Theory · Research architecture
Structure · Identity · Transformation · Constraints · Discovery
Five parts. Every experiment on this site belongs to one of them, declares which hypothesis it tests, and states what would falsify it.
Research architecture
- 01
Structure
What is the relational state, and what is it made of?
A modelled system is a set of objects and a relational state R holding a coupling R_ij for every pair. Effective separation is derived from R, never stored alongside it. If a quantity cannot be computed from R, it is not part of the structure.
- 02
Identity
What must stay the same for an object to remain itself?
Each tracked object carries an identity vector I. Modifying relations incident to that object applies strain to I. Identity deviation ΔI = ‖I′ − I‖₂ is the accounting for that strain, and every transformation is judged against a declared tolerance.
- 03
Transformation
Which changes of relational state are permitted, and at what cost?
A transformation is an ordered list of edits R → R′. Its cost is C_A = Σ|R′_ij − R_ij| and its efficiency is Γ_A = |ΔD| / C_A. Transformations are recorded step by step so any claimed result can be replayed exactly.
- 04
Constraints
What would make a result invalid?
Metric validity and numerical stability are audited and can genuinely fail. Causality, energy budget, geometric reconstruction and physical correspondence are reported NOT MODELED, because they are not modelled. A gate is never marked PASS by default.
- 05
Discovery
What does the model do that we did not tell it to do?
Bounded automated search proposes transformations without being told which relation to change. Failures, infeasible populations and unexpected candidates are first-class results and stay on the page.
Core definitions
Relational address
Two objects are close when their addresses make the derived separation small — not because they share a region of a background space. Changing an address means rewriting relations, not translating a point.
Relational neighbourhood
Neighbourhood similarity measures how strongly X belongs to a given group. In AT-0021 the Traveler's coordinates never change; only its membership in N(Origin) and N(Destination) does.
Transition operator
T is not a dynamical law and does not evolve anything in time. Today it is defined only through declared cost and constraint functions. Its structure — corridors, forbidden regions, invariants — is what AT-0017 onward is meant to probe.
Hypothesis records
Relational Distance Hypothesis
Adjacency Theory hypothesis- Claim
- Effective separation between two objects is a function of the relational state of the system rather than an independent primitive.
- How it is tested
- Reproduce known separations from R alone; show that distance responds to R in a stable, path-consistent way; find where the mapping breaks.
- What would falsify it
- A modelled system where two configurations share an identical relational state but demand different separations.
Identity Preservation Hypothesis
Adjacency Theory hypothesis- Claim
- An object's identity state can be held within tolerance while its relational state — including its neighbourhood — is substantially rewritten.
- How it is tested
- AT-0016 identity failure boundary; AT-0021 relocation under tolerance; strain accounting per transformation step.
- What would falsify it
- Every transformation that meaningfully changes N(X) also drives ΔI beyond any declared tolerance.
Discrete Transition Hypothesis
Adjacency Theory hypothesis- Claim
- It is worth investigating whether a transition model necessarily requires every classical intermediate spatial state, or whether a relational description permits transitions that do not enumerate them.
- How it is tested
- AT-0023 continuity challenge; path-independence tests; comparison of transformation histories under different orderings.
- What would falsify it
- Every admissible transformation, under every ordering, is found to pass through the full intermediate chain.
Experiment registry · AT-0011 → AT-0023
| id | experiment | gate | status | purpose |
|---|---|---|---|---|
| AT-0011 | Adjacency Threshold Detection | AT-2 | implemented | Locate the relational state at which effective separation first satisfies D ≤ ε, and record the cost of reaching it.ε is a chosen number in a graph model. Crossing it means nothing physical. |
| AT-0012 | Adjacency Stability | AT-2 | prototype | Determine whether a state satisfying D ≤ ε survives bounded perturbation of the relational state, or collapses immediately.Stability of a simulated configuration is not stability of anything real. |
| AT-0013 | Reversal and Hysteresis | AT-2 | specified | Reverse an accepted transformation step by step and test whether the return path costs the same as the outbound path.Path dependence in a greedy search is a property of the search, until shown otherwise. |
| AT-0014 | Adjacency Debt | AT-2 | specified | Measure whether reducing D between one pair systematically increases separation elsewhere in the same relational state.Any conserved-looking quantity here is a bookkeeping artefact until derived, not assumed. |
| AT-0015 | Multiple Simultaneous Adjacencies | AT-2 | specified | Attempt to satisfy D ≤ ε for several pairs at once and find where the demands become mutually unsatisfiable.Infeasibility in a toy optimiser is a statement about the optimiser. |
| AT-0016 | Identity Failure Boundary | AT-2 | implemented | Push the transformation until ΔI exceeds tolerance, and characterise where identity preservation fails.Identity here is a stipulated vector. Its failure boundary is a property of that stipulation. |
| AT-0017 | Conservation Surface Search | AT-3 | specified | Search the space of transformations for any quantity that stays invariant across accepted moves.An invariant of a model is not a conservation law. Deriving one would require the model to mean something first. |
| AT-0018 | Path Independence | AT-2 | specified | Test whether different transformation orders reaching the same final R also incur the same adjacency cost.Order effects are combinatorics, not thermodynamics. |
| AT-0019 | Self-Organised Adjacency | AT-3 | specified | Let local update rules act on the relational state without a global objective, and see whether low-separation structure emerges unbidden.Emergence in a cellular rule set is common and cheap. It carries no explanatory weight by itself. |
| AT-0020 | Discovery Sweep | AT-2 | implemented | Run the bounded optimiser across seeds and collect the candidate transformations, the failures, and the unexpected results.A ranked candidate list is a ranked list of edits to numbers. |
| AT-0021 | Relational Relocation | AT-2 | implemented | Change a tracked object's relational neighbourhood from Origin to Destination while preserving its identity vector. The object never moves.MATHEMATICAL / COMPUTATIONAL MODEL. PHYSICAL EVIDENCE: NONE. This is not relocation of anything. |
| AT-0022 | Minimum Relocation Cost | AT-2 | implemented | Grid-search the cheapest relational transformation that reaches the destination-similarity target inside the identity tolerance.A minimum over a chosen grid, under a chosen cost function. |
| AT-0023 | Continuity Challenge | AT-2 | implemented | Inspect the accepted transformation history and report whether the modelled transition passed through intermediate relational states.Absence of intermediate states in a graph model is not evidence that a physical transition can be discontinuous. |
| AT-0024 | Spectral Connectivity Response | AT-3 | implemented | Measure how algebraic connectivity λ₂ of the weighted Laplacian responds to bounded single-relation changes.λ₂ is a whole-graph observable, not a separation between two objects. |
| AT-0025 | Topology Robustness Sweep | AT-3 | implemented | Run one perturbation protocol across random, small-world, scale-free and lattice families to see what survives a change of topology.Robustness across toy families is not robustness across physical systems. |
| AT-0026 | Observable Competition | AT-3 | implemented | Rank geodesic, resistance, efficiency, λ₂ and spectral radius by measured response and noise robustness for one identical transformation.A high rank makes an observable responsive, not real. |
| AT-0027 | Scale Sweep | AT-3 | implemented | Repeat the probe protocol across bounded graph sizes and fit a scaling trend only when the log–log diagnostic supports one.Five bounded sizes cannot distinguish a power law from a logarithm. |
| AT-0028 | Noise and Adversarial Stability | AT-3 | implemented | Add bounded noise and randomised control changes, then check whether candidates persist, disappear or reverse sign.Survival is measured inside one generator and one observable. |
| AT-0029 | Invariant Search | AT-3 | implemented | Look for quantities that stay fixed under accepted transformations, separating invariants imposed by construction from invariants actually discovered.No invariant here is a conserved physical quantity; energy and causality stay NOT MODELED. |
| AT-0030 | Missing-Rule Search | AT-3 | implemented | Score candidate response laws O = F(R, I, T, Q) on response, cost, identity deviation, invariance and stability minus complexity, then attack the winner with seven countermodels.A promoted candidate is a candidate response law inside a toy model. It is not new physics. |