graphs.sgit.ai › the evidence estate, second edition

Four more graph vaults

The first edition analysed five published graph vaults and froze with them, so /v1/vaults/ is evidence and cannot take a new one. These four arrived with brief 47: they are rungs of a ladder a sibling project walks from the text of a law to a threat on a compute instance, and this site had never looked at them.

These 4 readings are second-hand, and the pages say so on themselves. This estate cannot open an encrypted vault. Every number on these pages is quoted from the page sgit.ai publishes about that vault, all 5 of which are carried whole in sources/ with their SHA-256 and the timestamp they were fetched at (2026-09-20), and the build fails if a quotation is not in its carried file byte for byte. The first edition’s five analyses were written by opening the vaults. These were not.

Vault	Its rung	Id
Standards Atlas: GDPR A regulation whose operative meaning is not in its text, modelled so that the rulings and the guidance are nodes rather than footnotes.	the law, and its interpretation	`4zv4bvmu`
Provenance is not conformance A conformance layer over a published standard, whose whole design is that an absent answer is a finding rather than a silence.	the standard, the evidence and the policy	`2wzct4k7`
Licence to Operate An agent's authority modelled as an insurance policy, and then spent, so that the gap between what it can do and what it may do has a price.	the policy, spent turn by turn	`posrhzp3`
Scaling Threat Modeling with Semantic Knowledge Graphs Eleven linked threat models stacked from the customer to the compute instance, so a flaw in one method can be traced to the revenue it threatens.	the system, all the way down to the compute instance	`0ict6flm`

The second cross-vault finding

The first edition produced one finding that no single vault contained: the capability scale, which only the comparison of five vaults made visible. This is a second, and it is of a different kind. The capability scale came out of comparing; this one comes out of a join, and the fact it returns is in neither of the two vaults it joins.

AIUC-1 publishes 1,126 crosswalks to external frameworks as text. The conformance layer resolves the EU AI Act ones into node ids in a second published vault, so a crosswalk stops being a string and becomes a traversal.

so a crosswalk stops being a string and becomes a traversal between two vaults

Sixty-two of them resolve, onto twenty-seven articles, and the rest are reported unresolved rather than forced into a shape they do not fit.

62 of the 1,126 resolve, at article level, onto 27 articles.

Then the join returns something neither vault knew alone: eight of those twenty-seven articles have since been amended, so the crosswalk was written against the text before the amendment.

8 of those 27 articles are amended by Regulation (EU) 2026/1744, so the crosswalk was published against the text before amendment.

Which is the point, stated by the vault in one sentence.

That is a finding you cannot reach with a document.

Why this is the argument rather than an example of it. Two organisations published two artefacts, in two vocabularies, neither aware of the other. Nobody merged them. Somebody named the edges between them, and the traversal then answered a question neither author had asked. That is meaning through connectivity producing a fact, which is the thing this site has spent three books claiming is possible.

The ladder these four are rungs of

Eleven altitudes, eleven ontologies, one grammar. Each rung names the published vault in which that level is a live graph, and every one is openable with the read key printed on its page. The diagram below is lifted from sgit.ai rather than redrawn, because redrawing a picture that is already correct produces two things to keep in step; the extraction is recorded in figures/CREDITS.md. Its links are live, which is why it is inlined here rather than embedded as an image.

The same grammar, a different ontology at each altitude

The correction this site took at v0.6.21, drawn. A four-node semantic graph; the Law node opened into a legal ontology of articles and paragraphs; one paragraph opened into a lexical ontology that has nothing in common with it two levels up. The footer is the whole claim: the grammar never changes; the ontology does.

Three panels: a semantic graph of four nodes; the Law node opened into a legal graph of articles and paragraphs; one paragraph opened into a lexical graph of defined terms.

And the link on which you leave one world for another

Inside one vocabulary, knowledge accrues one well-named edge at a time, and a register with ten thousand of them is a great deal of knowledge and not yet fractal. The fractal property is the link you jump on: out of the register into security operations, out of a suspicious DNS entry into the DNS estate, out of one record into a packet capture. Four worlds, four ontologies, one continuous path of named edges.

Four worlds in a row: a risk register, security operations, the DNS estate and a network capture, each a small graph in its own vocabulary, joined by a jump link on one node each.

For an agent

The machine surface is data/vaults.json, schema evidence-estate-v1: four vaults, every fact carrying the quotation it rests on and the carried file that quotation was checked against. The carried sources are sources/MANIFEST.json, each with its origin URL, SHA-256 and fetch timestamp. Nothing here was read out of a vault; it was read out of the pages listed in that manifest, on the date they record. The five first-edition vault analyses at /v1/vaults/ were written differently, by opening the vaults, and are frozen.