% =====================================================================
%  First-Principles Validation and Extension of the
%  S--I--C--T Category-Theoretic Reconstruction:
%  Functorial Stratification, Error Budgets, and the Limits of
%  Thermodynamic Obstructions to Ontological Collapse
%
%  Revised Technical Edition: Finite Factors and Reproducible Benchmarks
%  Roth Complexity Lab (rothcomplexity.org) -- 21 October 2026
%
%  LaTeX source reconstructed from the supplied revised PDF.
%  Compile with:  pdflatex paper_revised && bibtex paper_revised && pdflatex paper_revised && pdflatex paper_revised
%  (No external .bib is required; references are inlined with thebibliography.)
% =====================================================================
\documentclass[11pt,a4paper]{article}

\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{amsmath,amssymb,amsthm}
\usepackage{geometry}
\usepackage{booktabs}
\usepackage{array}
\usepackage{hyperref}
\usepackage{microtype}

\geometry{margin=2.4cm}

\newtheorem{theorem}{Theorem}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{finding}[theorem]{Finding}

\newcommand{\EST}{\textsc{Established}}
\newcommand{\GRD}{\textsc{Grounded}}
\newcommand{\HEU}{\textsc{Heuristic}}
\newcommand{\SPEC}{\textsc{Speculative}}

\title{First-Principles Validation and Extension of the \\
S--I--C--T Category-Theoretic Reconstruction: \\
Functorial Stratification, Error Budgets, and the Limits of \\
Thermodynamic Obstructions to Ontological Collapse \\[6pt]
\large Revised Technical Edition: Finite Factors and Reproducible Benchmarks}
\author{Autonomous First-Principles Science Panel (source audit) \\
Roth Complexity Lab (rothcomplexity.org) \\[4pt]
\small Revised extension prepared with AI-assisted numerical and proof review; \\
not externally peer-reviewed}
\date{21 October 2026}

\begin{document}
\maketitle

\begin{abstract}
Report A audited R\'oth's category-theoretic reconstruction of the S--I--C--T framework (Structure, Information, Cohesion, Transformation) and extended it with a forgetful-functor hierarchy, a quantisation-and-truncation error bound, a computability obstruction and a Landauer thermodynamic obstruction. This paper is a second-order audit: we re-derive every formal claim of Report A from stated axioms, execute its three benchmark protocols numerically, and classify each claim with the Lab's four-tier epistemic badge system. We confirm the mathematical core: the category axioms, isomorphism criterion, compact-embedding theorem, finite-memory obstruction, forgetful functoriality and the total error bound $E_n \le L^n \delta_0 + \varepsilon \sum_{k<n} L^k$ are all sound, and the error bound is shown to be near-sharp for explicit Euler on the harmonic oscillator once the one-step consistency constant is evaluated on reachable states. We correct three substantive overstatements concerning the Landauer obstruction, the computability obstruction, and the Rule-90 benchmark. The executed benchmarks add a floating-point non-injectivity observation for RK4 Lorenz-63 and a scheme-dependent rounding-drift measurement. A new exact finite-factor theorem and a corrected six-bit benchmark strengthen the mathematical boundary. The results do not demonstrate that physical reality is a simulation.
\end{abstract}

\paragraph{Keywords.} S--I--C--T; category theory; topological conjugacy; forgetful functors; approximate simulation; shadowing; Landauer principle; Bennett reversibility; structural realism; epistemic badges.

\section{Scope, method and epistemic badges}

\textbf{What was audited.} The Panel received Report A only. R\'oth's underlying manuscript was not supplied, so R\'oth's theorems are validated as restated in Report A. Where Report A's rendering is internally sufficient to reconstruct a proof we do so; where it is not, we say so.

\textbf{Epistemic badges.} Each claim carries one of four badges: \EST\ (proved from stated axioms or standard theorems); \GRD\ (proved under an explicitly named modelling assumption, or verified numerically); \HEU\ (a widely used rule of thumb with known exceptions); \SPEC\ (a conjecture or interpretation not supported by proof or measurement).

\section{Part I --- Clause-by-clause audit of Report A}

The principal category and error-propagation arguments survive under explicit assumptions. Proposition~3 (SICT$_{M,J}$ is a category), Theorem~4 (iso $\Leftrightarrow$ intertwining homeomorphism), Theorem~5 (compact embedding), Proposition~7 (finite memory), Corollary~8 (observational invariance without material identity), and Theorem~12/Corollary~13 (Lipschitz error propagation) are all \EST. Three substantive overstatements are corrected in Part~II: the Landauer obstruction prose, the computability obstruction's physical scope, and the Rule-90 benchmark's applicability.

\section{Part II --- Validated and corrected formal theory}

Throughout, $(M,\cdot,e)$ is a topological monoid ($\mathbb{N}_0$ or $\mathbb{R}_{\ge 0}$), $J$ an index set, and $\{Q_j\}_{j\in J}$ Hausdorff spaces of observable values.

\subsection{Objects and the two categories}

\begin{definition}[Enriched S--I--C--T system]
An enriched system is a tuple $U = (X, V, \Omega, A, (q_j)_{j\in J}, T, \mathcal{B}_A, \mu)$ such that: (S) $X \neq \emptyset$ and $V$ are topological spaces and $\Omega \subseteq V^X$ carries a specified topology; (C) $A = \bigcap_{k\in K} c_k^{-1}(1) \subseteq \Omega$ for a family of (not necessarily continuous) predicates $c_k : \Omega \to \{0,1\}$, $A \neq \emptyset$, and each $q_j : A \to Q_j$ is continuous; (I) $\mathcal{B}_A$ is the Borel $\sigma$-algebra of $A$ and $\mu$ a Borel probability measure; (T) $T : M \times A \to A$ is a jointly continuous monoid action with $q_j \circ T_m = q_j$ and $(T_m)_* \mu = \mu$.
\end{definition}

\begin{remark}
The continuity caveat on $c_k$ is not pedantry: a continuous map into the discrete space $\{0,1\}$ from a connected $\Omega$ is constant. Admissibility predicates are Borel, not continuous; continuity is reserved for the observables $q_j$.
\end{remark}

\begin{definition}[The categories SICT$_{M,J}$ and SICT$^{\mathrm{full}}_{M,J}$]
In SICT$_{M,J}$ (R\'oth) an object is $U = (X,V,\Omega,A,(q_j),T)$; a morphism $h : U \to W$ is a continuous map $h : A_U \to A_W$ with $h \circ T_{U,m} = T_{W,m} \circ h$ and $q^W_j \circ h = q^U_j$. In SICT$^{\mathrm{full}}_{M,J}$ objects are enriched systems and a morphism $\Phi = (\varphi_X, \varphi_V, h)$ consists of a homeomorphism $\varphi_X$, a continuous $\varphi_V$, and a continuous $h$ preserving the field action, the measure, the dynamics and the observables.
\end{definition}

\subsection{The forgetful functor and substrate non-uniqueness}

\begin{proposition}
Both SICT$_{M,J}$ and SICT$^{\mathrm{full}}_{M,J}$ are categories, and $U : \text{SICT}^{\mathrm{full}}_{M,J} \to \text{SICT}_{M,J}$, $U \mapsto (X,V,\Omega,A,(q_j),T)$, is a functor that is neither full nor faithful and is not injective on isomorphism classes. \EST
\end{proposition}

\begin{corollary}
An isomorphism in SICT$_{M,J}$ is an isomorphism of images under $U$. Since $U$ does not reflect isomorphisms, no conclusion about spatial carriers, ensembles, or substrate follows. \EST
\end{corollary}

\subsection{Exact equivalence}

\begin{theorem}[Isomorphism criterion and compact embedding]
\EST\ (1) $U_P \cong U_S$ in SICT$_{M,J}$ iff there is a homeomorphism $h : A_P \to A_S$ with $h \circ T_{P,m} = T_{S,m} \circ h$ and $q^S_j \circ h = q^P_j$ for all $m,j$. (2) If $A_P$ is compact, $A_S$ Hausdorff and $h : U_P \to U_S$ an injective morphism, then $Z = h(A_P)$ is compact, closed and $T_S$-forward-invariant, and $h : U_P \to U_S|_Z$ is an isomorphism.
\end{theorem}

\subsection{The error budget theorem}

\begin{theorem}[Total simulation error]
\EST\ Let $(A_P,d)$ be a metric space, $F : A_P \to A_P$ with $d(Fx,Fy) \le L\,d(x,y)$, and $E,G,D$ the encoder, update and decoder. Fix $x_0 \in A_P$, $z_0 = E(x_0)$, and let $R = \{G^k z_0 : k \ge 0\}$. Assume (Q) $d(D(E(x_0)),x_0) \le \delta_0$ and (C) $d(D(Gz), F(Dz)) \le \varepsilon$ for all $z \in R$. Then
\[
E_n := d(D(G^n z_0), F^n x_0) \le L^n \delta_0 + \varepsilon\, S_n(L), \qquad
S_n(L) = \sum_{k=0}^{n-1} L^k = \begin{cases} \frac{L^n - 1}{L - 1}, & L \neq 1, \\ n, & L = 1. \end{cases}
\]
\end{theorem}

\begin{remark}[Three regimes]
\EST\ $L < 1$: uniform bound $\delta_0 + \varepsilon/(1-L)$. $L = 1$: linear growth $n\varepsilon + \delta_0$. $L > 1$: exponential growth with horizon $n_{\mathrm{crit}} \approx \ln(1/\max(\delta_0, \varepsilon(L-1)^{-1}))/\ln L$. The hypothesis ``for all $z \in R$'' is load-bearing.
\end{remark}

\subsection{The cardinality and computability obstruction}

\begin{theorem}
\GRD\ Assume the continuum model: $A_P = [0,1]^d$ with Lebesgue measure $\lambda$. Let $A_S$ be any set of finite binary strings, or the set of Turing-computable sequences. Then no bijection $A_P \to A_S$ exists, and for any injective encoding $E : A^{\mathrm{rep}} \to A^{\mathrm{comp}}_S$ on $A^{\mathrm{rep}} \subseteq A_P$, $\lambda(A_P \setminus A^{\mathrm{rep}}) = 1$.
\end{theorem}

\begin{remark}
The badge is \GRD, not \EST, because the proof is elementary but the premise---that the physical state space is a continuum---is a property of the model $U_P$, not an axiom of physics. To promote it to a statement about the physical target is precisely the Tier~2 $\to$ Tier~4 inference R\'oth's programme forbids.
\end{remark}

\subsection{The erasure budget theorem (corrected Landauer obstruction)}

\begin{definition}[Logical erasure]
For a finite $A_S$, a map $G : A_S \to A_S$ and a distribution $\mu$ on $A_S$,
\[
\Delta H_{\mathrm{erase}}(\mu,G) := H(\mu) - H(G_* \mu) = H_\mu(Z \mid G(Z)) \quad \text{(bits)}.
\]
\end{definition}

\begin{theorem}[Erasure budget]
\EST\ Let $G : A_S \to A_S$ with $|A_S| \le 2^b$, $\mu_0$ a distribution, $\mu_{n+1} = G_* \mu_n$. Then: (1) $\Delta H_{\mathrm{erase}}(\mu_n,G) \ge 0$, with equality iff $G$ is injective on $\operatorname{supp} \mu_n$. (2) \emph{(Telescoping)} $\sum_{n=0}^{N-1} \Delta H_{\mathrm{erase}}(\mu_n,G) = H(\mu_0) - H(\mu_N) \le H(\mu_0) \le b$. (3) $\Delta H_{\mathrm{erase}}(\mu_n,G) \to 0$; it is eventually $0$. (4) If $\mu_0 = \delta_{z_0}$ then $\Delta H_{\mathrm{erase}}(\mu_n,G) = 0$ for all $n$. (5) Under an isothermal implementation with equal logical-state free energies and no ancillary record, $Q_{\mathrm{env}} \ge k_B T_{\mathrm{env}} \ln 2\, [H(\mu_0) - H(\mu_N)]$; this does not upper-bound actual heat. (6) If $T_{P,1}$ is a bijection and $G$ is not injective on $\operatorname{supp} \mu_n$ for some $n$, then $U_P \not\cong U_S$ in SICT$_{M,J}$.
\end{theorem}

\begin{remark}
Report~A's prose---``every simulation step must dissipate a strictly positive minimum heat''---does not follow, by items (2)--(4). Real hardware dissipation is implementation-level; by Bennett (1973), any computation of $G$ can be embedded in a logically reversible computation at the price of retaining auxiliary garbage bits, introducing a dissipation--memory trade-off rather than a universal per-step heating law.
\end{remark}

\section{Part III --- Executed benchmarks}

\begin{finding}
\GRD\ Shift conjugacy is verified exhaustively for $N = 6,7,8$. Rule~90 on a ring is 4-to-1 for even $N$ and 2-to-1 for odd $N$. The erasure sequence confirms Theorem~3.13(2)--(3) exactly.
\end{finding}

\begin{finding}[Sharpness of the reachable-set hypothesis]
\GRD\ With $\varepsilon$ evaluated on the unit energy shell, the bound $n\varepsilon$ is violated by the measured Euler trajectory errors; replacing it by $\sum_{k<n} \varepsilon(G^k z_0)$ gives equality to the displayed digits. A Cohesion failure (energy drift) is what invalidates the Transformation bound.
\end{finding}

\begin{finding}[Observed floating-point non-injectivity]
\GRD\ At $u_{\mathrm{ref}} = (-4.4085, -3.3242, 24.1048)$, displacing $x$ by 1, 2 or 3 ulp and integrating both trajectories with identical arithmetic, the displaced state becomes bit-identical to the reference after 8, 14 and 34 RK4 steps; 4-ulp and 8-ulp displacements collapse onto one trajectory after 191 steps. This is an implementation-specific observation supporting $U_S \not\cong U_P$ under an injective target flow map.
\end{finding}

\section{The equivalence taxonomy, amended}

\begin{itemize}
\item Tier~1: Approximate observational simulation (one-sided); $\sup_{n \le N} d_Q(q(DG^n E x_0), q(F^n x_0)) \le \eta$.
\item Tier~$1\frac{1}{2}$: Shadowing for every pseudo-orbit there is a true orbit with $\sup_n d \le \eta$.
\item Tier~2: Reduced isomorphism in SICT$_{M,J}$ (intertwining homeomorphism $h$).
\item Tier~3: Full isomorphism in SICT$^{\mathrm{full}}_{M,J}$.
\item Tier~4: Physical identity $W_P = W_S$ (Leibniz's law over all physical predicates).
\end{itemize}
The hierarchy is well-ordered ($4 \Rightarrow 3 \Rightarrow 2 \Rightarrow 1\frac{1}{2} \Rightarrow 1$). The ``ontological collapse'' is the inference $2 \Rightarrow 4$ or $3 \Rightarrow 4$, which is not derivable.

\section{Part IV --- New results: finite factors, reversible targets and falsification}

\begin{theorem}[Finite-factor permutation theorem]
\EST\ Let $A$ be any nonempty set, $F : A \to A$ surjective, and $Z$ a finite nonempty set. If $h : A \to Z$ is surjective and $G : Z \to Z$ satisfies $h \circ F = G \circ h$, then $G$ is a permutation of $Z$.
\end{theorem}

\begin{proof}
Take any $z \in Z$ and choose $a \in A$ with $h(a) = z$. By surjectivity of $F$, there exists $a' \in A$ with $F(a') = a$. Then $G(h(a')) = h(F(a')) = h(a) = z$. Thus $G$ is onto. A surjection of a finite set to itself is injective and hence bijective.
\end{proof}

\begin{corollary}
\EST\ If $F$ is invertible and a finite-state $G$ merges distinct states, there cannot exist any surjective exact factor map $h$ intertwining $F$ and $G$. This is independent of entropy, cardinality, temperature and implementation technology.
\end{corollary}

\begin{finding}[Thermodynamic benchmark repaired]
\GRD\ The replacement six-bit benchmark (reversible cyclic shift target, lossy bit-clearing simulator) tests exactly the reversible-target/irreversible-simulator asymmetry. Its total logical erasure is six bits and stops after six steps. The six-bit bound is a ceiling on state-level entropy loss, not on physical device heat.
\end{finding}

\section{Verdict and recommendations}

\textbf{Verdict on Report A.} Mathematical core: sound. The functorial stratification and the error budget are correct and useful (\EST). The computability obstruction is correct as mathematics and must be badged \GRD\ as conditional on the continuum model. The Landauer obstruction is correct as a theorem and overclaimed as physics; corrected to an erasure budget and a dissipation--memory trade-off. One benchmark misapplies a theorem; two benchmark heuristics are badged \HEU. No claim in Report~A is \SPEC; no claim required rejection outright.

\textbf{Verdict on R\'oth (as restated).} Every restated theorem is \EST. The reconstruction achieves what it claims: it blocks the inference from categorical isomorphism to ontological identity by making explicit which structure the morphisms carry.

\begin{thebibliography}{99}
\bibitem{roth2026} R\'oth, M. (2026). \emph{Structural Equivalence Without Ontological Collapse: A Category-Theoretic Reconstruction of S--I--C--T Systems, Simulations, and Physical Dynamics.} Roth Complexity Lab Working Paper, 21 October 2026.
\bibitem{reportA} Autonomous Truth-Seeking Science Panel (2026). \emph{First-Principles Validation Report \& Extended Academic Paper on~[1]} (``Report~A''), 21 October 2026.
\bibitem{maclane} Mac Lane, S. (1998). \emph{Categories for the Working Mathematician}, 2nd ed. Springer.
\bibitem{ahrens} Ahrens, B., North, P. (2019). Univalent foundations and the equivalence principle. In \emph{Reflections on the Foundations of Mathematics}, Springer.
\bibitem{walters} Walters, P. (1982). \emph{An Introduction to Ergodic Theory.} GTM 79, Springer.
\bibitem{lind} Lind, D., Marcus, B. (1995). \emph{An Introduction to Symbolic Dynamics and Coding.} Cambridge University Press.
\bibitem{rutten} Rutten, J.J.M.M. (2000). Universal coalgebra: a theory of systems. \emph{Theoret. Comput. Sci.} 249(1), 3--80.
\bibitem{milner} Milner, R. (1989). \emph{Communication and Concurrency.} Prentice Hall.
\bibitem{girard} Girard, A., Pappas, G.J. (2007). Approximate bisimulation relations for constrained linear systems. \emph{Automatica} 43(8), 1307--1317.
\bibitem{lax} Lax, P.D., Richtmyer, R.D. (1956). Survey of the stability of linear finite difference equations. \emph{Comm. Pure Appl. Math.} 9(2), 267--293.
\bibitem{hairer} Hairer, E., Lubich, C., Wanner, G. (2006). \emph{Geometric Numerical Integration}, 2nd ed. Springer.
\bibitem{shannon} Shannon, C.E. (1948). A mathematical theory of communication. \emph{Bell Syst. Tech. J.} 27, 379--423, 623--656.
\bibitem{landauer} Landauer, R. (1961). Irreversibility and heat generation in the computing process. \emph{IBM J. Res. Dev.} 5(3), 183--191.
\bibitem{bennett73} Bennett, C.H. (1973). Logical reversibility of computation. \emph{IBM J. Res. Dev.} 17(6), 525--532.
\bibitem{bennett82} Bennett, C.H. (1982). The thermodynamics of computation---a review. \emph{Int. J. Theor. Phys.} 21, 905--940.
\bibitem{bowen} Bowen, R. (1975). \emph{Equilibrium States and the Ergodic Theory of Anosov Diffeomorphisms.} LNM 470, Springer.
\bibitem{hyg} Hammel, S.M., Yorke, J.A., Grebogi, C. (1987). Do numerical orbits of chaotic dynamical processes represent true orbits? \emph{J. Complexity} 3, 136--145.
\bibitem{sauer} Sauer, T., Grebogi, C., Yorke, J.A. (1997). How long do numerical chaotic solutions remain valid? \emph{Phys. Rev. Lett.} 79, 59--62.
\bibitem{pilyugin} Pilyugin, S.Yu. (1999). \emph{Shadowing in Dynamical Systems.} LNM 1706, Springer.
\bibitem{tucker} Tucker, W. (2002). A rigorous ODE solver and Smale's 14th problem. \emph{Found. Comput. Math.} 2, 53--117.
\bibitem{brouwer} Brouwer, D. (1937). On the accumulation of errors in numerical integration. \emph{Astron. J.} 46, 149--153.
\bibitem{bekenstein} Bekenstein, J.D. (1981). Universal upper bound on the entropy-to-energy ratio for bounded systems. \emph{Phys. Rev. D} 23, 287--298.
\bibitem{chalmers} Chalmers, D.J. (2003/2005). The Matrix as metaphysics. In C.~Grau (ed.), \emph{Philosophers Explore the Matrix}, Oxford University Press.
\bibitem{frigg} Frigg, R., Nguyen, J. Scientific representation. \emph{Stanford Encyclopedia of Philosophy.}
\bibitem{ladyman} Ladyman, J. (1998). What is structural realism? \emph{Stud. Hist. Phil. Sci.} 29, 409--424.
\bibitem{newman} Newman, M.H.A. (1928). Mr.\ Russell's ``Causal Theory of Perception''. \emph{Mind} 37, 137--148.
\end{thebibliography}

\appendix
\section{Reproduction parameters}

\textbf{Added extension.} Script \texttt{extension\_benchmarks.py} requires Python 3.10+ and NumPy; its deterministic assertions verify the reversible target, the irreversible lossy map, the six-step entropy budget, the finite factor and the reachable-orbit Euler ratio. Results are in \texttt{extension\_results.json}.

\textbf{Environment.} Python 3, NumPy 2.5.3, IEEE-754 binary64 (binary32 where stated). Script: \texttt{benchmarks.py}; runtime $\approx 15$\,s.

\textbf{Benchmark 1.} Rings $N \in \{6,7,8\}$; all $2^N$ states enumerated. Shift $T(z)_i = z_{i+1 \bmod N}$; encoding $h(\sigma) = (\sigma+1)/2$. Rule~90: $z'_i = z_{i-1} \oplus z_{i+1}$ (periodic). Erasure: uniform $\mu_0$, $\mu_{n+1} = G_* \mu_n$, $H$ in bits, 12 steps.

\textbf{Benchmark 2.} $\Delta t = 0.01$, $(x_0,p_0) = (1,0)$, $E_0 = 0.5$. Explicit Euler $(x + \Delta t\, p,\; p - \Delta t\, x)$; implicit Euler $(1+\Delta t^2)^{-1}(x + \Delta t\, p,\; p - \Delta t\, x)$; St\"ormer--Verlet kick--drift--kick; exact rotation with $\cos\Delta t, \sin\Delta t$ in binary64.

\textbf{Benchmark 3.} Lorenz-63 $(10, 28, 8/3)$; RK4 with $h = 10^{-3}$; transient $t = 50$ from $(1,1,1)$ to $u_{\mathrm{ref}} = (-4.40848, -3.32421, 24.10480)$. Jacobian spectral norm sampled every 10 steps over $t \in [0,100]$: max $= 30.30$, mean $18.27$. Divergence: perturb $x$ by $\delta_0$, integrate to $t = 200$ or until $\|a-b\| \ge 1$. Ulp test: displacements of $k \cdot \mathrm{ulp}(x_{\mathrm{ref}})$, $\mathrm{ulp} = 8.88 \times 10^{-16}$; bit-identity checked at every step for $10^4$ steps.

\paragraph{Reporting discipline.} Every number in the tables and findings is copied from the script's output without rounding beyond the displayed digits. Two Lorenz table rows at $10^{-13}$ and $10^{-14}$ have been removed because they are absent from the attached executable script.

\end{document}