There are many possible "observations" that might emerge
in the writeups from students, based on their results.
Here are a few examples:

 - starting at the LEAF level takes a constant (deterministic)
   number of probes, for a given value of K

 - starting at the ROOT level takes a variable number
   of probes, depending on the number of items, and which ones

 - the ROOT level is always faster than the LEAF level
   on the examples given, but the opposite might be true
   if there were even more items in a customer's basket (e.g., 700)

 - if the basket has every possible item from the store,
   then LEAF level is 100% efficient, and ROOT is about 50%

 - in general, more items means more time slots and better efficiency,
   but it also depends on exactly where the items are in the tree

 - a worst case for ROOT level is having adjacent items
   like 0 and 1 at the leaf level, and nothing else. You have
   to probe all the way down to the leaf level to resolve this.

 - the number of successful probes is always the same as the
   number of items in the basket (this one is pretty obvious)

 - on the given baskets of goods, the number of collisions
   always exceeds the number of successes

 - the proposed MAC protocol for RFID scanning from the root
   is very scalable (i.e., costs increase logarithmically with
   the catalogue size M). For the leaf level, costs are linear
   with M.

 - if the common case is customers with very few items,
   then the ROOT level is always better, and very efficient

 - if a "time slot" were 10 milliseconds, for example,
   then customer 8 can be scanned in only 9.97 seconds.
   This is way faster than using a human cashier. The leaf
   level would always take 10.24 seconds, regardless of the
   number of items that customers buy.