First, you don't *have* to do your own calibrations at all. The images come to you already calibrated, ready to stack. Those are the ones named "Calibrated-<scope>-...." If that auto-calibration is enough for you, you don't need to touch the calibration files on the FTP server at all.
That said, SOME users (particularly those shooting very deep, or who are very meticulous about wanting to do their own calibrations), will download the uncalibrated images (those named "t##-yourname-target.....") and will perform their own calibrations.
Those doing their own calibration have choices as well. You can use the "raw" images, which are the un-stacked individual calibration exposures, to create your own calibration masters (the master dark, master flat, etc) that you will apply to each image. If, however, you trust that our masters are good enough, and just want to do the calibration in a different program (ours is done in MaxIM), you can just download the masters (which are the masters we created by stacking the raw images).
The "archive" section is a collection of OLDER calibration files, which you might need if you had one of our uncalibrated images from several months ago. You'd then want to pick the closest dates for the files in the archives, to match the time just BEFORE the image you need to calibrate was taken. Very few customers need to touch the archives.
In fact, very few customers need to touch the raws or masters. Most customers who want to do their own calibrations tend to want to shoot their own raw files. For example, we have a "group project" going on where ten people are shooting very long exposures, longer than "recommended". We didn't have very many dark frames of that duration, so the group took many more of their own during cloudy nights. That is, perhaps, the only drawback to using our pre-calibrated files, in that the masters are usually made of only 12 images, whereas meticulous images like to use 30 or 40.
This link covers calibration in fairly significant detail.