The test has two halves and both are worth running. If the inside has the
same types and verbs as the outside, all the way down, you have a hierarchy, and a folder
tree is the clean example. If the inside needs a different grammar, so that it is no longer
a semantic graph, the claim is false. In between, every zoom that opens a new vocabulary
joined by a named edge is the claim working.
- book1
- chapter17
- section165
- block822
- sentence1,782
- word27,002
This estate’s own making-of book at v0.2.0, taken apart six
levels and put back byte-for-byte or the build fails: 160 shards, one format, one
validator. That is scale invariance, and it is the honest label, because the
vocabulary never changes on the way down. By the first half of the test, this ladder on
its own is a hierarchy. The book says so about itself, in the chapter that states the
test. Open it and zoom it yourself →
So where is the fractal part? In the vocabulary changing. The same
pilot document on this site carries
two graphs at once: a core graph of sections,
blocks, sentences and words, and an extraction of concepts, claims, hypotheses and
objections joined by verbs like
departs-from and
licenses. Neither
is derived from the other; they share nodes by anchor. Open a word in one and you are in a
lexical world, open a claim in the other and you are in an argumentative one.
The chapter runs the test on this
estate and reports where it fails.
The meaning of a node is supplied by the ontology at the altitude where
it sits, which is why the same node means different things at different altitudes.
Article 9 is a binding provision in a graph of instruments, a container of paragraphs inside
the regulation, and a source of definitions to the paragraph that cites it. Same node,
different edges around it at each level.
Nobody is forced to conform. An organisation, a division, a team, a
person or a regulator can each define their own world in their own words and connect by
drawing edges rather than by adopting a shared schema. That is the consequence of
don’t merge vocabularies, bridge them: shared facts
owned by nobody, formulas per party, declared bridges between them.
The fullest worked demonstration is not on this site.
sgit.ai walks eleven altitudes across seven
published vaults, from the text of a law to a threat on a compute instance, each rung
modelled by its own author in its own vocabulary and every one openable with the read key on
its page. Markdown twin for agents:
index.md. Four of those
rungs are now read and gated here, in the evidence
estate, alongside the ladder itself.