• There are 21 files with units, with 2,819 lines of code.
• 0 very complex files (McCabe index sum 251+; 0 lines of code)
• 0 complex files (McCabe index sum 126-250; 0 lines of code)
• 4 medium complex files (McCabe index sum 51-125; 879 lines of code)
• 6 slightly complex files (McCabe index sum 26-50; 1,098 lines of code)
• 11 simple files (McCabe index sum 1-25; 842 lines of code)
• 1 main files have no units (e.g. languages without unit analysis, or files without functions) and are not classified.

The McCabe index of a file grows with its size: per 100 lines tells a large but plain file from a dense one, and most complex unit whether the complexity sits in one unit or is spread over many.

File	# lines	# units	McCabe index (sum)	per 100 lines	most complex unit
tree_util.py in lib	167	32	64	38	7
xf.py in lib	366	32	60	16	10
misc.py in lib	189	17	59	31	13
action_head.py in lib	157	27	53	34	7
torch_util.py in lib	136	18	48	35	18
action_mapping.py in lib	153	16	48	31	13
policy.py in lib	314	28	48	15	5
data_loader.py in root	142	6	43	30	22
util.py in lib	209	13	33	16	6
agent.py in root	144	9	29	20	9
run_inverse_dynamics_model.py in root	157	2	19	12	12
actions.py in lib	116	11	18	16	4
minecraft_util.py in lib	66	4	17	26	5
masked_attention.py in lib	88	6	14	16	3
inverse_dynamics_model.py in root	57	6	11	19	3
impala_cnn.py in lib	150	7	10	7	3
behavioural_cloning.py in root	86	2	9	10	8
normalize_ewma.py in lib	42	5	7	17	3
scaled_mse_head.py in lib	29	6	6	21	1
mlp.py in lib	26	3	5	19	2
run_agent.py in root	25	1	2	8	2

• File complexity is the sum of the McCabe indexes (conditional complexity) of the units (methods, functions...) in a file.
• Files are classified in five categories based on that sum: 1-25 (simple files), 26-50 (slightly complex files), 51-125 (medium complex files), 126-250 (complex files), 251+ (very complex files). The default bands are five times the unit complexity bands, as a file is a group of units.
• Only main code files with units are classified. Files without units (languages without unit analysis, or files without functions) have no complexity to measure and are left out rather than counted as simple.
• A very complex file holds many decisions, which often means it has too many responsibilities: a candidate to split, and a place where changes need more testing.