"Calibrate" intuition around percentage of support / uptake of features

[romainmenke]

In the topics on the definition of Baseline (numbers of versions supported, ...) we use percentages of support a lot.

Everyone seems to agree that "more than 90%" is good and that "less than 90%" is questionable.
But this doesn't seem based on reality or research, more a intuition based on past experience.

I think it could be useful to "calibrate" our intuition around these numbers.
I don't have an immediate suggestion for how this can be done.

This is directly related to #190

We first need to agree on a dataset.
Without all looking at the same data we might all be talking about a different 90% :)

If we do find a dataset it might be interesting to test our intuition.
Maybe 80% is perfectly fine?
Maybe 99% really is required?

[atopal]

Yes, usually we talk about something higher than 90% and lower than 98% as the threshold and that almost always refers to caniuse.com stats, even if that's implicit.

I can think of multiple ways to go about testing that, with any given data set:

• Check multiple features that are on a threshold boundary and survey developers how they feel about them
• Check usage counters for those multiple features that have crossed the threshold boundary and see if that leads to higher usage
• Expert review: check multiple features on a threshold boundary and categorize them accordingly

[romainmenke]

Sounds good!

[romainmenke]

This work has all been done