\section{Reproducibility, dependencies and scope}\label{sec:repro} \subsection{Computational dependencies} Companion version~1.1.0 supplies the article's sources, certificates and verifiers in the \href{https://github.com/aconsciousfractal/FCIG-Cyclotomic-Digit-Reversal-Thresholds}{project repository}. Run \texttt{python -B scripts/verify.py} from its root; repeat under \texttt{python -O -B} to check independence from assertions. All arithmetic is exact. Unit tests mutate certificates and exercise boundary cases. \path{MANIFEST_SHA256.txt} fixes the distributed files; hash agreement alone is not a mathematical check. The finite components have distinct roles: \begin{center}\small \begin{tabular}{c p{.33\linewidth} p{.46\linewidth}}\toprule ID&component&role in the argument\\\midrule F1&\path{period_recipe.json}&Redundant facet recipes for 15 and 30.\\ F2&\path{threshold_105.json}&Essential separators and polynomial lifts for order 105.\\ F3&\path{threshold_210.json}&Essential separators and polynomial lifts for order 210.\\ F4&\path{root_residue6.json}&Essential root-restricted comparison at order 210.\\ F5&\path{core/three_p.py}&Diagnostic replay of the deductive $3p$ proof and digit witnesses.\\ F6&\path{core/six_p/}&Regression checks of transition, boundary and multiplicity formulas.\\ F7&\path{core/six_p_uniqueness.py}&Regression checks of the bounded uniqueness classification.\\ F8&\path{core/scalar_transfer.py}&Diagnostic scalar comparisons, reversal and period-transfer witnesses.\\\bottomrule \end{tabular} \end{center} F2/F3 check the lower-bound separators and polynomial lifts specified in Section~\ref{sec:certified}. Their proofs also require the exhaustive shell reduction, not merely valid separators for selected candidates. F4 checks all 210 roots and the matching upper witness. Panel definitions, execution counts and toolchain information are in \path{docs/COMPUTATIONS.md} and \mbox{\path{REPRODUCTION_RECORD.md}}. For both $3p$ and $6p$, the prefix classification, shell reduction and noncritical bounds are proved in the text. The scalar $6p$ boundary proof does not import the F6/F7 certificates. Their replay is diagnostic, as are the finite F8 panels; none proves a conclusion for all primes or bases by sampling. F2--F4 retain their essential roles for the separately stated results. The witness audit reconstructs actual digit words, verifies the two evaluations modulo the complete cyclotomic value, and independently checks a nonzero polynomial remainder. A folded vector is not itself a digit word. Lemma~\ref{lem:linear-lift} supplies an $O(p)$ arithmetic construction of three-row potentials; explicit expansion into a word still takes time proportional to its length. \subsection{Scope and further questions} The general theorems hold for each fixed $t$. Their effective monoid construction is finite but is not advertised as a practical algorithm when $t$ varies. The companion does not implement that construction for arbitrary $t$; its fixed-order certificates use the smaller geometric reductions. The period $\rho_tL_t$ is sufficient, and need not be the least period. The exact families and the non-TU cases illustrate stronger period information. The first failure threshold, the first shell, the set of normalized algebraic multipliers, the folded vectors and the digit words are different objects. A classification of one does not automatically classify the others. In particular the multiplicity and uniqueness statements for $6p$ count normalized algebraic values; the uniqueness domain for $B\equiv2\pmod4$ starts at 22, even though its threshold formula already starts at 10. No uniqueness assertion is made there for bases 10, 14 or 18. The non-palindromic threshold corollary does apply at those bases: it needs one primitive value, not exhaustion of the minimum shell. Fixed-value counts under reversal do not count palindromic words; the word counts in Corollary~\ref{cor:nonpal-count} are lower bounds. Nor is a full minimum-fibre classification claimed at orders 105 and 210. Natural remaining questions include the smallest valid cutoffs, exact formulas for further prime products, and the relation between minor denominators and least coefficient periods. These are extensions of the same lattice--height--capacity problem, rather than prerequisites for the results proved here. The alphabet-height and collision interpretations likewise concern the same coefficient map. Passing from the full modulus to a proper divisor changes the problem: in base 4 the word \texttt{11} has value 5, a divisor of $\Phi_{10}(4)=205$, whereas the full-modulus threshold is 14. The eventual quasi-polynomial behavior and general witness selection have antecedents in Bogart--Goodrick--Woods~\cite{BGW}; the effective semilinear decomposition uses Chistikov--Haase~\cite{CH}. The cyclotomic lattice, matroid and total-unimodularity inputs are attributed in the relevant proofs. The contribution developed here is the explicit slope and threshold analysis with its geometric reductions, uniform formulas, and fully specified finite components. The literature comparison is targeted and does not certify priority. \paragraph{Use of computational and AI assistance.} OpenAI Codex and, through ChatGPT in a web browser, GPT-6 Pro and Astra contributed substantially to mathematical exploration, proof development and checking, literature comparison, software and exact computations, critical review, drafting and editing. The author remains responsible. The programs check the finite components described above; these checks are not formal verification or independent human peer review.