# The Interaction Surplus Framework (Papers A–E)

**A Unique Scalar Law for Cross-Domain Interaction, with Open-System Order Dynamics**

*Author: Michael Laurence Curzi (36N9 GENETICS LLC) · License: MIT (Attribution Required)*
*Operational encoding: `modules/vovina_interaction_surplus.py` · Empirically verified in `BEHAVIOR.md`*

---

## Abstract

We present the Interaction Surplus Framework, a five-part theory (Papers A–E) that
derives a **unique** scalar functional measuring the "surplus" generated when two
representations interact across a block-decomposed inner-product space, and an
**open-system dynamics** that turns a stream of such surplus into accumulated
*order*. The framework is not fitted; under four axioms it is the *only* admissible
law. It is wired into XERO as the stability ceiling against which every module is
benchmarked. All theorems below are machine-verified (`verify_axioms`) and all
dynamical claims are reproduced by the live behavioral probe.

---

## 1. Setup (Paper A)

Let `V = H₁ ⊕ H₂ ⊕ … ⊕ H_N` be a block-decomposed inner-product space (the default
`N = 22` mirrors the 22 modules / Hebrew letters / Tree-of-Life paths, but the
theory holds for any `N ≥ 2`). For unit vectors `x, y`, define the **geometric
interaction parameter**

```
u(x,y) = 1 − (x·y)²  =  ‖x ∧ y‖²  =  sin²θ  ∈ [0, 1].
```

`u = 0` means collinear (no interaction); `u = 1` means orthogonal (maximal).

## 2. The axioms and the uniqueness theorem

| Axiom | Statement |
|-------|-----------|
| **S1** Geometric dependence | `F(x,y) = f(u(x,y))` for some scalar `f` |
| **S2** Zero at zero | `f(0) = 0` |
| **S3** Affine effective count | `g(u) = e^{f(u)}` is affine in `u` |
| **S4** Normalization | `f(1) = ln N` |

> **Theorem 2.1 (Uniqueness).** The only functional satisfying S1–S4 is
> ```
> f(u) = ln(1 + (N−1)·u),     g(u) = e^{f(u)} = 1 + (N−1)·u.
> ```

`g(u)` is the **effective count**: the number of blocks effectively participating
in the interaction. `f` is its logarithm — an entropy/order measure.

## 3. Derived properties (Theorems 3.1–3.4)

```
f'(u)  = (N−1) / (1 + (N−1)u)      > 0   ⇒ strict monotonicity (T3.1)
f(u)   > 0   for u > 0                     ⇒ positivity         (T3.2)
f''(u) = −(N−1)² / (1 + (N−1)u)²   < 0   ⇒ strict concavity    (T3.3)
```

**Sharp Lipschitz bound:** `|f(u₁) − f(u₂)| ≤ (N−1)·|u₁ − u₂|`. The constant `N−1`
is attained at `u = 0` and is used throughout XERO as the **stability ceiling** —
no module may exhibit a marginal surplus slope exceeding `N−1`.

*Empirical check (N=22):* `f(0)=0`, `f(1)=ln 22=3.091`, `g(1)=22`, Lipschitz `=21`,
measured max slope `17.5 ≤ 21`, all sampled `f''<0`. ✔

## 4. Two-source decomposition (Theorem 4.1)

Any interaction splits into a **crossing** (widening) and a **divergence**
(deepening) part:

```
u = u_cross + u_div = β² + α²·sin²γ
max(f(u_cross), f(u_div)) ≤ f(u) ≤ f(u_cross) + f(u_div).
```

The `widening_ratio = f_cross /(f_cross+f_div)` tells us how much of an interaction
is *breadth* (new cross-domain coverage) versus *depth* (refinement within a
domain) — a directly actionable diagnostic for routing and curriculum.

## 5. Information-theoretic bridge & the falsifiable test (Papers C–D)

Because `f = ln g`, the surplus **is** a log-effective-count. Paper C issues a
**falsifiable prediction**: a transformer's empirical effective dimensionality,
measured by the **participation ratio**

```
PR = (Σ wᵢ)² / Σ wᵢ²
```

on its attention/mixture weights, should equal `g(u) = 1 + (N−1)u` for the block
geometry's interaction parameter `u`. This makes the framework testable on real
models (`flagship_prediction(u, N)` returns the predicted observables).

## 6. Open-system order dynamics (Paper E)

A stream of surplus is converted into accumulated **order** `Q` by the discrete law

```
Q_{t+1} = (1 − δ)·Q_t + η·F(U_t) − C_t
```

with retention-loss `δ ∈ (0,1]`, conversion efficiency `η ≥ 0`, surplus input `F`,
and dissipation cost `C`.

| Result | Statement |
|--------|-----------|
| **Growth (Cor. 3.2)** | `Q` grows iff `η·F > δ·Q + C` |
| **Min surplus (Prop. 4.1)** | `F_min = (δQ + C + r)/η` for target growth `r` |
| **Steady state (Thm. 3.5)** | `Q∞ = (η·F − C)/δ` (0 if `ηF < C`) |

### 6.1 Empirical validation (from the behavioral probe)

With constant input such that `ηF − C = ln N = 3.091`, the system converges to
`Q∞ = ln(N)/δ` with **geometric** relaxation factor `(1 − δ)`. Measured:

| δ | predicted `Q∞ = 3.091/δ` | measured `Q∞` | steps to 99% (`ln 0.01 / ln(1−δ)`) | measured |
|------|------|------|------|------|
| 0.05 | 61.82 | 61.82 | 90 | 90 |
| 0.10 | 30.91 | 30.91 | 44 | 44 |
| 0.25 | 12.36 | 12.36 | 18 | 17 |
| 0.50 | 6.18  | 6.18  | 9  | 7  |

The match is exact for the steady state and within rounding for relaxation — a
clean confirmation of Theorem 3.5.

## 7. The 27/33 fractal binding

Only `27/33 ≈ 0.818` of the available surplus is committed to action; `6/33 ≈
0.182` is held in reserve for the self-witness protocol. Thus the *operational*
budget is `f(u)·27/33` and the *reserve* is `f(u)·6/33` — coupling the surplus law
to XERO's self-verification quorum.

## 8. Role in XERO

- **Stability ceiling.** Every module is benchmarked against the Lipschitz bound
  `N−1`; exceeding it flags an unstable interaction kernel.
- **Self-proof.** `verify_axioms()` is XERO's machine proof that its interaction
  kernel *is* the unique surplus functional, not an approximation.
- **Self-evolution engine.** The open-system law is the mechanism by which XERO
  accumulates order from surplus-bearing interactions and decays toward a
  *designable* steady state — the quantitative backbone of its autonomy.

---

*Companion documents: `WHITEPAPER_01_CAPABILITIES_AUDIT.md`,
`WHITEPAPER_02_ARCHITECTURE.md`, `BEHAVIOR.md`.*