That’s not unsurprising when it comes to a lot of research. The bench scientists I know often struggle with yield rates and labs have special knowledge about their particular set up when it comes to getting stuff to work.
Even in robotics (my area), if you are watching a video of a robot doing something cool, there’s often a bunch of times they ran the same demo and it didn’t work for some (often largely irrelevant to the main idea) reason. I also remember, in an undergrad analog circuits class, we had to build an amplifier on a breadboard with certain performance specs (e.g., a fairly high cut off frequency, etc.). This ended up being fairly difficult due to the tolerances in the components to which we had access and breadboard parasitics. I recall getting a non-trivial performance boost by swapping out a dozen 2n2222’s until we found “a good one.” The gray beard professor laughed and said that’s an expected part of our practical education.
