Before starting an acquisition, you set the spectral width. That is an hard limit. What you truncate at this stage will be difficult (let's say impossible) to recover afterwards. Before printing you can further reduce the spectral width. I judge this as good practice because the zoomed-in spectrum is more readable. How can we define the limit? If the spectrum is not saved on disc, or if it is saved but the disc becomes unreadable, that limit will be another hard limit. If you keep the spectrum on your computer, it will be a soft limit. Tomorrow you can enlarge the spectral width again.
It's not my intention to bore you with trivial considerations. I still need to make people understand that the printed copy is not all that remains of the spectrum, it's just a copy, a translation. People with the CW mentality still don't understand which is the true spectrum. The true spectrum is the FID!
There is still the tendency to consider the processing as a phase in the generation of the spectrum. Like the clothes you are dressing are important but are not a part of you, processing is vital but it's not forever. The spectrum (namely the FID) can survive without processing, can go to bed naked and can dress new clothes every day, and still being the same spectrum.
Back in the old days when processing a 2D spectrum could take a whole hour, it was a necessity to send the spectrum to bed with its clothes on. Today it takes more time to change your real clothes than to reprocess an NMR spectrum. There is a page that shows how even a large FID can be read from disc, weighted, transformed, phase- and baseline- corrected, and displayed in a fraction of second. The whole process can be repeated more times per second. You can download the program from the same site and check it out. And you don't even need the last generation of computers!
While yesteryear we had to process our spectra, all we have to do now is to choose the appropriate processing. At the end of the day, instead of storing the processed spectrum, we store the processing parameters into a wardrobe. Next time we open that spectrum, the computer will automatically recover the parameters from the wardrobe and dress the FID, in a fraction of a second. But you can also choose a new dress. What's more important: you are never losing anything, but only if you have realized that the FID is the true spectrum.
If you try to store the transformed spectrum ("for not to lose my work") you are actually throwing away the most precious part, the original, and keeping the copy!
Beware that not all softwares are equal. Those in widest use, like VNMR and TopSpin, by default store the FID, the parameters and the transformed spectrum. iNMR and SwaN-MR store the FID and the parameters but not the transformed spectrum. Mestre-C stores the parameters and the processed spectrum. It's your responsibility to store the original FID. The additional weakness of Mestre-C is that you can't keep the parameters separated from the spectrum. Let's say you want to reprocess the FID with a different weighting function. You recover your original FID from your back-up storage and you can do what you want. Unfortunately you have to perform everything again, including phase correction, baseline correction, defining the integral regions, the display limits, etc... even if don't want to change them. It ain't to be this way and, fortunately, Mestre-C is an exception.
I need to digress now. At the end of 2005 I decided to adopt Mestre-C as my NMR software. I tested it during that year and started using it from 2006. (The idea was to keep the whole 2005 archive in SwaN-MR format and the whole 2006 archive in Mestre-C format). I knew that the rationale behind Mestre-C is the outrageous "undo" logic and I was suspicious. I discovered a command called "Reload" or something like that. After selecting it, the processed spectrum disppeared and the FID was displayed. I said to myself: "There must be a private storage of the FID into the Mestre-C document". After two months in which I had, with confidence, thrown away all my Bruker FIDs, I discoverd that the Reload had ceased to work. The reason is that it only worked as long as the FID file was not moved or removed (practically: only during the first session of usage). This is to say that even an expert and suspicious spectroscopist like me can make mistakes. The desperating thing is not the software itself, but the discovery the I was the first user (after 20000 earlier adopters) to complain! (Do you realize, now, why I am so boring with trivial considerations? I need to re-educate the whole world!).
To close the digression, the rationale behind Mestre-C is that you can reconstruct the FID with an inverse FT and the imaginary part with an Hilbert Transform. It's like generating the negative from the photograph: unnecessary, expensive, time-consuming, avoidable, risky, painful.
Last but not least: I hate GLP, but the original data must be preserved in its integrity.
You have seen a facet of what I call the old mentality, the "recorded music" mentality. In my "live" NMR, processing is to be repeated every time (by the computer), and I am the one who writes the score. If someday I want to add or remove an instrument, it will not be overdubbed, everything will remain authentic and live. Certainly the "recorded music" mentality is the daughter of the CW mentality ("if NMR existed before pulsed NMR, NMR can exists without the FID") that implicitly renounces to one of the advantages of FT-NMR, namely processing (you can process the same experiment in many ways, and it costs nothing). The pure CW mentality, still existing today, ignores even the possibility to inspect a spectrum on the screen (only printed paper matters). Maybe they don't even observe the printed spectrum, what matter is to store it into a file cabinet...