Diffusion Ordering Surfaces a Dominant Transport Direction on a 7-Axis Behavioral Primitive Basis

A first-instance test of MANIFOLD_DYNAMICS_v1 on atomic chemistry.

First published 2026-05-25 · First-instance test

Summary

Under a seven-axis behavioral primitive basis (ATOMIC_STRUCTURE_v1), diffusion-based manifold analysis of the basis-projected element set surfaces a dominant transport direction whose ordering correlates with the metallic↔non-metallic chemistry continuum. Under adversarial testing per BASIS_VECTOR_VALIDATION_PROTOCOL_v1 §3.5, two invariants survived and two sub-claims were honestly demoted. The visualization layer demonstrates a separate discipline: the renderer is a pure projection membrane consuming an immutable scene contract; the same contract + same seed produces a bit-identical PNG on re-render. This page describes both findings, the limits of what can be claimed, and what the visualization is and is not.

What the manifold surfaces

Starting from a Phase-9-characterized basis of seven semantic axes (I / T / E / S / C / R / P_s) applied to a seven-anchor element subset {H, He, Li, B, C, N, O}, we constructed a mutual-information graph over the element-vector neighborhoods and computed the diffusion-operator eigendecomposition. The first non-trivial eigenvector — the Fiedler vector of the resulting graph Laplacian — orders the seven anchors along a single axis. The same ordering correlates (Spearman ρ ≥ 0.99 across eleven chemistry-correlation tests in MANIFOLD_DYNAMICS_v1 §3.7; reproduced under §3.5.2 extended-iteration smoothing and §3.5.3 multi-element removal) with the metallic↔non-metallic property continuum across these seven elements: the metallic / metalloid end (Li, B, C) anchors one tail; the diatomic non-metals and noble gas (N, O, He) anchor the other; H bridges in between.

This is the L0 invariant of MANIFOLD_DYNAMICS_v1 and the most-tested + most-survived claim of the five-experiment first-instance sequence. The framing matters: we are not asserting that diffusion-on-this-basis “discovered chemistry.” We are reporting that a diffusion-based topology computation on a behavioral basis produced an ordering that correlates with one known chemistry continuum on a seven-element first-instance subset. The observation is empirical, the scope is narrow, and the falsification surfaces are explicit.

A second invariant (L6b) survived adversarial smoothing: the Fiedler vector itself does not smooth to a fixed point under 5,000 iterations across four kernels (arithmetic mean, Gaussian, Laplacian, Fiedler-weighted), with ρ = 0.998. This is a field-mechanics empirical signature — the manifold’s transport structure is preserved under iterated local averaging, the formal criterion for treating the basis as carrying topological information beyond simple parameter coordinates.

What adversarial testing did

Two sub-claims were demoted under the same testing pass. L6a (a stronger version of the smoothing claim about R values rather than the Fiedler vector) was demoted: under 5,000 iterations the recovery_fraction reached 0.9728, indicating a finite-iteration artifact masquerading as structural finding under the earlier 100-iteration test. L3-basin (the claim that Carbon was the centroid of a coherent attractor neighborhood {C, B, Si, P, As, Ge, Te, Se}) was demoted: when Carbon was removed and the manifold re-derived, the named “basin” members did not cluster together. Carbon’s status as a singleton attractor is preserved (three independent §3.7 tests; reproducibility under §3.5.3) — but only as a singleton, not as a centroid of a coherent collective.

The demotions are part of the work. They are the operational consequence of feedback_anti_confirmation_after_intuition_match (when a finding matches an intuition, demand more falsification, not less) and feedback_skewed_distribution_metric_ambiguity (when a sub-claim depends on a single metric over a skewed axis, decompose into a metric-set and accept split-claim outcomes rather than substituting a friendlier metric post-hoc). Honest application sharpened the primitive’s constitutional law from three claimed strong invariants to two empirically validated strong invariants.

The visualization

The static render at the head of this page is the output of a Babylon.js scene whose state is fully described by an immutable JSON scene contract. The renderer code makes zero decisions about state: positions, colors, and edges are read straight from the contract; no animation, no procedural effects, no generative interpretation. Same contract + same seed produces a bit-identical PNG across re-renders (verified: 0 differing pixels out of 1,024,000). When the visual axis convention was changed to put the dominant transport direction (Fiedler value) on screen-vertical, only the scene contract changed — the renderer was untouched. This is feedback_visualization_is_execution_governance applied: visualization is lawful state projection, not graphics improvisation.

What you see: seven labelled spheres at deterministic positions, colored by Fiedler value (blue at the metallic / metalloid end; red at the non-metallic / noble-gas end), connected by five edges from the induced mutual-information subgraph, partitioned by a horizontal plane at the Fiedler-sign boundary. The vertical ordering is the L0 finding made visible.

Methodology

The basis primitive (ATOMIC_STRUCTURE_v1) was constructed per BASIS_VECTOR_VALIDATION_PROTOCOL_v1 through eight phases of validation. The dynamics framework (MANIFOLD_DYNAMICS_v1, RATIFIED-FINAL-v1.0 on 2026-05-25) defines the ten-part executable-topology suite and the four-layer separation contract (Semantic / Latent / Topology / Singularity). The first-instance test is the five-experiment sequence §3.7 → §3.2 → §3.9 → §3.5. The visualization layer implements the scene-contract / renderer separation governed by feedback_visualization_is_execution_governance plus the discipline rules in the six doctrine memories cited below.

Honest limitations

This is a first-instance test on a seven-element subset of one domain (atomic chemistry). Cross-domain validation (Phase 10 of BASIS_VECTOR_VALIDATION_PROTOCOL_v1) is not yet done — MANIFOLD_DYNAMICS_v1 v1.0 explicitly defers it to v2.0+.
The seven-axis declaration is semantic; the underlying manifold collapses to approximately five latent dimensions under linear methods (Phase 9 result). The semantic-to-statistical relationship is captured by the four-layer separation contract, not by claiming axis-equals-dimension equivalence.
Two sub-claims were demoted under adversarial testing. Future amendments may demote others. The basis primitive remains amendable; published claims on this page are sized to remain defensible under amendment.
Temporal evolution (§3.4 of the dynamics framework) is deferred to v1.1+. The findings reported here describe the manifold’s static topology, not its evolution under driven dynamics.

What’s next

MANIFOLD_DYNAMICS_v1 v1.1+ work includes §3.4 temporal evolution M(t), a sub-basin probe for column-14 / column-15 elements to further narrow the C-INDIVIDUAL claim, §3.6 semantic↔latent translation machinery, and §3.10 executable-pipeline formalization. v2.0+ work is the Phase 10 cross-domain validation (apply the framework to non-chemistry domains). Until those phases land, this page does not extend its claims beyond the atomic-chemistry first-instance scope.