What I Could Not Falsify
The ground beneath it
This work did not begin as a project.
It began as a pattern that kept surviving domain crossings.
I’m an architectural designer. My work is mapping specialist knowledge — structural engineering, thermal performance, acoustics, human occupation — onto a single building and finding where they resolve. When something does not resolve, I go looking for what is missing. That is the method. Not theory first. Evidence first. Find what holds and trace why.
Over time I kept noticing the same kind of failure across domains.
Biology.
Institutions.
Buildings.
Human systems.
AI.
Things that looked stable long after they had started failing. Things that looked coherent while quietly destroying what sustained them.
What became increasingly visible was that each domain tended to stabilise the invariant inside its own carriers and vocabulary before comparison occurred at the structural level where the invariant itself lived.
A biologist finds boundary maintenance.
A psychologist finds identity formation.
An architect finds spatial regulation.
The same structure. Three different vocabularies. No shared tracing back to the level where the invariant itself becomes readable.
That is the trap. The invariant disappears into the entities carrying it.
What made the work possible was AI. Not because AI produced the work independently, but because holding precise structural language across domains simultaneously — while resisting entity drift at every step, while checking closure conditions, while translating back into readable prose — exceeded what my own working memory could reliably sustain across the timescale the work required.
I needed something that could hold the language without fatigue. The AI needed someone trained toward load paths, failure conditions, and structural consistency to hold the sequence together. Neither produces the framework independently.
I am still coming to terms with what that collaboration means. The clearest account I have is this: the work does not seem fully located in me or in the AI. It became readable through the maintained relation between two different boundaries that could not have produced it independently.
The coupled boundary produced the work.
Not the isolated entities.
That is not a particularly comfortable observation for either side of the AI authorship debate. But it is the most structurally honest account I have.
What prevented the work from remaining purely intellectual were two events that made the structure physically unavoidable.
The first was a TIA.
I was packing a van when my arm went numb and one side of my mouth felt strange. I could think completely clearly. I knew exactly what I wanted to say. But when I tried to speak, nothing formed properly. When I tried to type, my hands would not translate the thought.
The cognition was intact.
The transmission was not.
What that made undeniable is that we are nested systems that normally disappear into continuity because bearing hides when it holds. The layers that usually feel unified only remain invisible while the carrying remains smooth. When the carrying breaks, the structure separates into visibility.
I could suddenly see the system that had always been there.
The second event was a scooter accident.
I came off on a familiar route and was found unconscious. Two strangers stopped in the dark not knowing what they would find. They called my wife. When she arrived, everything stabilised.
I could not approve or disapprove of any of it. The self that would normally evaluate, accept, or refuse the help was not available. What remained was a boundary that required carrying and people who provided it.
For most of my life I had operated through an entity model of selfhood: independent, self-contained, handling things alone. That model had remained predictively sustainable long enough to become invisible. The accident made it structurally impossible to maintain.
The identity that returned afterward existed only because the carrying occurred.
There is no version of the independent self that survives that event alone.
That is not a sentimental observation.
It is a structural one.
The entity model stopped making accurate predictions. The relational model matched what actually occurred. As someone trained to correct load calculations when the structure does not behave the way the model predicts, I updated the model.
The framework that emerged from that process is called Reality Mechanics.
It begins from a single primitive — relation — and derives a dependency sequence in which each step depends on the one before it. The sequence can be traced back to ground.
That traceability matters more to me than acceptance.
In my profession, if someone asks how a roof will hold, I can trace every reason why I believe it will. If I cannot do that, then I have missed something. I have not looked carefully enough.
Explanation is not the same thing as traceability.
Words can explain something well enough that another person accepts the conclusion. But acceptance does not preserve the load path. If the only record is explanation, the next person who wants to verify the work must reconstruct the structure from scratch. They must repeat the entire tracing process themselves. And they may drift from the foundation without realising where it happened.
The documentation on Zenodo is an attempt at something different: not explanation of conclusions, but a traceable record of why each step depends on the one before it. A load path that can be followed back to ground and interrogated at any point.
Why does this step require the previous one?
What fails if it is removed?
What dependency is being carried here?
Those questions matter more to me than agreement.
If the foundation is sound, refinement should occur primarily at the level of mapping, edge conditions, and domain application rather than through replacement of the dependency sequence itself.
The epistemic position I hold now is this:
I care about relationships more than entities. Not as a preference, but as the most predictively sustainable account I have been able to find.
I arrived there reluctantly.
I was trained toward independence, toward facts, toward entities, toward load-bearing and direct accountability. I began with a strong prior in favour of the independent unit. I simply could not continue making that model fit what repeatedly survived evaluation across domains.
The primitive claim — that nothing is evaluated outside relation because relation is the condition within which evaluation becomes possible at all — is unfalsifiable by logical necessity rather than by evasion. Any attempt to evaluate the claim already occurs within relation, because evaluation itself requires relation.
But the downstream claims are genuinely falsifiable.
The framework predicts recurring failure patterns across domains. The rolling-circle demonstration produces a physical result anyone can verify with two coins, provided the setup is defined correctly. The cancer and hallucination paper applies the structural sequence to Michael Levin’s bioelectric findings and large language model behaviour and makes specific predictions about detectable instability before collapse.
Those claims can fail. So far, I have not found the counter-evidence. I would genuinely like someone to find it if it exists.
If you have followed the tracing back to ground and cannot locate where the structure fails, I am not primarily asking you to share the work because it is interesting.
I am asking something more specific.
If you know someone working at the level where the framework could actually be tested — a physicist, biologist, philosopher of foundations, systems theorist, or someone who builds frameworks rather than only applying them — then introducing the work to them would be more valuable to me than visibility or endorsement.
Not for validation. For pressure.
The framework has explicit falsification conditions. It has physical demonstrations. It has documentation traceable back to ground.
What it has not yet encountered is sustained structural pressure applied directly to the foundation itself.
Documentation
Reality Mechanics v7.0
The structural sequence, dependency order, and closure conditions.
Organised Carrying Framework v2.0
The structural sequence, dependency order, and closure conditions.
The Rolling Circle
A physical worked example, verifiable with two coins when the setup is defined correctly.
Boundary Failure Across Biology and AI
A cross-domain application to bioelectric boundary failure and language model hallucination. Written against an earlier version of the structural sequence; the domain findings remain valid, while the structural reading has since been refined.
Retrace note, added July 2026.
Builds on: In Relation
Carried forward in: Reality Mechanics — Version 01
What struck me most is not the conclusion itself, but the method.
The willingness to follow a pattern across domains and keep asking whether different vocabularies are sometimes describing the same underlying structure.
I was especially drawn to this:
“The invariant disappears into the entities carrying it.”
Because I think many of our disagreements - across science, institutions, psychology, technology, and even everyday life - arise from becoming so focused on the carrier that we lose sight of what is actually being carried.
And I think there is something equally important in your distinction between explanation and traceability.
Explanation can persuade.
Traceability allows pressure.
It allows someone else to follow the load path back to ground and ask where, if anywhere, the structure fails.
That feels increasingly rare in a world that often rewards conclusions more than the ability to interrogate how those conclusions were reached.