I wanted to update my post because, with the help of Shoestring, we managed to find that the PAL on location N5 from Caveman Ninja and Robocop 2 are finally identical, but also that they are identical to Edward Randy.
My board had issues I didn't expect... I will explain here how I managed to fix it.
First, I had a black screen due to a bad PAL on location N5. I replaced it by a hypothetically identical one from Caveman Ninja (or Robocop 2, they're the same).
Here is what I got with it plugged in (that's why I firstly suspected these boards were using different PALs):
Shoestring told me that it may be caused by an other failure than a wrong PAL.
I started looking for issues and noticed that bending the board made the garbled graphics changing, even making them better looking at some point.
So I suspected the two custom SMC labeled "55" having bad solder joints and reflowed the solder on them.
It then went way better. I had clean backgrounds in game, full intro with clean texts and pictures, title and Data East logo appearing but no sprites in game (there was a square on the bottom-right corner by the way that seemed containing garbled parts of sprites).
I then managed to find where the sprites part was located on the board and found two 6116 RAMs on locations N9 and M9 that had suspicious outputs (pulsing but weakly and at low voltage). Piggybacking a known good working compatible RAM (2018) made parts of garbled sprites randomly appearing on screen.
I then desoldered one of the two RAMs (at location N9). With no big surprise it was tested bad on my programmer. I soldered a socket and put a new RAM in place. There was still no sprites but the garbled square on the bottom-right corner was not there anymore. Outputs on the new RAM were looking pretty much better though with strong pulses at 5V. The RAM located at M9 still showed weak outputs so I replaced it with another good known RAM and sprites magically appeared !!
I played it to the 3rd level and it seems everything is working fine.
