SMARTHAUSThe Mathematically Governed AI Fabric
Investors ↗ Twenty minutes
CAIOOrchestration that cannot execute

It plans. It is structurally incapable of doing.

CAIO decides which services would satisfy a request and emits a plan carrying a hash chain back to the authority that justified it. It cannot execute the plan — not by policy, but because its schema rejects any plan that carries a result and its own tests fail if execution becomes reachable.

DecidesWhat would satisfy a request
SeamBetween deciding and acting
EnforcedThree independent ways
Contracts shippedTwenty-seven
HANDED ON · SOMETHING ELSE DECIDES planning only · direct execution not allowed INTENT id · fingerprint forbidden to read SERVICE CONTRACTS C1C2C3C4C5C6 DETERMINISTIC · NO MODEL HASH CHAIN · BACK TO THE AUTHORITY PLAN · REVIEWABLE 01 · C2# → intent02 · C5# → intent03 · C3# → intent + execution result + admission request 3 proof fields · false SCHEMA · VALID PLAN THE ACTION GATE decides whether it runs EXECUTION called by nothing NO PATH FROM CAIO SCHEMA · MANIFEST · TESTS: THREE INDEPENDENT ENFORCEMENTS NO LANGUAGE MODEL IN THE DECISION PATH

The word orchestration usually means one component that decides what to do and then does it. That arrangement is convenient and it removes the only natural place to put a gate: the moment between deciding and acting.

Here those are different components with a contract between them. The planner produces an artifact describing what would happen — which services, in what order, what context would be needed, what would need admitting at the action gate. Something else decides whether any of it runs.

There is an execution gateway in this codebase, built and real, with adapters for two dozen providers. It is quarantined out of the sealed path and called by nothing. That is deliberate, and this page explains it rather than hiding it.

If planning and execution live in the same component, there is nowhere to stand between them.

The position, in one sentence

Every other orchestrator can act. This one is built so that it cannot.

And proves it in three independent ways.

The schema forbids it

A plan containing an execution result is invalid by construction, and three proof fields are pinned to false as constants.

The runtime declares it

The manifest states planning only and direct execution not allowed, and the application refuses capabilities outside that set.

The tests enforce it

Both entry points are exercised with the execution machinery replaced, and the suite fails if either can reach it.

No language model in the decision path

Verified by import analysis. Model libraries exist only in the quarantined execution side.

And it refuses to read the meaning

The request arriving here carries an identifier and an opaque fingerprint. Nothing else.

Intent resolution happens upstream, in a different runtime, and produces a hash-bound artifact. This component is explicitly forbidden from branching on the content of that artifact — the instruction is written in the source, and enforced structurally by the fact that the routing object contains nothing to branch on.

So the component that chooses services cannot second-guess what the user meant. It acts on a decision that was made, recorded and hashed somewhere else.

That is the separation of powers we build to. One component decides meaning, another decides what would satisfy it, a third decides whether it may happen, and each one ships its own prohibitions.

The problem

The gate has nowhere to stand.

An orchestrator that decides and then acts has removed the only natural boundary in the system.

01

Deciding and doing in one component

There is no artifact to review, because the decision and the effect are the same event.

02

Two components interpreting the same sentence

They will eventually disagree without anyone noticing.

03

A plan nobody can read before it runs

The value of a purchase order is that the thing describing the action and the thing performing it are different.

How it works

Contract-driven selection, deterministic, no model.

Hash-bound intent
→
Intersect contracts
→
Score cost · risk · latency
→
Emit a plan
→
Admission requests

Twenty-seven service contracts ship with it, intersected against requirements and scored on declared cost, risk and latency. Deterministic, seeded tie-break, no model.

Every derived request carries the hash of the artifact that justified it, so a downstream item walks back to the intent decision. A plan touching governed side effects must carry an admission request for the action gate.

The seal

Three independent enforcements, and a quarantine.

Schema

A plan containing an execution result is invalid by construction. Three proof fields are pinned to false as constants.

Manifest

Declares planning only and direct execution not allowed. The application refuses capabilities outside that set.

Tests

Both entry points exercised with the execution machinery replaced. The suite fails if either can reach it.

And the quarantine: there is a working execution gateway in the same codebase with adapters for two dozen providers. It is out of the sealed path and called by nothing, and the previous release that could reach it is still the tagged one. Both facts are on this page rather than discovered later.

Evidence

Enforced in the running code.

EnforcedPlanning cannot executeSchema constraint, manifest declaration, and runtime tests against both entry points.
EnforcedNo model in the decision pathVerifiable by import analysis. Provider libraries appear only in the quarantined gateway.
EnforcedFive distinct refusals at ingressEach with its own status, each exercised.
EnforcedGoverned side effects cannot be planned silentlyA plan touching them must carry an admission request for the action gate.
EnforcedCanonical hashing throughoutSorted keys, canonical numbers, explicit failure on values that cannot be represented.

Hard questions

The question that is the whole product, and the one after it.

“An orchestrator that cannot execute is useless.”

It is useless on its own, which is the point. What it produces is a reviewable artifact: which services, in what order, what context they would need, what would have to be admitted at the action gate, and a reference chain back to the decision that justified it. Something else — a runtime you choose, or the action gate — decides whether any of it happens.

That is the same separation that makes a purchase order useful. The value is that the thing describing the action and the thing performing it are different, and the boundary between them is where governance can stand.

“Why not let it interpret the request itself?”

Because then you have two components interpreting the same sentence, and they will eventually disagree without anyone noticing.

Here the request carries an identifier and an opaque fingerprint, and the source instructs this component not to branch on either. The structural enforcement is better than the instruction: the routing object simply contains nothing to branch on.

In the fabric

Where CAIO sits, and what it hands on.

CAIO is the seam between deciding and acting. It consumes a hash-bound intent it is forbidden to read, and emits a plan it is incapable of running.

TAI
→
MAIA
→
CAIO
→
SAID
→
UCP
→
MGR
→
Effect

Beneath every step: RFS and NME hold state and meaning, and MAE on the Unified Calculus supplies the rules and their proofs.

The contract

What it promises the next component.

To UCP

Admission requests for every governed side effect the plan would touch.

To the reviewer

A reviewable artifact: which services, in what order, what context, what would need admitting.

To MAIA

A reference chain — every derived request carries the hash of the artifact that justified it.

Every component is a product in its own right and works without the others. The contract is what makes them compose when you want them to, not a dependency that makes you take all of it.

The thesis

Mathematics as the nervous system of AI.

Everything here descends from one argument: that the integrating substrate for artificial intelligence should be mathematics itself — not another orchestration layer, not a better prompt, and not a policy document.

Each part of a modern AI system works. The joins between them do not. Vision, language, planning and retrieval are each remarkable and they are integrated through hand-built pipelines and brute-force scaling. The thesis proposes a shared mathematical space that components write into and read from through operations defined once and behaving the same way for all of them — a nervous system rather than a bundle of wires.

Guarantees become measurable. Every property claimed has a quantity attached. Measure it and either the implementation holds or it is broken; there is no third answer.

The foundation is reusable across customers. The calculus, the construction engine, the control plane and the receipts are common. Your rules, connectors, integrations and authority model are yours.

The ladder, in order

Each rung was built from the one before it.

Rung 01
The thesis
The origin.
Rung 02
Mathematical Autopsy
The method.
Rung 03
MAE
The engine that runs it.
Rung 04
Unified Calculus
The foundation it builds on.
Rung 05
The components
What you actually deploy.

That order is why the components share a foundation instead of being a suite assembled after the fact, and it is why a refusal at the action boundary can be traced back through a proof to a sentence somebody wrote.

The paper

Openly licensed, so you can check the argument.

Open

Mathematics as the Nervous System of AI: A Unified Field Operator Framework for Distributed Cognition. Philip Siniscalchi, v9, 27 August 2026, CC-BY-4.0.

Falsifiable

It separates conformance — does the implementation obey the mathematics it claims — from superiority over alternatives, and refuses to let the first stand in for the second.

Bounded

No claims about consciousness or sentience. The biological analogies are engineering inspiration, not identity claims. Theoretical extensions are labelled as a roadmap, never as capability.

Start at one seam

Twenty minutes, on your own machine.

Connect your own coding agent, ask it to do something that writes, deny it, verify in your own environment that nothing happened, then read the ledger.

Book the twenty minutes