gistfile1.txt

# Looking to reduces rows of board game data to
# primitives (for eventual JSON output).

# each cell can be either unoccupied, or occupied by N 
# armies of a given color. What data format do you prefer,
# and why?
#----------------------------------------------------------

# 1) nil indicates empty territory
[[:white, 2], nil, [:black, 3]] 

# 2) :empty indicates empty territory
[[:white, 2], :empty, [:black, 3]] 

# 3) somewhat homogenous. nil indicates count being non-applicable
[[:white, 2], [:empty, nil], [:black, 3]] 

# 4) fully homogenous. 0 indicates no units of either color present
[[:white, 2], [:empty, 0], [:black, 3]] 

# your own representation (leave a note in the comments)

In that case I would go with number 4. I like constancy in my data and api. If the empty cells are going to take up too much space then I would go with 1 to save space.

I go for #4, after all the advantages already said, I'd add the self-documented style.

As you said is a serialization structure what we are defining here, so would be possible to think it is gonna be drunk up by external systems, better then that the info has an implicit way to be understood.