Documenting what failed and what did not work is equally important as documenting things which worked out ...

Indeed. Essential to the scientific method---and at the heart of the crisis in scientific research.

I see this problem all the time in data science.

I remember my prof. reminding me this when I was writing my thesis: "If you find out that the experiment comes out negative, its still a valid result. We want to prove a hypothesis and if we find out its false that gives the next guys doing the research good idea about what to do next"

Another similar concept exists in game design and agile prototyping which says "fail faster" meaning trying a lot of things and failing is a much more useful learning experience than trying to slowly get something right