> What command are you running, '3dDeconvolve -xrestore'?

Yep.

> The error comes because it did not write the number of bytes that
it expected to. It is not clear to me how that could be a bogus
message.

I see. I considered it bogus because clearly the perms are fine if something was written, and I'm surely not out of disk space. Haven't checked out the contrasts, actually.

> By "bogus short output", you mean that the output is simply not
of type float, or is that a reference to the error from above?

To the error, mostly. Thanks for the pointer to that thread. I'm now less worried about the round-off for the contrasts themselves, but still haven't thought through how that affects downstream analyses.

> Note that if you are using -xrestore with 3dDeconvolve, then
adding -float might not be the best option at this point. Then
your dataset would be a mixture of shorts and floats (since the
purpose is to not re-create the entire dataset).

Is that problematic? I'm guessing each sub-brik makes it clear what data type (and scaling etc.) it has?

Looking again, it seems my original 3dDeconvolve was actually run with -float.