alt Hacker Newsreally enjoyed this. question: when benchmarking typesetting systems, how do you normalize for 'design intent' vs algorithmic optimization (eg manual kerning/ligature choices)? are you comparing them on the same input markup with default settings, or tuning each system to best effort? the tuning knob count can swamp raw perf.
Thank you (author here). I deliberately used the default settings. Typst mentions settings for the line breaking system in the manual (for example you can set the penalties for changing interword spacing between lines), the speedata publisher can switch on margin kerning.
There might be settings to speed up things, so if any of the authors provide me with the settings, I'd be happy to implement those.