On Sun, Oct 2, 2022 at 9:20 PM Federico Fiorani <deferoci(a)gmail.com&gt; wrote: Federico, I think you have explained it quite well. After "smpl 1 5" we still have a dataset of length 10, but (in effect) with a marker saying that only observations 1 to 5 are currently accessible. So, as you say, when the series s is saved to a bundle, we store the whole thing, to "preserve information". With "smpl --no-missing", however, gretl creates a parallel dataset holding just 5 (non-contiguous) observations. Saving the dataset to file in this context would save only the 5 selected observations. What should happen when storing the series into a bundle? I don't think there's any definitive answer. Basically it's a matter of "Don't do that!". Use lists or matrices rather than putting series into bundles, if the sample is going to change between accesses to the bundle (other than just by changing the starting or ending observation). Allin