• There are 14 files with 3,300 lines of code.
• 1 very long files (1,339 lines of code)
• 2 long files (1,251 lines of code)
• 0 medium size files (0 lines of code)
• 4 small files (588 lines of code)
• 7 very small files (122 lines of code)

File	# lines	≈ tokens to read	# changes (1y)	≈ lines edited per change	lines read per line changed
viz.py in circuit_sparsity	1339	~7k–20k	7	~20	~50
gpt.py in circuit_sparsity/inference	709	~3k–10k	12	~7	~100
train.py in circuit_sparsity	542	~3k–10k	1	~700	~1
registries.py in circuit_sparsity	155	~1k–3k	3	~30	~5
tinypython.py in circuit_sparsity/tiktoken_ext	27	under 1k	2	~10	~3
clear_cache.py in root	24	under 1k	1	~50	~1

Tokens are lines × 7 to 14 (analysis.fileReadsForChanges). Lines edited per change are the file's lines added and deleted divided by its commits over the whole history. Agents read about 2,000 lines at a time: when a 17,000-line file was split into modules, the tokens an agent read for the same change fell by 83% (The Economic Benefit of Refactoring, 2026).

There are 4 files with lines longer than 120 characters. In total, there are 12 long lines.

File	# lines	# units	# long lines
viz.py in circuit_sparsity	1339	34	6
per_token_viz_demo.py in root	143	6	4
single_tensor_pt_load_slice.py in circuit_sparsity	172	1	1
gpt.py in circuit_sparsity/inference	709	43	1

• File size measurements show the distribution of size of files.
• Files are classified in four categories based on their size (lines of code): 1-100 (very small files), 101-200 (small files), 201-500 (medium size files), 501-1000 (long files), 1001+(very long files).
• It is a good practice to keep files small. Long files may become "bloaters", code that have increased to such gargantuan proportions that they are hard to work with.