Architecture
Architecture
How independently governed things compose without losing the boundary at which one kind of authority becomes another.
Foundations asks what exists and what laws hold over it. These questions are different: they are about passage. What may cross from one governed object to another, on whose authority, for what purpose, and with what left behind.
The problem is not integration. Enterprise systems have every reason to seek shared meaning, common identifiers and reusable metrics. The problem is where integration is made authoritative — because coupling that erases a boundary does not remove the transfer of authority across it. It removes the ability to govern that transfer as its own act.
Separate the worlds. Govern the crossings. Compose through the crossings.
Three crossings, not three stages. A result does not pass through them in order, and together they do not claim to cover the subject. They are three places where authority moves and where an architecture can quietly stop representing that it did.
Between worlds
How may independently governed objects and jurisdictions compose?
What it opens
- When two systems agree they mean the same supplier, what else — if anything — has been agreed?
- If a business definition and its analytical identity are one object, where did the transfer of authority go?
- What is left of a governed relation once the architecture declares its two sides to be one thing?
Where we are
The claim is a negative one, and it is load-bearing: a crossing can be governed as a crossing only while the worlds on its two sides remain distinct. Merge them and the transfer of authority still happens — it simply stops being representable as a separate, challengeable act.
Permissionless reference is not permissionless authority.
A shared identifier can be strong evidence for a relation. It does not establish the chain that runs from same referent through same meaning to same standing for use, because each of those is a separate claim that someone has to make.
Ways in
- Governing Across Ontological Worlds The general rule, and the three recurring ways architectures break it.
- The Three Worlds of Analytics The jurisdictions this rule is most often applied to. Read first if you like — that is pedagogy, not dependency.
What is unsettled
A crossing is defined here as a governed relation — endpoints, purpose, evidence, authority, permitted consequences, revision. What is not settled is what CARRIES it. A declaration can be authoritative in one environment and become mere data when it is serialised and handed to another, and nothing yet says how a receiving side establishes that identity, evidence and standing survived the passage.
Through the service
When is an analytical service legitimate?
What it opens
- Does the purpose a person arrived with survive being turned into a request?
- What must a served result carry for the answer built on it to be defensible?
- What should a system do when it can compute a number but cannot stand behind it?
Where we are
The subject is the legitimacy of the whole passage — purpose, request, result, answer, consequential use — rather than the correctness of any one step in it. Servability is a gate inside that question, not the question.
Ways in
- Analytical Governance The doorway to the current record, with the worked example.
What is unsettled
Standing is the least settled part. A result that is legitimate here is not automatically legitimate once it crosses into another system, and what a result must carry with it to keep its standing is an open question rather than a solved one.
By an agent
How can an intelligent agent traverse a governed system without becoming the source of its own authority?
What it opens
- Which of the boundaries a person used to stand at were ever written down?
- What is the difference between reaching an object and being permitted to use it?
- If every local step is valid, what stops an upstream settlement travelling further than its grounds?
Where we are
AI did not create these passages. It changed the conditions under which they are traversed. The people at the boundaries were called translators; they were doing something else — carrying authority distinctions the systems themselves did not represent. An agent crosses all of them in one continuous motion.
Let intelligence traverse broadly. Let authority traverse only through governed passages.
This permits more capable agents, not weaker ones. A system can safely expose more of its legitimate request space once reaching an object is no longer confused with authority to make every use of it.
Ways in
- Do Not Let Your AI Agent Govern Itself Authority separation applied directly to agents.
- The Statistical Bridge The passage from governed evidence to a claim it licenses. Also on Foundations — it belongs to both.
What is unsettled
An agent may propose a relation that does not yet exist. What it may not do is ratify its own proposal. Who ratifies, on what evidence, and how quickly enough to be usable, is an organisational question this work states rather than answers.
How these relate
One general result and two of the places it bites hardest. The general result is that merging independently governed objects destroys the very act you would want to hold someone to. That applies vertically — between business meaning, governed data and material execution — and horizontally, between two business domains that legitimately mean different things by the same word. Neither direction is a special case of the other.
So the reading order below is a movement, not a dependency. Nothing here has to be read first for the next thing to make sense, and a work that helps you see another one is not thereby a chapter of it.
Some works sit here and on Foundations
That is intended. The Statistical Bridge establishes what must exist before inference is possible and governs the passage from governed evidence to a claim it licenses. It is on both shelves because a reader might reasonably look for it on either, and neither placement is the truer one. The shelves are there to help you find things, not to rule on what a work fundamentally is.
Columna is the next room
Columna is one system built so that these boundaries survive being implemented. It is not offered as proof — the argument does not depend on it, and a working system is not an argument. It is offered the other way round: here is a principle, here is something built to it, and here is enough of its behaviour exposed that you can check.
Every deposited work, with its current version, date and record, is on the research corpus page. This page is the discussion; that page is the record.