The engine

DRF-MWU

Dynamic Recursive Field model with Modulated Weights Update

A dual-substrate spatial reasoning architecture. It pairs neural transformers with deterministic architectural rules, so its coherence is a property of the structure it reasons on — not a behaviour it had to learn. GeoRadX is its first instantiation.

01  /  What it is

Not a model with rules bolted on. A structure that reasons.

Most machine-learning systems are a single learned function: data goes in, a trained network produces an output, and whatever regularities it holds live inside opaque weights. DRF-MWU is built differently. It is a hybrid transformer — learned attention working inside a fixed architectural frame — and it runs as two paired substrates that do different, complementary work.

The first is the kernel: the on-device engine that performs a read. It is deterministic and versioned. The second is AIDAAutonomous Intelligence Dynamics Architecture — a cloud arm that handles learning and re-authoring, separately from the device. The kernel reasons; AIDA improves what the kernel reasons with. Neither is the whole system, and keeping them apart is deliberate.

What makes it a spatial reasoner is that it does not treat an input as a picture to classify. It treats it as a field to be reasoned over, position by position — which is why, in GeoRadX, a finding stays pinned to where it sits rather than collapsing into one label for the whole image.

02  /  Coherence by construction

Some behaviour you train for. Some you build in.

A model that can't guess can't write fiction.

In a conventional model, faithfulness to the input is something you hope training instilled and then test for afterwards — with evaluations, guardrails, and checks that catch inventions after they happen. The model can produce a finding the input never supported; you are working to make it unlikely.

DRF-MWU closes that door earlier. The input — the patient's actual field — is read-only by architectural position. The reasoning parts of the system attend to it, but they cannot transform it. Inventing a finding is not a behaviour the structure permits, so nothing downstream has to audit for one.

This is the difference between a system that is usually faithful and one that is structurally unable to be otherwise. Hallucination is closed here by construction, not by a check applied after the fact — and that property does not degrade as the model scales or the domain changes, because it was never a learned behaviour to begin with.

03  /  Empty at baseline

It ships with nothing learned in it.

At baseline, DRF-MWU carries zero parameters. It is pure architecture — the reasoning structure, and nothing trained. Parameters exist only once it has been deployed into a specific domain; medical imaging, in GeoRadX's case. Before that, there is nothing to overfit, nothing to un-learn, and nothing carried over from someone else's data.

The consequence a practitioner will notice is how little data it takes to stand the deployed system up — hundreds of films, not hundreds of thousands — because the structure is already doing the work that a conventional model has to learn from scratch. The baseline is architectural, not statistical.

Baseline
Architecture only
Pure reasoning structure. No trained weights, no domain, nothing to un-learn.
deploy to
a domain
Deployed
Calibrated engine
Parameters attach to the structure — tuned to one domain, on hundreds of examples, not millions.
04  /  Two substrates

One reasons on the device. One learns in the cloud.

The split is the point. The kernel that reads a patient's film has to be deterministic and stable — the same input gives the same read, and it does not quietly rewrite itself mid-use. Learning is a different kind of work, and it happens elsewhere: AIDA re-authors the engine between deployments and ships the result as a versioned update, so improvement never destabilises a device that is in use.

The kernel

ON DEVICE · DETERMINISTIC
  • Performs every read locally — no cloud in the loop
  • Same input, same output — versioned, not self-modifying
  • Light enough to run on existing edge hardware, offline

AIDA

IN CLOUD · LEARNS & RE-AUTHORS
  • Handles all learning off the device
  • Re-authors the engine and ships it as a versioned update
  • Carries the longitudinal and cross-population work the device can't
kernel reads on device  →  measurements sync to AIDA  →  AIDA re-authors  →  new version ships back to kernel
05  /  Why it holds

The architecture is the advantage — and it isn't visible from outside.

Three properties fall out of the design, and together they are the position.

Low data
Because the structure carries the baseline, a domain is stood up on hundreds of examples, not hundreds of thousands. That makes deployments possible where the enormous labelled datasets conventional models need simply do not exist — and it makes each new region or clinic calibration cheap.
Runs at the edge
A deterministic kernel with no cloud in its reading loop runs on hardware clinics already own, without stable power or a stable connection. The reach isn't a feature added on; it's what a light, self-contained engine allows.
Defensible
The engine is held as a trade secret, not filed. A patent trades disclosure for twenty years and, for deep-tech architecture, would be near-impossible to police from the outside. Kept as a secret, the mechanism stays unobservable — a deployed engine reveals its behaviour, never its construction.

Conventional imaging model

Training data10⁵–10⁶ labelled
New regionretrain / large set
Runscloud / GPU
Faithfulnesstested after

DRF-MWU

Training data~10² per domain
New region~350 films
Runsedge / phone, offline
Faithfulnessby construction

The proof isn't a paper. It's the engine, running.

We're keeping the mechanism a trade secret and letting the working system speak for it. GeoRadX reads chest X-rays on a phone, offline, calibrated to the population in front of it — that is DRF-MWU, deployed. See it do the thing.