The following table compares the solution of the program with
the best solution known to the author (referred to as 'human solution'
from now on).

Listed are
- the number of notes in the piece
- the 'cost' of playing the piece when applying the 'cost-
  evalution-function' to the human solution
- the best cost the program can comeup with
  (calculated with the 'Hack' Domain, cutoff as stated)
- the FIRST cost the program comes up with

Values in [brackets] refer to NO SCR's, they are of interest only
in the last column of this table.

                     cost    best cost  first cost
Problem     # notes  theory  program    program 
----------- -------  ------  ---------  -------------
c-major        15      25      25 (1)    27      [41]
2c-major       29      43      45 (1)    45      [58]
hans           13      21      21 (1)    31      [45]
sleep          37     109     109 (1)   118     [160]
beethoven      26      74      74 (1)    82      [82]



The second table compares the performance of the system with and
without SCR's. Shown are
- the first two columns (problem, # notes) as above
- the # of nodes and cpu time required to find the FIRST solution
  the domain can come up with. To do so, we set MOST-COST in the
  problem file to 999. Values in [brackets] refer to NO SCRs
- the # of nodes and cpu time required to find the BEST solution
  the domain can come up with. Setting MOST-COST in the problem
  file to this values results in terminating the search process
  when the first solution of this cost (or smaller) has been found.

                     find-first-solution      find-best-solution (2)
Problem     # notes  nodes    time in sec     nodes     time
----------- -------  -------- -----------     --------- ---------
c-major        15    34 [34]   3  [3]         283 [*]   32  [*]
2c-major       29    62 [62]   6  [7]         [*] [*]   [*] [*]
hans           13    30 [30]   6  [3]         204 [*]   24  [*]
sleep          37    78 [78]  32 [11]         250 [*]   61  [*]
beethoven      26    56 [56]  20  [6]         197 [*]   27  [*] 


Note the OBVOIUS trade-off when using SCRs:
 - the first cost the domain comes up with decreases significantly
 - without SCRs we cannot find the best solution within 1000 nodes,
   with SCRs we can. Still: If the machine would be infinitely large
   and infinitely fast, we could just scan the whole tree and wouldn't
   need any SCRs due to our 'cost'-function.

-----------------------------------------------------------------------------
(1) node cutoff after 1000 nodes
(2) best solution according to table above
[*] Node cutoff after 1000 nodes

