I've made a few changes in https://next.observablehq.com/@fil/fcumsum-benchmark : - show intermediate results - add the ratio (compared to the best) Feel free to merge some of them if you like.

That was fast. Awesome! I will merge them back.

I added warmup trials.

Excellent, thanks! I've merged back (note: I put the warmup # below the # of trials, and fixed a typo)

I moved the utils into a new notebook so that the template is lighter. I also fixed a few bugs I found along the way.

We should fix the table's columns widths otherwise the display tends to be jaggy

Done. Thanks for the suggestion.

This would be awesome if it dynamically loaded experiments from 3rd party notebooks (e.g. https://observablehq.com/@tomlarkworthy/svg-to-gif)

I would like to use your benchmar utils to replace my bad ad-hoc solution here https://observablehq.com/d/507ab3f69d80ffb9?ui=classic However, since it's a sorting benchmark I need to randomize the input before each run. I can imagine other types of benchmarks where preparation, or clean-up after the run, is necessary. How about adding support for optional before/after properties to the experiment objects? They could be callbacks (like run is now) that lets us prepare data or maybe do clean-up afterwards (I don't need the latter but I can imagine there being use-cases for it)

(I'll try quickly adding them myself, doesn't sound too complicated)

Thanks for the suggestion. I added it.