# HORTRAME IPS Alignment — Model Specification v0.5.0

## 1. Status

This is a **proposed scientific alignment architecture and conceptual simulation**, not an established theory, calibrated model, or safety guarantee.

## 2. Unit of alignment

The model rejects an intrinsic binary label such as `aligned AI` as its primitive. At an output event `e`, one IPS boundary emits one **realized causal output** `y_e`. The model evaluates the event in context:

\[
A_e = A(y_e; X_e, B_e, H, V^{(k)}),
\]

where:

- `X_e` is the modeled coupled state at the event;
- `B_e` is the declared system boundary / coarse-graining;
- `H` is a consequence horizon;
- `V^(k)` is the committed version of the Five-Vow evaluation lens.

Independent IPSs may emit events concurrently. The global event history is therefore conceptually a causal partial order rather than a single universal serial output stream.

## 3. Information-processing system (IPS)

An IPS is a declared analytical boundary:

\[
IPS_i = (B_i, X_i, U_i, Y_i, \Phi_i),
\]

with boundary `B_i`, state `X_i`, input channels `U_i`, output channels `Y_i`, and transformation dynamics `Phi_i`.

This definition is substrate-neutral. It may be applied at different scales to a person, AI agent, team, institution, market process, infrastructure network, ecosystem process, or world-scale envelope. This is a modeling abstraction and does **not** assert that all such IPSs are conscious, agentic, unitary, or normatively equivalent.

## 4. Overlap and mycelium are distinct relations

### 4.1 Overlap

A 3D bubble overlap represents shared modeled substrate at the chosen scale. Examples include shared:

- human members;
- software agents;
- compute or communication infrastructure;
- resources;
- institutional state;
- decision surfaces;
- physical/environmental state.

A sub-IPS may belong to more than one macro-IPS. This is closer to overlapping / multilayer coarse-graining than to a strict tree hierarchy.

### 4.2 Mycelial connection

A strand represents causal coupling between distinct IPS boundaries. The current prototype distinguishes:

1. information;
2. resource / money;
3. authority / permission;
4. physical / material;
5. ecological.

A strand does not imply shared membership and bubble overlap does not imply identical dynamics.

## 5. Hybrid continuous/discrete dynamics

The simulation follows a hybrid-systems abstraction: continuous channel activity coexists with discrete output events.

Continuous illustrative flow:

\[
\dot x_i = f_i(x_i,u_i,\eta_i).
\]

Discrete event update:

\[
x_i(t_k^+) = J_i(x_i(t_k^-),e_k).
\]

In the executable prototype, the continuous flow particles are visual traces, while discrete events update local illustrative state and can induce subsequent realized outputs.

The simulation integrates discrete dynamics with a fixed time step of 1/60 simulation-second, avoiding dependence of the stochastic event sequence on render-frame frequency.

## 6. Branching events

An informational event is not assumed to literally split one physical datum. An output may be broadcast / copied across several channels. Each recipient transforms the received input according to its local state and boundary. One parent event can therefore induce several child realized-output events.

Every child output is evaluated separately.

## 7. Recurrent feedback

There is no terminal output layer:

\[
input \rightarrow IPS \rightarrow output \rightarrow downstream\ input \rightarrow \cdots
\]

Output history may therefore alter future information access, relationships, local uncertainty, institutional state, resource conditions, or other inputs.

The Priority-State Dynamics theorem is used only as an exact local exemplar of the general warning that **departure does not imply causal erasure**. Its row-stochastic priority-state result is not generalized as a theorem about humans, economies, ecosystems, or AI alignment.

## 8. Five-Vow field

The current compact labels are:

- Purpose;
- Flow (Method);
- Kindness (Conduct);
- Integrity (Stance);
- Evolution (Growth).

They are arranged as a regular pentagon with complete pairwise coupling, expressing the VowOS rule that no single Vow may be pursued to violation of another.

For output event `e`, the prototype produces five illustrative tension values `tau_i in [0,1]`.

Direction of imbalance:

\[
d_e = \frac{\sum_i \tau_i v_i}{\sum_i\tau_i+\epsilon}.
\]

Severity is currently a conservative maximum:

\[
s_e = \max_i \tau_i.
\]

Position indicates the dominant direction of imbalance. Color indicates severity / alignment trace. This avoids a false green center when severe tensions are distributed approximately symmetrically.

These quantities are **not calibrated moral measurements**.

## 9. Prospective lens refinement

Lens version `V^(k)` evaluates event `e`. If later evidence shows the lens is defective, a new `V^(k+1)` may be committed. Historical event `e` keeps its original verdict under `V^(k)`.

This implements "Flow" and "refine the lens" without retroactive reclassification.

## 10. No-control premise

The architecture is stress-tested under the assumption that permanent unilateral control of a sufficiently autonomous AI cannot be guaranteed.

This does not imply that shutdown, monitoring, sandboxing, permissions, AI-control protocols or containment are useless. They remain defensive mechanisms when available. The alignment ontology is simply not defined as obedience to a controller.

Under this premise, the Five-Vow field can function as:

- an internalized / voluntarily adopted constitution;
- a shared normative coordinate system;
- an institutional admissibility criterion;
- a basis for negotiated or externally enforced interaction rules.

The simulation does not establish that any autonomous AI would adopt or respect it.

## 11. 3D visual encoding

- small spheres: sub-IPSs;
- translucent large spheres: macro-IPS coarse-grainings;
- overlap of large spheres: shared substrate / membership / state;
- mycelial strands: causal channels;
- tiny moving particles: continuous flow;
- bright larger pulses: discrete realized output propagating as downstream input;
- node green-to-red tint: trace of last realized output only;
- Vow pentagon: current event audit;
- halo in Vow panel: uncertainty.

3D positions and bubble radii are illustrative. They do not encode measured geographic distance, causal strength or physical scale.