robospect.py returns large EQWs, possibly because of environment conflict

[mwanakijiji]

I have bumped into a bug with robospect.py (tag v0.76) when attempting to run it within a python package environment that is also intended to accommodate another piece of software (of my own). 

I have an environment in which both Robospect and my own software work; and another environment in which Robospect produces EQW values that are far too large: ~100s AA (I think the 'mAA' in the .robolines file is meant to be 'AA'.) I have tried installing packages for these two pieces of software in different orders, and have used identical versions of numpy, scipy, and matplotlib (which are mentioned in the robospect.py setup.py file) but I can't find a pattern.

Attached are 
1.   Robospect outputs from runs with "good" and "bad" values in the .robolines files. As you can see, the .robospect and .pdf plots between the two are identical (as far as I can tell). Somehow the EQW values that get printed to the .robolines files are messed up very late in the process. (The .log files are interrupted at the end, because I ctrl-C'ed the runs.)
2.
Archive.zip
   Outputs from robospect.py command python ./setup.py test. There are 4 tests that fail in the "good" environment and 5 in the "bad" environment.