Development Active

Neuroca, Inc.

Proposing the Void Dynamics Model (VDM): real-time, physics based intelligence. Driven by intrinsic motivation, guided by novelty, habituation, and homeostasis. May be one of the first "curious" models. The VDM becomes capable of powerful cross domain reasoning, and currently runs autonomously, indefinitely on entry level consumer notebooks, zero training.

The model becomes more performant over time while decreasing in size. Thousands of test runs have been conducted, each between 1,000 - 100,000 neurons (10,000 - 2 million synapses). Strong, cross-domain, causal-based reasoning and inverse scaling have been observed in 1,000 neuron models after only the first 3 minutes of runtime, on an Acer Aspire notebook.

"Interesting things like intelligence love to show up in dynamic, evolving systems.
There needs to be an opportunity cost, and you don't see that in static systems.
The trouble isn't in creating dynamic systems, it's in creating the right dynamics."
- Justin Lietz, Founder & CEO

Overview

Void Dynamics are implemented as event‑driven, bottom up, emergent learning rules that push and pull on a neural substrate.

VDM anchors this in a proven Reaction–Diffusion baseline and a verified single‑site logistic invariant (Q_FUM). Current validations span hydrodynamics (LBM → NS), Memory Steering, and goal directed structural plasticity (GDSP + SIE).

These results are reproducible, with falsifiable scripts, logs, explicit acceptance criteria, and rigorous physics benchmarks.

This model trades scale and compute for time, where learning increases with time while scale and total cost of ownership decrease.

Current Status
Reaction–Diffusion BaselinePROVEN
Q_FUM Logistic InvariantVerified
Core ArchitectureDevelopment
Hydrodynamics (LBM → NS)Validated
Last Updated: 2025.08.26
Methodology

Sparsity + Emergence = Performance, no exceptions.

My philosophy from the beginning has been simple. Start from domain agnostic observations of nature's elegance. I saw how unpredictable systems that manage to balance chaos and order just right showed up everywhere. I noticed that it was not the chaos, nor the order that created the emergent phenomena. It was the space in between the interaction of these two concepts that seemed to produce the most interesting results.

The Void Dynamics Model is build entirely on this principle, voids create a kind of pressure, or a call to action. Either there is something to be filled, or something to remove. Rivers will let inefficient tributaries dry up to redirect flow to more efficient paths because there is no choice. This is a physics principle where systems evolve to minimize energy expenditure.

It may be more accurate to say that this is a model of opportunity cost in action more than it is a model of intelligence. It might just so happens that intelligence seems to always find itself strung up in opportunity cost.

Reaction–Diffusion Baseline

PROVEN

Canonical leading‑order continuum model; Fisher–KPP front speed and linear dispersion validations with published acceptance criteria.

Q_FUM Invariant Checks

PROVEN

Per‑site logistic constant of motion with ΔQ ≤ 1e−8 numerical verification for the discrete on‑site law.

Hydrodynamics (LBM → NS)

Validated Benchmarks

D2Q9 BGK with Taylor‑Green vortex and Lid‑driven cavity benchmarks; criteria specified for ν_fit vs ν_th and ‖∇·v‖₂.

Memory Steering

Validated

Refractive‑index routing n(x,t)=exp[ηM(x,t)] with boundedness, linear step response, and target convergence acceptance tests.

Structural Plasticity (GDSP) + SIE

Operational

Event‑driven growth/pruning and intrinsic valence (novelty/TD/habituation) integrated with sparse connectome runtime.

Event‑Driven Telemetry

Operational

Void‑walker announcers, non‑interference guards, KPIs, and CI hygiene for deterministic, sparse‑first measurement.

Current focus

Physics: Map emergent intelligence through Void Dynamics to the rigor of physics (RD and Q_FUM), advance the EFT/KG branch with publishable derivations, accurately predict behaviors of memory steering, further strengthen Tachyon Condensation derivation, and aim to maintain a transparent validation pipeline.

Intelligence Model: Complete my public release roadmap by integrating convergent reasoning with the existing divergent reasoning, upgrade the Universal Transduction Decoder / Universal Temporal Encoder to upscale undersanding and precision measurements of per-millisecond neural activity analysis, refine and maintain my philosophy for unidirectional, sparsity + emergence architecture, and attempt to drive the model to its limits in efficiency, reasoning, and safe autonomy.

Empirical Foundation

Current validated foundations.

We enumerate what has been validated and what criteria are used. The items below reflect reproducible results mapped explicitly to some area of the model's architecture, with explicit acceptance thresholds—no projections or marketing hype.

PROVEN
RD Branch
ΔQ ≤ 1e−8
Q_FUM Invariant
TG/Cavity criteria
LBM Benchmarks
Acceptance met
Memory Steering

Reaction–Diffusion Baseline

PROVEN

Leading‑order continuum model with Fisher–KPP front speed and linear dispersion validations meeting published acceptance criteria.

Methodology: Front speed c_front = 2√(Dr) and σ(k)=r−Dk²; error bounds and R² thresholds as specified.
Implications: Establishes a rigorous continuum anchor for FUVDM dynamics.

Q_FUM Logistic Invariant

Verified

Exact single‑site constant of motion for the autonomous on‑site logistic ODE verified numerically to machine precision.

Methodology: ΔQ ≤ 1e−8 with step‑size scaling consistent with time‑stepper order.
Implications: Provides a conservation structure for per‑site trajectories.

Hydrodynamics (LBM → NS)

Validated

Operational reduction via D2Q9 BGK and Chapman–Enskog; validates Taylor‑Green vortex decay and Lid‑driven cavity divergence norms.

Methodology: ν_fit within 5% of ν_th on ≥256²; max ‖∇·v‖₂ ≤ 1e−6.
Implications: Demonstrates a fluid regime consistent with Navier–Stokes under stated criteria.

Memory Steering

Validated

Refractive‑index routing n(x,t)=exp[ηM(x,t)] with boundedness, linear step response, and target convergence.

Methodology: Acceptance framework on [0,1] bounded M, |p_fit−p_pred| ≤ 0.02, convergence to canonical target.
Implications: Adds a non‑interfering, slower steering layer for routing bias.

Documentation

A consolidated index of derivations, benchmarks, and acceptance criteria will be published here. Until then, the summary document in the repository captures scope and current status.

Research

Research publications (drafts)

Draft companion papers to be submitted to arXiv. Links below provide repository context and direct PDF access.

VDM RD baseline — validated methods and QA invariants

Draft (preprint)

A logarithmic first integral for the logistic on‑site law in Void Dynamics

Draft (preprint)
Status: drafts pending arXiv submission; repository links include context and change history.