My contrast is 1 0 0, which I think reflects mean FD for the whole group (as I demeaned the covariates).

The outputs of Z-values or t-values seem reasonable, as mean FD should be positive across the brain. However, Null_contributions and 1-fwe outputs seem incorrect.
Z-values range from 13 to 35.4
t-values: 15.8 - 228.1
abs-effects: 0.1 - 1.4
Null_contributions: 0 - 5000 (one fixel is 5000, others 0)
1-fwe: 0.0002 - 0.0002

I tried other contrasts for the 2 covariates (e.g. 0 1 0, 0 0 1), the values of null_contributions outputs are all within reasonable range (lower than 10).
What could be the possible reason for the incorrect outputs when I look at the global intercept? Any suggestion would be highly appreciated!

My contrast is 1 0 0, which I think reflects mean FD for the whole group (as I demeaned the covariates).
The outputs of Z-values or t-values seem reasonable, as mean FD should be positive across the brain.

While technically true, it does make me question exactly what you’re trying to achieve here. This is equivalent to stating that your null hypothesis is that mean FD is zero across the brain; which doesn’t seem like a reasonable null hypothesis to me. This is why the t-values & Z-values are universally large.

What could be the possible reason for the incorrect outputs when I look at the global intercept?

I would want to ensure that the hypothesis being tested is reasonable, and the shuffling strategy is appropriate for the model, before digging too deeply into this. But the fact that the distribution is reasonable for evaluating the correlation between FD and those other variables suggests that it’s not a fixel-fixel connectivity issue.