is probably the way to go. At the time that I was implementing error recovery for Tree-sitter, that felt beyond my "complexity budget", since batch error recovery was already not trivial, and the batch approach was good enough for the features I was building on GitHub/Atom. Someday, I could imagine augmenting it with history-sensitive behavior. I am thankful that Lukas has helped to "pave that path" so thoroughly.