Decompilation success discrepancy between GUI and headless project creation

[adamvsmith22]

OS=Windows,
Ghidra Versions=11.3.2, 11.4.2
JDK Version=21.0.6
JEP=4.2
Ghidrathon=4.0.0

I've encountered a consistent discrepancy in decompilation success rates between projects created via the Ghidra GUI and those created using analyzeHeadless.bat. Specifically, when I create and analyze a PowerPC ELF binary using the GUI with default settings, my script (which uses decomp_interface.decompileFunction(func, ...)) achieves a 100% decompilation success rate across all functions.

However, when I create the same project and import the same binary using analyzeHeadless.bat, approximately 5% of functions fail to decompile. The error reported is typically:

Low-level Error: Overlapping input varnodes

I've verified that the same functions succeed when analyzed via the GUI. I'm wondering:

Are there default analysis options or heuristics applied in the GUI that are not triggered in headless mode?

Is there a recommended way to replicate GUI analysis behavior in headless workflows?

Could this be related to instruction recovery or function boundary detection differences?

Any guidance on aligning headless analysis with GUI behavior would be greatly appreciated.