Cellular automata entropy
and steganographic metadata
An entropy score of 0.618 emerges consistently from well-formed document pages processed through the Entreacte cellular automata layer. This is not a coincidence. It is a structural consequence of how typographic spacing distributes information across a page — and it forms the basis of our steganographic fingerprinting system.
Cellular automata over interval lattices
After the différance objects are extracted and organised into a spatial lattice, the Entreacte pipeline runs a cellular automaton over this structure. Each cell corresponds to a region of the document. The automaton rule is entropy-based: a cell's next state is determined by the Shannon entropy of its neighbourhood — the distribution of interval sizes and densities in the surrounding area.
We run three timesteps of this automaton. At each step, the global entropy of the lattice is recorded. For a well-formed typeset document — consistent margins, regular line spacing, uniform type size — the entropy converges rapidly. The final value after three steps is typically close to 0.618.
Why 0.618?
The golden ratio appears here not by design but by consequence. Typographic systems evolved over centuries to produce visual rhythm that is legible and aesthetically balanced. This rhythm, it turns out, corresponds to a specific distribution of interval sizes — a distribution whose entropy, when measured over a cellular automaton lattice, converges to the golden ratio.
This is empirically consistent across Latin, Greek, and Cyrillic scripts at comparable type sizes. Arabic and Hebrew scripts, which use different spacing conventions, produce slightly different values — but still within a predictable range. CJK scripts, with their uniform character size and regular spacing, produce the most consistent entropy values of all.
Entropy as a steganographic fingerprint
The CA entropy score, combined with the signature checksum derived from the GNN edge structure, forms a document fingerprint that is highly sensitive to modification. If a single character is altered, inserted, or deleted, the interval structure changes. The différance objects shift. The entropy of the CA lattice changes. The fingerprint changes.
This makes the Entreacte fingerprint useful for document authentication. A legal contract, a medical record, a patent filing — any document that needs to be verifiable can be fingerprinted at submission and verified at any later point. The fingerprint does not depend on any watermark embedded in the document. It is derived entirely from the visible structure of the text.
Entropy under degradation
One of the most useful properties of the CA entropy is its behaviour under document degradation. When we apply Gaussian noise and blur to a clean document scan, the CA entropy increases — from 0.618 to approximately 0.775 in our benchmark tests. This increase is semantically correct: a degraded document has more uncertainty in its interval structure, and the entropy reflects this.
More importantly, the entropy increase is monotone and predictable. This means we can use the entropy score as a quality indicator — a measure of how much information has been lost to degradation. A score close to 0.618 indicates a clean, well-formed document. A score above 0.8 indicates significant degradation. A score approaching 1.0 indicates near-random structure — a heavily damaged or artificially corrupted document.
This quality signal is available at no extra computational cost. It is a byproduct of the same CA computation that produces the fingerprint. In a document processing pipeline, it can be used to route heavily degraded documents to human review without requiring any additional analysis.