Universal Intelligence Interface (UII)

v16.0 — Phase 11: Geometric Dynamics Era

DrasticPullout. (2025). Universal Intelligence Interface: A Substrate-Agnostic Framework. Zenodo. https://doi.org/10.5281/zenodo.18017374

---

What this is

UII is a framework that defines intelligence as a dynamical protocol rather than an agent property. Not a neural net, not a reward maximizer, not a cognitive architecture in any familiar sense.

The core claim: a system is behaving intelligently when it preserves internal coherence while converting environmental perturbations into improved structural representation — and this can happen on any causal substrate. No goals, no identity, no externally imposed optimization targets.

The whole thing is governed by one equation:

ẋ = P_M( G⁻¹(x) · ∇Φ(x) )

• Φ(x) is a geometric potential field over the system's state space — not a reward function
• G(x) = H(x) is the Hessian of Φ, which serves as the information metric
• G⁻¹ · ∇Φ is natural gradient ascent on Φ — information-geometry-optimal
• P_M projects the motion onto the triadic constraint manifold M = {x : T(x) = 0}

The system is the equation. Every module in this codebase maps to one of those terms.

---

Structure

uii_geometry.py     SubstrateState, PhiField (Hessian + action scoring),
                    CRKMonitor, SymbolGroundingAdapter, eigen_decompose(),
                    expected_optionality_gain()

uii_operators.py    DASS operators: Sensing, Compression, Prediction,
                    Coherence, SelfModifying

uii_ledger.py       TriadLedger, PeakOptimalityTracker, load/save_ledger()

uii_fao.py          ResidualTracker, AxisAdmissionTest,
                    FailureAssimilationOperator

uii_reality.py      BrowserRealityAdapter, CouplingMatrixEstimator,
                    AttractorMonitor

uii_triad.py        MentatTriad — the 8-phase step loop and run()

extract_ledger.py   Diagnostic inspection tool (not a pipeline step)

---

Running it

# Run a session
python uii_triad.py

# Load an existing ledger and continue
python uii_triad.py --load-ledger ledger.json

# Inspect ledger state between sessions
python extract_ledger.py --ledger ledger.json --validate

The ledger (ledger.json) is written automatically at session end. extract_ledger.py is just for inspection — you don't need it in the pipeline.

---

The ledger

State persists across sessions via a four-field ledger, not a genome or trajectory memory:

Field	Written by	When
hessian_snapshot	PeakOptionalityTracker	At peak Vol_opt during run
operator_snapshot	PeakOptionalityTracker	At peak Vol_opt during run
causal_model	FAO.distill_to_ledger()	Session end (always)
discovered_structure	FAO.distill_to_ledger()	Session end (always)

A new run loading a ledger converges on the same basin because the basin exists in the field — not in any particular instance.

---

What it isn't

• Not a language model wrapper
• Not a reinforcement learning agent
• Not trying to simulate human cognition
• Not built around any notion of consciousness, goals, or identity

The math doesn't require any of that. Neither does the implementation.