The BWTT works with "projects", like many applications. A BWTT project consists, primarily, of a list of "jobs". You see this list in the Job List.
The BWTT "job" encapsulates everything you need to get a particular set of tracking done on a particular video file. The job consists of a video file and one or more "tasks".
A task consists of a frame range, a list of tracking "methods" to run on the video file, and the parameters of those methods, as well as a set of results (if the task has been run already).
The simple case
Most jobs will consist of only one task. Thus, simple projects will consist just of a list of jobs, each of which represents a video file and the tracking you want to do it on it.
You will encounter multiple tasks when you want to run a second (or third, etc.) lot of tracking on a video, and integrate it with the tracking that has already completed.
You might do this, for instance, (a) to extend existing tracking because you changed your mind about the frame range, (b) to repeat some tracking in a region of the video where the parameters you chose did not work out well or (c) to add (say) whisker tracking to a job that you had originally set up to do only (say) snout tracking.
In general, you can keep adding tasks to a job until the sum of all those tasks provides all the successful tracking that you need. You can delete failed tasks completely, or just invalidate their results over some frame range.
Tasks do not have to be related, so you can add snout tracking at one end of a video and object tracking at the other end. However, if two sets of tracking are entirely unrelated, consider creating separate jobs rather than separate tasks, since "integrating over tasks" only really makes sense if all the tasks are contributing to a single aim.
For a detailed description of how to handle multiple tasks, see Multiple Tasks.
Comparison with earlier versions
Earlier versions of the BWTT had a project that consisted of a list of "videos" (rather than "jobs"). The new system is much the same, at least if you don't use multiple tasks, so you can think of the "job" as the "video", more or less. However, the new system provides the flexibility to extend and integrate multiple sets of tracking to generate the full set of tracking that you require.
In various places, 2/98 normalisation is used to "equalise" an image histogram. This normalisation makes a linear transformation of the image luminance such that the 2nd and 98th percentile of the luminance distribution of the image lie at 2% and 98% luminance.