The missing symbol sweep in docs/libreoffice-gap.md answers a different question and is EXHAUSTED. Every wall since it was written has been invisible to it: a C function that exists, prints STUB and returns nil an ObjC method that exists and calls NSUnimplementedMethod a selector no class here implements at all None of those three is an undefined symbol, so comparing symbol tables cannot find any of them. backingScaleFactor and accessibilityDisplayShouldReduceMotion were both missing SELECTORS on classes that exist, and CTFontCopyTable was a present symbol that answered nil 782 times in one run. This reads the markers WE ALREADY WRITE in our own sources and intersects them with what the application actually references: its imported symbols, and the selector names in its __objc_methname section, which is exactly the set of selectors it can send. Against LibreOffice it reports 24 stubbed C functions it imports and 74 stubbed methods whose selector it uses, each with the file to edit. The third list, selectors nothing here defines, is a HEURISTIC and says so: an application also sends selectors to its own classes, so it over reports and is printed last for that reason. docs/libreoffice-worklist.txt is the current output, kept so the next session starts from a list rather than from a run. Also here: the colour space patch that the GUI needs. O2ColorSpaceCreateWithName handled only sRGB and Display P3 and returned NULL for everything else, which is not a degraded colour but a caller that cannot make a context. LibreOffice asks for GenericGrayGamma2_2 while bringing its GUI up and dies on it. The gamma is NOT modelled and the comment says so: each name maps to the device space with the right number of components, so a 2.2 gamma grey and a linear grey come out the same. That is a colour accuracy difference against a process that does not start.