Yes, looking at this output, the shim is by unfortunately by far not sufficient enough to provide reliable estimates. MRS requires some attention to the shim quality - which can mean spending extra time getting it right, especially on older Siemens systems this may sometimes even include manually optimizing the shim currents. If all datasets look like this, I would not recommend placing too much trust into the outcome. Poor data quality is also quite a likely explanation why Gannet fails to properly process or model your data. The display on the console is not really a great substitute for looking at your data (or judging the quality), since its options are relatively limited.
The desire to get as many datasets as possible out of an hour of scan time is understandable, but the risk is always that all datasets end up with sub-optimal quality. An additional complication can arise from running edited MRS immediately after fMRI or DTI acquisitions, since these can induce drift on some systems, and that drift can’t be corrected properly depending on its extent.
I recommend that you eat up the loss of 1.5 datasets (which very likely won’t be usable at all) and save the next 18.5 by spending more time on the shim, and exporting TWIX data. I’m also happy to assist reviewing or setting up future protocols that you may want to design. MRS is tricky, edited MRS even more so, and it’s easy to get tangled up in the (not necessary consistent or helpful) literature, especially when it’s the first time venturing out into MRS world. For MEGA-PRESS of GABA, the classic Mullins paper is a good starting point; for MRS in general, I’ll gladly point you to the list of consensus papers that has come out over the last couple of years. We’re currently working on an evidence-based “Beginner’s Guide to MEGA-PRESS of GABA”, but that will take a few more months, I expect.
Do not hesitate to ask any further questions that you may have - providing a path for MRS newcomers to reach out for advice was one of the main motivations to create this forum.