Enthropy Systems

Invariant Limits of Complex Systems

Independent research into structural constraints on global consistency, degradation, and cyclic behavior in large-scale systems.

Quadrant plot: mean pairwise inconsistency is statistically identical between fault and null while the frustration radius separates the populations.
Paper 1. Mean pairwise inconsistency is statistically identical between fault and null (x-axis) — the frustration radius (y-axis) is what separates them. This is the certificate pairwise monitoring cannot see.

What this is

Enthropy Systems is an independent research effort focused on identifying invariant constraints that govern the behavior of large-scale, heterogeneous systems.

Current focus: distribution-free certificates for heterogeneous sensor and calibration systems, the operator-algebraic rigidity that explains why they hold, and their formal (Lean 4) verification — backed by reproducible numerics on data from two fusion machines (MAST tokamak, LHD stellarator).

Papers

Six preprints, one program: structural invariants under lossy translation. The applied end is a calibration certificate that catches faults pairwise monitoring provably cannot see; the theoretical end is the operator-algebraic obstruction that survives information loss; the method end is formal proof used as an active debugging instrument. Preprints, July 2026.

Frustration-radius quadrant plot separating faulted from null sensor networks.
Paper 1 · Applied · MAST + LHD validation

Frustration Certificates for Independently Calibrated Sensor Systems

A distribution-free certificate for calibration inconsistency, validated on two fusion machines under one untuned pipeline. Key result: a drift fault that leaves every pairwise residual statistically clean (0.218 vs 0.223 null) is detected at AUC 0.976–1.000 and localized to the faulted instrument.

In plain terms: every pair of sensors can agree while the network as a whole is inconsistent. Cycle frustration measures exactly the part pairwise checks cannot see — and comes with a proof of when it certifies a fault.

Torus flux plot: operator distance plateaus while defect and Frobenius distance decay.
Paper 2 · Theory · Reproducible numerics

Stability of Almost-Flat Lossy Cocycles: an Index Obstruction Survives Rank Loss

Stability of almost-representations extended to non-invertible targets. Key result: on the torus, a topological index keeps the system at fixed operator distance from consistency even as every local defect vanishes — through rank loss.

In plain terms: when translations between components destroy information, some global defects still cannot be smoothed away locally. The obstruction is topological, and it survives the information loss.

Crossover plot: local signature falls through the detection threshold while the removal floor stays flat.
Paper 3 · Synthesis · Detection vs removal

Undetectable but Unremovable: Capacity versus Topology in Lossy Networks

A topological charge whose local signature falls below every detection threshold while its removal cost stays pinned at a proved floor. Key result: a quantitative crossover scale m* separating the visible regime from the stealth regime, with a falsification map for every load-bearing claim.

In plain terms: past a certain system size, a real global fault can hide below every local monitor's noise floor — yet no local repair can remove it. Detection capacity and topological protection are different budgets.

Paper 4 · Method · Lean 4, machine-checked

Formalized Mathematics as a Debugging Tool

Lean 4 formalization as an active research instrument, not an after-the-fact certificate. Key result: the proof assistant caught two errors in opposite directions — falsifying a plausible invariant by constructing the counterexample, then deleting a step earlier drafts treated as half the open problem.

In plain terms: the machine did not just check the mathematics; it corrected the research direction twice. The full formalization is public.

Trapped topological charge frustrating a loop on the torus so it cannot relax to zero.
Paper 5 · Theory · Spectral vs topological

Three Gaps, One Index

The m-uniform rigidity of a lossy magnetic pair is topological, not spectral. Key result: three independently measured gaps collapse onto one index obstruction, and the rigidity constant is resolved by codimension — exactly 1 at codimension ≥ 2.

In plain terms: three phenomena that look like separate stability effects are one and the same integer doing the work. Knowing which one lets you predict the constant.

Cross-era vs within-era frustration radius and certificate firing rates around the real MAST changepoint.
Paper 6 · Audit · Pre-registered, real archive fault

What a Quality Flag Does Not Encode

A pre-registered certificate audit of the documented MAST magnetics changepoint at shot 16493. Key result: the certificate stays silent on the real boundary (AUC 0.485) while the planted positive control fires (AUC 0.95) — so the archive's quality flag encodes a decision, not a measured inconsistency.

In plain terms: an honest negative, reported in full. Running the instrument against a real archive event shows precisely what data-quality flags do and do not mean — both directions of that answer are useful.

About Enthropy Systems

Enthropy Systems is an independent research imprint. It publishes work on structural invariants under lossy translation, outside any university or institutional program. Authorship belongs to the authors and is stated on each paper.

The current program: certificates for the faults that pairwise monitoring provably cannot see — stated distribution-free, explained by operator-algebraic rigidity, checked in Lean 4, and validated on live diagnostics from two fusion machines. Every formal claim in the program compiles, carries a citation, or declares its epistemic status.

The Lean formalization is public: github.com/GoodRoyal/LossyCocycles.

Working together

These methods transfer: detecting the calibration and consistency faults that pairwise monitoring misses — in fusion diagnostics, sensor networks, navigation and timing systems — verifying the composition of complex or neuro-symbolic systems, and reasoning about the invariant limits of large-scale infrastructure.

If your problem looks like ours — global consistency, silent degradation, or cyclic behavior in a system too large to check by hand —

Get in touch