Small experiments / 001
Structure changes
the description.
Take a 64-bit sequence. Change its pattern. Watch a simple run-length code become useful—or spectacularly unhelpful.
The input sequence
8 × 8 = 64 BITSRead left to right, top to bottom. Click a bit to toggle.
Keyboard: arrow keys to move; Space or Enter to toggle.
Exact reconstruction verified
Original sequence
A fixed-size binary block.
Run-length payload
1 starting bit + 6 × 4 runs.
This encoding benefits from long runs.
What is actually being counted?
This experiment uses a deliberately simple binary format. Store the first bit. Then store each run’s length minus one in six bits, which represents lengths from 1 to 64. Successive runs alternate between zero and one. The decoder knows the block contains exactly 64 bits.
For four runs, the payload is 1 + 6 × 4 = 25 bits. For an alternating sequence, every bit is a separate run: 1 + 6 × 64 = 385 bits. Both decode exactly to their original input.
These are bit counts for an agreed format, not actual browser storage or a downloaded file size. The decoder, the known block size, format identification, and file-container overhead are not counted. A general compressor would also need a way to identify the format or fall back to raw data.
Simple to us, difficult for this code
The alternating sequence is highly regular. A different representation could say “repeat 01.” This run-length encoder cannot exploit that kind of structure. Its failure is a limitation of the chosen code, not proof that the sequence is incompressible.
The experiment makes one modest point: compression is always relative to a representation and its costs. It does not measure semantic understanding, generalization, or intelligence.