\documentclass[11pt,a4paper,reqno]{amsart} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage{amsmath,amssymb,amscd,mathtools,mathrsfs} \usepackage{booktabs,array,longtable,needspace} \usepackage{microtype} \usepackage[hidelinks]{hyperref} \hypersetup{pdftitle={A proof of the HLRV conjecture},pdfauthor={Qiyue Tang},pdfsubject={Mixed Hodge polynomials of generic semisimple character varieties}} \usepackage[margin=2.5cm]{geometry} \linespread{1.08} \allowdisplaybreaks[2] \raggedbottom \emergencystretch=1.5em \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{convention}[theorem]{Convention} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\Q}{\mathbf{Q}} \newcommand{\C}{\mathbf{C}} \newcommand{\Z}{\mathbf{Z}} \newcommand{\A}{\mathbf{A}} \newcommand{\Gm}{\mathbf{G}_m} \newcommand{\OO}{\mathcal{O}} \newcommand{\LL}{\mathbb{L}} \newcommand{\TT}{\mathbf{T}} \newcommand{\Th}{\vartheta} \newcommand{\La}{\Lambda} \newcommand{\IC}{\operatorname{IC}} \newcommand{\MHM}{\operatorname{MHM}} \newcommand{\MMHM}{\operatorname{MMHM}} \newcommand{\MHS}{\operatorname{MHS}} \newcommand{\MMHS}{\operatorname{MMHS}} \newcommand{\pH}{{}^{p}\!\mathcal{H}} \newcommand{\Gr}{\operatorname{Gr}} \newcommand{\Sym}{\operatorname{Sym}} \newcommand{\Spec}{\operatorname{Spec}} \newcommand{\Tot}{\operatorname{Tot}} \newcommand{\Tr}{\operatorname{Tr}} \newcommand{\RHom}{\operatorname{RHom}} \newcommand{\Hom}{\operatorname{Hom}} \newcommand{\Ext}{\operatorname{Ext}} \newcommand{\Cone}{\operatorname{Cone}} \newcommand{\fib}{\operatorname{fib}} \newcommand{\cofib}{\operatorname{cofib}} \newcommand{\Crit}{\operatorname{Crit}} \newcommand{\Perf}{\operatorname{Perf}} \newcommand{\Pic}{\operatorname{Pic}} \newcommand{\Bun}{\operatorname{Bun}} \newcommand{\Quot}{\operatorname{Quot}} \newcommand{\Hilb}{\operatorname{Hilb}} \newcommand{\Conf}{\operatorname{Conf}} \newcommand{\GL}{\operatorname{GL}} \newcommand{\rk}{\operatorname{rk}} \newcommand{\id}{\operatorname{id}} \newcommand{\supp}{\operatorname{supp}} \newcommand{\im}{\operatorname{im}} \newcommand{\coker}{\operatorname{coker}} \newcommand{\wt}{\operatorname{wt}} \newcommand{\sgn}{\operatorname{sgn}} \newcommand{\Log}{\operatorname{Log}} \newcommand{\Exp}{\operatorname{Exp}} \newcommand{\pt}{\mathrm{pt}} \newcommand{\red}{\mathrm{red}} \newcommand{\vir}{\mathrm{vir}} \renewcommand{\ss}{\mathrm{ss}} \newcommand{\sslash}{\mathbin{/\mkern-6mu/}} \newcommand{\reg}{\mathrm{reg}} \newcommand{\nil}{\mathrm{nil}} \newcommand{\DR}{\mathrm{DR}} \newcommand{\cS}{\mathscr{S}} \newcommand{\cI}{\mathscr{I}} \newcommand{\cJ}{\mathscr{J}} \newcommand{\cP}{\mathscr{P}} \newcommand{\cA}{\mathcal{A}} \newcommand{\cM}{\mathcal{M}} \newcommand{\eps}{\varepsilon} \newcommand{\into}{\hookrightarrow} \newcommand{\xto}[1]{\xrightarrow{#1}} \newcommand{\bbone}{\Q} \newcommand{\abs}[1]{\lvert #1\rvert} \newcommand{\norm}[1]{\lVert #1\rVert} \title[A proof of the HLRV conjecture]{A proof of the HLRV conjecture} \author{Qiyue Tang} \address{Zhili College, Tsinghua University, Beijing 100084, China} \email{tangqy24@mails.tsinghua.edu.cn} \date{November 14, 2026} \subjclass[2020]{14D20, 14C30, 14H60, 05E05} \keywords{Character variety, mixed Hodge polynomial, Higgs bundle, Macdonald polynomial} \begin{document} \begin{abstract} We prove the Hausel--Letellier--Rodriguez-Villegas conjecture on the mixed Hodge polynomials of generic semisimple $\GL_n$ character varieties. \end{abstract} \maketitle \tableofcontents \clearpage \section{Introduction} \label{intro:history} We prove the Hausel--Letellier--Rodriguez-Villegas formula for the compactly supported mixed Hodge polynomial of a generic semisimple $\GL_n$ character variety. Theorem~\ref{main:hlrv} treats every genus, rank, multiplicity profile, and generic determinant component. It identifies the weight and cohomological gradings with the two variables of the Macdonald generating function. The proof also gives the independence of the polynomial from the generic eigenvalues and the nonnegativity of its coefficients in the prescribed sign convention. \subsection{Character varieties and the mixed Hodge conjecture} The geometry underlying the problem begins with Hitchin's study of the self-duality equations and the associated algebraically completely integrable system \cite{HitchinSD,HitchinIS}. The moduli of Higgs bundles carries a proper map recording the characteristic polynomial of the Higgs field. Its general fibers are abelian varieties, and its singular fibers encode the topology of the total space. Nonabelian Hodge theory relates this space to a moduli space of local systems; Simpson developed both the correspondence and the algebraic moduli theory \cite{Simpson,bettiSimpson}. Prescribed local monodromy at punctures corresponds to parabolic data on the Higgs side. The two algebraic structures give different filtrations on the same underlying cohomology. The Betti moduli space has Deligne's mixed Hodge structure, whereas the Hitchin morphism gives a perverse filtration on Higgs-bundle cohomology. Even when the latter cohomology is pure, the weight filtration on the character variety contains additional information. Hausel and Thaddeus connected the Hodge theory of these moduli spaces with mirror symmetry and Langlands duality \cite{HT}. This supplied a geometric setting for comparisons between explicit invariants of different character varieties. Hausel and Rodriguez-Villegas formulated an explicit mixed Hodge polynomial conjecture for the character variety with primitive central twist \cite{HRV}. Their finite-field character calculation, together with Katz's point-counting theorem in the appendix to that paper, determined its $E$-polynomial. They verified the full mixed Hodge formula in rank two and formulated the curious hard Lefschetz symmetry suggested by the answer. The partition sum contains a product indexed by the boxes of a Young diagram; its factors involve the arm and leg lengths and the genus of the curve. Hausel, Letellier, and Rodriguez-Villegas extended this formula to punctured curves with generic semisimple local monodromies \cite{HLRV}. Each puncture contributes an independent alphabet and a modified Macdonald polynomial. Pairing the plethystic logarithm with the complete symmetric functions extracts the prescribed eigenvalue multiplicities. Their arithmetic harmonic analysis establishes the $E$-polynomial formula and a related quiver-variety calculation. Conjecture~1.2.1 of \cite{HLRV} predicts the full mixed Hodge polynomial, its eigenvalue independence, and the resulting positivity. Section~\ref{main:statement} fixes the genericity condition and all symmetric-function conventions used here. The subsequent work of Hausel, Letellier, and Rodriguez-Villegas proved connectedness of every nonempty generic character variety and identified a specialization of the kernel with the Kac polynomial of a comet-shaped quiver \cite{HLRVII}. Their cohomological interpretation of Kac polynomials and quiver Donaldson--Thomas invariants established nonnegative coefficients for those invariants \cite{HLRVKac}. The mixed Hodge conjecture connects this quiver calculation with the pure compactly supported part of the character variety. \subsection{Counting formulas and filtration comparisons} Several parts of the generating function have been established by different methods. Schiffmann counted indecomposable vector bundles and stable Higgs bundles over finite fields \cite{Schiffmann}. Mozgovoy and Schiffmann developed counts of semistable twisted Higgs bundles and type $A$ quiver bundles \cite{MozSchiff}. Mellit proved the polynomiality and integrality of the Hausel--Letellier--Villegas kernels \cite{MellitIntegral}. He then obtained the predicted Poincar\'e polynomials in the central-twist case \cite{MellitHiggs} and for generic semisimple punctured character varieties \cite{MellitPoincare}. In the punctured calculation, affine Springer fibers and local factorization explain the occurrence of modified Macdonald polynomials. The $E$-polynomial and Poincar\'e polynomial are two specializations of the mixed Hodge polynomial. The former takes an alternating sum over cohomological degrees; the latter sums over weights. The full formula requires the joint distribution of these two degrees. Mellit's survey \cite[Sections 4.8--4.9]{MellitSurvey} describes these specializations and records the two-variable formula as the remaining conjecture. The $P=W$ conjecture of de Cataldo, Hausel, and Migliorini gives a geometric interpretation of the weight filtration in terms of the Hitchin map. They proved it in rank two \cite{dCHM}. De Cataldo, Maulik, and Shen then established it in genus two for arbitrary rank and coprime degree, using degenerations from moduli of sheaves on abelian surfaces \cite{dCMS}. Maulik and Shen proved it for coprime $\GL_n$ Higgs moduli in every rank on curves of genus at least two \cite{MSPW}. Hausel, Mellit, Minets, and Schiffmann constructed a proof using surface Hecke operators and a deformation of the algebra of polynomial Hamiltonian vector fields \cite{HMMS}. Their operator construction applies in every genus; the parabolic geometric application in Section~8 of that work fixes genus greater than one and complete flags. The argument below verifies the geometric inputs for all genera and passes from complete to partial flags by regular Gysin maps and flag integration. Maulik, Shen, and Yin obtained another proof of coprime $\GL_n$ $P=W$ from Fourier transforms for compactified Jacobian fibrations \cite{MSY}. Tautological classes provide the connection between these methods. Markman's diagonal-class calculation gives generators for the cohomology of moduli spaces of sheaves on symplectic surfaces \cite{Markman}. Shende determined the weights of the universal character-variety classes \cite{Shende}. Their normalized Chern indices can therefore be compared with both perverse degree and Betti weight. Borri has also realized the twisted character variety as a component of the character variety of a root curve and used that realization to compute tautological weights \cite{Borri}. In this paper the finite-cover argument is performed on the full orbifold local-system stack and retains the isotropy summands at every puncture. Mellit's toric stratifications prove Hodge--Tate cohomology and curious hard Lefschetz for generic semisimple character varieties \cite[Theorems 1.5--1.6]{MellitToric}. His Grothendieck--Springer construction also permits extraction of arbitrary semisimple multiplicities from an auxiliary complete flag. We use the corresponding strata and localization maps in families over all the generic eigenvalue parameters. Su's work proves the weak geometric $P=W$ statement for very generic character varieties, where at least one local conjugacy class is regular semisimple, through their cell decompositions and dual boundary complexes \cite{Su}. These results describe complementary parts of the geometry at the Hitchin and Betti boundaries. \subsection{Stable pairs and the proof} The curve-counting approach has a prior formulation in the work of Chuang, Diaconescu, and Pan on BPS states and $P=W$ \cite{CDP}. Chuang, Diaconescu, Donagi, and Pantev developed its parabolic version using stable pairs on local root stacks \cite{CDDP}. They proposed the refined BPS identification with Hitchin-theoretic invariants and the local Macdonald factors needed for the HLRV formula. Carlsson and Rodriguez-Villegas studied related symmetric-function identities through vertex operators \cite{CRV}. The Hilbert-scheme calculation in \cite{CDDP} already proves a related $K$-theoretic Macdonald identity; the cohomological stable-pair comparison is the additional step used here. Monavari computed local stable-pair invariants using double nested Hilbert schemes \cite{MonavariNested} and developed the refined local Donaldson--Thomas theory of curves \cite{MonavariRefined}. We carry out the cohomological comparison and local calculation needed for this program. The comparison begins with two Calabi--Yau threefolds over an ordinary curve: one has a sufficiently positive line-bundle direction, and the other has the canonical and trivial directions. A quotient of the canonical-plus-trivial bundle produces the quadratic potential connecting them. We construct its oriented critical comparison, including families over nonreduced bases. Filtered extension correspondences give the stability-change maps, and the Hecke action is constructed on the whole comparison arrow. The support argument separates the one-packet contribution from products of smaller ranks. Its proof retains the finite permutation action needed for the symmetric-power formula. The other calculation uses the attracting loci of stable-pair moduli. Partitions describe the transverse monomial ideals, and reverse plane partitions record modifications along the curve. The deformation character determines each shift and Tate twist. Closed attracting filtrations and purity recover the resulting cohomology objects. Section~\ref{ordinary:construction} gives this ordinary-curve argument in full, including the central-twist mixed Hodge formula and the separation of its two gradings. For a root curve, the construction keeps rank, Euler characteristic, and every root-character multiplicity. The primitive contribution is obtained in the Euler characteristic corresponding to the Betti parameters. The local coefficient at a root marking is initially an independently defined symmetric function. Its two-pointed genus-zero pairing gives the Cauchy identity. A colored local calculation gives finite capacity, boundary invisibility, and dominance inequalities for every prefix. These identities and the value at the one-letter alphabet characterize the modified Macdonald polynomial. The final comparison uses normalized universal Chern and flag classes. The Hecke algebra identifies their index filtration with centered perversity, including the zero-residue and partial-flag specializations. At finite-order unitary monodromy, an orbifold realization identifies this filtration with the Betti weight filtration. Relative toric strata and Springer sign extraction give constancy in each cohomological degree and each weight over every generic determinant component. Each such component contains a suitable finite-order unitary point. The stable-pair identity is retained in the Grothendieck group of Hodge structures until both cohomological indices have been recovered. Purity makes the weights of the different compactly supported degrees distinct at a fixed perverse index. We can then extract those degrees separately and apply the hook substitution. This yields the full mixed Hodge formula of Theorem~\ref{main:hlrv}. The final section records the duality, signs, and shifts and derives positivity and the pure-part formula. \section{The character-variety formula} \label{main:statement} We prove the mixed Hodge formula of \cite[Conjecture 1.2.1]{HLRV}. We begin by fixing its geometric data and symmetric-function conventions. Fix integers $g\geq0$, $m\geq1$, and $n\geq1$. For each $p\in\{1,\ldots,m\}$ let $\mu^p=(\mu^p_1,\ldots,\mu^p_{\ell_p})$ be a partition of $n$. Choose pairwise distinct numbers $\xi_{p,1},\ldots,\xi_{p,\ell_p} \in\C^*$, and let $\mathcal C_p\subset\GL_n(\C)$ be the semisimple conjugacy class with these eigenvalues and multiplicities. The tuple is \emph{generic} if \begin{equation}\label{main:genericity} \prod_{p,a}\xi_{p,a}^{\mu^p_a}=1, \qquad \prod_{p,a}\xi_{p,a}^{\nu^p_a}\ne1 \end{equation} whenever $0\leq\nu^p_a\leq\mu^p_a$ and $\sum_a\nu^p_a=r$ for every $p$, with $0c$. The symbol $\cP_{r,\chi}(c)$ denotes the corresponding nonzero-section surface window. The potential is \begin{equation}\label{eq:quadratic-potential} g_q(E,\Theta)=\frac12 \langle\alpha_q,\Tr(\Theta^2)\rangle. \end{equation} It factors through a function $g_{\cA}$ on $\cA_r^L$. For a rank partition $\nu=(r_1,\ldots,r_s)$ of $r$, let $\Sigma_\nu\subset\cA_r^L$ be the image of spectral addition. There are two different Serre subcategories of the perverse heart $\MMHM(\cA_r^L)$ in the argument. The category $\cS_r$ consists of objects whose simple constituents have strict supports among the $\Sigma_\nu$. It includes the single rank-$r$ semistable unframed coefficient. The category $\cI_r$ is the Serre subcategory generated by the perverse constituents of the finite-addition convolutions of the framed and unframed coefficients occurring in the proper flag correspondences, with at least two positive-rank factors, all of rank less than $r$. In taking these generators we first perform the indicated shifts, Tate twists, and source vanishing-cycle operations, and then take perverse cohomology and its simple constituents. In particular $\cI_1=0$. For either category $\cJ$, let $D_{\cJ}$ denote the complexes whose perverse cohomology objects all belong to $\cJ$. Direct images from the intermediate stacks can be unbounded on one side. Statements about them are always made in a fixed perverse degree. Grothendieck classes are taken at the bounded proper endpoints. A finite-type smooth presentation gives a common lower standard bound on the ordinary pullbacks to its \v Cech nerve. The $m$th totalization fiber has this bound increased by $m$; therefore only finitely many levels contribute in any fixed degree. On the finite-dimensional Hitchin base, the same observation gives a finite Postnikov range for each fixed perverse degree of a finite convolution. All Grothendieck classes used later belong to bounded proper stable-pair or Hitchin endpoints. \subsection{Established inputs} For a sufficiently positive effective $L$, the positive Hitchin support theorem and cohomological $\chi$-independence hold in the derived category of mixed Hodge modules \cite[Theorems 0.1 and 0.4]{MSChi}. The smooth-stack integrality theorem of \cite[Theorem 10.2.7 and Lemma 10.2.9]{BDIKP}, with its commutative cotangent orientation, applies to the fixed-slope positive Higgs stack. Section~\ref{sec:source} verifies its hypotheses and fixes the scalar stabilizer parameter. The integral localization theorem of \cite[Corollary 7.19 and Remark 7.20]{KPS} supplies the morphism for filtered oriented $3$-Calabi--Yau objects. We use its chart-level monodromic Hodge realization together with the specialization exchange of \cite{Descombes}. The smooth descent and the comparison of orientations are included below. The conductor transversality criterion of \cite[Theorem 4.16(3)]{MSV} supplies smoothness of the relative Hilbert schemes on the reduced positive spectral locus. Finally, smooth semiprojective Higgs moduli have pure cohomology \cite[Corollary 1.3.2]{HRVPurity}. The $P=W$ theorem for coprime $\GL_n$ Higgs moduli is \cite[Theorem 0.2]{MSPW}. The rational tautological classes have the Hodge types stated in \cite{Shende}; together with tautological generation, they identify the resulting weight pieces with Tate structures. These facts are used only after the two stable-pair calculations and their comparison have been proved. \section{Critical models and their orientations}\label{phys:section}\label{sec:physical} We first construct the comparison of coefficients used below. All determinants in this section are graded determinants. Products of determinants therefore include their Koszul commutativity maps. The derived category of perfect complexes is used with its stable enhancement, so that a fiber, a map of triangles, and their compatibilities refer to the corresponding enhanced constructions. \subsection{The framed critical model} Let \(N\) be a rank-two bundle on \(C\), with a fixed identification \(\det N=K_C\), and choose an exact sequence \begin{equation}\label{phys:extension} 0\longrightarrow K_CL^{-1}\longrightarrow N \xrightarrow{q}L\longrightarrow0. \end{equation} Its signed extension class is denoted by \(\alpha_q\in H^1(C,K_CL^{-2})\). The sign is fixed by writing a transition between local lifts of the quotient coordinate as \(X_j=X_i-c_{ij}\Theta\), with \(c_{ij}\) representing \(\alpha_q\). Put \(S=\Tot_C L\), with projection \(p:S\to C\). The map \(\Tot_C N\to S\) is the \(K_S\)-torsor with class \(\alpha_q y\), where \(y\) is the tautological section of the pullback of \(L\). Let \(\cP_L\) be the derived stack of surface pairs \(\OO_S\xrightarrow{s}G\), with \(G\) perfect and properly supported, and write \[ J_G=\fib(\OO_S\to G). \] We work on the locus on which the ordinary sheaf \(G\) is finite over \(C\). It can then be described by its curve pushdown \(E\), the endomorphism \(\Theta:E\to E\otimes L\), and its section. The regular function used on this stack is \begin{equation}\label{phys:potential} g_q(E,\Theta,s) =\frac12\,\Tr_C\!\left(\alpha_q\Tr(\Theta^2)\right). \end{equation} Here \(\Tr_C:H^1(C,K_C)\to\C\) is Serre trace. Formula \eqref{phys:potential} defines a function on the derived stack as well: use a Čech or Thom--Whitney representative of \(\alpha_q\) and the universal trace. Its derivative in a bundle and field direction \((b,v)\) is \(\Tr_C(\alpha_q\Tr(\Theta v))\), since the additional term \(\Tr(\Theta[b,\Theta])\) vanishes. The physical enhancement of a stable pair on \(X=\Tot_C N\) is the moduli of perfect complexes \[ I=\fib(\OO_Y\to F),\qquad I|_D\simeq\OO_D, \quad Y=\mathbf P_C(\OO_C\oplus N),\quad D=\mathbf P_C N. \] The boundary is framed by the indicated isomorphism. Whenever a determinant trivialization is also used, its restriction agrees with the determinant of this boundary framing, with the specified homotopy of agreement. On a formal neighborhood of this locus, boundary framing agrees with fixing the determinant: near \(\OO_D\), the perfect-object stack is the derived Picard stack, and \(R\Gamma(Y,\OO_Y)\to R\Gamma(D,\OO_D)\) is an equivalence. Thus restriction to \(X\), followed by extension with the fixed determinant, also identifies the formal problems for the two compactifications used below. The stable-pair locus is an open condition on the classical truncation. Pure one-dimensional sheaves in this locus have locally free curve pushdown: a curve-torsion submodule would generate a zero-dimensional submodule under the finite support algebra. \begin{proposition}\label{phys:critical-equivalence} The boundary-framed physical enhancement is naturally equivalent, on the indicated locus, to \(\Crit(\cP_L,g_q)\). The equivalence is defined over connective Artin differential graded algebras, and respects maps of objects, their homotopies, direct sums, and finite exact diagrams. \end{proposition} \begin{proof} Compactify the torsor over \(S\) as \(\pi:Z=\mathbf P_S(V)\to S\), using the lines convention, where \[ 0\longrightarrow K_S\longrightarrow V\longrightarrow\OO_S \longrightarrow0,\qquad H=\OO_Z(D_\infty). \] The projective bundle formulas give \begin{equation}\label{phys:p1-identities} R\pi_*H=V^\vee,\qquad R\pi_*H^{-1}=0,\qquad R\pi_*H^{-2}=K_S[-1]. \end{equation} The relative diagonal has resolution \[ 0\longrightarrow H_1^{-1}H_2^{-1}\det(V)^\vee \longrightarrow\OO_{Z\times_S Z} \longrightarrow\OO_{\Delta_Z}\longrightarrow0. \] For a framed perfect object \(I\), define \[ G=R\pi_*(IH^{-1})[1],\qquad U=R\pi_*(IH^{-2}K_S^{-1})[1],\qquad Q=GK_S^{-1}. \] Applying this diagonal resolution and then tensoring by \(H\) recovers \begin{equation}\label{phys:diagonal-reconstruct} I\simeq\fib(\pi^*U\longrightarrow\pi^*G\,H). \end{equation} Adjunction identifies the arrow with \(U\to GV^\vee\). Since \(H|_{D_\infty}=K_S^{-1}\), the framing identifies its boundary restriction with \(U\to Q\). Thus the framed object is equivalently a map of triangles \[ \begin{CD} \OO_S @>>> U @>>> Q @>>> \OO_S[1]\\ @V{s}VV @VVV @V{\id}VV @V{s[1]}VV\\ G @>>> GV^\vee @>>> Q @>{c_G}>> G[1]. \end{CD} \] If \(e:Q\to\OO_S[1]\) is the upper connecting map, the middle arrow includes the specified homotopy \(s[1]e\simeq c_G\). In cone coordinates, write \(\varphi\) for a degree-zero cochain in \(\RHom_S(Q,G)\), and view \(e\) as a degree-one cochain in \(\RHom_S(Q,\OO_S)\). The equations are \begin{equation}\label{phys:homotopy-equations} d\varphi+se=c_G,\qquad de=0. \end{equation} Conversely these equations construct the upper triangle, its map to the lower triangle, and the fiber \eqref{phys:diagonal-reconstruct}. Both composites are the unit and counit of the displayed diagonal kernel. This also proves the assertion on mapping spaces and higher homotopies, after every derived base change. The tangent of \(\cP_L\) at a pair is \(\RHom_S(J_G,G)\). Surface duality with proper support identifies its shifted cotangent with \(\RHom_S(GK_S^{-1},J_G[1])\). Consequently \eqref{phys:homotopy-equations} describes the derived zero stack of the cotangent section \((c_G,0)\). In the monic resolution \[ [\,p^*E\otimes L^{-1}\xrightarrow{y-\Theta}p^*E\,] \] multiplication by \(y\) is chain homotopic to multiplication by \(\Theta\). The torsor cocycle \(\alpha_q y\) therefore represents the covector \(\Tr_C(\alpha_q\Tr(\Theta\,d\Theta))\), including bundle directions by the trace identity preceding the proposition. It is \(dg_q\). In a local splitting of the quotient, $X:E\to E\otimes A$ is the transverse field, and $\eps:E\to K_C$ is the dual framing field. Write $\psi$ for the cotangent field dual to $\Theta$, and $\rho_{\mathrm{cot}},\rho_{\mathrm{phys}}$ for the relation generators in the cotangent and physical models. The cone convention in \eqref{phys:homotopy-equations} gives \begin{equation}\label{phys:dictionary} \psi=-X,\qquad e=-\eps,\qquad \rho_{\mathrm{cot}}=-\rho_{\mathrm{phys}}. \end{equation} The last sign follows from the first two: they negate the moment expression, so they also negate its degree-minus-one relation variable. This proves the claimed equivalence. \end{proof} For the split bundle \(L\oplus K_CL^{-1}\), the function is zero. For \(N=K_C\oplus\OO_C\) and the fixed positive quotient \(q_0\), it is the function \eqref{phys:potential}. These two uses have the same cone convention and the same dictionary \eqref{phys:dictionary}. \subsection{The full deformation complex and its trace} We record a model that determines both the closed form and its orientation. It is useful to allow a framing bundle \(R\) in place of \(\OO_C\). On a curve and bundle chart, write \(V_0\) for the curve pushdown and \(W\) for the two normal directions, with \(\Delta=\det W\). The universal coefficient algebra has ordinary coordinates \(\Theta,X,s,\eps\), and all matrix entries of an odd coordinate \(\rho\), with \[ |\rho|=-1,\qquad d\rho=[\Theta,X]+s\eps. \] For homogeneous coefficients and homogeneous matrix maps we use \((a\otimes f)(b\otimes h)=(-1)^{|f||b|}ab\otimes fh\). On the relative projective plane, let \([z_0:z_1:z_2]\) be the homogeneous coordinates, with \(D=(z_0=0)\). The universal perfect complex has terms \[ M^{-1}=V_0\Delta^{-1}\OO(-2),\quad M^0=(V_0\otimes W^\vee)\OO(-1)\oplus R,\quad M^1=V_0\OO \] and arrows \begin{align} \alpha&=(z_0X-z_2,\;-(z_0\Theta-z_1),\;z_0^2\eps),\nonumber\\ \beta&=(z_0\Theta-z_1,\;z_0X-z_2,\;s),\qquad \gamma=-z_0^2\rho:M^{-1}\longrightarrow M^1. \label{phys:monad} \end{align} The matrix degree of \(\gamma\) is two and its coefficient degree is minus one. Hence it is part of the total differential, and \(d\gamma+\beta\alpha=0\). Projection to \(R\) gives a framing on \(2D\): near \(D=(z_0=0)\) the remaining two-variable Koszul complex is contractible, since one of \(z_1,z_2\) is invertible. For objects \(i,j\), the strict morphism model \(\widetilde{\mathcal C}_{ij}\) has generators \[ (g,r)\ \text{in degree }0,\qquad (a,b,u,v)\ \text{in degree }1,\qquad w\ \text{in degree }2. \] The source and target types are \(g,a,b,w:V_i\to V_j\), \(r:R_i\to R_j\), \(u:R_i\to V_j\), and \(v:V_i\to R_j\), with the normal twists in \eqref{phys:monad}. Its differential is \begin{align} D(g,r)&=(g\Theta_i-\Theta_jg,\ gX_i-X_jg,\ gs_i-s_jr,\ r\eps_i-\eps_jg) -g\rho_i+\rho_jg,\nonumber\\ D(a,b,u,v)&=aX_i-X_ja+\Theta_jb-b\Theta_i +u\eps_i+s_jv,\qquad Dw=0. \label{phys:hom-differential} \end{align} The last two terms of \(D(g,r)\) belong to the relation slot. Indeed, applying the second line to the four field entries gives \(g([\Theta_i,X_i]+s_i\eps_i) -([\Theta_j,X_j]+s_j\eps_j)g\), which is canceled by their coefficient differential. The map to monad morphisms sends \((g,r)\) to \(\operatorname{diag}(g,g,g,r,g)\), a field to minus the derivative of \eqref{phys:monad}, and \(w\) to the lower-to-top entry \(-z_0^2w\). Denote it by \(\kappa\). Its field entries, in the order lower, middle, frame, top, are \[ (-z_0b,z_0a,-z_0^2v),\qquad (-z_0a,-z_0b,-u). \] The only product of positive-degree generators that can be nonzero is \begin{equation}\label{phys:field-product} (a_2,b_2,u_2,v_2)(a_1,b_1,u_1,v_1) =w(-a_2b_1+b_2a_1-u_2v_1). \end{equation} Degree-zero products and their actions are ordinary rectangular composition. Equations \eqref{phys:hom-differential} and \eqref{phys:field-product} follow by multiplication of the displayed matrices. Thus \(\kappa\) is a morphism of differential graded categories. \begin{lemma}\label{phys:monad-hom} The morphism \(\kappa\) is a quasi-isomorphism on every universal chart and after arbitrary derived base change. The relative boundary-framing complex is its subcomplex \(r=0\). \end{lemma} \begin{proof} All entries in the matrix Hom complex have projective-plane twists between \(-2\) and \(2\), whose higher cohomology vanishes. At zero fields the monad is \(\OO\otimes R\oplus i_*V_0[-1]\). The Koszul resolution of the point \(i\) gives the four Hom blocks \[ \begin{array}{c|c} \text{source and target} & \text{nonzero cohomology}\\ \hline i_*V_i[-1],\,i_*V_j[-1]& \Hom(V_i,V_j)\otimes\bigwedge^kW\quad(k=0,1,2)\\ \OO R_i,\,i_*V_j[-1]&\Hom(R_i,V_j)\quad(k=1)\\ i_*V_i[-1],\,\OO R_j&\Hom(V_i,R_j)\Delta\quad(k=1)\\ \OO R_i,\,\OO R_j&\Hom(R_i,R_j)\quad(k=0). \end{array} \] For the first block, differentiation of the two Koszul equations gives the two exterior generators, and the lower-to-top map gives their exterior product. The two framing maps give respectively the restriction and the top Koszul class. Thus the displayed map \(\kappa\) identifies every generator with the corresponding class in this table. Combine simultaneous dilation in the two normal directions with the central \(V_0\)-gauge action. Each of \(\Theta,X,s,\eps\) has weight one, and \(\rho\) has weight two. The cone of \(\kappa\) is a perfect equivariant complex on the whole affine derived chart. If its classical support were nonempty, the positive dilation would put the origin in its closure. Its derived fiber at the origin is zero by the calculation above. Derived Nakayama then makes that support empty. This argument is made before imposing a stability open, and its quasi-isomorphism can subsequently be pulled back to any derived base. Boundary restriction is exactly \((g,r)\mapsto r\), proving the last assertion. \end{proof} Write \(\mathcal C_{ij}\) for \(r=0\). For arbitrary framing rank this means the fiber of the entire boundary restriction. For rank-one framing on the diagonal it is also the scalar trace fiber. \begin{lemma}\label{phys:proper-trace} The proper relative Serre trace on \(\mathcal C\) is the cochain pairing \begin{align} q_{\mathcal C}(g,w)&=q_{\mathcal C}(w,g)=\Tr(gw),\nonumber\\ q_{\mathcal C}((a,b,u,v),(a',b',u',v')) &=-\Tr(ab'-ba')-\Tr(v'u)+\Tr(vu'). \label{phys:raw-pairing} \end{align} The last two traces are taken on the respective framing spaces. After the shift \(T=\mathcal C[1]\), it agrees under \eqref{phys:dictionary} with the canonical critical pairing. \end{lemma} \begin{proof} Let \(\pi\) be the projection from the relative projective plane to its curve-and-bundle chart, and let \(\pi_D\) be its boundary restriction. Put \(G_j=R\pi_*\RHom(M,M(-jD))\). The second-order framing gives \(G_2\simeq G_1\), because its boundary cone is \(\operatorname{End}(R)\otimes R\pi_{D*}\OO_D(-D)=0\). The pairing in question is the specific product \[ G_1\otimes G_1\simeq G_2\otimes G_1 \longrightarrow G_3 \xrightarrow{\Tr,\ \mathrm{Serre}}\Delta[-2]. \] We compute its sign in the ordered cover \((z_0\ne0,z_1\ne0,z_2\ne0)\). Put \(x=z_1,y=z_2\), and order the five monad slots as lower, middle one, middle two, frame, top. At the origin the nonzero differential entries are \[ d_{21}=-y,\quad d_{31}=x,\quad d_{52}=-x,\quad d_{53}=-y. \] The contractions away from the point are \[ h_x=\frac{E_{13}-E_{25}}{x},\qquad h_y=-\frac{E_{12}+E_{35}}{y}, \] where \(E_{ij}\) is the elementary matrix. They satisfy \([d,h_x]=[d,h_y]=P_V\), the projector onto the unframed summand. Use the total differential \(\delta_{\mathrm{\check Cech}}+(-1)^p[d,-]\) in Čech degree \(p\). The two normal cocycles in \(G_1\) are \(T=E_{31}-E_{52}\) and \(X=-E_{21}-E_{53}\). A \(G_2\) lift of \(T\) has components \[ T/z_0,\qquad b_{01}=Th_x/z_0,\quad b_{02}=Th_y/z_0, \quad b_{12}=0,\qquad c_T=-h_x(b_{01}-b_{02}). \] Explicitly, \[ b_{01}=\frac{E_{33}+E_{55}}{z_0x},\qquad b_{02}=-\frac{E_{32}}{z_0y},\qquad c_T=\frac{-E_{13}+E_{25}}{z_0x^2} -\frac{E_{12}}{z_0xy}. \] Multiplication then gives \([d,b_{0i}]=-T/z_0\), \([d,c_T]+b_{01}-b_{02}=0\), and \[ \operatorname{str}(c_TX)=-\frac1{z_0xy}. \] For the section cocycle the corresponding components are \[ -E_{54}/z_0^2,\quad -E_{24}/(z_0^2x),\quad -E_{34}/(z_0^2y),\quad c_u=-E_{14}/(z_0^2xy). \] The dual section in \(G_1\) is \(v_1=-z_0E_{41}\), and \(\operatorname{str}(c_uv_1)=-1/(z_0xy)\). For the gauge class the top component of its lift is \(c_g=-h_xh_y/z_0^2=E_{15}/(z_0^2xy)\); multiplication by the relation \(-z_0E_{51}\) has supertrace \(+1/(z_0xy)\). Here the supertrace is \(-m_{11}+m_{22}+m_{33}+m_{44}-m_{55}\). Reversing the two field factors reverses their sign, whereas reversing gauge and relation does not. The ordered projective Serre trace sends \(1/(z_0xy)\) to \(+1\); this is the ordered residue convention of \cite[Proposition 5.2.3]{NayakSastry}. Tensoring these calculations with arbitrary rectangular Hom spaces gives exactly \eqref{phys:raw-pairing}. There is also a direct differential check. If \(P\) is minus the field pairing in \eqref{phys:raw-pairing}, cyclic trace gives \(P(Dg,z)=\Tr(gDz)\). For opposite gauge inputs \(g,h\), the additional ghost terms cancel as \[ \Tr\bigl((-g\rho_i+\rho_jg)h +g(-h\rho_j+\rho_ih)\bigr)=0. \] These identities verify the cochain relation with all ghosts. To see that the origin calculation determines the entire universal pairing, assign generator degree \(i\) the section-module weight \(-i\). The target \(\Delta[-2]\) has degree two and weight \(-2\). A pairing cochain of degree \(-k\), with \(r\) ghosts and ordinary polynomial degree \(n\), would satisfy \[ r=2+k-i-j,\qquad n+2r=2-i-j, \quad\text{hence}\quad n+r+k=0. \] For \(k>0\) there is no such cochain; for \(k=0\) only constants on complementary generator degrees occur. Finite semifreeness and exact invariants for the reductive frame group justify this calculation in derived Hom. Thus the comparison and its descent are fixed by the origin computation. The resulting universal maps pull back to every test algebra. Finally décalage, with \(s\) denoting suspension, gives \(\omega(su,sv)=(-1)^{|u|}q_{\mathcal C}(u,v)\). The tangent field form is consequently \(\Tr(d\Theta\wedge dX)+ds\wedge d\eps\), the derivative of \(-\Tr(X\,d\Theta)-\eps\,ds\). For a coordinate of degree \(a\), the canonical raw cross pairing of its cotangent generator with the base generator is \((-1)^{a+1}(-1)^a=-1\) times evaluation. The whole dual-half reversal in \eqref{phys:dictionary} therefore matches the gauge--relation block as well. \end{proof} \subsection{Closed forms and the determinant square} We specify the closed lift of the trace just calculated. In the convention of \cite[Theorem 2.12]{PTVV}, the leading component of \(\mathrm{Ch}_2\) is \(\tfrac12\Tr(\mathrm{mult})\). We therefore use \begin{equation}\label{phys:closed-normalization} \omega_{\Perf}^{\Tr}=2\mathrm{Ch}_2. \end{equation} On \(Y\), the given volume has \(K_Y=\OO_Y(-3D)\). The framed mapping construction of \cite[Theorem 3.3]{Spaide}, with extra divisor \(D\), applies: the lift of the specified framing from \(D\) to \(2D\) is étale because \(R\Gamma(D,\OO_D(-D))[1]=0\). Its framing isotropic path is zero, since the ideals of \(D\) and \(2D\) multiply to zero on \(3D\). The other path is the boundary triangle \(K_Y\to\OO_Y\to\OO_{3D}\). This constructs a closed \((-1)\)-shifted form whose leading pairing is precisely the proper product in Lemma~\ref{phys:proper-trace}. The factor in \eqref{phys:closed-normalization} does not change the coefficient \(1/2\) of \eqref{phys:potential}: the latter is its quadratic primitive coefficient. Let \(\operatorname{pr}:\mathcal C_{ij}\to\mathcal B_{ij}\) be the projection to the surface framing complex, forgetting \(b,v,w\). It is a strict functor. Its kernel has \(b,v\) in degree one and \(w\) in degree two, and has square zero by \eqref{phys:field-product}. The pairing \[ (b,v),(a,u)\longmapsto\Tr(ba)+\Tr(vu),\qquad w,g\longmapsto\Tr(wg) \] identifies this kernel with \(\mathcal B_{ji}^\vee K_C[-2]\). Writing \(\langle-,-\rangle\) for this pairing, the two bimodule identities are \[ \langle kh,b\rangle=\langle k,\operatorname{pr}(h)b\rangle,\qquad \langle hk,b\rangle=(-1)^{|h|(|k|+|b|)} \langle k,b\operatorname{pr}(h)\rangle. \] Here \(k\) lies in the kernel and \(h,b\) are homogeneous composable entries. In generator degree two they are rectangular trace identities, using \eqref{phys:field-product}; all other cases vanish. Thus, after curve sections and Serre duality, there is a natural paired triangle \begin{equation}\label{phys:paired-triangle} B_{ji}^{\vee}[-3]\longrightarrow C_{ij} \longrightarrow B_{ij},\qquad B_{ij}=R\Gamma(C,\mathcal B_{ij}),\quad C_{ij}=R\Gamma(C,\mathcal C_{ij}). \end{equation} It extends to finite twisted sums and cones by their block-matrix differentials. In particular it applies when a cofiber is a finite-length sheaf. The leading pairing comparison of Lemma~\ref{phys:proper-trace} extends to the full closed forms. Both forms have weight two for normal dilation. If \(\delta\) is their difference and \(h_0\) is its leading homotopy, then \(\delta-Dh_0\in F^3\DR\). For the Euler field \(\mathcal E\), Cartan's formula gives \begin{equation}\label{phys:cartan} H_0=h_0+\tfrac12\iota_{\mathcal E}(\delta-Dh_0) \in F^2\DR,\qquad DH_0=\delta. \end{equation} This uses the full de Rham mixed complex. In a formal presentation with invariant ideal \(J\), the filtration \(\prod_p J^{\max(m-p,0)}\Omega^p\) makes the operation continuous. Construct \eqref{phys:cartan} on the equivariant formal diagram, with its transition maps, before restricting to a completion at a moving point. For a diagonal object put \(B=B_{ii}\). The half line is \(\det B\); its square is given by \eqref{phys:paired-triangle}, duality, and the canonical critical factor \((-1)^{\operatorname{vdim}\cP_L}\) in \cite[Example 3.20]{KPS}. This defines the orientation used throughout the paper. The bimodule structure and the determinant double-sequence identity \cite[Lemma 2.4]{KPS} give its direct-sum and finite-flag compatibilities. \begin{proposition}\label{phys:orientation-square} Let \(\Psi\) be the equivalence of Proposition~\ref{phys:critical-equivalence}, or one of the supported equivalences constructed below, and let \(M_A,M_B\) be the orientation lines of its source and target. Its specified half-line comparison \(\gamma:M_A\to\Psi^*M_B\) satisfies \begin{equation}\label{phys:same-square} o_A=\det(d\Psi^*)\,o_B\,\gamma^{\otimes2}, \qquad d\Psi^*:\Psi^*\mathbb L_B\longrightarrow\mathbb L_A. \end{equation} Here \(o_A,o_B\) are the orientation square maps and \(\mathbb L_A,\mathbb L_B\) are cotangent complexes. \end{proposition} \begin{proof} First consider \eqref{phys:critical-equivalence}. Write its paired triangle as \(K\to C\to B\), with \(h:K\xrightarrow{\sim}B^\vee[-3]\). The cotangent triangle is \[ B^\vee[-1]\longrightarrow C^\vee[-1] \longrightarrow K^\vee[-1]. \] Identify its last object with \(B[2]\) by the inverse of \(h^\vee[-1]\), using the same shifted double-dual convention as in \cite[Appendix B.1]{KPS}. The physical and critical triangles have identical first arrows. On the last term, the dual of the equality \(h_{\mathrm{crit}}\Psi_K=h_{\mathrm{phys}}\), proved in Lemma~\ref{phys:proper-trace}, is \[ \Psi_K^\vee[-1]h_{\mathrm{crit}}^\vee[-1] =h_{\mathrm{phys}}^\vee[-1]. \] Consequently these are equivalent diagrams of fiber sequences. Taking their graded determinants, and inserting the same critical factor on both sides, proves \eqref{phys:same-square}. This keeps the direction of the cotangent derivative. For example, scaling \(h\) by \(c\) changes the raw square by \(c^{\chi(B)}\), and negating the whole dual half changes its derivative determinant by \((-1)^{\chi(B)}\); these are the same factor. Dualizing and shifting inverts the full determinant map. For a finite exact diagram, apply this proof to the paired bimodule triangles \eqref{phys:paired-triangle} on every entry of its double Hom grid. Naturality of duality and determinant double-sequence coherence give the asserted square for the resulting half-line map. The supported case and the identification of its full map with the derivative of the specified \(\Psi\) are spelled out in Lemma~\ref{phys:supported-determinant}. \end{proof} \subsection{The fixed positive quotient} We compare this orientation with the canonical one for \(N=K_C\oplus\OO_C\). For two fixed even-degree quotients \(q_i:N\to L_i\), put \(A_i=\ker q_i\), and assume their kernels are generically transverse. In the full deformation complex the dual-section and relation terms form a common isotropic subcomplex \(J\). Its orthogonal quotient is \[ J^\perp/J=\operatorname{End}(E)\otimes N[-1], \] and the two isotropic halves have quotients \(\operatorname{End}(E)\otimes A_i[-1]\). These are filtrations of paired complexes. \begin{lemma}\label{phys:fixed-quotient} The two orientation roots differ by \(\mathcal C_{\mathrm{res}}^{\,r^2}\), where \(\mathcal C_{\mathrm{res}}\) is one fixed quadratic line depending only on the two quotients. Choosing its quadratic trivialization once gives a square-preserving comparison, compatible with direct sums and flags. Its factors cancel canonically between a finite modification and its same-rank baseline. \end{lemma} \begin{proof} Refine the filtration by \(J\), the two isotropic halves, and \(J^\perp\). Determinant double-sequence coherence cancels the common hyperbolic part of their squares. The remaining ratio is the Serre linking form on the fixed finite module \(N/(A_0+A_1)\), of length \(m=\deg L_0+\deg L_1-2(g-1)\). Choose a local ordered frame with \[ A_1=\OO v_1,\quad A_0=\OO(uv_1+t^av_2),\quad v_1\wedge v_2=dt, \] where \(u\) is a unit. Decomposing \(xv_2\) along these two lines gives the linking form \[ b_{\mathrm{res}}(x,y)=\operatorname{Res} \frac{uxy\,dt}{t^a}. \] Adding an integral lift does not change the residue. In the basis \(1,t,\ldots,t^{a-1}\), its determinant is \((-1)^{a(a-1)/2}u(0)^a\). Tensoring with the trace form on \(\operatorname{End}(E)\) gives \[ \det b_E =(-1)^{mr(r-1)/2}(\det b_{\mathrm{res}})^{r^2}. \] The length \(m\) is even, so the additional sign is one. A change of frame acts by \(\operatorname{Ad}(g)\), whose determinant over the finite support ring is one; its norm is also one. Thus this is an equality of quadratic line maps over nonreduced bases, independent of \(E\) and of all its fields. The difference of the critical factors has exponent \(r^2(\deg L_1-\deg L_0)\), which is even. For a direct sum, opposite cross-Hom blocks give the factor \(\mathcal C_{\mathrm{res}}^{2r_1r_2}\), contracted by its quadratic form. The interchange sign is \((-1)^{mr_1r_2}=1\). The common determinant filtration gives coherence for further blocks and refinements. A modification and its baseline have the same rank, so their factors cancel by evaluation, including when the defect meets the residue divisor. \end{proof} \subsection{Support gluing and the derivative} For the fixed-point calculation use the equivariant projection \[ q_K:K_C\oplus\OO_C\longrightarrow K_C,\qquad S_K=\Tot_C K_C. \] Transport the fixed positive-quotient orientation by Lemma~\ref{phys:fixed-quotient}. Both quotient degrees are even, and their kernels are generically transverse: otherwise the component $\OO_C\to L$ of the positive quotient would vanish, forcing a surjection $K_C\to L$ contrary to $\deg L>\deg K_C$. The fixed quadratic factor has exponent \(r^2\) and cancels between a modification and its same-rank baseline. All Hom, section, support, and duality maps below are equivariant for the torus \(H\) defined next. Their determinant grids are taken weight by weight: weight zero supplies the orientation of the fixed enhancement, and the moving weights supply the localization index. Let \(Z_\lambda\subset\Tot_C(K_C\oplus\OO_C)\) be the monomial thickening specified by a partition \(\lambda\), and let \(I_0=\fib(\OO\to\OO_{Z_\lambda})\). Use the volume-preserving torus \[ H:(\theta,z)\longmapsto(a\theta,a^{-1}z). \] Consider a finite modification \(0\to\OO_{Z_\lambda}\to F\to\widetilde T\to0\). Ordinary \(H\)-fixed families of these modifications are identified by curve duality with quotients of the dualizing module \(\omega_{Z_\lambda}\), where \(\omega_{Z_\lambda}=\Hom_C(\OO_{Z_\lambda},K_C)\). Let two étale coordinate maps, over a smooth parameter scheme, identify selected neighborhoods of the finite defect support with neighborhoods on a reference curve. Their normal lifts have the form \[ (t,\theta,z)\longmapsto (f(t),\,\theta/f'(t),\,z), \] so they preserve \(dt\wedge d\theta\wedge dz\). The coordinate maps on ordinary quotient schemes will be denoted by \(\mathsf A,\mathsf B\). \begin{lemma}\label{phys:support-gluing} On the selected support branch there is a formal equivalence of the specified \(H\)-fixed physical enhancements whose truncation is \(\mathsf B\mathsf A^{-1}\) on every ordinary Artin base. It is constructed on all connective Artin bases by restriction and gluing with fixed determinant. \end{lemma} \begin{proof} For a rank-one perfect object define \[ \mathscr A_I=\fib(R\mathcal Hom_Y(I,I)\xrightarrow{\operatorname{str}}\OO_Y). \] The identity has supertrace one, so this is the strict trace kernel in a locally free model. Near \(\OO_D\), the perfect-object stack equals the derived Picard stack formally: derived Nakayama and Tor-amplitude zero make the deformation an invertible module. Moreover \(R\Gamma(Y,\OO_Y)\simeq R\Gamma(D,\OO_D)\). The determinant and boundary restriction square therefore identifies the boundary-framed formal problem with the fixed-determinant formal problem, including all higher automorphisms. If \(Z\) is the underlying finite defect support and \(U=Y-Z\), the modification problem is \begin{equation}\label{phys:modification} \bigl(\Perf(Y)\times_{\Pic(Y)}\{\det I_0\}\bigr) \mathop{\times}_{\Perf(U)\times_{\Pic(U)}\{\det I_0|_U\}} \{I_0|_U\}. \end{equation} It has deformation complex \(R\Gamma_Z\mathscr A_I\). An étale map on the selected finite support is universally injective and proper there, hence a closed immersion onto that support. Delete the other sheets over it. This gives an elementary distinguished square with the open complement. Descent for \(\Perf\) and \(\Pic\) gives an equivalence of \eqref{phys:modification} on the two sides. Its inverse is the Mayer--Vietoris homotopy fiber of the difference of restriction maps; the same fiber on mapping complexes gives full faithfulness and the higher cells \cite[Tag 08GS]{Stacks}. This construction uses the open complement in the derived ambient. For the monomial baseline, its transverse ideal has scalar endomorphisms, local \(\Ext^1\) equal to the Hilbert tangent representation, and no higher local Ext, by the two-variable Hilbert--Burch resolution. The two tangent characters at a box are \(a^{h_b}\) and \(a^{-h_b}\); none is zero-weight. After removing scalars, the \(H\)-invariant baseline complex on the complement and its overlaps is therefore acyclic. Curve sections and localization in curve points preserve these characters. For a square-zero extension of an Artin base, the obstructions, choices, and all automorphisms for forgetting the complement identification are shifts of this acyclic complex. Induction over small extensions and the Postnikov tower proves that forgetting it is a formal equivalence. It remains to identify the ordinary map. In an ordinary Artin family, the curve pushdown \(E\) is locally free. The monomial annihilator acts by zero away from the defect. A local curve parameter, or its monic characteristic polynomial, is a nonzerodivisor on the free curve module over an Artin ring. The annihilator consequently acts by zero everywhere. The section gives \(0\to\OO_{Z_\lambda}\to F\to\widetilde T\to0\) with base-flat finite cokernel. Relative curve duality gives its quotient of \(\omega_{Z_\lambda}\). Selected-sheet pullback commutes with these operations and with the section. Hence the truncation is exactly \(\mathsf B\mathsf A^{-1}\) on ordinary Artin families. \end{proof} \begin{lemma}\label{phys:supported-determinant} For the equivalence in Lemma~\ref{phys:support-gluing}, the algebraic half-line comparison defined by the four supported Hom and section maps has square \eqref{phys:same-square} for the derivative of that same equivalence. \end{lemma} \begin{proof} Write the surface modification triangle as \(G_0\to G\to T\). On the equivariant surface \(S_K\), put \[ B_G=\fib\bigl(\RHom_{S_K}(G,G)\to R\Gamma(S_K,G)\bigr), \quad M(G)=\det B_G,\quad M_H(G)=\det(B_G^H). \] Superscripts $H,+,-$ denote the invariant, positive-weight, and negative-weight parts for $H$. The map is precomposition by the section. For the paired triangle \eqref{phys:paired-triangle}, the positive-weight part of the physical cotangent complex has determinant \[ \det(\mathbb L_{\mathrm{phys}}^{>0}) =\det(B_G^-)\otimes\det(B_G^+). \] The localization orientation \(M(G)\otimes\det(\mathbb L_{\mathrm{phys}}^{>0})^{-1}\) therefore identifies with \(M_H(G)\), using the graded regrouping maps of \cite[Section 6.5]{KPS}. The double Hom grid is \(H\)-equivariant before taking determinants. Its full determinant identity is \begin{equation}\label{phys:four-hom} \frac{M(G)}{M(G_0)} = \det\RHom(T,G_0)\otimes\det\RHom(T,T) \otimes\det\RHom(G_0,T)\otimes(\det R\Gamma T)^{-1}. \end{equation} Its comparison maps are supported adjunction on \(\RHom(G_0,T)\), its Serre dual on \(\RHom(T,G_0)\), the support-equivalence unit and counit on \(\RHom(T,T)\), and composition of pushforwards on \(R\Gamma T\). They are algebraic and compatible with base change. Taking weight zero in each grid entry and then the ordered determinant product defines \(\gamma\) on the lines \(M_H\). The physical triangle is \(I_0\to I\to J\) with \(J=\widetilde T[-1]\), since the cofiber of the map of source sections is zero. Represent \(I\) as the twisted sum with differential \[ D_I=\begin{pmatrix}d_0&\kappa_0\\0&d_J\end{pmatrix}. \] For a degree-\(k\) endomorphism \(M=\begin{pmatrix}a&b\\c&e\end{pmatrix}\), its differential has blocks \begin{align*} (\delta M)_{00}&=d_0a-(-1)^kad_0+\kappa_0c,\\ (\delta M)_{0J}&=d_0b-(-1)^kbd_J +\kappa_0e-(-1)^ka\kappa_0,\\ (\delta M)_{J0}&=d_Jc-(-1)^kcd_0,\\ (\delta M)_{JJ}&=d_Je-(-1)^ked_J-(-1)^kc\kappa_0. \end{align*} Graded cyclicity cancels the extension terms in the supertrace. The strict trace kernel has a three-step filtration by the upper-right block, the trace-zero upper triangular blocks, and the whole kernel. Its graded pieces are \begin{equation}\label{phys:trace-filtration} R\mathcal Hom_Y(J,I_0),\qquad \mathscr A_{I_0}\oplus R\mathcal Hom_Y(J,J),\qquad R\mathcal Hom_Y(I_0,J). \end{equation} The middle identification sends \((a_0,e)\) to \((a_0-\operatorname{str}_J(e)\id_{I_0},e)\). The extension differential is retained in the filtered complex. After supported sections and \(H\)-invariants, the baseline term in \eqref{phys:trace-filtration} is acyclic and the other terms are perfect with finite support, so their determinants are defined. Restriction and gluing preserve the displayed filtration. Consequently its determinant map is the determinant of the derivative of the functor in Lemma~\ref{phys:support-gluing}. To compare with the proper Hom grid, use the natural maps from supported to proper sections on its surviving finite-support terms. The possible trace-projector correction on \(\RHom(J,J)\) is \(1-uv\), where \(u(1)=\id_J\) and \(v=\operatorname{str}_J\). After proper sections, these maps run between \(R\Gamma(Y,\OO_Y)\) and \(R\Gamma(Y,R\mathcal Hom_Y(J,J))\), both perfect over the parameter base. Their composite on the first complex is \(vu=\rk(J)\operatorname{id}=0\). The graded Sylvester identity \(\det(1-uv)=\det(1-vu)\) gives determinant one, even on nonreduced bases. The paired triangle \eqref{phys:paired-triangle}, applied to the same exact arrow diagram, has surface halves \[ \RHom(T,G_0),\quad \RHom(T,T),\quad \fib(\RHom(G_0,T)\to R\Gamma T). \] Its full half determinant is \eqref{phys:four-hom}; its weight-zero half determinant is the same formula with invariants in every Hom and section term. Applying the proof of Proposition~\ref{phys:orientation-square} to this grid identifies the square of \(\gamma\) with the derivative determinant just computed. The three full surface Hom Euler pairings are zero, and the section term has Euler characteristic \(d=\operatorname{length}T\). If \(m_0,m_+,m_-\) are the Euler changes of the half complex in weight zero, positive weights, and negative weights, respectively, then \[ m_0+m_++m_-=-d. \] The fixed determinant square retains the dual-half reversal \((-1)^{m_0}\). The moving weight pieces retain their corresponding reversal factors; their product with the fixed factor is \((-1)^d\). These factors are independent of the even residue length in Lemma~\ref{phys:fixed-quotient}. \end{proof} \subsection{The comparison over the full parameter ring} The preceding construction also compares the closed forms. On full endomorphism complexes, stable Koszul local cohomology gives the supported product. The boundary section is invertible on the support. Denote the maps from supported sections to \(R\Gamma\RHom(I,I(-jD))\) by \(J_j\): multiply by the inverse of the specified boundary section to the power \(j\), then apply the support counit. Multiplicativity of this counit gives \(J_i(a)J_j(b)=J_{i+j}(ab)\), with its composition homotopies in the enhanced category. The stable Koszul model represents these maps and products. Restricting the inputs to the trace kernel only after taking these products gives the proper pairing of Lemma~\ref{phys:proper-trace}. The selected étale maps preserve the volume. Their canonical bundle maps and duality counits therefore compare this pairing on the full moving family, including directions that move the support. The residue change-of-generators, localization, and base-change identities in \cite[\S\S4,10]{NayakSastry} apply to these counits. Locally a monic characteristic polynomial in the curve coordinate, together with high powers of the two normal coordinates, gives a bounded flat Koszul model. Lifting its coefficients to a connective presentation gives the same construction after derived base change. Different lifts compute the same support functor and its already defined duality counit. Normal dilation gives both closed forms weight two. Equation \eqref{phys:cartan} consequently constructs their full closed homotopy on the common parameter diagram. The following argument upgrades this statement to the particular ordinary maps used in coefficient descent. \begin{proposition}\label{phys:parameter-square} On a neighborhood of the punctual locus, the maps \(\mathsf A,\mathsf B\) identify the algebraic \(d\)-critical sections and the oriented vanishing-cycle coefficients. The comparison holds over the entire smooth moving-support and coordinate parameter space. \end{proposition} \begin{proof} The ordinary map \(\mathsf A\) is étale. Its unique derived étale lift relative to the specified physical enhancement is identified with the lift in Lemma~\ref{phys:support-gluing}: their ordinary maps agree, and restriction--gluing identifies their lifting spaces and mapping complexes on all connective Artin test bases. On a selected completion this makes the \(\mathsf A\)-side the fixed physical formal problem extended over \(R=\C[[s_1,\ldots,s_e]]\). Choose one minimal critical model at its reference point and extend scalars: \begin{equation}\label{phys:minimal-model} \mathcal A_R=R[[x_1,\ldots,x_m]][e_1,\ldots,e_m], \quad |e_j|=-1,\quad de_j=I_j(x)\in(x)^2. \end{equation} Put \(I=(I_1,\ldots,I_m)\subset R[[x_1,\ldots,x_m]]\). The existence of such a model is the formal critical-chart statement of \cite{BBJ}; it is chosen once over \(\C\). Represent the two specified maps on cofibrant complete models, calling the representatives \(\alpha,\beta\). Let \(f_A,f_B\) be their chart potentials and \(\lambda_A,\lambda_B\) the Liouville one-forms of their critical models. Set \[ F=\alpha^*f_A-\beta^*f_B,\qquad \ell_0=\alpha^*\lambda_A-\beta^*\lambda_B. \] The preceding closed homotopy makes \(F-\ell_0+H_0\) closed in the full relative de Rham complex. Its augmentation to \(R\) is zero: both potentials are normalized to vanish on their reduced critical loci, and the moving section is a critical family over the reduced ring \(R\). Give \(x,e,d_{\DR}x,d_{\DR}e\) Euler degree one and \(R\) degree zero. On positive Euler degree \(v\), \(\iota_{\mathcal E}/v\) contracts the augmented de Rham differential. The internal differential raises that degree, by \eqref{phys:minimal-model}. The perturbation series \[ h_{\mathcal E}(1+dh_{\mathcal E})^{-1} \] therefore converges in the combined adic and Hodge filtration and contracts the full augmented complex. The zero- and one-form components of a primitive satisfy \[ F=d\Psi_0,\qquad -\ell_0=d_{\DR}\Psi_0+d\psi_1. \] A degree-zero chart coordinate pulls back to a function of \(x\) over \(R\); hence \(\ell_0\) has no \(d_{\DR}e_j\) component. Degree constraints give \[ \Psi_0=\sum_jL_je_j,\qquad \psi_1=\sum_{i,j}M_{ij}e_i\,d_{\DR}e_j +\sum_{i,j,k}N_{ijk}e_ie_j\,d_{\DR}x_k. \] Comparing \(d_{\DR}e_j\) coefficients yields \(L_j+\sum_iM_{ij}I_i=0\). Consequently \begin{equation}\label{phys:ideal-square} F=-\sum_{i,j}M_{ij}I_iI_j\in I^2 \quad\text{in }R[[x]]. \end{equation} This is a single identity over the full parameter ring. For original charts with acyclic Hessian variables \(y\), put \(J=(I,y)\). A class modulo \(J^2\) has the form \(f_0(x)+\sum y_af_a(x)\). If its differential vanishes modulo \(J\), its \(dy_a\) coefficients give \(f_a\in I\). Thus \eqref{phys:ideal-square} detects equality in the original charts, with their Hessian factors retained. The product parameter directions contribute the identical factor \(K_{\mathrm{par}}\) to both half lines. The maps are over the parameter base, so their derivative matrices have an identity parameter block and introduce no additional determinant. The two \(d\)-critical sections and the half-line map of Lemma~\ref{phys:supported-determinant} existed algebraically before completion. Faithful flatness of local completion detects equality of their sections of \(\OO/J^2\), and then equality of the two existing square maps on the reduced base. The identification of the critical cotangent determinant with the \(d\)-critical canonical line is \cite[\S\S3.5--3.6]{KPS}, with the same Hessian and critical factor used above. Equation \eqref{phys:same-square} is therefore the oriented \(d\)-critical comparison, proving the assertion about vanishing-cycle coefficients. \end{proof} \section{Monodromic coefficients and localization}\label{phys:mmhm-section} The intermediate comparisons take place in monodromic mixed Hodge modules. We give the descent details needed for the stack correspondences. For an oriented critical chart \((U,f)\) the normalized coefficient is \[ \Phi_{(U,f)} =\phi_f^{\mathrm{mon}}(\Q_U)\La^{-\dim U}, \] tensored with its orientation torsor. Vanishing cycles are normalized to be perverse exact and to satisfy \(\phi_0=\id\). All products in this formula are monodromic products. The ordinary Hodge category embeds exactly and symmetrically monoidally in this category \cite[\S2.1]{DM}. \subsection{An explicit coefficient projector} Let \(\pi:V\times\A^1\to V\), and let \(B_V\) be the stable geometrically monodromic Hodge category on \(V\times\A^1\). The constants form the image of \(\pi^*\). Since \(\id\to R\pi_*\pi^*\) is an equivalence, the functor \begin{equation}\label{phys:projector} J_VF=\cofib(\pi^*R\pi_*F\to F) \end{equation} is the reflector onto \(\ker R\pi_*\). Indeed, \(R\pi_*J_VF=0\), and for \(R\pi_*K=0\) adjunction gives \(\operatorname{Map}(J_VF,K)\simeq \operatorname{Map}(F,K)\). Thus the quotient by constants is this explicit kernel. This projector also identifies the heart. If \(j:\Gm\into\A^1\), \(i:\{0\}\into\A^1\), the triangle \[ \Q_{\A^1}\longrightarrow i_*\Q \longrightarrow j_!\Q_{\Gm}[1] \] and the change of coordinates \((u,v)\mapsto(u+v,u)\) give \begin{equation}\label{phys:DM-projector} J_VF\simeq R(\id_V\times +)_*(F\boxtimes j_!\Q_{\Gm}[1]). \end{equation} The first map of the resulting triangle is exactly the counit in \eqref{phys:projector}. Equation \eqref{phys:DM-projector} is the perverse-exact projector of \cite[\S2.1, Lemmas 2.1--2.6]{DM}. Its image has the usual monodromic perverse heart. Both formulas commute with smooth pullback, and membership in the kernel is smooth local. They therefore descend through the ordinary smooth nerve of an Artin stack, using the Hodge six operations of \cite{Tubach}. The nilpotent and semisimple monodromy operators remain in this construction, as do the Hodge and weight filtrations. \subsection{The integral localization map} For a derived stack $X$, define the filtered and graded stacks by \begin{equation}\label{phys:filtered-stacks} \operatorname{Filt}(X)=\operatorname{Map}([\A^1/\Gm],X), \qquad \operatorname{Grad}(X)=\operatorname{Map}(B\Gm,X). \end{equation} Restriction at zero and evaluation at one give $\operatorname{Grad}(X)\xleftarrow{\operatorname{gr}} \operatorname{Filt}(X)\xrightarrow{\operatorname{ev}}X$. We use the component specified by the chosen integral grading. For a one-parameter torus action $\mu$, write $X^\mu$ for the fixed locus and $X^{\mu,+},X^{\mu,-}$ for the attracting loci as the parameter tends to zero and infinity. Their limit and evaluation maps are \[ X^\mu\xleftarrow{p^\pm}X^{\mu,\pm} \xrightarrow{\eta^\pm}X. \] Distinguish the functors \[ \operatorname{Loc}_\mu =(p^+)_*(\eta^+)^!\simeq(p^-)_!(\eta^-)^*, \qquad A_\mu=(p^+)_!(\eta^+)^* =\operatorname{Loc}_{\mu^{-1}}. \] The first is used for filtered-stack attachment, and the second for compact attracting strata. At a fixed point, the index convention is \[ I_\mu= \bigl(\rk^+H^0(T_X)-\rk^-H^0(T_X)\bigr) -\bigl(\rk^+H^{-1}(T_X)-\rk^-H^{-1}(T_X)\bigr), \] as in \cite[Definition 7.21]{KPS}. For the surface-half model \eqref{phys:paired-triangle}, it is \(\chi(B^-)-\chi(B^+)\). \begin{proposition}\label{phys:integral-localization} For the oriented \((-1)\)-shifted symplectic stacks used here, the integral map is an isomorphism in monodromic mixed Hodge modules \begin{equation}\label{phys:integral-map} \zeta_X: \Phi_{\operatorname{Grad}(X)}\La^{-I_\mu} \xrightarrow{\ \sim\ } R\operatorname{gr}_*\operatorname{ev}^!\Phi_X. \end{equation} It is compatible with products, smooth pullback, and matching refinements of an ordered finite flag. For a scheme with a torus action, its compact attracting version is \begin{equation}\label{phys:compact-attractor} (p^+)_!(\eta^+)^*\Phi_X \simeq\Phi_{X^\mu}\La^{I_\mu}. \end{equation} \end{proposition} \begin{proof} The stacks in use are covered by finite-type quotient charts, are quasi-separated, and have affine stabilizers. Their orientations and their product and flag maps are supplied by \eqref{phys:paired-triangle} and Proposition~\ref{phys:orientation-square}. These are the stack hypotheses of \cite[Corollary 7.19]{KPS}. We describe the mixed-Hodge construction of its map; this also fixes the normalization. On an equivariant critical chart, use the specialization exchange constructed by units and counits in \cite[Theorem 3.7, Proposition 6.2]{Descombes}. Its staircase inverse follows from the adjunction triangle identities. Tensor it with the finite-torsor map of \cite[(6.75)]{KPS}. The normalization on a moving hyperbolic quadratic block is the compact Thom map of the same repelling axis as in \cite[Theorem 7.9]{KPS}. Switching that axis would introduce \((-1)^r\) in rank \(r\). The scalar is checked on the normalized quadratic point, where the constructed maps lie in the perverse heart. All other maps are transformations in the Hodge six-functor system. For a self-dual filtration, locally choose a splitting and write \(q=q_{\mathrm{gr}}(1+N)\), with \(N\) filtration-raising and self-adjoint. The finite polynomial \((1+N)^{-1/2}\) orthogonalizes it. Orthogonal splittings form a torsor under the filtration-raising orthogonal unipotent group, an iterated affine-bundle torsor. Pullback to it is fully faithful, since its direct-image adjunction unit is an equivalence. The constructed map and all its comparison squares descend uniquely. The interpolation \(\exp(t\log u)\) between splittings has determinant one, so it contributes no orientation character. For smooth descent, let \(q:U\to X\) be an equivariant smooth chart and let \(q_\mu:U^\mu\to X^\mu\) be its fixed-locus map. Let the relative tangent of \(q\) have weight-zero, positive-weight, and negative-weight ranks \(d_0,d_+,d_-\), and let \(d=d_0+d_++d_-\). For a smooth map \(h\) of relative dimension \(e\), purity is \(h^!=h^*[2e](e)\). Applied to the fixed and attracting maps, it gives \[ q_\mu^*\operatorname{Loc}_X \simeq\operatorname{Loc}_\mu q^*[2d_-](d_-) =\operatorname{Loc}_\mu q^*\La^{-2d_-}. \] With normalized smooth pullback \(q^\dagger=q^*\La^{-d}\), this becomes \[ q_\mu^\dagger\operatorname{Loc}_X \simeq\operatorname{Loc}_\mu q^\dagger \La^{d_+-d_-}. \] The indices obey \(I_U=I_X+d_+-d_-\), so both the shift and Tate factor of the local map descend. The common charts and their fixed overlaps are \cite[Proposition 5.27, Corollary 5.28]{KPS}. On every level \(p_m\) of their smooth nerve, the two ordinary pullbacks are the same invertible shift and twist of perverse heart objects. Hence their mapping spaces have \(\pi_r=\Hom(A,B[-r])=0\) for \(r>0\). The smooth comparison squares, which are purity and Beck--Chevalley squares, give equal maps on double overlaps. Their compositions give the higher equalities. The discrete mapping spaces then descend the isomorphism and its inverse through the full nerve. Degeneracy compatibility is obtained by pullback of these identities, without applying smooth purity to a nonsmooth diagonal. For two ordered cuts, write their block weights as \(a_i,b_i\) in the same order. Then \((a_i-a_j)(b_i-b_j)\ge0\) on every cross-Hom block. This checks matching on the full cotangent complex, including relation and obstruction terms. Finitely many ordered cuts admit a common multi-torus chart by the construction of \cite[Lemma 5.30]{KPS}. On this locus the multi-filtered square is Cartesian and its excess map is the identity. Chart comparisons reduce to ordinary Beck--Chevalley by \cite[Lemma 7.4]{KPS}. Thus the product and matching-refinement constructions in \cite[Propositions 7.10--7.12, Lemmas 7.14--7.15]{KPS} are the same Hodge transformations, with determinant associators from \eqref{phys:paired-triangle}. This proves \eqref{phys:integral-map}. Using the inverse action changes the index sign and gives \eqref{phys:compact-attractor}. \end{proof} \begin{lemma}\label{phys:closed-counit} Suppose a filtered correspondence \(G\xleftarrow{q}F\xrightarrow{p}U\) occurring in \eqref{phys:integral-map} has classical evaluation a closed immersion \(i:F_{\mathrm{cl}}\into U_{\mathrm{cl}}\). Let \(h_U:U\to B\) and \(h_G:G\to B\) satisfy \(h_Up=h_Gq\). Its induction map, formed from the adjoint of \(\zeta\), is, after that isomorphism, \[ Rh_{U*}(i_*i^!\Phi_U\longrightarrow\Phi_U). \] Its cofiber is \(Rh_{U*}Rj_*j^*\Phi_U\), where \(j\) is the complementary open immersion. \end{lemma} \begin{proof} Put \(K_G=\Phi_G\La^{-I}\), and let \(c:q^*K_G\to p^!\Phi_U\) be the adjoint of \(\zeta:K_G\to Rq_*p^!\Phi_U\). The triangle identity is \(Rq_*c\circ\eta_q=\zeta\). Consequently induction is the composite \[ Rh_{G*}K_G\xrightarrow{Rh_{G*}\zeta} Rh_{U*}Rp_*p^!\Phi_U \xrightarrow{Rh_{U*}\epsilon_p}Rh_{U*}\Phi_U. \] Here \(p_*=p_!\) because its classical map is the stated closed immersion. The last arrow is therefore its counit. The closed/open localization triangle proves the cofiber assertion. No properness of \(h_U\) is needed. \end{proof} \subsection{One-sided constructibility on finite-type windows} We will apply the preceding operations on bounded finite-type stack windows. If \(V\to\mathcal X\) is a finite smooth atlas, \(p_m:V_m\to\mathcal X\) its ordinary nerve, and \(f_m:V_m\to B\) the maps to a finite-dimensional base, then a common lower standard bound \(b\) for \(p_0^*K\) is a lower bound for every \(p_m^*K\). Ordinary pullback is standard \(t\)-exact and \(Rf_{m*}\) preserves this lower bound. For \(C^m=Rf_{m*}p_m^*K\), the normalized term \(N^mC\) is a retract of \(C^m\), and the totalization tower has fiber \[ \fib(\operatorname{Tot}_m C\to\operatorname{Tot}_{m-1}C) \simeq (N^mC)[-m]\in D^{\ge b+m}. \] Thus each fixed degree depends on a finite partial totalization. Those finite terms have constructible cohomology. The finite amplitude difference between the standard and perverse structures on \(B\) gives finite constructible perverse cohomology and a one-sided bound. In a specified output range, finite convolutions can likewise be computed after a finite Postnikov truncation. \section{Harder--Narasimhan restrictions} \label{sec:comparison} We first construct the restriction maps that will be inverted in a perverse quotient. All constructions in this section are made separately on $X_0$ and $X_1$. The identification of either physical moduli problem with its critical presentation, including its orientation, is the one constructed in Section~\ref{sec:physical}. \subsection{The framed heart and its numerical bounds} Let $X=\Tot_C(N)$ be either threefold in \eqref{eq:physical-rows}, and let $Y=\mathbf P_C(\OO_C\oplus N)$ with infinity divisor $D_\infty=\mathbf P_C(N)$. In the usual tilted heart on $Y$ consider \begin{equation}\label{cmp:heart} \mathscr A_X=\langle\OO_Y, \operatorname{Coh}_{c,\leq1}(X)[-1]\rangle_{\mathrm{ex}}. \end{equation} The framing is an isomorphism along $D_\infty$, retained over the test base. In framing rank one, the weak slope of a framed object is a real parameter $t$. For every nonzero unframed coherent sheaf $F$, define $\mu(F[-1])=\chi(F)/\rk_C(F)$ when $\rk_C(F)>0$, and define $\mu(F[-1])=+\infty$ when $\rk_C(F)=0$. Weak semistability means $\mu(J)\leq\mu(Q)$ for every exact sequence $0\to J\to I\to Q\to0$. The heart criterion applies because $\operatorname{End}(\OO_Y)=\C$ and \[ \operatorname{Hom}(\OO_Y,F[-1])=0, \qquad \operatorname{Hom}(F[-1],\OO_Y) =\Ext^1_Y(F,\OO_Y)=0. \] The latter vanishing follows from codimension at least two and the local-to-global Ext spectral sequence. Thus the heart criterion of \cite[Proposition 3.6]{TodaPT} applies. To check noetherianity, consider a chain of epimorphisms. Its nonnegative framing and curve ranks eventually stabilize, so its remaining kernels are zero-dimensional objects shifted by $[-1]$. The ordinary $H^1$ sheaves form an eventually constant chain of quotients, after which the $H^0$ sheaves form an ascending chain inside the fixed reflexive hull of the first one. Coherence terminates this chain. The corresponding decreasing-chain condition for weak Harder--Narasimhan filtrations follows because a remaining nonzero zero-dimensional quotient has slope $+\infty$. These are the two chain conditions in \cite[Proposition 2.12]{TodaPT}. The boundary framing also removes infinitesimal Picard directions. For the projection $\pi:Y\to C$, relative projective-space cohomology gives \begin{equation}\label{cmp:boundary-trace} R\pi_*\OO_Y(-D_\infty)=0. \end{equation} This equality is preserved by derived base change. Hence the trace deformation complex preserving the specified boundary trivialization vanishes. Write $\mathscr X_{r,\chi}(c)$ for the physical window whose projected $L$-Higgs bundle is pure with minimum Euler slope greater than $c$. Its nonzero-section open is denoted by $\mathscr X_{r,\chi}^{\times}(c)$. \begin{lemma}\label{cmp:cyclic-bound} A compact cyclic structure sheaf $\OO_Z$ of curve rank $e$ satisfies \[ \chi(\OO_Z)\geq a_e. \] Every final $L$-Higgs Harder--Narasimhan quotient of its pushdown is again the pushdown of a cyclic physical quotient. Consequently its minimum Euler slope is at least $\alpha_e$. A positive-chamber semistable framed factor of curve rank $e$ has pure projected bundle with minimum Euler slope at least $\alpha_e$. \end{lemma} \begin{proof} Write $\Hom_L$ for morphisms of $L$-Higgs bundles. The negative line $A=K_CL^{-1}$ preserves the projected Higgs filtration. At a strictly decreasing cut $E'\subset E$, the off-diagonal transverse field lies in $\Hom_L(E',(E/E')\otimes A)=0$. In the nonsplit row, two local transverse lifts differ by a multiple of the projected field $\Theta$; their off-diagonal blocks therefore agree. The vanishing is global. The same argument applies to an equal-slope subobject of a semistable factor. Filter the cyclic module of $\OO_Z$ by monomials of total degree at most $j$ in the two normal variables and saturate this filtration on $C$. Let its positive successive ranks be $r_0,\ldots,r_s$. They occur consecutively: a stage at which the cyclic span stops growing is already preserved by both variables. Thus $s\leq e-1$, $\sum r_j=e$, and \[ \sum_{j=0}^{s}jr_j\leq\frac{e(e-1)}2. \] In either row, $\Sym^j(N^\vee)$ has a filtration with line quotients of degree at least $-j\ell$. Its minimum degree slope is therefore at least $-j\ell$. A rank-$r_j$ quotient has degree at least $-j\ell r_j$, and saturation increases degree. Therefore \[ \deg\pi_*\OO_Z\geq-\frac{\ell e(e-1)}2, \qquad \chi(\OO_Z)\geq-he-\frac{\ell e(e-1)}2=a_e. \] By the negative-line argument, a final Higgs quotient is a physical quotient. A quotient of a cyclic structure sheaf is another cyclic structure sheaf. Applying the preceding bound to its rank $f\leq e$ gives slope at least $\alpha_f\geq\alpha_e$. The coherent projection of an object with $H^0(I)=I_Z$ and $H^1(I)=Q$ is the exact sequence \begin{equation}\label{cmp:projection-sequence} 0\longrightarrow\pi_*\OO_Z\longrightarrow E_I \longrightarrow\pi_*Q\longrightarrow0. \end{equation} It follows by applying $R\pi_*(-D_\infty)[1]$ to the truncation triangle and using \eqref{cmp:boundary-trace}. Finite-$t$ semistability forces $E_I$ to be pure. Indeed its maximal zero-dimensional submodule lifts uniquely through the coherent framing row because $\Hom_S(TK_S^{-1},\OO_S)=\Ext^1_S(TK_S^{-1},\OO_S)=0$ on the smooth surface $S=\Tot_C(L)$. The lift is preserved by the field, and would give an unframed subobject of infinite slope. Every positive-rank quotient factor of $Q$ has slope at least $t$. Equation~\eqref{cmp:projection-sequence} now gives the lower bound $\min(\alpha_e,t)=\alpha_e$ when $t>0$. \end{proof} Purity and a lower minimum-slope bound are extension closed. Thus every extension of the good graded factors in the lemma lies in the prescribed window. This applies to the entire attracting fiber. The same assertion holds for the immediate negative chamber $t=-\epsilon$, $0<\epsilon<1/r$: a finite slope with denominator at most $r$ that is at least $-\epsilon$ is nonnegative. \subsection{Complete coherent flags} The diagonal resolution of Section~\ref{sec:physical} realizes framing rank at most one by coherent diagrams on $S$ with row \begin{equation}\label{cmp:coherent-row} 0\longrightarrow\OO_S\longrightarrow U \longrightarrow FK_S^{-1}\longrightarrow0, \end{equation} where $F$ denotes the surface projection of the physical sheaf. The diagrams also include the compatible global field arrows. The nonsplit torsor changes the extra arrow but not this exact row. Exact sequences and flat families are termwise exact; in the split case this is also the construction of \cite[Theorem 4.5.7 and Corollary 4.5.9]{TodaCat}. \begin{lemma}\label{cmp:complete-receiver} For fixed graded Hilbert polynomials and a fixed position of the framing, compatible complete coherent flags in \eqref{cmp:coherent-row} have a representable projective evaluation map to the middle object. The receiver includes nonsaturated subobjects and torsion in the graded quotients. \end{lemma} \begin{proof} Use the fixed compactification $\overline S=\mathbf P_C(\OO_C\oplus L)$. The compactly supported quotient $F$ extends by zero, as do its field targets, and the rank-one part of $U$ extends as $\OO_{\overline S}$. Form the product of the relative projective flag-Quot schemes of these coherent terms. The flag of $\OO$ is zero before the framing position and is all of $\OO$ afterwards. Require preservation by every field arrow and containment of this framing by setting the composites to the universal quotient sheaves equal to zero. These are closed incidence conditions. For example, a common sufficiently positive twist on the bounded flag-Quot family makes them zero entries of matrices between base-change-compatible vector bundles. The incidence conditions imply row exactness. Before the framed step, the flag term of $U$ has support of dimension at most one, so it intersects the torsion-free $\OO_{\overline S}$ trivially. Its map to the corresponding term of $FK_S^{-1}$ is injective; the prescribed Hilbert polynomials make it an isomorphism on every fiber. After the framed step, divide by the contained $\OO$ and apply the same argument. Flatness of the flag terms and quotients upgrades these fiberwise identifications to exact rows after arbitrary base change. All changes from $\OO$ are supported on $\supp F$. Thus the boundary flag is the fixed zero or $\OO$ flag. The resulting closed incidence subscheme is projective over the middle object. Its field arrows are induced by those of the middle diagram, including on a nonsaturated inclusion. \end{proof} For the rank-zero framed semistable factor at finite $t$, set $a_0=0$. This factor is the vacuum $\OO_Y$, with Euler characteristic $\chi_0=0$. \begin{lemma}\label{cmp:finite-types} For fixed $r,\chi,c$, a finite list of coherent receivers suffices for the Harder--Narasimhan decomposition in a sufficiently large positive chamber. A single chamber parameter can be chosen after the complete receivers, including their torsion quotients, have been formed. \end{lemma} \begin{proof} Choose a rational $\alpha\leq\min(c,\alpha_r,0)$. If $E'$ is a pure rank-$f$ subobject of the projected middle bundle, its torsion-free quotient has Euler characteristic at least $\alpha(r-f)$. Hence \begin{equation}\label{cmp:prefix-bound} \mu(E')\leq\frac{\chi-\alpha(r-f)}{f},\qquad b=\max_{1\leq f\leq r}\frac{\chi-\alpha(r-f)}f. \end{equation} For $t>\max(b,0)$ the first factor is framed; an earlier unframed factor would violate \eqref{cmp:prefix-bound}. Write its class as $(1,e,\chi_0)$ and the successive unframed classes as $(0,r_j,\eta_j)$. They satisfy \begin{equation}\label{cmp:finite-list} \begin{gathered} a_e\leq\chi_0\leq\chi-\alpha(r-e),\qquad \eta_j>cr_j,\qquad \sum_jr_j=r-e,\qquad \chi_0+\sum_j\eta_j=\chi. \end{gathered} \end{equation} This is a finite list. The strict inequalities are retained in the list; $\alpha$ is only a bounding device. The projected window is bounded. Its Higgs Harder--Narasimhan numerical list is finite, and successive slopes in the ordinary bundle filtration of a semistable $L$-Higgs factor have gaps at most $\ell$. In particular a rank-$f$ semistable Higgs factor of Euler slope $\mu$ has underlying minimum degree slope at least $\mu+h-(f-1)\ell$. The coherent extensions and compatible arrows in \eqref{cmp:coherent-row} form finite-type families, being represented by bounded perfect relative Hom complexes. Form the projective receivers of Lemma~\ref{cmp:complete-receiver} for the finite list \eqref{cmp:finite-list}. Their graded coherent sheaves form bounded families. For a common large twist $k$, the length of a maximal zero-dimensional subsheaf $T_0(G)$ is at most $h^0(G(k))$, uniformly bounded. For each possible length, its possible positions are covered by the relative Quot scheme with the complementary Hilbert polynomial. This avoids requiring the maximal torsion subsheaf itself to be flat in the family. Absorb torsion backwards in each flag; each packet moves left through at most $r$ positive-rank steps. Repeating the finite Quot construction still gives bounded families. If $H$ is a pure Higgs subobject of $G/T_0(G)$, then \[ \chi(H(k))\leq h^0(H(k)) \leq h^0((G/T_0(G))(k))\leq h^0(G(k)), \] since $H^1(T_0(G))=0$. Thus all slopes occurring after torsion absorption and Higgs refinement have a common upper bound $b_{\mathrm{fl}}$. Choose a generic $T$ above $b$, $b_{\mathrm{fl}}$, and the final stable-pair thresholds for the finitely many framed classes. These thresholds can be obtained directly: if $t>\chi-a_r$, a positive-rank part of $H^1(I)$ would have Euler characteristic at least $t$, contradicting \eqref{cmp:projection-sequence} and Lemma~\ref{cmp:cyclic-bound}. The remaining $H^1(I)$ is zero-dimensional, and $\Ext^2(Q,\OO_X)=0$ on the smooth threefold lifts the truncation triangle to an ordinary pair. Absence of zero-dimensional subobjects makes its sheaf pure. Conversely the pure subobject slopes of a bounded stable-pair family are uniformly bounded. Thus $T$ is a stable-pair chamber simultaneously for every receiver under consideration. \end{proof} \subsection{The closed-stratum counit} Order the finite types first by increasing framed curve rank $e$, then by decreasing anchored tail polygon, choosing a linear extension of this order. Its vertices are \[ (e,\chi_0),\quad (e+r_1,\chi_0+\eta_1),\quad\ldots,\quad(r,\chi). \] The initial height $\chi_0$ is part of the polygon. Absorbing torsion of length $d$ into the preceding flag member raises the preceding vertex by $d$ and leaves later vertices fixed; this includes the initial vertex. Higgs refinement replaces the result by its concave majorant. If the framed prefix is unstable, its first framed factor has smaller curve rank and is earlier in the primary order. The uniform choice of $T$ ensures that every unframed tail slope is less than $T$. Consequently every bad displayed flag lies in an earlier true Harder--Narasimhan type. Delete the union of the earlier proper images and call the resulting open $U_\tau$. Its receiver has exactly the good decreasing semistable flags. Conversely, a good flag is the unique Harder--Narasimhan flag. This identifies the open substacks, since an open substack is determined by its geometric points. Evaluation is now proper and universally injective. Its relative tangent is filtered by the forbidden $\Hom$ groups between decreasing semistable factors, and these vanish. It follows that evaluation is a representable closed immersion \begin{equation}\label{cmp:closed-hn} i_\tau:S_\tau\into U_\tau. \end{equation} This also gives the unique relative flag over nonreduced bases. Use $U_\tau$ also for the derived open of the physical window with this classical truncation. Choose ordered integral weights for this flag type, with weight zero on its unique framed graded piece. Let $G_\tau$ be the open product of its derived good graded-factor stacks. Using \eqref{phys:filtered-stacks}, define \[ \mathfrak F_\tau=\operatorname{Filt}(U_\tau)_{\tau} \mathop{\times}_{\operatorname{Grad}(U_\tau)}G_\tau, \qquad G_\tau\xleftarrow{q}\mathfrak F_\tau\xrightarrow{p}U_\tau. \] The framing at infinity stays fixed in this mapping-stack problem. Proposition~\ref{phys:critical-equivalence}, applied to universal finite exact diagrams, identifies its truncation with the coherent flag rows and compatible fields constructed above. Every extension of the good factors lies in the cutoff window by Lemma~\ref{cmp:cyclic-bound}. Its decreasing semistable flag is its Harder--Narasimhan flag, so it lies in the current open $U_\tau$. Conversely a filtered object with these good graded terms has that coherent flag. These are open conditions on the common exact-diagram problem, determined by geometric fibers; hence the identification holds over derived test bases. The Rees construction realizes the chosen weights, and its classical evaluation is $t_0(p)=i_\tau$ in \eqref{cmp:closed-hn}. Let $K_{G_\tau}$ denote the product DT coefficient with the moving weight normalization in integral localization. The oriented localization map of Proposition~\ref{phys:integral-localization}, equivalently \cite[Corollary 7.19]{KPS} with its monodromic Hodge construction, is \begin{equation}\label{cmp:zeta} \zeta:K_{G_\tau}\xrightarrow{\sim}Rq_*p^!\Phi_{U_\tau}. \end{equation} The good product orientation is the one obtained from the same physical orientation; its moving determinant and all refinement maps are those of \cite[Proposition 8.34]{KPS}. \begin{lemma}\label{cmp:counit} Let $h_{U_\tau}$ and $h_{G_\tau}$ be the spectral maps to $\cA_r^L$, using spectral addition on the graded factors of $G_\tau$. After Hitchin direct image, the induction map of the good factor coefficient is the closed-stratum costalk counit \begin{equation}\label{cmp:costalk} Rh_{U_\tau,*}i_{\tau,*}i_\tau^!\Phi_{U_\tau} \longrightarrow Rh_{U_\tau,*}\Phi_{U_\tau}. \end{equation} Its cofiber is the direct image from the next open. \end{lemma} \begin{proof} Let $c:q^*K_{G_\tau}\to p^!\Phi_{U_\tau}$ be the adjoint of $\zeta$. The triangular identity says exactly that $Rq_*c\circ\eta_q=\zeta$. Since $h_{U_\tau}p=h_{G_\tau}q$, the integrated induction map is \[ Rh_{G_\tau,*}K_{G_\tau} \xrightarrow{Rh_{G_\tau,*}\zeta} R(h_{U_\tau}p)_*p^!\Phi_{U_\tau} \xrightarrow{Rh_{U_\tau,*}\epsilon_p} Rh_{U_\tau,*}\Phi_{U_\tau}. \] Mixed Hodge modules are insensitive to derived nilpotent thickenings of the underlying map. With $t_0(p)=i_\tau$, its second arrow is \eqref{cmp:costalk}. The localization triangle for the complementary open $j$ has cofiber $Rh_{U_\tau,*}Rj_*j^*\Phi_{U_\tau}$. Products, refinements, and restrictions use these same units and counits, so their comparison cells are retained. \end{proof} In the split row, apply $\phi_{g_{\cA}}$ after these maps have been constructed. The trace potential is additive under finite spectral addition. Moreover the split coefficients are equivariant for a normal dilation under which $g_{\cA}$ has weight two. After the degree-two Kummer cover they are locally acyclic in the function direction away from zero. Thus the relevant vanishing cycles of a sum are supported where every summand has value zero. Proper exchange for finite addition and monodromic Thom--Sebastiani give exactly the lower-rank source images used in $\cI_r$. \subsection{Vacuum removal and finite wall transport} For $i=0,1$, let $\mathscr X_{i,r,\chi}^{\times}(c)$ be the physical nonzero-section window in the row $X_i$, with its map $h_i$ to $\cA_r^L$. Set \[ W_{i,r,\chi}^{\times}(c)=R(h_i)_* \bigl(\Phi_{\mathscr X_{i,r,\chi}^{\times}(c)}\otimes\La^\chi\bigr). \] We suppress $i$ in assertions proved separately in both rows. Let $R_{r,\chi}$ denote the corresponding stable-pair restriction, with the same coefficient $\Phi\La^\chi$. \begin{proposition}\label{cmp:restrictions} For every $c\leq c_r$, the restriction to the stable-pair chamber has cofiber in $D_{\cI_r}$. The same is true for restriction to every generic $t>\max(\chi/r,0)$ and for the nested-window restriction \begin{equation}\label{cmp:rho} \rho_\chi(c_1,c_2): W_{r,\chi}^{\times}(c_1)\longrightarrow W_{r,\chi}^{\times}(c_2),\qquad c_1\max(\chi/r,0)$, the wall stack has no $s=0$ object: its vacuum subobject would have slope $\lambda$ and its unframed quotient slope $\chi/r$. A rank-zero framed factor would be the vacuum and would force $\lambda=\chi/r$. Therefore both ranks are strictly between $0$ and $r$, so all closed terms lie in $D_{\cI_r}$. The octahedral triangle transports \eqref{cmp:large-vacuum} through the finitely many walls between $t$ and $T$. Two cutoffs $c_10\qquad(1\leq ec_r$. This proves containment for both complete intrinsic chamber stacks. \end{proof} \begin{lemma}\label{cmp:endpoints} The core surface window is empty for $\chirM_r$. For the ordinary source modulo $\cS_r$, the upper vanishing endpoint is instead $\chi>-a_r$. \end{lemma} \begin{proof} Since $\ell$ is even, $\alpha_r$ is an integer. A rational slope of denominator at most $r$ below $\alpha_r$ is at most $\alpha_r-1/r$. Thus $\mu_{\min}>c_r$ is equivalent to $\mu_{\min}\geq\alpha_r$. If $\chih+(r-1)\ell$. Increase this bound to the projective framing threshold. Consider a semistable physical object in the immediate chamber $t=\chi/r+\epsilon$, $0<\epsilon<1/r^2$. If its projected Higgs bundle $E$ were unstable, its highest Higgs Harder--Narasimhan factor $H$ would satisfy $\mu(H)>\chi/r$. The denominators are at most $r$, so $\mu(H)>t$. The preceding bound, applied to the semistable bundle $H$, gives $H^1(C,H)=0$. Serre duality then gives $\varepsilon|_H=0$ for the dual framing $\varepsilon:E\to K_C$. The moment relation makes the off-diagonal transverse field $H\to(E/H)A$ a Higgs morphism. In the nonsplit row the local transverse lifts differ by a multiple of $\Theta$, so their off-diagonal blocks agree. Negativity of $A$ and the decreasing Higgs slopes force this morphism to vanish. Thus $H$ is an unframed physical subobject of slope greater than $t$, a contradiction. Consequently $E$ is semistable, $H^1(C,E)=0$, and $\varepsilon=0$. If $s=0$, the vacuum subobject and its full unframed quotient would give $t\leq\chi/r$, a contradiction. Thus $s\neq0$. The same negative-$A$ Hom vanishing preserves every equal-slope subobject. The remaining framed stability condition is the Joyce--Song condition: the nonzero section is contained in no proper Higgs subbundle of the same slope. Denote this pair space by $\cP^{\mathrm{JS}}$. The ordinary pair stack is therefore the smooth stable framed open, projective over the positive semistable Hitchin moduli; the projective framing construction is described in \cite[Definition 2.11 and (2.37)]{PToda}. The relative framing map is smooth because $H^1(E)=0$, and the positive Higgs stack is smooth. The framed stability kills stabilizers. Thus its Hitchin map is proper and its two critical presentations are $T^*[-1]\cP^{\mathrm{JS}}$ and $\Crit(\cP^{\mathrm{JS}},g)$ with the orientations of Section~\ref{sec:physical}. On this open $\delta$ and the proper exchange $e_h$ are isomorphisms. Proposition~\ref{cmp:restrictions} then gives the upper endpoint for $K_\chi$. The ordinary-source endpoint is needed only in the split row. Use the full source, with the zero section retained, and work modulo $\cS_r$. Its restriction to a sufficiently large positive chamber is an isomorphism in this quotient by Proposition~\ref{cmp:restrictions}. The same proper-receiver argument transports it across every wall between that chamber and $-\epsilon$, where $0<\epsilon<1/r$. Here one permits the rank-zero framed factor $\OO$ and its single rank-$r$ unframed complement. The latter has allowed support by the independent Proposition~\ref{source:unframed-support}; every other framed factor has smaller rank and is covered by the source induction. At the zero wall the purity argument is unchanged: if the unframed quotient has average zero, its nonnegative finite slopes and its torsion length force all slopes and that length to vanish. All walls in this bounded interval have the finite numerical lists and complete coherent receivers constructed above. The transport uses the independent rank-$r$ unframed support theorem and the smaller-rank source induction. Lemma~\ref{cmp:split-reflection} places both immediate chambers in the core and identifies their normalized direct images by \eqref{cmp:normalized-reflection}. If $\chi>-a_r$, then $-\chi0$, its special fiber vanishes on $D$ and contradicts \eqref{cmp:full-frame}; hence $m=0$. Its determinant is a section of a degree-zero line, nonzero on $D$, and therefore nowhere zero on the special fiber. It is an isomorphism, and uniqueness follows from torsion-freeness over the DVR. Fields and sections extend compatibly for the same reason. Universally, the diagonal can also be described as the zero locus of the boundary obstruction of the prescribed frame in $R^1\pi_*\mathcal Hom(E,F)(-D)$; the vanishing of $R^0$ makes the extension unique. This proves the closed-diagonal statement over arbitrary bases and also removes stabilizers. In the framed product of the initial level-two and final level-zero endpoints over $T$, take the scheme-theoretic closed union $\widetilde Z$ of the two word images and the marked graph $\widetilde{\mathfrak D}$. A word is proper over its final endpoint; its map to the product is proper because the initial endpoint is separated. Each image, their union, and the graph are therefore proper over the final endpoint. Quotient by the common diagonal frame group $G=\operatorname{Res}_{D/T}\GL_r$: \[ Z=[\widetilde Z/G],\qquad \mathfrak D=[\widetilde{\mathfrak D}/G]. \] The right map $b:Z\to X_{\chi,0,T}$ is representable proper, as is checked after the final frame atlas. The parameter $T$ is retained in both endpoints and in this proper pushforward. The intrinsic maps and homotopies will descend along $T\to C\times C$ before the proper curve trace is applied. The graph $\mathfrak D$ is smooth of dimension $q=d_\chi+\dim T-1$. Smooth descent applies to $G$, including its unipotent part. The equalities of functions and residue coordinates descend to the scheme-theoretic images and their union. If $a$ is the left map, then on the carrier itself \begin{equation}\label{cmp:carrier-identities} a^*F_{\mathrm{initial}}=b^*F_{\mathrm{final}}, \qquad a^{-1}S_{\mathrm{initial}} =b^{-1}S_{\mathrm{final}} \end{equation} as functions and closed schemes. For ordinary Hodge Borel--Moore theory use the mapping spectrum \[ \operatorname{BM}_q(Z)= \operatorname{Map}(\Q_Z,\omega_Z\LL^q). \] Each derived word is a derived base change of the regular diagonal of the smooth framed middle endpoint. Both are quasi-smooth of virtual dimension $q$. The fundamental class is the coordinate-class specialization, localization boundary, and inverse Thom map; its composition and base-change paths are those of \cite[\S2, Constructions 3.1, 3.4, 3.6 and Theorems 3.12--3.13]{Khan}. In this regular-diagonal situation the binary path is also the double-deformation diagram of \cite[Sections 3.2.19--3.2.21]{DJK}. Swapping the two odd coordinate classes contributes $-1$, and swapping the two odd localization boundaries contributes $-1$. They cancel. This fixes the binary path, including its sign. Transport the class by the nil-invariant equivalence induced by $t_0W\to W$ and properly push the classical truncation to $Z$. The bicartesian equivalence above identifies the two classes off $\mathfrak D$, with these binary paths. Localization lifts their difference to $\operatorname{BM}_q(\mathfrak D)$. Smooth purity identifies its underlying mapping space with \[ \operatorname{Map}(\Q,R\Gamma(\mathfrak D,\Q)). \] There is no negative cohomology, so the space is discrete with components $H^0(\mathfrak D,\Q)$, one scalar per connected component. This reduction takes place in Hodge theory. On the dense source open where $\Theta(y)$ is regular semisimple and every eigencomponent of $s(y)$ is nonzero, removal first is impossible. Creation followed by removal has one branch for each eigenline. On its $j$th chart the condition is exactly \begin{equation}\label{cmp:eigenline-equation} (y-x)s_j(y)=0. \end{equation} The second factor is a unit, so the branch is smooth of dimension $q$. Its quasi-smooth virtual dimension is also $q$; hence the obstruction space vanishes there, and its fundamental class has complex multiplicity $+1$. Evaluation surjectivity for the meromorphic bundles makes this open meet every component of the smooth graph. The eigenline cover has degree $r$, and its trace is $r$. We conclude \begin{equation}\label{cmp:bm-relation} [W_+]-[W_-]=r[\mathfrak D] \quad\hbox{in }\pi_0\operatorname{BM}_q(Z), \end{equation} with the specified binary composition paths. The localization lift is locally constant, so the dense calculation determines it. \subsection{Lifting the class to the entire arrow} Let $X\xleftarrow a Z\xrightarrow bY$ be such a carrier with $b$ proper and $X,Y$ smooth. If $d=\dim X$, dualizing purity and proper adjunction give \begin{align}\label{cmp:bm-map} \operatorname{BM}_q(Z) &\simeq\operatorname{Map}(a^*\omega_X, b^!\omega_Y\LL^{q-d})\nonumber\\ &\simeq\operatorname{Map}(b_*a^*\omega_X, \omega_Y\LL^{q-d}). \end{align} Pushing a class properly inserts the pullback unit followed by the proper counit. Thus \eqref{cmp:bm-map} represents the original convolution maps and their binary paths. Let $i_X:S_X\into X$ and $i_Y:S_Y\into Y$ be the residue-zero inclusions, and put $\Gamma_X=i_{X,*}i_X^!$ and $\Gamma_Y=i_{Y,*}i_Y^!$. Write $K_X,K_Y$ for the ambient dualizing normalizations, absorbing the indicated Tate factors. Put $A_0=b_*a^*\Gamma_XK_X$ and $A_1=b_*a^*K_X$. By \eqref{cmp:carrier-identities}, $A_0$ is supported on $S_Y$. The mapping spectrum of squares to $\Gamma_YK_Y\to K_Y$ is \begin{equation}\label{cmp:arrow-coreflector} \operatorname{Map}(A_0,\Gamma_YK_Y) \mathop{\times}\limits^h_{\operatorname{Map}(A_0,K_Y)} \operatorname{Map}(A_1,K_Y). \end{equation} The first map to the base is an equivalence, since $\Gamma_Y$ is the coreflector to supported objects. Consequently \eqref{cmp:arrow-coreflector} projects equivalently to $\operatorname{Map}(A_1,K_Y)$. Its inverse produces the supported map and its square homotopy with contractible choice. It is additive and preserves the binary composition paths. Write $\phi_X,\phi_Y,\phi_Z$ for vanishing cycles of the common function on the three spaces, and $a_0,b_0$ for the restrictions of $a,b$ to its zero fibers. Apply vanishing cycles to the entire square. Its source is $\phi_X\Gamma_XK_X$. The critical correspondence uses \[ b_{0,*}a_0^*\phi_XK_X\longrightarrow b_{0,*}\phi_Za^*K_X\xrightarrow{\sim} \phi_Yb_*a^*K_X\longrightarrow\phi_YK_Y. \] The first exchange is used only in its forward direction; only the proper exchange along $b$ is inverted. In the residue presentation, let $V$ be the smooth ambient, $S=V|_{t_0\cP}$, $i:S\into V$, and $p:S\to t_0\cP$ the dual-residue vector bundle. Use $\mathscr D_\chi=\omega_V\LL^\chi$. The supported arrow is \begin{equation}\label{cmp:ambient-counit} \phi_F(i_*i^!\mathscr D_\chi)\longrightarrow\phi_F\mathscr D_\chi. \end{equation} Both vertices are supported on $S$, because derivatives in the dual residue variables are the residue equations. The source identification uses \[ i^!\omega_V=\omega_S=p^*\omega_{t_0\cP}\LL^{-e}, \qquad\phi_{p^*g}p^*\simeq p^*\phi_g, \qquad\id\simeq Rp_*p^*. \] If $d_0=B_r+2e$, its pushdown is \begin{equation}\label{cmp:counit-source-normalization} Rp_*\phi_{p^*g}i^!\mathscr D_\chi =\phi_g\omega_{t_0\cP}\LL^{\chi-e} =\phi_g\Omega_{\cP}\LL^{\chi/2}\LL^{-d_0/2}. \end{equation} Its target is $R\pi_{g,*}\Phi_{\Crit(\cP,g)} \LL^{\chi/2}\LL^{-d_0/2}$. Thus \eqref{cmp:ambient-counit} is the local $\delta$ in \eqref{cmp:delta}, with the same common factor at both vertices. Closed base change identifies its supported correspondence with the ordinary $\Omega_{\cP}$ correspondence after this vector-bundle pushdown. Finally the nonproper exchange $e_h$ is the right mate of the forward pullback exchange. If $h:X\to B$ and $h_0:X_0\to B_0$ are the maps before and after the zero fiber, then on $K$ the mate is \[ \phi_Bh_*K\longrightarrow h_{0,*}h_0^*\phi_Bh_*K \xrightarrow{\operatorname{Ex}_h} h_{0,*}\phi_Xh^*h_*K\longrightarrow h_{0,*}\phi_XK. \] The outer maps are the pullback unit and counit. Composition of forward exchanges and the adjunction triangle identities determine its composition homotopies. For the ordinary action of a correspondence $u$, use those same homotopies in \[ T(u):h_{X,*}K_X\xrightarrow{\eta_a} (h_Xa)_*a^*K_X=h_{Y,*}b_*a^*K_X \xrightarrow{u}h_{Y,*}K_Y, \] its mate identities give \[ e_{h_Xa}=(h_{X,*}e_a)(e_{h_X}a_*),\qquad e_a\phi_X(\eta_a)=a_*\operatorname{Ex}_a(\eta_a\phi_X). \] They identify the square for $e_h$ with the square already obtained for the local $\delta$. The complete diagram has rows \[ \phi_{g_{\cA}}Rh_*\Omega\longrightarrow Rh_*\phi_g\Omega\longrightarrow R(h\pi_g)_*\Phi, \] and includes the two square homotopies. Composing the horizontal arrows is an exact functor on enhanced diagram categories. It takes \eqref{cmp:bm-relation} to a relation on precisely $\Delta$, and then on its cofiber. \subsection{Independence of the auxiliary presentation} Enlarging the divisor adds an acyclic equation-variable block $[W\xrightarrow d E]$. On a splitting torsor, a triangular change of equations and its dual reduce the added potential to $\langle d(c),\zeta\rangle$ with support $c=0$. Differences between extensions $\beta_D$ belong to the residue ideal: the principal-part expansion of $\Tr\Theta^2$ expresses them as linear combinations of the residue coordinates. Translating the corresponding dual variables removes them. For one hyperbolic pair write $c=z_2+\mathrm i z_1$, $\zeta=z_2-\mathrm i z_1$. Then $c\zeta=z_1^2+z_2^2$. Projection of the positive real thimble to $\zeta$ preserves orientation, whereas projection to $c$ reverses it. Thus the raw support-counit scalar for a rank-$k$ block in the compact dimensional-reduction convention is $(-1)^k$. Equivalently this is $\binom{2k}{2}\equiv k\pmod2$ in the duality normalization of \cite[Lemma 3.4]{Kinjo}. The grouped orientation form is \[ (-1)^{k(k-1)/2}(-\mathrm i/2)^k dc_1\cdots dc_k\,d\zeta_1\cdots d\zeta_k. \] For invertible $d$, the contraction root contains $\det d$; its square is the Hessian determinant with this same parity. Changes of bundle frame act inversely on the dual half, so the total determinant is one. In our presentations $e=mr(r+1)$ and every enlargement rank are even. Hence the raw stabilization scalar is $+1$, and the existing local $\delta$ and its binary paths are preserved. Splittings and potential extensions form affine torsors. On the fixed intrinsic endpoint objects their pullback is fully faithful, by $\id\simeq Rp_*p^*$ and the projection formula. The preceding transformations therefore descend the maps and their homotopies. To compare a previous one-point divisor to the two-point family, use their sum, which need only avoid the active mark of that one operator. Descend the one-step comparison first, then compose on the whole $C\times C$. Changing bounded bundle opens is handled by supported open excision of \eqref{cmp:ambient-counit}. Both vertices are supported on the complete residue-zero window. Restriction to an open containing it is fully faithful on the objects and their mapping spaces. On a Hecke diagram first restrict the proper target, then take its full inverse image. If a further source excision is needed, delete the proper image of its bad closed inverse image; that image misses the target residue support. This gives a common refinement preserving the proper arm, fundamental class, coreflector square, and binary path. The three directed presentations consequently define the same intrinsic arrows and restriction squares. \begin{proposition}\label{cmp:weyl-relation} There are degree-zero creation and removal morphisms on the arrows $\Sigma_\chi(c)$, \[ E_\chi^c:\Sigma_\chi(c-1)\longrightarrow\Sigma_{\chi+1}(c), \qquad F_\chi^c:\Sigma_{\chi+1}(c-1)\longrightarrow\Sigma_\chi(c), \] natural for the cutoff restrictions, with relation \begin{equation}\label{cmp:weyl-buffer} F_\chi^c E_\chi^{c-1} -E_{\chi-1}^c F_{\chi-1}^{c-1} =r\rho_\chi(c-2,c). \end{equation} The same holds on $K_\chi(c)$ and on the normalized full ordinary source \begin{equation}\label{cmp:ordinary-source} \mathcal A_\chi(c)=Rh_* \omega_{t_0\cP_{r,\chi}^{\mathrm{all}}(c)} \LL^{\chi+B_r/2}, \end{equation} where the zero section is allowed. \end{proposition} \begin{proof} At a fixed mark creation has target $\Sigma_{\chi+1}\LL^{-1}$; removal already has degree zero. Indeed the ambient dimensions in \eqref{cmp:ambient} and \eqref{cmp:bm-map}, together with the common factor $\LL^{-d_0/2}$ in \eqref{cmp:counit-source-normalization}, give these twists. The class relation on $C_x\times C_y$ is therefore \[ f_\chi^c(y)e_\chi^{c-1}(x) -e_{\chi-1}^c(x)f_{\chi-1}^{c-1}(y) =r\rho_\chi(c-2,c)\boxtimes\operatorname{cl}(\Delta_C). \] Fix $y$ and integrate the moving mark $x$ by the proper trace of the whole curve $R\Gamma(C,\Q)\to\LL$. It cancels $\LL^{-1}$ and sends $\operatorname{cl}(y)$ to $1$. This gives \eqref{cmp:weyl-buffer}. The graph term is the inclusion of the level-zero endpoint into level two, so its action is the restriction $\rho$. Naturality follows by proper target base change and the preceding supported excision. The enhanced cofiber functor gives the same relation on $K_\chi$. For the ordinary source use the same smooth meromorphic ambient with $s=0$ allowed. The regular-semisimple open used in \eqref{cmp:eigenline-equation} is still dense in every component. Apply the coreflector but not vanishing cycles. Its pushed coefficient is $\omega_{t_0\cP}\LL^{\chi-e}$, differing from \eqref{cmp:ordinary-source} by the single common factor $\LL^{-B_r/2-e}$. This cancels on every edge. The same trace gives the ordinary-source relation, without requiring the holomorphic pair stack itself to be smooth. \end{proof} \subsection{Finite-band vanishing} \begin{lemma}\label{cmp:weyl-nilpotence} Let $V=\bigoplus_{a\leq\chi\leq b}V_\chi$ in a $\Q$-linear abelian category. If degree-zero morphisms $E,F$ raise and lower $\chi$ and satisfy $[F,E]=r\id$ for an integer $r>0$, then $V=0$. \end{lemma} \begin{proof} $E$ is nilpotent, and $[F,E^m]=\sum_{j=0}^{m-1}E^j[F,E]E^{m-1-j} =mrE^{m-1}$. Starting with $E^m=0$ and decreasing $m$ gives $\id=0$. \end{proof} \begin{theorem}\label{cmp:comparison} For every $r,\chi$ and $c\leq c_r$, the cofiber of the original comparison arrow \eqref{cmp:delta} belongs to $D_{\cI_r}$. \end{theorem} \begin{proof} Fix a perverse degree $j$, and let $q_r:\MMHM(\cA_r^L)\to\MMHM(\cA_r^L)/\cI_r$ denote the Serre-quotient functor on the perverse heart. Proposition~\ref{cmp:restrictions} makes every cutoff restriction an isomorphism. Use its unique inverse to transport the operators in Proposition~\ref{cmp:weyl-relation} to \[ V_\chi=q_r\pH^jK_\chi(c_r). \] Restriction naturality makes their relation $[F,E]=r\id$. Lemma~\ref{cmp:endpoints} leaves a finite characteristic band, independent of $j$. Lemma~\ref{cmp:weyl-nilpotence} gives $V_\chi=0$ for all $\chi$. All cutoffs follow by their restrictions. Only the finite band and its adjacent values were needed for the three directed bundle levels; iteration of $E$ now occurs inside the quotient category. \end{proof} \begin{theorem}\label{cmp:source-support} Every perverse constituent of the normalized split stable-pair direct image belongs to $\cS_r$. \end{theorem} \begin{proof} We prove by induction on $r$ the support assertion for every full ordinary window $\mathcal A_\chi(c)$, $c\leq c_r$, and for its complete chamber and wall coefficients with $t>-1/r$. The rank-zero framed factor is the vacuum. The positive unframed factors have allowed supports by Proposition~\ref{source:unframed-support}. In the full-source Harder--Narasimhan restrictions, every smaller framed-rank factor has allowed supports by this induction. For $e-1/r$ is contained in $t>-1/e$; nested cutoff restriction compares a lower-rank factor at cutoff $c$ with its own core. Finite spectral addition sends the concatenated partition supports to the corresponding $\Sigma_\nu$. The only one-factor rank-$r$ unframed term is exactly the independent positive input. Thus source restrictions are isomorphisms in the quotient by $\cS_r$. The lower empty endpoint and the source reflection of Lemma~\ref{cmp:endpoints} leave the band $[a_r,-a_r]$. Apply the ordinary-source relation of Proposition~\ref{cmp:weyl-relation} and Lemma~\ref{cmp:weyl-nilpotence} in each perverse degree. This proves support for the full core source, and nested restriction gives it for every $c\leq c_r$. For $t>-1/r$, the slope-denominator argument of Lemma~\ref{cmp:split-reflection} and the cyclic bound place the complete chamber and wall stacks inside the core. Their restrictions from the full window are successive closed-stratum costalk triangles from the proper-receiver wall argument in the proof of Proposition~\ref{cmp:restrictions}. Each closed term has either a smaller positive framed rank, or the framed vacuum and a rank-$r$ unframed factor. The former is covered by the induction hypothesis; the latter by Proposition~\ref{source:unframed-support}. Finite addition and the Serre property place every closed term in $D_{\cS_r}$. The triangles therefore give support for these chamber and wall coefficients, completing the induction. Dimensional reduction identifies the ordinary source \eqref{cmp:ordinary-source} with the split physical coefficient. Taking a sufficiently large positive chamber gives the stable-pair assertion. \end{proof} \begin{corollary}\label{cmp:pt-difference} Let $R_{0,r,\chi}$ and $R_{1,r,\chi}$ be the normalized stable-pair direct images from $X_0$ and $X_1$. Then \begin{equation}\label{cmp:bounded-pt-difference} [R_{1,r,\chi}]-[\phi_{g_{\cA}}R_{0,r,\chi}] \in K_0(\cI_r). \end{equation} \end{corollary} \begin{proof} The available morphisms are the original window comparison $\phi W_0\to W_1$ and the two restrictions $\phi W_0\to\phi R_0$, $W_1\to R_1$. Their cofibers lie in $D_{\cI_r}$ by Theorem~\ref{cmp:comparison} and Proposition~\ref{cmp:restrictions}. They form a zigzag of isomorphisms in each perverse quotient. The stable-pair endpoints are bounded proper direct images by Proposition~\ref{fock:properness}. Taking their Grothendieck classes and applying the Serre localization sequence gives \eqref{cmp:bounded-pt-difference}. \end{proof} \section{The positive source identity} \label{sec:source} We first compute the split stable-pair coefficient on $X_0$. Let $h_{0,r,\chi}:P_{r,\chi}(X_0)\to\cA_r^L$ be the positive spectral map, and set \begin{equation}\label{source:R-L} R^L_{r,\chi}=R(h_{0,r,\chi})_* (\Phi_{P_{r,\chi}(X_0)}\otimes\La^\chi). \end{equation} Dimensional reduction and the ordinary-source argument of Theorem~\ref{cmp:source-support} put every perverse cohomology object of \eqref{source:R-L} in $\cS_r$. Choose any Euler characteristic $\chi_*$ coprime to $r$. Write $M_L(r,\chi_*)$ for the smooth stable positive Higgs moduli space, and let $h_L:M_L(r,\chi_*)\to\cA_r^L$ be its Hitchin map. Put \begin{equation}\label{source:positive-perverse} P_i^L=\pH^iR(h_L)_*\Q_{M_L(r,\chi_*)}[B_r+1]. \end{equation} This Hodge module is independent of $\chi_*$ by \cite[Theorem 0.4]{MSChi}. Define \begin{equation}\label{source:b-gamma} b_r=\frac{\ell r(r+1)}2-rh,\qquad \gamma_r=1+rh+\frac{\ell r(r-1)}2. \end{equation} Thus $\dim\cA_r^L=b_r$, $\gamma_r$ is the arithmetic genus of a rank-$r$ spectral curve, and $b_r+\gamma_r=B_r+1$. Let $x$ be the formal variable recording Euler characteristic. \begin{theorem}\label{source:positive-formula} The complexes $R^L_{r,\chi}$ are ordinary pure Hodge complexes of weight zero. In the coefficientwise completed Grothendieck group of mixed Hodge modules on the positive Hitchin bases, \begin{equation}\label{source:positive-series} \begin{split} 1+\sum_{r\geq1,\chi}R^rx^\chi[R^L_{r,\chi}] =\Exp_{\operatorname{add}}\!\left( \sum_{r\geq1}R^r\sum_i\sum_{u,v\geq0} x^{i+1+u+v} [P_i^L[-i-1-2v]\otimes\TT^{v-B_r/2}] \right). \end{split} \end{equation} Only $u,v$ are constrained to be nonnegative; $i$ runs through the finite perverse range in \eqref{source:positive-perverse}. The exponential includes the derived symmetric-group actions. \end{theorem} \subsection{The unframed input and its hypotheses} We begin with the unframed support statement used in Section~\ref{sec:comparison}. \begin{proposition}\label{source:unframed-support} For any fixed Euler slope, every individual perverse direct-image coefficient of the split semistable unframed moduli stack has strict supports among the spectral-addition images $\Sigma_\nu$. The rank-$r$ primitive is the positive coarse intersection complex, whose proper Hitchin direct image is independent of the Euler characteristic as an object of $D^b\MHM(\cA_r^L)$. \end{proposition} \begin{proof} Fix an Euler slope $\mu$ and let \[ \mathfrak N_\mu=\{0\}\sqcup \coprod_{r>0,\,\chi/r=\mu} \mathfrak M_L^{\ss}(r,\chi). \] We apply the smooth linear-stack integrality theorem \cite[Theorem 10.2.7]{BDIKP} to this stack and verify its hypotheses. The objects are compactly supported pure sheaves on $S=\Tot_C(L)$, or equivalently $L$-Higgs bundles on $C$. Semistable objects of slope $\mu$ form an abelian category. For example, if the image of a morphism between two such objects were not saturated, saturation would strictly raise its slope, while the quotient inequality for the source gives slope at least $\mu$. This contradicts semistability of the target. Kernels, images, cokernels, and extensions therefore remain in the same category. Direct sum and the scalar action are its coherent operations. For two objects $E,F$ in this category, the deformation complex is \[ R\Gamma\bigl(C,[\mathcal Hom(E,F) \longrightarrow\mathcal Hom(E,F)\otimes L]\bigr). \] Riemann--Roch and surface duality give \begin{equation}\label{source:euler-ext} \chi_S(E,F)=-\ell\rk(E)\rk(F),\qquad \Ext_S^2(E,F)=\Hom_S(F,E\otimes A)^\vee=0. \end{equation} The second equality uses $\deg A<0$ and semistability at the same slope. Hence $\mathfrak N_\mu$ is smooth and its numerical Euler pairing is symmetric. We also check the stronger pointwise orthogonality condition. At a closed polystable point write $E=\bigoplus_j S_j\otimes V_j$, where the $S_j$ are pairwise nonisomorphic stable objects of ranks $r_j$. Then \[ \dim\Ext_S^1(S_i,S_j)=\delta_{ij}+\ell r_ir_j. \] The stabilizer is $\prod_j\GL(V_j)$. The opposite summands $\Hom(V_i,V_j)\otimes\Ext^1(S_i,S_j)$ and $\Hom(V_j,V_i)\otimes\Ext^1(S_j,S_i)$ have equal multiplicity and admit an invariant symmetric cross-pairing, using the trace pairing on the two Hom spaces. For $i=j$, use trace on $\operatorname{End}(V_i)$ and any nondegenerate symmetric form on the multiplicity space. The gauge summands are orthogonal by the same trace pairing. Thus the stabilizer representation is pointwise almost orthogonal in the sense of \cite{BDIKP}. The semistable spectral Quot construction supplies a good moduli space $p:\mathfrak N_\mu\to M_\mu$, with affine diagonal. At fixed $(r,\chi)$ the stack is of finite type: consecutive ordinary Harder--Narasimhan slopes of a semistable $L$-Higgs bundle differ by at most $\ell$, which bounds the underlying bundles. Direct sum is quasi-compact. Indeed a decomposition over a fixed family is a tuple of complementary idempotents in its relative endomorphism space; the idempotent equations are closed, and at a fixed target rank there are only finitely many numerical decompositions. For each fixed finite character-support face, the graded-object stack is a finite product of these eigensheaf stacks. Its total-object morphism is therefore quasi-compact. This is the facewise finiteness condition of \cite[\S 2.3.7]{BDIKP}. The perfect cotangent complex has finitely many weights on each such face. The theorem is locally finite in the numerical components. We keep the union over all ranks when applying it, so that the stack remains closed under direct sum. The torus Chern parameters can be chosen on the stack itself. For $p_0\in C$, the line $\det(E|_{p_0})$ has scalar weight $r$; its rational Chern class divided by $r$ restricts to the generator of $H^2(B\Gm,\Q)$. On a graded stack the same construction for each eigensheaf gives a basis of torus parameters. Finite automorphisms of the face permute these lines. By \cite[Proposition 2.9]{MSChi}, the rank-$r$ stack is smooth irreducible of dimension $B_r$ and its coarse space is irreducible of dimension $B_r+1$, for every $\chi$. On the dense smooth spectral locus, the stack is the relative Picard stack and has exactly the scalar stabilizer. Scalar rigidification is generically an isomorphism onto the coarse moduli space. The BPS identification in \cite[Corollary 7.2.4 and Theorem 10.2.7]{BDIKP} consequently identifies the primitive with the coarse intersection complex, also when $(r,\chi)$ is not coprime. Let $\eta=(r,\chi)$ run through the nonzero numerical components with $\chi/r=\mu$, let $M_\eta$ be their coarse moduli spaces, and let $\Sym_{\oplus}$ denote the symmetric-power operation formed using direct sum. With $\IC^{\vir}_Y=\IC_Y\otimes\Th^{-\dim Y}$, the theorem gives the cohomological induction isomorphism \begin{equation}\label{source:cohi} \Sym_{\oplus}\!\left( \bigoplus_{\eta\neq0}\IC^{\vir}_{M_\eta} \otimes H^*(B\Gm)_{\vir}\right) \xrightarrow{\sim} Rp_*\IC^{\vir}_{\mathfrak N_\mu}, \quad H^*(B\Gm)_{\vir} =\La\otimes\bigoplus_{v\geq0}\TT^v[-2v]. \end{equation} It is formed from the lowest-perverse primitive inclusion, the specified Chern parameters, and the smooth Hall induction maps. It supplies object-level support. The primitive normalized by scalar rigidification differs from the unnormalized first perverse coefficient by its half-Tate factor. In the split physical threefold the unframed critical stack is $T^*[-1]\mathfrak N_\mu$. Negativity in \eqref{source:euler-ext} makes the ordinary transverse field zero; the derived statement follows from the physical model in Section~\ref{sec:physical}. Its half/dual-half triangle identifies the fixed orientation with the cotangent orientation. Numerical symmetry and \cite[Lemma 10.2.9]{BDIKP} make this orientation commutative. Dimensional reduction gives \[ \pi_*\Phi_{T^*[-1]\mathfrak N_\mu} =\omega_{\mathfrak N_\mu}\otimes\LL^{B_r/2} =\IC^{\vir}_{\mathfrak N_\mu}. \] Since $\mathfrak N_\mu$ is smooth, this is the zero-function critical chart. The critical and smooth induction maps agree by \cite[\S\S 6.2.3 and 8.1.7]{BDIKP}. The coarse Hitchin maps are proper. Spectral addition is finite, as proved below, and coarse direct sum is finite as well: its composite with the Hitchin map is proper, while a polystable object has only finitely many allocations of its Jordan--H\"older multiplicities among a fixed list of summands. Properness and quasi-finiteness give finiteness. Finally \cite[Theorem 0.4]{MSChi} identifies the positive coarse Hitchin direct image of $\IC$ with the full-support intermediate extensions of the exterior-power spectral variations, independently of $\chi$, in $D^b\MHM$. Applying the proper Hitchin pushforward to \eqref{source:cohi} therefore gives precisely the supports $\Sigma_\nu$. At each fixed perverse degree only finitely many degrees of the torus factors contribute. The single rank-$r$ unframed coefficient belongs to $\cS_r$. \end{proof} \subsection{Finite spectral addition and reduced spectral curves} \begin{lemma}\label{source:finite-addition} For every rank partition $\nu=(r_1,\ldots,r_s)$ of $r$, spectral addition $\prod_j\cA_{r_j}^L\to\cA_r^L$ is finite. Its image $\Sigma_\nu$ is irreducible and has a dense locus of reduced nodal curves with smooth irreducible components of the specified ranks. The number of cross-component nodes and the codimension of this image are both \begin{equation}\label{source:cross-nodes} e_\nu=\ell\sum_{i0$ apply $\theta$. Every target term remains nonzero after this operation: its new box lies in $\lambda$ and its height is at least $m_{ij}$. Thus the coefficient propagates to column $j-1$. At column zero the source has zero $\theta$-image whereas every target still survives; the coefficient is zero. Induction on $j$ proves $\Hom(K,M/K)^{H,R}=0$. A common fixed quotient is supported at $d[0]$ and is killed by $t^d$, by Cayley--Hamilton. It lies in a projective finite-module Quot fiber. The residual torus has a limit there, which is monomial. A nonconstant residual orbit would have positive-dimensional closure through this limit, contradicting the vanishing of its fixed tangent. This proves the classification and reducedness. Differentiating the Artin-family identification above identifies the invariant tangent with $\Ext^1(I,I)_0^{H,R}$. Stability kills the trace-free $\Ext^0$, and proper-support Calabi--Yau duality kills the invariant $\Ext^2$ and $\Ext^3$. The derived fixed tangent is therefore acyclic. For $d=0$ the same assertion follows before taking curve cohomology: the trace-free local self-Ext complex of the partition ideal has only the nonzero hook characters in \eqref{fock:hilbert-hook-tangent}. Its normalized oriented coefficient is the unit. The orientation comparisons of Section~\ref{sec:physical} give the unit for every common fixed point. \end{proof} \subsection{Closed attracting filtrations and punctual purity} We use the compact form of oriented hyperbolic localization. For a torus action $\mu$, write $i^+:Y^+\to Y$ and $p^+:Y^+\to Y^0$. With the orientation from Section~\ref{sec:physical}, it is \begin{equation}\label{fock:compact-localization} p^+_!(i^+)^*\Phi_Y\simeq\Phi_{Y^0}\otimes\La^{I_\mu}. \end{equation} Here $I_\mu$ is the index of Proposition~\ref{phys:integral-localization}, namely \[ (\rk^+H^0T_Y-\rk^-H^0T_Y) -(\rk^+H^{-1}T_Y-\rk^-H^{-1}T_Y). \] This is the localization theorem for the inverse action in the convention $p_*\eta^!$ of \cite{KPS}. On a smooth zero-potential chart the index is $d_+-d_-$. Compact contraction contributes $\La^{2d_+}$, the ambient normalization contributes $\La^{-d_0-d_+-d_-}$, and the fixed normalization is $\La^{-d_0}$. Their quotient is exactly the exponent in \eqref{fock:compact-localization}. The chart maps, orientation torsors, and their descent in $\MMHM$ were fixed in Section~\ref{sec:physical}. \begin{lemma}\label{fock:closed-filtration} Let $Y$ be a finite-type separated $\Gm$-scheme, let $V$ be an affine representation of strictly positive weights, and suppose that $Y\to V$ is proper and equivariant. If $Y$ has a linearized ample line bundle, it has a finite closed filtration whose differences are entire attracting schemes over open-and-closed unions of fixed components. This description holds on arbitrary ordinary bases. \end{lemma} \begin{proof} A power of the ample line gives a relative projective embedding. Choose homogeneous generators of the finite equivariant module defining it. Their free cover yields a closed equivariant embedding $Y\into\mathbf P(W)\times V$ for a finite-dimensional representation $W$. Order its distinct weights $w_1<\cdots0}$. Thus the sum in \eqref{fock:index-increment} is \begin{equation}\label{fock:kernel-sum} 1+2(L_{c+1}-L_c)+2\mathbf1_{c=0}. \end{equation} Counting row endpoints gives $L_c-L_{c+1}=\mathbf1_{c\in\mathcal B_\lambda}-\mathbf1_{c<0}$. Since $\alpha+j>0$, \eqref{fock:index-increment} reduces to $\Delta I=\sigma(c)$. Restricting \eqref{fock:virtual-character} to $H$ kills its quadratic term in $\mathsf Z_m$. A legal increment has difference $F_c(u)-F_c(u^{-1})$, with \[ F_c(u)=u^{-c}\bigl(1-(1-u^{-1})(1-u)\overline{\mathsf Z_\lambda}\bigr). \] The kernel $2\sgn(D)-\sgn(D-1)-\sgn(D+1)$ is supported at $D=-1,1$ with values $-1,1$. Therefore \begin{align} \Delta I_H&=\sgn(-c)+L_{c-1}-L_{c+1} =\tfrac12\bigl(\sigma(c-1)+\sigma(c)\bigr),\nonumber\\ I_H&=\tfrac12\sum_cd_c\bigl(\sigma(c-1)+\sigma(c)\bigr),\quad I_R=\tfrac12\sum_cd_c\bigl(\sigma(c)-\sigma(c-1)\bigr). \label{fock:two-indices} \end{align} The first index is constant on each graded-dimension component. Passing from compact affine cohomology to punctual cohomology changes $I_R$ to $-I_R$. After the factor $\La^a[d]$ in \eqref{fock:relative-object}, the punctual exponent and degree are \begin{equation}\label{fock:punctual-indices} J=I_H-I_R=\sum_cd_c\sigma(c-1),\qquad \text{degree}=J-d. \end{equation} We use the diagonal-trace rim-hook identity \cite[Theorem 0.2 and Section 6]{Sulzgruber}. In the present notation it asserts that reverse plane partitions are enumerated, with their individual diagonal sums, by independently repeated rim hooks. The hook of a box of content $c$, arm $a$, and leg $l$ occupies the diagonal interval $[c-l,c+a]$. That interval contains $l+1$ Maya members and $a$ nonmembers; its right endpoint is a member and the preceding left endpoint is a nonmember. Hence \begin{equation}\label{fock:hook-index} \sum_{v=c-l}^{c+a}\sigma(v-1)=l-a-1=e_b, \qquad e_b-h_b=-2(a_b+1). \end{equation} Equations \eqref{fock:punctual-indices}--\eqref{fock:hook-index} and the already proved punctual purity give the punctual graded cohomology object as the symmetric object on the generators \begin{equation}\label{fock:punctual-generators} v^{h_b}\Th^{e_b}[2(a_b+1)],\qquad b\in\lambda. \end{equation} The intrinsic weight $k+d$ separates degree $k$ before any global symmetric-power operation is taken. \subsection{Moving support and finite symmetric actions} The punctual calculation must descend with its Hodge structure and its finite permutation actions. Let $x,y$ be two \'etale coordinates on an open set $U\subset C$. On a neighborhood $V$ of the diagonal in $U_t\times U_p$, discard the other sheets of the two \'etale equivalence relations. Then \[ X=x(t)-x(p),\qquad Y=y(t)-y(p)=Xv(t,p), \] where $v$ is a unit. Over $B=U_p\times\A^1_s$ form \begin{equation}\label{fock:coordinate-dnc} W=\Spec\bigl(\OO(V)[s,\tau]/(X-s\tau)\bigr), \qquad \upsilon=\tau v(t,p). \end{equation} It has two \'etale maps to $\A^1_\tau\times B$ and $\A^1_\upsilon\times B$. At $s=1$ they are the original coordinate maps; at $s=0$ their ratio is the linear change $\upsilon=(dy/dx)(p)\tau$. Cotangent lifting preserves the relative threefold volume and the transverse monomial module. For a fixed $d$, the support-cycle fiber $d$ times the distinguished section is the constant punctual Quot scheme over $B$. This holds on nonreduced bases: formal \'etaleness and Cayley--Hamilton identify the quotients killed by the $d$th power of the section coordinate. The supported fixed-determinant gluing construction in Proposition~\ref{phys:parameter-square} applies to these two maps. On the completed parameter ring $R=\C[[p-p_0,s-s_0]]$ its reference branch is constant. The full-parameter calculation there gives equality of the two potentials modulo the square of the common critical ideal, for these same maps. Its algebraic four-Hom line map has square equal to the determinant of the same cotangent derivative. Faithful flatness of completion therefore identifies the existing algebraic oriented critical structures over $B$. Tensor the line map with $\id_{K_B}$ to retain the smooth-base factor. Proper pushforward of the punctual comparison gives a map between pullbacks of one punctual cohomology complex over $U\times\A^1$. Pullback from $U$ is fully faithful, by the affine-line cohomology adjunction. The two endpoint maps are equal. At $s=0$ this is the linear coordinate action. The physical orientation map gives an action over $\Gm$, and its induced maps on cohomology objects are morphisms of constant variations. They equal the identity at $1$, and hence everywhere on the connected group. Thus all moving coordinate transitions are identity on the punctual Hodge variations. On an ordered configuration space the finite defect decomposes canonically by support idempotents, also over nilpotent bases: $T=\bigoplus_iT_i$. Use a single baseline $G_0$ and the triangles $G_0\to G_i\to T_i$. Put \[ M(G)=\det\fib\bigl(\RHom_{S_K}(G,G) \longrightarrow R\Gamma(S_K,G)\bigr). \] Use the equivariant surface and transported orientation of Lemma~\ref{phys:supported-determinant}. The following grid is \(H\)-equivariant before taking determinants. Its weight-zero part compares the fixed orientations; its moving parts supply the index and permutation factors below. The supported Hom and section triangles give the line map \begin{equation}\label{fock:baseline-factorization} M(G)M(G_0)^{-1}\simeq \bigotimes_i\bigl(M(G_i)M(G_0)^{-1}\bigr). \end{equation} The cross complexes $\RHom(T_i,T_j)$ vanish canonically for $i\ne j$; the volume trace splits by the same idempotents. Taking weight zero in this grid before determinants gives the oriented fixed-coefficient comparison, with the determinant functor's regrouping maps. The merge morphism is \'etale on the labeled branch, since the idempotents lift uniquely over infinitesimal extensions. When two reference curves have baselines of different parity, divide first by their baseline point coefficients. The quadratic-point calibration in \cite[Lemma 3.4]{Kinjo} and \cite[Section 7.8]{KPS} is a rational unit map of the associated orientation torsors. The compact Thom and product-thimble maps have scalar $+1$. This fixes the cancellation in \eqref{fock:baseline-factorization}. The ordered-configuration variations are consequently constant; their cohomology still retains the topology of the configuration space. We check the symmetry used in their descent. In the $J$-model of monodromic Hodge modules, let $r(x)=x^2$, let $\iota(x)=-x$ be its deck involution, and put \[ \Th=e_-r_*\Q_{\A^1}[1],\qquad e_-=(1-\iota)/2. \] Its global cohomology vanishes. Its $k$th convolution power is the full-sign summand of $(\sum x_i^2)_*\Q_{\A^k}[k]$. A smooth quadratic fiber retracts equivariantly onto $S^{k-1}$; its vanishing cohomology has determinant character. A coordinate permutation has its sign on this cohomology, and the shifted $[1]$ factors have the same Koszul sign. The product is $+1$. These objects lie in the perverse heart, where rational realization is faithful. This proves that $\Th$ is even and $\La$ is odd, as stipulated in \eqref{eq:tate-conventions}. For a relative defect packet, let $m_0,m_+,m_-$ be the Euler ranks of the surface half in weights zero, positive, and negative. The four-Hom comparison gives \begin{equation}\label{fock:packet-parities} m_0+m_++m_-=-d,\qquad a=m_--m_+,\qquad \kappa=m_0+2m_+,\qquad \kappa+a=-d. \end{equation} The graded orientation exchange contributes parity $\kappa$. After the normalization $\La^a[d]=\Th^a[d-a]$, the extra exchange exponent for two packet colors is \[ \kappa_i\kappa_j+(d_i-a_i)(d_j-a_j) =2d_id_j+2a_ia_j. \] For repeated packets the remaining exponent is \begin{align*} \kappa-a^2+d^2 &=m_0(m_0+1)+2m_++2m_0(m_++m_-)+4m_+m_-. \end{align*} Both are even for all integer Euler ranks. The permutation group relations themselves come from determinant and Thom--Sebastiani coherence. The odd finite-defect correction $(-1)^d$ is multiplicative and has ratio one under exchange and refinement. Thus the finite symmetric actions are the ordinary derived actions on \eqref{fock:punctual-generators}. In particular both charge twists $F_d\mapsto F_d\otimes\Th^{-d}$ and $F_d\mapsto F_d\otimes\Th^d[-2d]$ are symmetric monoidal. Thom--Sebastiani products are formed in $\MMHM$. Choose local critical functions vanishing on their reduced critical loci and shrink to the zero-critical-value situation of \cite[Theorem 5.3]{SaitoTS}. \subsection{Global purity and proof of the relative formula} Write $A_a=Rf_{d,a,*}\Phi_{d,a}$ before twisting by its index. The global support map is proper, since the dualizing Quot scheme on the finite thickening of the projective curve is projective. At a support cycle $\sum d_ip_i$, proper base change and \eqref{fock:baseline-factorization} identify its coefficient with the external product of the punctual coefficients, summed over the compatible local indices. The punctual result shows that every stalk $H^j(i_x^*A_a)$ is pure of intrinsic weight $j$. It follows that $\wt(A_a)\leq0$. More explicitly, on a smooth stratum of dimension $s$, a lisse perverse coefficient is a variation shifted by $[s]$ and its perverse weight is its fiber weight plus $s$. Applying this to a stratification adapted to the weight filtration proves the bound. The untwisted $\Phi_{d,a}$ is self-dual, and proper pushforward preserves duality. Thus the lower weight bound also holds: \begin{equation}\label{fock:relative-purity} \wt(A_a)=0,\qquad \wt(L_{\lambda,d})=d. \end{equation} We have used self-duality before the twist by $\La^a$. Pure polarizable monodromic Hodge modules are semisimple. One can see this directly in the weight-stable Serre quotient defining $\MMHM$: a pure quotient object of weight $w$ has the pure ambient lift $\Gr^W_w M$. The latter is semisimple, and its image under an exact Serre quotient remains semisimple. This also splits the same-weight extensions used in the punctual purity argument. Stratify $\Sym^d C$ by support multiplicities. On the ordered finite cover of a stratum, the preceding coordinate and product comparisons give the constant punctual objects with their computed stabilizer actions. The right-hand side of \eqref{fock:relative-series} has the same description: it assigns box occupancies at each support point, with total weighted multiplicity $\sum h_br_b$. Taking finite rational invariants is exact. Equivariant tensor-power filtrations of short exact sequences show that passage to full $K_0$ classes preserves these projectors. For a finite group quotient of $C^s$, the invariant direct image of its constant object is the constant object of the symmetric product, including at diagonals: the unit followed by the projector is an isomorphism on the finite orbit fibers. Therefore finite closed/open localization on the base gives \begin{equation}\label{fock:class-before-promotion} [L_{\lambda,d}]= \left[\bigoplus_{d(\mathbf r)=d} (\mu_{\mathbf r})_*\Q_{C_{\mathbf r}} \otimes\Th^{E(\mathbf r)}[2A(\mathbf r)]\right]. \end{equation} Each box on the right has weight $e_b+2(a_b+1)=h_b$. Hence both sides of \eqref{fock:class-before-promotion} are pure of weight $d$. The component of weight $d+k$ in their classes is exactly $(-1)^k[\pH^k]$. Semisimplicity identifies the corresponding perverse objects. A term on the right has perverse degree $R(\mathbf r)-2A(\mathbf r)$ because $\mu_{\mathbf r}$ is finite and $C_{\mathbf r}$ is smooth of dimension $R(\mathbf r)$. This proves Theorem~\ref{fock:relative-theorem}. Twisting by the symmetric charge $\Th^{-d}$ changes the box generator to \begin{equation}\label{fock:ordinary-charge} v^{h_b}(\delta_{h_b})_*\Q_C(a_b+1)[2(a_b+1)]. \end{equation} Thus the constituents of $\Th^{-d}L_{\lambda,d}$ lie in the ordinary Hodge-module image, with trivial auxiliary monodromy. This conclusion retains the $H^1(C)$ and all its exterior powers under proper integration. \subsection{Properness of the positive spectral map} The positive spectral map $P_{r,\chi}(X)\to\cA_r^L$ records the characteristic polynomial of $az+b\theta$. We will use its properness both for the comparison endpoint and for a closed filtration of the whole nil locus. \begin{proposition}\label{fock:properness} The positive spectral maps on $P_{r,\chi}(X_0)$ and $P_{r,\chi}(X_1)$ are proper. The restriction of the latter to $N_{r,\chi,\red}$ is equivariant for \eqref{fock:antidiagonal} and the positive grading on $\cA_r^L$, and admits an $H$-linearized ample line bundle. \end{proposition} \begin{proof} The closed immersions $f_s$ of \eqref{source:scalar-base-map} and the finite spectral addition of Lemma~\ref{source:finite-addition} show that, for each rank partition, the map \begin{equation}\label{fock:cycle-cover} \prod_i(\cA_{r_i}^{K_C}\times\A^1)\longrightarrow\cA_r^L \end{equation} is finite. On every compact irreducible curve in $X$, the function $z$ is constant, and its projection to $\Tot(K_C)$ is a spectral divisor finite over $C$. Thus the finitely many maps \eqref{fock:cycle-cover} parameterize all compact effective cycles of rank $r$, with their multiplicities. Given a DVR family in $\cA_r^L$ and a generic stable pair, after a finite extension its generic cycle lifts to one of these finite maps. Its lift extends over the integral-closure DVR and has a limiting cycle entirely in $X$. The projective stable-pair moduli on $Y=\mathbf P_C(\OO_C\oplus K_C\oplus\OO_C)$ extend the generic pair. Separatedness of its projective cycle identifies the limit with the cycle just constructed. Purity of the limiting sheaf prevents an additional isolated support component at infinity. The extended pair is therefore supported in $X$. Boundary avoidance descends through the finite extension, proving properness. For $X_0=\Tot_C(L\oplus A)$, the pullback of $A$ to the normalization of any compact irreducible curve has negative degree. Its transverse coordinate is consequently zero. Every compact cycle lies in $\Tot_C(L)$ and is a finite spectral divisor. The same DVR argument, now with the ordinary positive spectral coefficients as the cycle parameters, proves properness for this row as well. At a geometric nil point, the commuting $\theta,z$ are simultaneously triangularizable over the curve's generic field after extension, and the diagonal entries of $z$ vanish. Hence \begin{equation}\label{fock:nil-equivariance} \operatorname{char}(az+b\theta)=\operatorname{char}(b\theta),\quad \operatorname{char}(au^{-1}z+bu\theta)=\operatorname{char}(bu\theta). \end{equation} These are identities of regular functions on the reduced nil locus, and give the asserted positive-base equivariance. The $H$ action extends to $Y$. An $H$-linearized polarization induces linearizations on the projective section factor and the Pl\"ucker line in the stable-pair GIT construction \cite{ST}. It commutes with the framing linear group. Passing to the invariant graded ring and a divisible Veronese gives a linearized ample line on the projective stable-pair compactification. Its restriction gives the required line on the nil locus. \end{proof} \subsection{The whole nil locus} Proposition~\ref{fock:properness} and Lemma~\ref{fock:closed-filtration} give a finite closed filtration of $N_{r,\chi,\red}$ by whole attracting schemes over unions of complete fixed components. These are the reductions of the ambient attractors. Indeed, on an arbitrary ordinary ring the $j$th characteristic coefficient of $z$ in an attracting family has Laurent form $u^{-j}c_j$. Extension through zero forces $c_j=0$, by injectivity of $R[u]\to R[u,u^{-1}]$. On a reduced test base the orbit extension is reduced and therefore factors through $N_{r,\chi,\red}$; the converse is immediate. The universal reduced attractors and their limit maps consequently agree. Hodge modules and their six operations are insensitive to nilpotent thickenings, so \eqref{fock:compact-localization} computes the layers with exactly the coefficient \eqref{source:nil-coefficient}. The baseline Euler characteristic and $H$-index are \begin{align} \chi_0(\lambda) &=\chi\!\left(\bigoplus_{(i,j)\in\lambda}K_C^{-j}\right) =-h\sum_{(i,j)\in\lambda}(2j+1) =-h\norm{\lambda}^2,\label{fock:baseline-chi}\\ I_0(\lambda) &=\sum_{b\in\lambda}\chi(K_C^{a_b+1}) =h\sum_{b\in\lambda}(2a_b+1) =h\norm{\lambda}^2=-\chi_0(\lambda). \label{fock:baseline-index} \end{align} The positive transverse hook line is $K_C^{a_b+1}$, explaining the second formula. For fixed $\lambda$, the defect is $d=\chi-\chi_0(\lambda)$. The corresponding complete attracting layer has compact cohomology \begin{align} &R\Gamma(D_{\lambda,d,a},\Phi_{d,a}\otimes\La^{\chi+I_0+a}) =R\Gamma(D_{\lambda,d,a},\Phi_{d,a}\otimes\La^{d+a}), \nonumber\\ &\hspace{22mm}\bigoplus_aR\Gamma(D_{\lambda,d,a}, \Phi_{d,a}\otimes\La^{d+a}) =R\Gamma(\Sym^dC,\Th^d[-2d]L_{\lambda,d}). \label{fock:whole-layer} \end{align} The latter is pure of complex weight zero by \eqref{fock:relative-purity}. The symmetric charge changes each generator of \eqref{fock:relative-series} to \begin{equation}\label{fock:normalized-box} \Th^{h_b+e_b}[2(a_b+1)-2h_b] =\TT^{l_b}[-2l_b]. \end{equation} Hence the perverse cohomology and the proper cohomology of the normalized layer are ordinary pure Hodge objects, with summands $(\mu_{\mathbf r})_*\Q_{C_{\mathbf r}}(-L(\mathbf r))[-2L(\mathbf r)]$. For fixed $r,\chi$ there are finitely many partitions, and each partition fixes $d$. Each remaining projective Quot scheme has finitely many connected components; its index and the weight of the linearized ample line are locally constant. Every filtration layer therefore selects an open-and-closed union of complete components. Its proper cohomology is a direct summand of \eqref{fock:whole-layer}, so has the same purity and ordinary coefficient properties. In the compact-cohomology sequences of successive closed filtration steps, the boundary maps join weights differing by one and vanish. The remaining extensions of a fixed weight split in the pure polarizable heart. This gives \eqref{fock:nil-cohomology}, including its degree and auxiliary monodromy assertions, and proves Theorem~\ref{fock:nil-theorem}. \begin{remark}\label{fock:grading-distinction} The perverse degree in \eqref{fock:perverse-formula} belongs to the defect support map to $\Sym^d C$. The Hitchin perverse degree is extracted from the source primitive using $x$ in Section~\ref{sec:recovery}. Theorem~\ref{fock:nil-theorem} supplies the ordinary Hodge cohomology and its purity in each degree. \end{remark} \section{Hodge recovery and the character-variety formula} \label{sec:recovery} We combine the two calculations before taking any numerical realization. All formal variables in this section are even. For a bounded Hodge complex $K$, its class is $[K]=\sum_j(-1)^j[H^j(K)]$. Let \begin{equation}\label{rec:curve-zeta} \zeta_C(y)=\sum_{m\geq0}[H^*(\Sym^mC)]\,y^m =\frac{\bigwedge_{-y}H^1(C)}{(1-y)(1-\TT y)}. \end{equation} The last identity is the symmetric-product formula for a curve \cite{Macdonald}; it is an identity of Hodge classes, with the derived symmetric action on $H^1(C)$ giving the exterior numerator. \subsection{The identity in full Hodge classes} Define the series of the original restricted physical coefficients by \begin{equation}\label{rec:nil-series-definition} Z_{\nil}(R,x)=1+\sum_{r\geq1}\sum_\chi R^rx^\chi[R\Gamma_c(N_{r,\chi},K_{r,\chi})]. \end{equation} Theorem~\ref{fock:nil-theorem} gives \begin{equation}\label{rec:nil-fock-series} Z_{\nil}(R,x)=\sum_\lambda R^{|\lambda|} x^{-h\norm{\lambda}^2} \prod_{b\in\lambda}\zeta_C(x^{h_b}\TT^{l_b}). \end{equation} The degreewise ordinariness and purity proved above place these classes in $K_0(\MHS)$. Choose an integer $d$ coprime to $r$ with $\zeta=\exp(2\pi i d/r)$, and set $\chi_*=d-r(g-1)$. Let $M_K$ be the smooth moduli space of stable $K_C$-Higgs bundles of rank $r$ and degree $d$, equivalently of Euler characteristic $\chi_*$. This is the degree matching the fixed twist in \eqref{eq:character-variety}. The source formula permits this choice of its coprime parameter. Let $h_K:M_K\to\cA_r^{K_C}$ be its Hitchin map, and put \begin{equation}\label{rec:hitchin-perverse-objects} P_i^K=\pH^i Rh_{K,*}\Q_{M_K}[2N_r],\qquad A_i=[R\Gamma_c(\cA_r^{K_C},P_i^K)]. \end{equation} The scalar primitive of Section~\ref{sec:scalar}, restricted to the characteristic-nil locus, is \begin{equation}\label{rec:scalar-primitive} x^{i+1+u+v}P_i^K[-i-2v]\otimes\TT^{v-N_r}, \qquad u,v\geq0. \end{equation} Proper base change and compact integration of the finite spectral addition maps preserve its symmetric-power operations. Taking the plethystic logarithm of the source identity, summing $u,v$, and using the shift in \eqref{rec:scalar-primitive} give \begin{equation}\label{rec:primitive-identity} [R^r]\Log Z_{\nil}(R,x) =\frac{x\TT^{-N_r}}{(1-x)(1-x\TT)} \sum_i(-x)^iA_i. \end{equation} Both sides are now ordinary. The exact left inverse to the ordinary embedding into the monodromic category gives an injection on their Grothendieck groups, so the source identity gives \eqref{rec:primitive-identity} in $K_0(\MHS)$. Equivalently, define \begin{equation}\label{rec:recovered-series} F_r(x)=x^{-1}\TT^{N_r}(1-x)(1-x\TT) [R^r]\Log Z_{\nil}(R,x). \end{equation} Then \begin{equation}\label{rec:perverse-class-identity} F_r(x)=\sum_i(-x)^i [R\Gamma_c(\cA_r^{K_C},P_i^K)]. \end{equation} The variable $x$ records Euler characteristic; the source primitive identifies its exponent $i$ with the Hitchin perverse coefficient in \eqref{rec:perverse-class-identity}. All these operations are coefficientwise finite. For the finite rank range chosen at the start, the positive-source coefficients and the stable-pair endpoints vanish below $a_r$. For $r=\sum r_i$, \begin{equation}\label{rec:lower-bound-addition} \sum_i a_{r_i}-a_r=\ell\sum_{i2N_r$ \cite[Section 4.1]{BBD}. The ordinary weight bound for smooth varieties gives $2k\geq D$ whenever $\beta_{D,k}\ne0$. Both exponents in \eqref{rec:barh-betti} are nonnegative; replacing $w$ by $-w$ cancels the displayed sign. Relative hard Lefschetz for the proper Hitchin map gives \[ \beta_{D,k}=\beta_{D+2(N_r-k),\,2N_r-k}. \] It interchanges the two exponents in \eqref{rec:barh-betti} and preserves the parity of $D$. \end{proof} \providecommand{\End}{\operatorname{End}} \section{Root curves and the positive comparison}\label{root:geometry} \subsection{Numerical classes and a positive length} Let $C$ be a smooth projective connected complex curve of genus $g$, let $p_1,\ldots,p_m$ be distinct points, and let $s_1,\ldots,s_m$ be positive integers. Throughout this section \[ \nu:\mathcal X=C[\sqrt[s_1]{p_1},\ldots,\sqrt[s_m]{p_m}] \longrightarrow C \] is the tame root stack, $E_p$ is its reduced root divisor, and $K=K_{\mathcal X}$. Thus $\nu^*p=s_pE_p$. A local chart at $p$ is $[\Spec\C[t]/\mu_{s_p}]$, with coarse parameter $u=t^{s_p}$. All equivariant complexes below are formed on this chart and then passed to invariants. In particular, invariants are taken before determinants, traces, or cohomology of a complex. Choose the character convention for which a bundle $E$ has profile $\boldsymbol\mu=(\mu_{p,c})_{p,0\leq c0, \] which proves the required nesting. There is one additional point for coherent wall factors. If $T_0$ is their torsion and the positive-rank factors have slopes $\mu_j\geq\lambda$, equality at the wall gives \[ \lambda f=q(T_0)+\sum_j f_j\mu_j \geq q(T_0)+\lambda f. \] Lemma~\ref{root:positive-mass} forces $T_0=0$. Thus the argument eliminates also the nontrivial-character simples with zero ordinary Euler characteristic. A finite-rank-zero framed factor at finite framing parameter is the vacuum: on a smooth surface chart a finite sheaf $T$ has $\Hom(T,K_S^{-1})=\Ext^1(T,K_S^{-1})=0$; exact finite-group invariants preserve these vanishings. The physical reflection constructed below is \[ X^t_{r,\chi,\boldsymbol\mu} \simeq X^{-t}_{r,-\chi,\boldsymbol\mu^\vee},\qquad B^t_{r,\chi,\boldsymbol\mu} \simeq B^{-t}_{r,-\chi,\boldsymbol\mu^\vee}\otimes\LL^\chi. \] Together with the lower bound it gives the finite source interval $[A_r,-A_r]$. It does not identify two equal profiles unless the profile is self-dual. For the other endpoint use the elementary Higgs slope estimate \[ \mu_{\min}^{\rm orb}(E) \geq \frac{q(E)}r+\frac{k}{2}-(r-1)\ell. \] It follows by inserting the successive Harder--Narasimhan quotients: a gap larger than $\ell$ makes the upper part $L$-Higgs invariant. Thus $H^1(E)=0$ as soon as $q(E)/r>k/2+(r-1)\ell$. Choose adjacent framing chambers with distance less than $1/(4S_0r^2)$. If an underlying positive Higgs object at this endpoint were unstable, its highest Harder--Narasimhan part would have vanishing $H^1$, zero dual framing, and be preserved by the negative $A$-field. It would destabilize the framed object. The remaining framing map is therefore a smooth projective Joyce--Song framing over the semistable positive moduli, and its comparison is an isomorphism. This gives an upper endpoint for the error. The proof is about objects and correspondences, so the same endpoint works in every perverse degree. \end{proof} \subsection{Physical maps and their orientations on a root curve} We spell out the root descent of the physical construction, because descent of determinant lines alone would not identify the maps used in the comparison. For $S_L=\Tot_{\mathcal X}L$ and a compactly supported spectral sheaf $G$, let \begin{equation}\label{root:surface-half} B_{\mathcal X}(G)= \fib\bigl(\RHom_{S_L}(G,G)\longrightarrow R\Gamma(S_L,G)\bigr), \qquad M_{\mathcal X}(G)=\det B_{\mathcal X}(G). \end{equation} The arrow is composition with the section. The same definition with two entries gives the off-diagonal halves. \begin{proposition}\label{root:physical-descent} The critical equivalence, paired deformation triangle, orientation square, supported determinant comparison, and parameter square of Proposition~\ref{phys:critical-equivalence} are valid over $\mathcal X$. They are equivalences and commutative diagrams of the enhanced complexes, compatible with arbitrary Artin base change, direct sums, and finite-support modifications. Their orientation is $M_{\mathcal X}$ in \eqref{root:surface-half}. \end{proposition} \begin{proof} The construction of~\ref{phys:critical-equivalence} uses the relative diagonal resolution of $\mathbf P^1$, the monic two-term spectral resolution, evaluation against the framing, composition of Hom complexes, and the relative Serre trace. Each is functorial for equivariant sheaves on a curve chart. In characteristic zero the invariant functor is exact, commutes with tensoring by a trivial base algebra, and commutes with the dual of a perfect complex. Apply it to this complete diagram. In particular the triangle is \[ B_{\mathcal X}(G_j,G_i)^\vee[-3] \longrightarrow C_{\mathcal X}(G_i,G_j) \longrightarrow B_{\mathcal X}(G_i,G_j). \] This establishes the pairing and the equivalence of derived mapping spaces, including maps between objects over nonreduced Artin bases. The local Calabi--Yau coordinates $(t,y,z)$ have root weights $(1,-1,0)$, so $dt\wedge dy\wedge dz$ is invariant. The trace has exactly the ordinary normalization. With $u=t^s$, the pullback of $F(u)du$ is $sF(t^s)t^{s-1}dt$. Stack integration contributes $1/s$ times cover integration. These two factors cancel, giving \begin{equation}\label{root:trace-normalization} \nu_*K_{\mathcal X}=K_C \end{equation} with its Serre trace, and residue one for the pullback $s\,dt/t$ of $du/u$. No scalar square root of a root order enters this construction. The closed-form comparison is equally equivariant. The normal Euler vector field commutes with the root group. In the notation of the proof of~\ref{phys:parameter-square}, the Cartan homotopy is \[ H_0=h_0+\tfrac12\iota_{\rm Eu}(\delta-Dh_0). \] It may be taken before invariants. The derivative of this same cotangent equivalence gives the determinant-square map after invariants. This proves compatibility with the specified maps in the square. The virtual dimension is recomputed on the invariant complex. For $0\to G_0\to G\to T\to0$, use the four-Hom grid with entries $\RHom(T,G_0)$, $\RHom(T,T)$, $\RHom(G_0,T)$, and $R\Gamma T$. Its cones, traces, and derivative determinant maps are again equivariant. The traceless correction has determinant one: in a finite free presentation it is $\det(1-uv)=\det(1-vu)$, and the latter is one on the physical rank-zero cofiber. This proves the supported comparison. Two sign points require retaining more information than in the ordinary curve. First, on the canonical surface the Euler form is symmetric and its self-pairings are even, since the $\Ext^1$ pairing is symplectic. Hence \begin{equation}\label{root:supported-parity} \chi(B(G)) -\chi(B(G_0)) =2\chi_{S_K}(G_0,T)+\chi_{S_K}(T,T)-\chi(T) \equiv-\chi(T)\pmod2. \end{equation} Individual surface pairings need not vanish at a root. The Euler characteristic on the right is computed after invariants. Second, for two distinct scalar packets write $C_{21}=C_{12}^\vee[-2]$ and $A_{21}=-A_{12}^\vee$ for their cross differential. Put $L_{12}=\det C_{12}$, of parity $c=\chi(C_{12})\bmod2$. The two cross cones have determinant lines $L_{12}\otimes L_{12}^\vee$ and $L_{12}^\vee\otimes L_{12}$. Graded interchange contributes $(-1)^c$, while $\det(-A_{12}^\vee)=(-1)^c\det A_{12}$. Consequently their acyclic trivializations satisfy \[ \tau_{21}\circ\beta=\tau_{12}. \] This identity is functorial over the parameter ring; it follows degree by degree for free complexes and then from determinant coherence for exact filtrations. Taking inverses gives precisely the comparison of the two specified orientation roots. Thus there is no extra character in packet permutations even when a cross-profile Euler pairing is odd. Finally, a change of positive quotient has residual quadratic space $(\End(V)\otimes R_{\rm res})^{\mu_s}$. Its equivariant change-of-frame group has reductive part $\prod_c\GL(V_c)$ and connected unipotent radical. Its determinant character into $\{1,-1\}$ is trivial. The residual orientation line is therefore constant for a fixed profile; ordered color blocks and the graded determinant functor give its direct-sum compatibility. Ordinary elementary modifications preserve this profile. Root modifications use the ratio of the two constant lines together with the same determinant maps. For the comparison with the canonical projection, the residual module is $K/(aA)=K|_{D_a}$, where $D_a=(a=0)$ avoids the root points and has even length $\ell$. The ordinary even-residue interchange calculation therefore applies directly. These constructions glue by equivariant descent, completing the proof. \end{proof} \subsection{The moving root Hecke correspondence} Fix an ordinary point $y\in C$. The creation point $x$ moves on the whole coarse curve $C$. An elementary quotient is annihilated by the pullback of the ideal of its coarse graph, has invariant Euler characteristic one, and has regular character at a root. More explicitly, over a base algebra $R$ near a root of order $s$ it is an $R$-flat module over \begin{equation}\label{root:hecke-local-algebra} R[t]/(t^s-a) \end{equation} whose character summands all have rank one; $a$ is the moving coarse coordinate. Multiplication by $t$ need not be zero when $a=0$. Require that the $L$-field act as a scalar $\lambda$ on this quotient and that the section vanish in it. Its kernel is locally free by the fiberwise torsionfree criterion on the regular root chart. Moreover \[ [E_+|_p]-[E_-|_p]=(1-z)[\mathrm{Reg}]=0, \] so the profile is unchanged and both $q$ and $\chi$ increase by one. The correspondence is the schematic closure of this incidence over ordinary $x$, with the fundamental multiplicities of that dense incidence. \begin{proposition}\label{root:hecke} With these elementary quotients the whole-arrow Weyl relation of Proposition~\ref{cmp:weyl-relation} holds on the root curve: \begin{equation}\label{root:weyl} F_\chi^c E_\chi^{c-1}-E_{\chi-1}^cF_{\chi-1}^{c-1} =r\,\rho_\chi(c-2,c). \end{equation} It holds before taking a cone, perverse cohomology, or a Grothendieck class, and is compatible with the physical orientation of Proposition~\ref{root:physical-descent}. \end{proposition} \begin{proof} Use the three bounded bundle levels of the ordinary construction, with a one-unit buffer in $q$. For their finite collection of Harder--Narasimhan types choose a coarse divisor $D$ disjoint from the active graphs and large enough that \[ H^1(\End(E)L(D)(-J))=H^1(E(D)(-J))=0 \] for the thick union $J$ of the graphs. Exact $\nu_*$ and ordinary Serre vanishing give this uniformly. The framed bundle stack maps smoothly to its $J$-jets: the obstruction lies in $H^2$ on a curve. A moving root jet has frame group the units in an invariant finite locally free endomorphism algebra, an open smooth group scheme. The two displayed vanishings make the field and section restrictions jointly surjective vector bundle maps. The residue variables at $D$ are independent of the active jets. Hence every correspondence chart is a smooth pullback from the local jet incidence. For $x\ne y$, the two ordered Hecke words have a common four-bundle receiver. If $U\to T_x,T_y$ are their quotients, put \[ E_1=\ker(U\to T_x),\quad E_2=\ker(U\to T_y),\quad J_0=\ker(U\to T_x\oplus T_y). \] Then $U=E_1\amalg_{J_0}E_2$. The derived cross-Homs of the two quotients vanish, so this bicartesian diagram and its fields and sections are unique up to a contractible space of choices. The trace functions agree as functions on the whole scheme. The residue support equations agree as well, since they are restrictions along relative Cartier divisors. On the smooth jet presentation the variables for $x$ and $y$ are independent. Flat base change commutes with the schematic closure of the ordinary-$x$ incidence. In particular, if $A_x,A_y$ are the jet algebras and $R_x,R_y$ their incidence algebras, then the two orders have the same derived tensor product, namely the smooth pullback of $R_x\otimes_\C R_y$ over $A_x\otimes_\C A_y$. There is no cross Tor term. Their refined Borel--Moore cycles therefore agree off $x=y$, including when $x$ is a root point. For clarity, the proper carrier on which this comparison is made can be constructed without using a special-fiber cycle. For a fixed final pair $(E,\Phi,s)$ let $H=\nu^*(x+y)$ and let $\sigma_H$ be its canonical section. Consider the intermediate kernels \begin{equation}\label{root:hecke-carrier} E(-H)\subset G\subset E(H),\quad \chi(G)=\chi(E),\quad\operatorname{profile}(G)=\boldsymbol\mu, \quad\Phi(G)\subset G\otimes L, \end{equation} with $\sigma_Hs$ factoring through $G(D)$. Relative to $\mathcal G$ the finite module $E(H)/E(-H)$ has modified length $4rS$ and $E(H)/G$ has length $2rS$. These lengths are constant also at a root and on the collision diagonal. Lemma~\ref{root:positive-mass} makes the carrier projective; the last conditions are closed. Each Hecke word maps properly to it, by multiplying its initial bundle by $\sigma_H$. The difference of their cycles is supported on the endpoint graph $x=y$, $G=\sigma_HE$. The prescribed degree makes its coefficient a degree-zero class on each graph component. At a dense point with regular semisimple $\Phi_y$ and nonzero section in each eigenline, removal first is absent and creation then removal has $r$ transverse branches, each of multiplicity $+1$. Thus the coefficient is $r$. Jet surjectivity proves that this computation meets every graph component. The collision takes place at the ordinary point $y$, so its orientation is exactly the local orientation computed in Proposition~\ref{cmp:weyl-relation}. This cycle equality gives the full arrow equality. In the notation of the supported coreflector \eqref{cmp:arrow-coreflector}, for an object $A_0$ supported on the residue support $S_Y$, the natural map \[ \operatorname{Map}(A_0,\Gamma_{S_Y}K_Y) \longrightarrow\operatorname{Map}(A_0,K_Y) \] is an equivalence of mapping spaces. It preserves the homotopy just constructed. Forward vanishing-cycle exchange, its proper inverse, and the right mate therefore lift the entire difference arrow and its cofiber. To descend from a chosen residue divisor, choose two sufficiently large ordinary divisors $D_1,D_2$ with disjoint supports, both avoiding the roots and $y$. Their admissible chart opens cover the moving curve. Compare both with $D_1+D_2$. The same $H^1$ vanishings choose splittings zero on the active jets, with an affine space of choices. The added hyperbolic block has rank $\deg(D'-D)r(r+1)$, which is even, and its counit is $+1$. Full frame changes are handled by smooth descent. The enhanced coherences glue before taking proper trace over $C$. That trace cancels the $L^{-1}$ factor and takes the class of $y$ to one. This proves \eqref{root:weyl}. \end{proof} \begin{corollary}\label{root:hall-comparison} The conclusions of Theorems~\ref{cmp:comparison} and~\ref{cmp:source-support} hold for the full root numerical class. All their arrows retain the specified physical orientations and finite permutation actions. \end{corollary} \begin{proof} Use the finite intervals of Lemma~\ref{root:windows} and the whole-arrow relation \eqref{root:weyl}. The nilpotence argument of Lemma~\ref{cmp:weyl-nilpotence} is an argument in a bounded enhanced diagram; it depends only on these two facts and on the nonzero scalar $r$. It applies separately to every profile and to its dual profile. Lemma~\ref{root:positive-mass} gives the finite graded-face data required for the bounded Hall diagrams. Proposition~\ref{root:physical-descent} supplies all the root orientation squares used in their maps. \end{proof} \section{Positive root Hitchin support and canonical primitives} \label{root:support-section} \subsection{Full support in every genus} For a fixed profile write $M^L_{r,\chi,\boldsymbol\mu}$ for the coarse moduli of semistable $L$-Higgs bundles and \[ h_L:M^L_{r,\chi,\boldsymbol\mu}\longrightarrow \cA_r^L=\bigoplus_{j=1}^r H^0(C,L_0^j) \] for the positive Hitchin map. The weights are rational; replacing the root orders by common multiples realizes any rational weights. The profile keeps its ordered character labels under this replacement. We use $\IC$ with its perverse normalization. Set \begin{equation}\label{root:positive-dimensions} \begin{split} b_r&=\dim\cA_r^L=\frac{\ell r(r+1)}2+r(1-g),\\ \gamma_r&=1+r(g-1)+\frac{\ell r(r-1)}2,\qquad B_r=\ell r^2. \end{split} \end{equation} Thus $b_r+\gamma_r=B_r+1$. The character terms in $\chi(\End E)$ and $\chi(\End E\otimes L)$ cancel, because $L$ has trivial inertia. \begin{theorem}\label{root:positive-support} For sufficiently large $\ell$ the following statements hold for every $g\geq0$, every profile, and every Euler characteristic. The moduli $M^L_{r,\chi,\boldsymbol\mu}$ is irreducible and has a dense stable locus. Every simple constituent of \[ \pH^iRh_{L*}\IC_{M^L_{r,\chi,\boldsymbol\mu}} \] has full support $\cA_r^L$. These perverse Hodge modules are independent of $\chi$, with the identifications obtained from the degree-independent Jacobian variation on the smooth spectral locus and intermediate extension. \end{theorem} \begin{proof} First suppose that the parabolic numerical type admits weights for which semistability is stability. This always holds after the auxiliary full-flag refinement used at the end of the proof; a divisible partial-flag type need not admit such weights itself. Choose weights in such a chamber. The obstruction space is dual to $\Hom_{\rm Higgs}(E,E\otimes A)$, which vanishes because $A=KL^{-1}$ has negative degree. The stable coarse space is therefore smooth of dimension $B_r+1$. The same argument works when the distinct marked points vary over $U=\Conf_m(C)$ and proves smoothness relative to $U$. The Hitchin map is proper: after the spectral construction the objects are semistable sheaves with fixed modified Hilbert polynomial on the projective finite spectral curve, and the valuative extension followed by semistable reduction is unique on the coarse moduli. Tameness and the generating bundle \eqref{root:generating-bundle} give the same construction for root sheaves. We require a uniform fiber-dimension bound. Choose a reduced divisor $D'\in|L_0K_C^{-1}|$ containing all marked points. On the nilpotent fiber the weakly parabolic residues preserve their partial flags and are nilpotent. Refine each flag to a full flag strictly lowered by its residue, and add full flags at the other points of $D'$. Perturb the weights by sufficiently small generic numbers. This preserves stability of the old objects: the possible subprofiles are finite and their integral subdegrees give a positive gap from zero in each strict inequality. The resulting full strongly parabolic nilpotent moduli maps onto the original nilpotent fiber. Its dimension is at most \begin{equation}\label{root:nil-dimension} 1+r^2(g-1)+\deg(D')\frac{r(r-1)}2=\gamma_r. \end{equation} Here is the isotropic argument, which also covers $g=0,1$. Resolve a nilpotent stratum by a filtration on which the field strictly lowers degree. A tangent deformation of the bundle preserves this filtration, whereas the field deformation lowers it. Their composite is strictly filtration-lowering and has trace zero. Thus the Liouville one-form on the strongly parabolic cotangent moduli restricts to zero on this resolution. Its differential, the trace symplectic form, also restricts to zero. On a dense smooth locus of each image component this makes the component isotropic, so its dimension is at most half the dimension of the stable strongly parabolic moduli. Riemann--Roch with a full flag at each point of $D'$ gives that half-dimension as the left side of \eqref{root:nil-dimension}. Filtration ranks and degrees are bounded on the semistable locus, so finitely many such strata suffice. This is also the filtration-and-trace proof of the parabolic nilpotent isotropy statement in \cite[Section 3.2]{rootBGL}; it uses no lower bound on the genus. Properness and positive-weight scaling of the Hitchin base imply that every fiber has dimension at most $\gamma_r$. Let $P$ be the identity component of the ordinary spectral Picard acting by tensor product. Its multidegree on every reduced spectral component is zero. It therefore preserves parabolic degree, every subobject inequality, and the root-character profile. Stabilizers are affine. Indeed, on the normalization of every reduced component the sheaf has positive rank. A line bundle stabilizing it has a torsion power by taking its determinant, so the stabilizer image in the product of the normalization Jacobians is finite. The kernel of the map from the spectral Picard to that product is affine. Its Tate module has the polarization induced from the normalization Jacobians. We use the weak abelian support inequality in the following precise form. For a proper map from a smooth space with a polarizable relative Picard action and affine stabilizers, and with fiber dimension bounded by the arithmetic genus $\gamma_r$, a strict support $Z$ satisfies \[ \operatorname{codim}Z\leq\delta_Z, \] where $\delta_Z$ is the dimension of the affine part of the generic Picard. Equivalently its abelian dimension $\beta_Z$ satisfies $\beta_Z\leq\gamma_r-\operatorname{codim}Z$. We also use the integral spectral-curve inequality $\beta_W\geq\gamma_r-b_r+\dim W$ on an integral locus. Both follow from the exterior-algebra action of the Tate module and the integral $\delta$-regularity estimate; these are the support inputs in \cite{rootCL,rootCLAdd} and \cite{MSChi}. Their numerical comparison is given here to specify their application to moving partial flags. Let $W$ be the projection of $Z\subset\cA_r^L\times U$ to $\cA_r^L$. At its generic point write the spectral polynomial as a product of distinct integral factors of degrees $n_1,\ldots,n_s$, with multiplicities $m_1,\ldots,m_s$, so that $r=\sum_i m_i n_i$. The finite multiplication map gives a finite cover of $W$ with projections $W_i$ to the integral degree-$n_i$ Hitchin bases. The abelian part is the product of their normalization Jacobians. The two preceding inequalities give \begin{equation}\label{root:support-inequalities} \sum_i(\gamma_{n_i}-b_{n_i}+\dim W_i) \leq\beta_W \leq\gamma_r-b_r+\dim W \leq\gamma_r-b_r+\sum_i\dim W_i. \end{equation} The last step uses only $\dim W\leq\sum_i\dim W_i$; the projections need not be independent. Since $\gamma_n-b_n=1+n(2g-2-\ell)$, cancellation gives \begin{equation}\label{root:support-numerical} 1-s\geq\left(r-\sum_i n_i\right)(\ell-2g+2). \end{equation} As $\ell>2g-2$, this forces $s=1$ and $m_1=1$. The integral inequality then forces equality throughout the dimension estimate for $Z$, so $Z=W\times U$. Now move the marked points away from the branch and singular loci of the generic curve in $W$. Over this open set the choice of profile is the finite etale allocation cover with fiber \begin{equation}\label{root:allocation-cover} \prod_p\mathfrak S_r\big/\prod_c\mathfrak S_{\mu_{p,c}}. \end{equation} The spectral sheaf is locally a line on each of these branches; prescribing its characters changes neither its coarse Jacobian nor its ordinary Euler characteristic. The direct image is therefore the ordinary positive direct image tensored with this finite allocation variation. Here the ordinary integral-locus support statement is needed in every genus. In its precise form, for a smooth projective curve of any genus, an effective twist of degree $>2g-2$, and any integer spectral degree, the proper compactified Jacobian over the integral spectral locus has smooth total space and every perverse direct-image constituent is the intermediate extension from the smooth spectral locus. This is the elliptic support theorem in \cite[Sections 2 and 4, Theorem 7.1]{rootCL}; the all-base coprime form is \cite[Theorem 9.1]{rootCL} with the correction in \cite{rootCLAdd}. Its genus hypotheses therefore apply also for $g=0,1$. The support assertion for rational Hodge modules follows from this underlying rational perverse support assertion and the pure Hodge decomposition theorem. One can also reduce the arbitrary-degree integral statement to its coprime form locally, which makes its use here transparent. The smooth locus of the universal integral spectral curve is a smooth surjective morphism. Etale locally around any point of its base, choose a section through a smooth point avoiding all marked branches. Its divisor defines a relative degree-one line bundle. Tensoring by an integral power identifies the compactified Jacobians of any two Euler characteristics. Integral spectral support makes all their rank-one torsionfree sheaves stable, so this is an isomorphism of the whole moduli problem in this neighborhood, with no stability wall. Choose the second Euler characteristic coprime to $r$ and apply the all-genus coprime support theorem. The allocation cover \eqref{root:allocation-cover} is finite etale in the neighborhood under consideration; after a further etale pullback it is a disjoint union of copies of that neighborhood. On each copy root descent is precisely the ordinary compactified-Jacobian problem just described, and tensoring at the chosen ordinary section leaves all root characters unchanged. Etale pullback and finite-group projectors commute with intermediate extension. Thus this allocation cannot produce a proper strict support. It follows that $W=\cA_r^L$, proving full support over $\cA_r^L\times U$. Restriction to fixed marked points preserves that conclusion. The source is smooth over $U$. A nonzero covector in a $U$-direction has nonzero pullback, so the proper direct-image estimate for singular support shows that the slice at a fixed point of $U$ is noncharacteristic. Its shifted pullback is perverse exact and takes the intermediate extension just found to the intermediate extension of the restricted variation. Thus it creates no new strict support. We next prove the irreducibility needed at nongeneric weights, without assuming the existence of stable underlying bundles. The stack of root bundles of a fixed rank, degree, and profile is a partial-flag fibration over the ordinary bundle stack, hence is irreducible. On a Harder--Narasimhan stratum let $E_1,\ldots,E_v$ be the semistable pieces in strictly descending order of slope. Its codimension is \begin{equation}\label{root:hn-codimension} c=\sum_{i0$; if $c=0$, the excess is zero. Thus every positive codimension stratum contributes dimension strictly less than $B_r$ to the Higgs stack. Only finitely many such strata meet Higgs semistability. Successive slope gaps are at most $\ell$, since a larger gap makes the upper part Higgs invariant; the total spread is at most $(r-1)\ell$. Fixed total degree and the finite set of profiles then bound all the degrees. The Higgs stack is smooth of pure dimension $B_r$. There are stable points for every $\chi$ and every profile: take a smooth connected spectral curve unramified over the marks, allocate the characters, and take a line of the desired Euler characteristic. Its integral spectral support permits no proper invariant generic subspace. Every component therefore meets the unique codimension-zero Harder--Narasimhan stratum of the irreducible bundle stack. On that stratum the Higgs stack is a vector bundle and its semistable open is irreducible. The whole semistable stack, and hence its coarse space, is irreducible with dense stable locus. The codimension-zero stratum can be unstable on $\mathbf P^1$. Finally, at a weight wall introduce an auxiliary ordinary full invariant flag and perturb its weights by a sufficiently small generic amount. The new semistable objects are old semistable objects, and their forgetful map $\pi:M_{\rm new}\to M_{\rm old}$ is projective over the Hitchin base. Over the smooth unramified spectral locus it is finite etale of degree $r!$. Its image is dense and closed, hence all of the irreducible old space. The trivial summand of the generic permutation local system extends, by the decomposition theorem, to a direct summand $\IC_{M_{\rm old}}$ of $R\pi_*\Q_{M_{\rm new}}[B_r+1]$. Full support of the new direct image therefore implies full support of the old IC direct image. On the smooth spectral allocation cover the fibers are Jacobian torsors of varying degree. Their cohomology is the exterior algebra of the same $H^1$ variation and is independent of the degree. The allocation system also depends only on the profile. Full support and uniqueness of pure intermediate extension now give the asserted isomorphisms of perverse Hodge modules for all $\chi$, including wall classes. \end{proof} \subsection{Active profiles and the target Euler characteristic} Fix a target profile $\boldsymbol\mu$ of rank $n$. A profile $\boldsymbol\beta$ is called active if $\beta_{p,c}\leq\mu_{p,c}$ for all $p,c$, and the sums $\sum_c\beta_{p,c}$ have the same value $r$ at every marked point. The active set is finite and is closed under taking summands of an active total. \begin{lemma}\label{root:active-weights} Within a prescribed target determinant hyperplane, one can choose rational weights for which every active profile is stable whenever it is semistable at Euler characteristic $\chi_*=1$. \end{lemma} \begin{proof} Let the weights be $\alpha_{p,c}$ and write the target hyperplane as $\sum_{p,c}\mu_{p,c}\alpha_{p,c}=-d_0$. For an active rank-$r$ profile $\boldsymbol\beta$ and a proper rank-$r'$ subprofile $\boldsymbol\beta'$, the slope wall at $\chi_*=1$ is \begin{equation}\label{root:active-wall} (r\boldsymbol\beta'-r'\boldsymbol\beta) \mathbin{\cdot}\boldsymbol\alpha =r'-r\chi'. \end{equation} In a bounded weight chamber only finitely many integers $\chi'$ occur. The sum of the entries of its normal at each marked point is zero, whereas that of the target normal is $n$. Consequently a nonzero wall normal is not proportional to the target normal. If the wall normal is zero, its right side is nonzero because $0\gamma_a} +\sum_{(a,b)\in A_\lambda}\ind_{\gamma_b<\gamma_a}. \end{equation} \begin{proposition}[Closed index]\label{vert:closed-index} For every partition, every root order, and every legal height array, \begin{equation}\label{vert:closed-index-equation} J(m)=-d_0(\lambda,s)+\sum_{b\in\lambda}\sigma_b(r_b-j_b) +2\dinv_\lambda(\gamma). \end{equation} \end{proposition} \begin{proof} Write $\epsilon(e)=\ind_{s\mid e}$. The finite-difference kernel obtained by applying the first expression in \eqref{vert:indices} to \eqref{vert:finite-character} is \begin{align*} A_s(D,e)={}&\epsilon(e) \bigl(\sgn_{\rm lex}(D,e)-\sgn_{\rm lex}(D-1,e)\bigr)\\ &+\epsilon(e-1) \bigl(-\sgn_{\rm lex}(D+1,e-1)+\sgn_{\rm lex}(D,e-1)\bigr). \end{align*} It is zero outside $D=-1,0,1$. At these three values it is, respectively, \begin{equation}\label{vert:kernel} \begin{split} A_s(-1,e)&=-\epsilon(e-1)(1+\sgn(e-1)),\\ A_s(0,e)&=\epsilon(e)(1+\sgn(e)) +\epsilon(e-1)(\sgn(e-1)-1),\\ A_s(1,e)&=\epsilon(e)(1-\sgn(e)). \end{split} \end{equation} The $H$-kernel $A_s^H$ is obtained by replacing every $\sgn_{\rm lex}(D,e)$ by $\sgn(D)$; its three values are $-\epsilon(e-1)$, $\epsilon(e)-\epsilon(e-1)$, and $\epsilon(e)$. Here is the finite-character calculation leading to this kernel. Raise the height at $a=(i,j)$ by one, and let $\alpha=m_a+1$. Set $z=t_1^{-i}t_2^{-j}t_3^\alpha$. Replacing $\mathsf Z_m$ by $\mathsf Z_m+z$ in \eqref{vert:finite-character} gives \begin{align*} \Delta v={}&z-\kappa\overline z\\ &-(1-t_1)(1-t_2) \bigl[-t_3\mathsf Z_\lambda\overline z+ \overline{\mathsf Z_\lambda}z +(1-t_3)(z\overline{\mathsf Z_m} +\mathsf Z_m\overline z+1)\bigr]. \end{align*} For another box $(l,k)$ put $D=k-l-(j-i)$ and $e=\alpha+j-k$. Its baseline contributes $-A_s(D,e)$; its occupied levels contribute the telescoping sum \[ \sum_{\beta=1}^{m_{lk}} \bigl(-A_s(D,e-\beta)+A_s(D,e-\beta+1)\bigr) =A_s(D,e)-A_s(D,e-m_{lk}). \] The source contributes $\epsilon(\alpha+j)\sgn_{\rm lex}(-c_a,\alpha+j)$. The constant $-P=-(1-t_1)(1-t_2)(1-t_3)$ contributes $1$ if $s=1$, and $-1$ otherwise. The corresponding $H$-constants are $0$ and $-1$. The moving self term always has $(D,e)=(0,1)$; its kernel is exactly the relevant constant, so the two cancel before summing the other boxes. This also treats $s=1$. Construct $m$ from zero in reverse row-major order. During the construction of the height at $(i,j)$, every southeast box has its final height and every northwest box still has height zero. Every box on one of the three contributing diagonals is comparable to $(i,j)$. Substitution in $2A_s^H-A_s$ gives the following contributions to $\Delta J$, in addition to the source $f(c_a)\epsilon(\alpha+j)$, where $f(c)=1$ for $c<0$ and $f(c)=-1$ for $c\geq0$: \[ \begin{array}{c|cc} D&\text{northwest box }(l,k)&\text{southeast box }(l,k)\\ \hline 0&2\epsilon(\alpha+j-k-1)&-2\epsilon(\alpha+j-k-m_{lk})\\ 1&-2\epsilon(\alpha+j-k)&0\\ -1&0&2\epsilon(\alpha+j-k-m_{lk}-1). \end{array} \] The endpoint signs in this table follow directly from legality. For $D=0$, the northwest and southeast kernel arguments are at least $2$ and at most $-1$, respectively. For $D=1$ they are at least $1$ and at most $-1$, and for $D=-1$ at least $2$ and at most $0$. Thus the possible zero argument occurs precisely where the accompanying sign factor vanishes. Using $\sum_{\alpha=1}^{M}\epsilon(\alpha+a) =\lfloor(M+a)/s\rfloor-\lfloor a/s\rfloor$ and combining the two appearances of each pair yields \begin{align}\label{vert:floor-index} J(m)={}&\sum_b f(c_b) \left(\left\lfloor\frac{\ell_b}{s}\right\rfloor -\left\lfloor\frac{j_b}{s}\right\rfloor\right)\\ &+2\sum_{(a,b)\in S_\lambda} \left(\left\lfloor\frac{\ell_b-\ell_a-1}{s}\right\rfloor -\left\lfloor\frac{j_b-j_a-1}{s}\right\rfloor\right) \nonumber\\ &-2\sum_{(a,b)\in A_\lambda} \left(\left\lfloor\frac{\ell_b-\ell_a}{s}\right\rfloor -\left\lfloor\frac{j_b-j_a}{s}\right\rfloor\right). \nonumber \end{align} We identify both its linear part and its constant. Give the attacking graph the directed edges $b\to a$ for $(a,b)\in S_\lambda$ and $a\to b$ for $(a,b)\in A_\lambda$. For a box $b=(i,j)$, \begin{equation}\label{vert:graph-maya} \sigma_b=f(c_b)+2\bigl(\outdeg(b)-\indeg(b)\bigr). \end{equation} Indeed, northwest same-diagonal boxes number $\min(i,j)$, and northwest boxes of content $c_b+1$ number $\min(i,j+1)$. Their difference is $-\ind_{i>j}$. The difference between the numbers of southeast boxes of content $c_b-1$ and $c_b$ is \[ \#\{h\geq1:\lambda_{i+h}=j+h\} =\ind_{c_b-1\in\mathcal B_\lambda}. \] There is at most one such $h$, since row lengths decrease while $j+h$ increases; a Maya equality with row index at most $i$ is impossible because $(i,j)$ lies in the diagram. These two differences prove \eqref{vert:graph-maya}. Hence the coefficient of $r_b$ in \eqref{vert:floor-index} is exactly $\sigma_b$. For the constant use the legal array $m_b^0=(s-1)j_b$, for which $r_b=j_b$ and all $\gamma_b$ vanish. Put $A_s(k)=k-\lfloor k/s\rfloor$. Substitution in \eqref{vert:floor-index} gives \begin{equation}\label{vert:baseline-sum} \frac{J(m^0)+d_0}{2} =\sum_{c_b<0}A_s(j_b) +\sum_{(a,b)\in S_\lambda}A_s(j_b-j_a-1) -\sum_{(a,b)\in A_\lambda}A_s(j_b-j_a). \end{equation} In the last sum, move the northwest box one step left whenever its column is positive. This is a bijection onto $S_\lambda$, and preserves the displayed argument of $A_s$. Its inverse is legal because the southeast box forces the requisite box in the longer northwest row. The remaining pairs have northwest column zero and correspond bijectively to boxes of negative content, with argument $j_b$. Thus \eqref{vert:baseline-sum} vanishes. Finally, \[ \left\lfloor\frac{\ell_b-\ell_a-1}{s}\right\rfloor =r_b-r_a-\ind_{\gamma_b\leq\gamma_a},\qquad \left\lfloor\frac{\ell_b-\ell_a}{s}\right\rfloor =r_b-r_a-\ind_{\gamma_b<\gamma_a}. \] Subtracting the baseline in \eqref{vert:floor-index} proves \eqref{vert:closed-index-equation}. \end{proof} \subsection{Stable alphabets and the normalized series} The inequalities on $m$ have a useful equivalent form. The integers $\ell_b=m_b+j_b$ increase strictly in rows and weakly in columns. Consequently $r$ is a reverse plane partition; within an equal-$r$ level, $\gamma$ increases strictly in rows and weakly in columns. Conversely these conditions imply the inequalities on $\ell$ and hence on $m=\ell-j$. Nonnegativity follows from the first column and the successive strict row increments. Thus this description depends only on the order of the alphabet, not on its cardinality. Let $q,t$ be algebraically independent. For an ordered alphabet $X_s=x_0+\cdots+x_{s-1}$ set $X_s^\gamma=\prod_bx_{\gamma_b}$. \begin{definition}[Punctual vertex]\label{vert:definition} The normalized root vertex is the stable homogeneous series whose value on $X_s$ is \begin{equation}\label{vert:G-series} G_\lambda[X_s;q,t]= \sum_{\substack{r\in\RPP(\lambda),\ 0\leq\gamma_b0$, this is not $1$. The sum on the type is a monomial divided by $\prod_i(1-q^{-a_i}t^{b_i})$. There are finitely many types and color assignments. Every increase of $u_i$ raises the original height degree by $|U_i|$, so the same geometric sums converge in that completion. Deletion of a letter retains the order of the others and all defining comparisons, proving stability. \end{proof} \subsection{The LLT expansion and the one-letter value} For $r\in\RPP(\lambda)$ put $\lambda_{\leq h}=\{b:r_b\leq h\}$ and $\rho_h=\lambda_{\leq h}/\lambda_{j$ a pair with the first content one larger is inverted under the same comparison. The content of a transposed box is measured so that it equals the original $j-i$. This specifies all attacks and agrees with the usual inversion convention. \begin{proposition}\label{vert:llt} With \[ W_\lambda(r;q,t)=\prod_{\sigma_b=-1}q^{j_b-r_b} \prod_{\sigma_b=1}t^{r_b-j_b},\qquad \kappa(r)=\#\{(a,b)\in S_\lambda:r_a=r_b\}, \] one has \begin{equation}\label{vert:llt-expansion} G_\lambda[X;q,t]=\sum_{r\in\RPP(\lambda)} W_\lambda(r;q,t)(qt)^{\kappa(r)} \mathcal G_{\nu(r)}[X;qt]. \end{equation} In particular $G_\lambda\in\Q(q,t)\otimes\Lambda_{|\lambda|}$. Furthermore, \begin{equation}\label{vert:one-letter} G_\lambda[1;q,t]=D_\lambda(q,t)^{-1},\qquad F_\lambda[1;q,t]=1. \end{equation} \end{proposition} \begin{proof} Transposition changes the levelwise row-strict, column-weak condition into the ordinary semistandard condition. On different levels the southeast member of a same-content pair lies in the earlier component, and gives exactly the first term in \eqref{vert:dinv}. For an adjacent-content pair the northwest member lies in the later component and has content one larger; its inversion is the second term. All possible attacks on these diagonals are comparable, so this list is exhaustive. On the same level a same-content southeast box has a strictly larger entry: an order-convex path inside that level contains a horizontal step. Every such pair contributes one. A same-level pair in $A_\lambda$ contributes zero. Hence $\dinv=\Inv_{\rm LLT}+\kappa(r)$ and \eqref{vert:llt-expansion} follows. The symmetry theorem for inversion LLT polynomials \cite[Theorem 3.3]{HHLVertex} gives symmetry term by term in the height completion. Lemma~\ref{vert:rationality} then places the stable symmetric function in the stated field. For one letter, the equal-level strict row condition says $r_b=j_b+u_b$ for a unique $u\in\RPP(\lambda)$, and $\dinv=0$. Thus the one-letter sum is the ordinary diagonal-weighted reverse-plane-partition series. Under the rim-hook insertion bijection, each box supplies an independent nonnegative multiplicity; its hook traverses $a_b+1$ negative Maya signs and $l_b$ positive signs. The diagonal-hook calculation \eqref{fock:hook-index}, applied to the weights $q^{-1}$ and $t$, gives the factor $(1-q^{-a_b-1}t^{l_b})^{-1}$ for this multiplicity. Multiplying the independent geometric series proves \eqref{vert:one-letter}. \end{proof} \section{Supersymmetrization at fixed paired heights}\label{vert:paired-section} We next prove the complete dominance condition for $G_\lambda[(t-1)X;q,t]$. It is useful to make the finite polynomial at each paired height explicit. Put $T=qt$ and write $E_\lambda$ for the directed graph used in \eqref{vert:graph-maya}. \subsection{Signed letters and the finite polynomial} For $X=x_1+\cdots+x_s$, use the signed residue order \[ \bar1<\bar2<\cdots<\bar s\eta_R&L\text{ positive}\\ U,D\text{ in one column}&\eta_U<\eta_D&D\text{ negative}\\ U,D\text{ in one column}&\eta_U\geq \eta_D&U\text{ positive}. \end{array} \end{equation} All zero-height boxes are added to $N$. A conflicting pair of forced signs means that the contribution is zero. Write $\outdeg_{>\eta_b}(b)$ for the number of outgoing neighbors whose absolute color exceeds $\eta_b$, and similarly for other inequalities and incoming neighbors. Define \begin{align}\label{vert:fixed-p-data} D_0(\eta)&=\#\{u\to v\text{ in }E_\lambda:\eta_u\geq \eta_v\},\nonumber\\ B_b(\eta)&=\ind_{\sigma_b=-1} +\outdeg_{>\eta_b}(b)-\indeg_{\geq \eta_b}(b)\\ &=\ind_{c_b\geq0} -\outdeg_{\leq \eta_b}(b)+\indeg_{<\eta_b}(b),\nonumber\\ M_\lambda(p;q,t)&=\prod_{\sigma_b=-1}q^{j_b-p_b} \prod_{\sigma_b=1}t^{p_b-j_b}. \nonumber \end{align} The equality of the two formulas for $B_b$ is \eqref{vert:graph-maya}. \begin{definition}\label{vert:K-definition} For a nonnegative reverse plane partition $p$, put \begin{equation}\label{vert:K-sign-sum} K_{\lambda,p}[X;T]= \sum_{\eta:\lambda\to\{1,\ldots,s\}} X^\eta T^{D_0(\eta)} \prod_{b\in P}T^{B_b(\eta)} \prod_{b\in N}(-1) \prod_{b\in U}\bigl(T^{B_b(\eta)}-1\bigr), \end{equation} where $X^\eta=\prod_b x_{\eta_b}$, with the convention just stated for conflicting forced signs. This is a finite homogeneous polynomial in the $x_i$ with coefficients in $\Z[T,T^{-1}]$. \end{definition} \begin{lemma}\label{vert:paired-identity} The exact plethystic identity is \begin{equation}\label{vert:paired-identity-equation} G_\lambda[(t-1)X;q,t] =\sum_{p\in\RPP(\lambda)}M_\lambda(p;q,t) K_{\lambda,p}[X;qt]. \end{equation} Equivalently, put all positive letters before all negative letters, in the order \begin{equation}\label{vert:boundary-alphabet} s<\cdots<1<\bar1<\cdots<\bar s. \end{equation} Then $K_{\lambda,p}$ is the signed-tableau sum on the equal-$p$ skew components with weight \begin{equation}\label{vert:boundary-weight} X^\eta(-1)^{\#\{b:\varepsilon_b=0\}} T^{D'(\eta,\varepsilon)+\#\{b:c_b\geq0,\ \varepsilon_b=1\}}, \end{equation} where $D'$ is the inversion count on $E_\lambda$ in \eqref{vert:boundary-alphabet}, including equal negative pairs. The zero-height component is entirely negative. \end{lemma} \begin{proof} Inspection of the intervals with endpoints $2sp-\eta$ and $2sp+\eta-1$ proves \eqref{vert:unary-rules}. For example, if $\eta_L\leq \eta_R$, a positive letter at $R$ lies at or below either allowed letter at $L$, with the positive equality forbidden in a row; hence $R$ is negative. If $\eta_L>\eta_R$, the analogous obstruction forces $L$ positive. The column argument reverses the equality rule. These are all restrictions, since the tableau conditions are adjacent conditions. Consequently every unforced sign may be chosen independently. For an edge $u\to v$, its old inversion indicator is $\varepsilon_u$ if $\eta_u<\eta_v$, and $1-\varepsilon_v$ if $\eta_u\geq \eta_v$. Thus \[ D=D_0+\sum_b \bigl(\outdeg_{>\eta_b}(b)-\indeg_{\geq \eta_b}(b)\bigr) \varepsilon_b. \] Replacing $r_b$ by $p_b-\varepsilon_b$ in the height weight and including the letter factor $t^{\varepsilon_b}$ gives \[ M_\lambda(p;q,t)(-1)^{\#\mathrm{negative}} T^{D+\sum_{\sigma_b=-1}\varepsilon_b}. \] The exponent of each positive sign is exactly $B_b(\eta)$. Summing a forced positive, forced negative, or free sign gives, respectively, $T^{B_b}$, $-1$, or $T^{B_b}-1$. This proves \eqref{vert:paired-identity-equation}. Changing the residue order to \eqref{vert:boundary-alphabet} changes the inversion on an edge $u\to v$ by $\varepsilon_u-\varepsilon_v$. Therefore \[ D+\sum_{\sigma_b=-1}\varepsilon_b =D'+\sum_b\bigl(\ind_{\sigma_b=-1} +\outdeg(b)-\indeg(b)\bigr)\varepsilon_b =D'+\sum_{c_b\geq0}\varepsilon_b. \] The equal-$p$ order in \eqref{vert:signed-position} is already \eqref{vert:boundary-alphabet}, proving the last assertion. Every rearrangement is finite in each original height degree by Lemma~\ref{vert:rationality}. \end{proof} \section{The signed lattice and its intertwiners}\label{vert:lattice-section} In this section $\tau=T^{-1}$ is the vertex parameter; it is distinct from the Macdonald parameter $t$. There are finitely many ordered path colors. A state is a subset of these colors, identified with its zero-one vector. Sums and differences of states in an exponent are vector sums and differences. Define \[ \phi(U,V)=\sum_{ih'$; their path colors occur in this order in $\phi$. Fix a box $b\in\rho_h$ at site $u=i_b-j_b$, and let $r$ be the row of its signed letter in \eqref{vert:boundary-alphabet}. For the lower plateau put \[ a_0(u)=\ind_{u\in\{\alpha'_{h',j}-j:j\geq0\}},\qquad n_v^{h'$ accounts for every summand of $C$ and every cross-plateau attack. Within one plateau, signed strip monotonicity makes each same-content pair contribute one to $D'$ and each adjacent-content pair contribute zero; there is no same-color summand in $\phi$. Hence $D'(f)+C(f)$ is independent of the signed filling. Evaluation on the all-negative row-major filling identifies this constant with the specified $m_{\lambda,p}$ and proves \eqref{vert:coinv-normalization}. \begin{proposition}[Lattice realization]\label{vert:lattice} Let $Z_{\lambda,p}(X;\tau)$ be the above finite path partition function with row parameters \begin{equation}\label{vert:step-parameters} a_i(u)=\begin{cases}x_i/\tau,&u\leq0,\\x_i,&u>0, \end{cases} \qquad b_i(u)=-x_i. \end{equation} When $p=0$ occurs, impose the vacuum Maya state of that last color between the positive and negative row blocks. Then \begin{equation}\label{vert:K-lattice} K_{\lambda,p}[X;T]=T^{m_{\lambda,p}} Z_{\lambda,p}(X;T^{-1}). \end{equation} \end{proposition} \begin{proof} The path encoding is a bijection with the signed fillings in Lemma~\ref{vert:paired-identity}. The product of the row monomials supplies $X^\eta(-1)^{\#\mathrm{negative}}$. The local exponent is $\tau^{C(f)}=T^{D'(f)-m_{\lambda,p}}$ by \eqref{vert:coinv-normalization}. A positive box has $c_b\geq0$ precisely when its site is $u\leq0$; the extra factor $\tau^{-1}$ in \eqref{vert:step-parameters} supplies the remaining $T$ for that box. This proves \eqref{vert:K-lattice}. The zero-height restriction says that its positive skew shape is empty, which is exactly the imposed midpoint vacuum. \end{proof} \begin{proposition}[Fixed-height symmetry]\label{vert:fixed-p-symmetry} For every $p$, the polynomial $K_{\lambda,p}(X;T)$ is symmetric in $X$. A chosen set of $k$ variables can be moved to the innermost $k$ matched pairs in \eqref{vert:nested-order}, while retaining all other rows. \end{proposition} \begin{proof} Exchange two neighboring variable pairs by exchanging their positive rows with $R$ and their negative rows with $R''$. Equation~\eqref{vert:RLL} moves these crossings through every column. On the left all positive parameters are multiplied by the same $\tau^{-1}$; hence their ratios are unchanged. The positive crossing is therefore identical on the two sides of the parameter cut. The negative crossing is identical there as well. The empty horizontal boundaries are fixed, with coefficient $1$, so the complete partition function is unchanged. If a zero-height color is present, the positive rows must preserve its vacuum state. Equivalently, their horizontal occupations of that color all vanish. A same-sign positive crossing preserves the sum of these occupations in the two rows, and a negative crossing does not act on them. Thus the paired permutation commutes with this restriction and the same argument applies. The boundary scalar in \eqref{vert:K-lattice} is independent of the variables. Adjacent pair exchanges generate all variable permutations, including one that places any chosen set innermost. \end{proof} \section{Fusion with its full null submodule}\label{vert:fusion-section} \subsection{A rank-one identity for arbitrary incoming states} For a binary set $B$ define \begin{equation}\label{vert:fused-vector} v_B(a,b)=\sum_{I\subset B} a^{|I|}b^{|B-I|}\tau^{\phi(I,B)}\,|I,B-I\rangle, \qquad Z_d(a,b)=\prod_{r=0}^{d-1}(b+a\tau^r). \end{equation} The coefficient sum of $v_B$ is $Z_{|B|}$: each color in $B$ contributes $b+a\tau^r$, where $r$ is the number of larger colors of $B$. Let $\mathcal K(a,b)$ be the vector space with basis $e_B$, embedded in $P_aN_b$ by $e_B\mapsto v_B(a,b)$. This embedding is defined whenever $b\ne0$, including all specializations below, because its all-negative coordinate is $b^{|B|}$. \begin{lemma}[Raw two-row identity]\label{vert:raw-all} The incoming horizontal states $I,J$ need not be disjoint. Every nonzero product $L_a(A,I;M,K)L'_b(M,J;C,L)$ has $K\cap L=\varnothing$. Writing $D=K\sqcup L$, for every $K\subset D$ one has \begin{align}\label{vert:RAW-ALL} &\sum_M L_a(A,I;M,K)L'_b(M,J;C,D-K)\nonumber\\ &\qquad=\ind_{A\cap I=\varnothing}\ind_{J\subset C} \ind_{A+I+J=C+D}\, \tau^{\phi(D,C-J)} a^{|K|}b^{|D-K|}\tau^{\phi(K,D)}. \end{align} Thus every fixed incoming split has its outgoing vector on the line spanned by $v_D(a,b)$. \end{lemma} \begin{proof} The negative rule gives $L\subset M$, while the positive rule gives $M\cap K=\varnothing$, proving disjointness. The intermediate state is forced to be $M=A+I-K$. Under the three displayed conditions it is binary for every $K\subset D$. Indeed if $D_i=1$, conservation and $J_i\leq C_i$ force $A_i+I_i=1$ and $C_i=J_i$; if $D_i=0$, there is no split choice at that color. The two exponents satisfy \[ \phi(K,A+I)+\phi(D-K,C-J) =\phi(D,C-J)+\phi(K,A+I-C+J) =\phi(D,C-J)+\phi(K,D). \] The monomial factors are $a^{|K|}b^{|D-K|}$, proving \eqref{vert:RAW-ALL} in every number of colors. \end{proof} \begin{lemma}[Parameter jump]\label{vert:jump} Suppose $A+B=C+D$ with $B,D$ binary. A two-row column of parameters $(a,b)$, applied to the incoming vector $v_B(a_0,b_0)$, has outgoing $D$-component $\mathcal J(A,B;C,D)v_D(a,b)$, where \begin{align}\label{vert:JUMP} \mathcal J(A,B;C,D) ={}&b_0^{|B|}\tau^{\phi(D,C-B)} \prod_{i\in B-C}\left(\frac{a_0}{b_0}\tau^{h_i}\right) \prod_{i\in B-D} \left(1+\frac{a_0}{b_0}\tau^{h_i}\right),\\ h_i={}&\#\{j\in B:j>i\}+\#\{j\in D:j\ell}(a,b) =\operatorname{span}\{e_B:|B|>\ell\} \subset\mathcal K(a,b) \] is preserved by every vertical matrix element of a paired column. Its quotient is spanned by $|B|\leq\ell$. More precisely, take nonnegative integers $\ell_1\leq\ell_0$ and parameters $b_i=-a_i\tau^{\ell_i}$ with $a_i\ne0$. For fixed vertical input $A$ and output $C$, the identity gluing followed by the first paired column with parameters $(a_1,b_1)$ defines a map \[ T^{A,C}_{1\leftarrow0}:\mathcal K(a_0,b_0) \longrightarrow\mathcal K(a_1,b_1),\qquad e_B\longmapsto \sum_{\substack{D\text{ binary}\\A+B=C+D}} \mathcal J_{a_0,b_0}(A,B;C,D)e_D, \] where $\mathcal J_{a_0,b_0}$ denotes the coefficient $\mathcal J$ in \eqref{vert:JUMP}. This map descends to \[ \mathcal K(a_0,b_0)/\mathcal H^{>\ell_0}(a_0,b_0) \longrightarrow \mathcal K(a_1,b_1)/\mathcal H^{>\ell_1}(a_1,b_1). \] \end{proposition} \begin{proof} Write $\iota_i(e_B)=v_B(a_i,b_i)$. The parameter-jump identity gives the equality of raw two-row maps \[ \iota_1 T^{A,C}_{1\leftarrow0} =M_{P_{a_1}N_{b_1}}(A,C)\iota_0. \] Here the matrix entry of the one-column operator on the right, from $|I,J\rangle$ to $|K,L\rangle$, is \[ \sum_M L_{a_1}(A,I;M,K)L'_{b_1}(M,J;C,L). \] For a transition with $|B|>\ell_0$ and $|D|\leq\ell_1$, scan the ordered colors from smallest to largest. The integer \[ \#\{\text{unscanned colors in }B\} +\#\{\text{scanned colors in }D\} \] starts above $\ell_0$ and ends at or below $\ell_0$. It therefore has a downward crossing to height $\ell_0$. Such a crossing occurs at a color $i\in B-D$, and its new height is $h_i=\ell_0$. At $b_0=-a_0\tau^{\ell_0}$ the corresponding factor in \eqref{vert:JUMP} is zero. Hence $T^{A,C}_{1\leftarrow0}(\mathcal H^{>\ell_0}) \subset\mathcal H^{>\ell_1}$, proving descent. Taking identical input and output parameters proves invariance at a fixed capacity. The argument uses the unnormalized vectors $v_B$, including when $Z_{|B|}=0$. \end{proof} \begin{corollary}[An innermost pair]\label{vert:inner-invisible} For a matched pair with left parameters $(x/\tau,-x)$ and right parameters $(x,-x)$, every left fused state of degree at least two is annihilated by the cut-and-right-tail observer with empty horizontal boundary. This is an identity for each choice of vertical boundary states, including when further rows surround the pair. \end{corollary} \begin{proof} The left capacity is one, and its high submodule is invariant up to the cut. The first right column realizes the map of Proposition~\ref{vert:one-pair-quotient} with $(a_0,b_0)=(x/\tau,-x)$, $\ell_0=1$, and $(a_1,b_1)=(x,-x)$, $\ell_1=0$. Thus \eqref{vert:JUMP} sends a high state into the new capacity-zero high submodule, which remains invariant through the right tail. The empty horizontal boundary kills it. The argument is valid separately for every intermediate vertical state, so additional rows may be summed afterward. \end{proof} \subsection{Moving an external row past a fused pair} The positive train and negative train are the two-crossing maps \begin{equation}\label{vert:trains} \begin{split} \mathcal B^+_{a,b;c}:\quad P_aN_bP_c&\xrightarrow{\,R'_{b,c}\,}P_aP_cN_b \xrightarrow{\,R_{a,c}\,}P_cP_aN_b,\\ \mathcal B^-_{c;a,b}:\quad N_cP_aN_b&\xrightarrow{\,R'_{c,a}\,}P_aN_cN_b \xrightarrow{\,R''_{c,b}\,}P_aN_bN_c. \end{split} \end{equation} Only the outside spectral parameter occurs in their possible mixed-crossing denominators. \begin{lemma}[Image preservation at resonance] \label{vert:fused-image} The trains restrict to maps \[ \mathcal K(a,b)P_c\longrightarrow P_c\mathcal K(a,b),\qquad N_c\mathcal K(a,b)\longrightarrow\mathcal K(a,b)N_c. \] They intertwine columns, also at $b=-a\tau$, for a generic independent outside parameter $c$. \end{lemma} \begin{proof} For generic $a,b$, the image of $R'_{b,a}$ is exactly $\mathcal K(a,b)$: on total set $B$ its output is $v_B(a,b)/Z_{|B|}(a,b)$. The two auxiliary braid identities \eqref{vert:RRR-types} show that both trains preserve this image. Equation~\eqref{vert:RLL} gives column intertwining. To specialize, use the unnormalized generators $v_B(a,b)$. Different $B$ occupy different total-color sectors, and the all-negative coefficient $b^{|B|}$ is nonzero at $b=-a\tau$. They therefore give a regular constant-rank subbundle through this specialization. The external trains are regular there over the rational function field in $c$. In each output sector extract the putative scalar by dividing its all-negative coordinate by $b^{|D|}$. Proportionality of every other coordinate to this scalar is a rational identity regular at the specialization, so it continues to hold. The same reasoning applies to intertwining. The specialization uses only unnormalized $v_B$ and regular external trains; the normalized own-pair crossing may have a pole at this parameter. \end{proof} \begin{lemma}[High-submodule descent for both trains] \label{vert:train-high} At $b=-a\tau$, both maps in Lemma~\ref{vert:fused-image} preserve the high submodule $\mathcal H=\operatorname{span}\{e_B:|B|\geq2\}$ of the inner pair. At $b=-a$, they preserve the nonempty-state submodule and induce the identity on the inner vacuum quotient. \end{lemma} \begin{proof} By Lemma~\ref{vert:fused-image}, it suffices to prove that every high input has zero coefficient in the vacuum and singleton output sectors. Rescaling the inner basis permits $a=1$. First take $b=-\tau$ and positive outside parameter $c=z$. Write the input split as $I\subset B$, $J=B-I$, and the outside state as $H$. To read the output inner set $D$, of degree $d\leq1$, choose its positive component empty. Its negative component is then $D$, whose coefficient in $v_D$ is nonzero. The intermediate outside positive state is $L=H+J-D$. The two crossing formulas give, before dividing by that nonzero output coefficient, \begin{equation}\label{vert:positive-train-sum} (-1)^{|J|}\tau^{d-|H|} \tau^{\phi(I,B)+\phi(L,D)} \frac{(z;\tau)_{|H|+|J|-d}} {(z/\tau;\tau)_{|H|+|J|}}. \end{equation} The nonzero-state restrictions are $J\cap H=\varnothing$, $I\cap H\subset D$, and binary $L$. They will also be used in the cases below. For $d=0$, conservation forces $B\cap H=\varnothing$, and all $I\subset B$ are allowed. The Pochhammer ratio is \[ \frac{1-z\tau^{|H|+|J|-1}}{1-z/\tau}. \] Consequently the split sum is a linear combination of \[ \prod_{j\in B}(\tau^{r_j}-1),\qquad \prod_{j\in B}(\tau^{r_j}-\tau),\qquad r_j=\#\{b\in B:b>j\}. \] The largest color kills the first product; since $|B|\geq2$, the second-largest kills the second. For $D=\{i\}$ the Pochhammer ratio in \eqref{vert:positive-train-sum} is $1/(1-z/\tau)$, independent of the split. Up to a constant, the exponent assigned to a free color $j\in I$ is $r_j-\ind_{ji}$, with factor $\tau^{r_j+\ind_{j>i}}-\tau$. If $i$ is outside $B$, the largest color above $i$, or the second-largest if all are below $i$, has exponent one. If $i\in B$, its split is forced and the remaining exponents are the numbers of larger colors in $B-\{i\}$ plus one. Their largest color again has exponent one. The free set is nonempty because $|B|\geq2$. This proves negative descent. For capacity zero put $b=-a$ instead. Only a vacuum output is visible. In either train the two Pochhammer factors cancel, leaving, up to a nonzero common factor, \[ \sum_{I\subset B}(-1)^{|B-I|}\tau^{\phi(I,B)} =\prod_{j\in B}(\tau^{r_j}-1). \] This vanishes for $B\ne\varnothing$. For $B=\varnothing$ the surviving coefficient is $1$ and the outside state is unchanged. This proves the final assertion. \end{proof} We record an explicit quotient matrix, which will also make the genericity of the positive train transparent. Let the inner pair have $a=1,b=-\tau$ and let the outside positive parameter be $z$. On a binary total set $S=\{i_1<\cdotsi). \end{aligned} \end{equation} For $d=0$ the matrix is $1$. To verify the formula, choose output inner positive state empty in the two terms of the input singleton in \eqref{vert:positive-train-sum}, and divide by its negative coefficient $-\tau$. The other output split has the same scalar by Lemma~\ref{vert:fused-image}. Factoring $\tau^{i-d}$ from row $i>0$ leaves lower-right block $Q=I+(1-\tau^{-1})N$, where $N$ has entries $1$ strictly below the diagonal. Since $Q^{-1}(1,\ldots,1)^{\mathsf t} =(1,\tau^{-1},\ldots,\tau^{-(d-1)})^{\mathsf t}$, its Schur complement is $\tau^{-d}(1-z/\tau)$. Hence \begin{equation}\label{vert:fundamental-determinant} \det B_S(z)=\tau^{-d(d+1)/2}(1-z/\tau)^{-d}. \end{equation} If one color $i$ is repeated in the total charge, the block is one dimensional and has coefficient $\tau^{\#\{h\in H:h1$, expand the last factor as $P_kN_k$, move $P_k$ successively to the left of $\mathcal K_{k-1},\ldots,\mathcal K_1$ using $\mathcal B^+$, and apply $E_{k-1}$ to the remaining $k-1$ fused factors. In formulas, with neighboring positions followed through the successive swaps, \begin{equation}\label{vert:E-recursion} E_k=(\id_{P_k}\otimes E_{k-1}\otimes\id_{N_k}) (\mathcal B^+_{1,k}\cdots\mathcal B^+_{k-1,k} \otimes\id_{N_k}). \end{equation} The rightmost indicated train acts first. The final expansion of each remaining $\mathcal K_i$ is part of $E_{k-1}$; thus \eqref{vert:E-recursion} is an equality of typed linear maps, with the tensor positions changing at each successive train. \begin{lemma}\label{vert:E-intertwining} The map $E_k$ preserves total color charge, takes tensor vacuum to raw vacuum, and intertwines every left-column matrix of the adjacent-pair and nested models. Every left-vacuum reachable vector of the nested model lies in $\im E_k$. \end{lemma} \begin{proof} Conservation holds at every local weight and crossing. Their vacuum entries are $1$. Lemma~\ref{vert:fused-image} and \eqref{vert:RLL} allow each crossing in \eqref{vert:E-recursion} to pass through a left column, giving intertwining by induction. Apply it successively to a word of left columns and use the vacuum assertion. This expresses the raw vector as $E_k$ of the corresponding adjacent-pair vector; no injectivity is required. \end{proof} \begin{proposition}[Kernel of the right pairing] \label{vert:E-kernel} Every cut-and-right-tail observer annihilates \begin{equation}\label{vert:high-tensor-ideal} E_k\left(\sum_{i=1}^k \mathcal K_1\otimes\cdots\otimes\mathcal H_i \otimes\cdots\otimes\mathcal K_k\right). \end{equation} The assertion holds for each vertical matrix element, with arbitrary surrounding positive and negative rows. \end{proposition} \begin{proof} Induct on $k$. For $k=1$ use Corollary~\ref{vert:inner-invisible}. For an old factor $i0$. Let $\Pi$ project onto the subspace in which every positive horizontal $g$-bit vanishes. Since these bits are nonnegative, this is also the projection onto total positive $g$-occupation zero. \begin{lemma}[One projected pair]\label{vert:projected-pair} For the adjacent pair the outgoing fused lines are $\widehat v_D=\Pi v_D$. At left capacity one their degree-at-least-two part is invariant and invisible to the right observer. Their coefficient sums are \begin{equation}\label{vert:projected-norm} \epsilon(\widehat v_D)= \begin{cases} Z_{|D|}(a,b),&g\notin D,\\ b\,Z_{|D|-1}(a\tau,b),&g\in D. \end{cases} \end{equation} \end{lemma} \begin{proof} Apply \eqref{vert:RAW-ALL} and then $\Pi$. If $g\in D$, it is forced into the negative component. As the largest color, it gives \[ \widehat v_D(a,b) =b\,v_{D-\{g\}}(a\tau,b) \quad\text{with a negative }g\text{-bit appended}. \] This proves \eqref{vert:projected-norm}. At $b=-a\tau$ the remaining colors see capacity zero. On a left column the vertical input $g$-bit is one. For a projected incoming vector containing $g$, the positive rule forces its intermediate bit to one, and negative conservation forces the outgoing horizontal $g$-bit to remain one. The other colors have effective parameters $(a\tau,b)$; Proposition~\ref{vert:one-pair-quotient} keeps every nonempty set of them in its capacity-zero high submodule. If the input omits $g$, ordinary high-to-low vanishing applies and survives output projection. At the first right column the positive vertical $g$-bit is zero. An incoming negative $g$-arrow can only be absorbed vertically, and $C_g-J_g=0$. When the old high vector contains $g$, its remaining colors form a nonempty capacity-zero vector $v_{D-\{g\}}(a\tau,b)$. Its effective old parameters $(a\tau,b)=(x,-x)$ equal the new right parameters, so its vacuum coefficient vanishes by \eqref{vert:JUMP}. Positive-degree right output is an ordinary capacity-zero high vector and is killed by the empty boundary. When the old vector omits $g$, the ordinary jump proof applies. This proves the assertion pointwise in the other vertical colors. \end{proof} \begin{proposition}[Postponing the single projection] \label{vert:postponed-projection} The left-capacity theorem holds with the zero-height restriction. Its raw comparison map is \begin{equation}\label{vert:projected-E} \widehat E_k=\Pi E_k \end{equation} on the full tensor product of unprojected $\mathcal K_i$. For the right pairing it factors through $\bigotimes_i\mathcal F_i$. \end{proposition} \begin{proof} Let $M_L(A,C)$ be the unprojected nested left-column operator and $m$ the vertical $g$-bit at the midpoint. Conservation through the positive block gives \[ \text{positive output count} =\text{positive input count}+1-m \ \geq\ \text{positive input count}. \] Thus \begin{equation}\label{vert:Pi-column} \Pi M_L=\Pi M_L\Pi. \end{equation} On the $\Pi$-space, a surviving output forces $m=1$, which is exactly the midpoint vacuum constraint. The projected column is $\widehat M_L=\Pi M_L\Pi$. Intertwining gives \[ \widehat M_L\widehat E_k =\Pi M_L\Pi E_k=\Pi M_LE_k =\Pi E_k M_{\rm adjacent}. \] The projected map takes vacuum to vacuum, hence contains every projected reachable state in its image. The kernel induction of Proposition~\ref{vert:E-kernel} survives the final projection. For an old high factor apply the full positive trains first; they preserve that factor. The final projector factors into its parts on the new outer positive row and on the old positive block. The inductive projected inner vanishing applies for every vertical boundary. For the new high factor use the simultaneous same-sign permutation $\Xi$ before $\Pi$. It preserves total positive $g$-occupation, so \begin{equation}\label{vert:Pi-Xi} \Xi\Pi=\Pi\Xi. \end{equation} Cancel its positive-positive crossings as in the unprojected proof. The remaining full negative trains preserve the ordinary high submodule. Only at the end apply $\Pi$; the innermost vectors become $\widehat v_B$ with $|B|\geq2$, invisible by Lemma~\ref{vert:projected-pair}. Equation~\eqref{vert:Pi-Xi} transfers this vanishing to the original pairing. Thus the same tensor quotient proves capacity $k$. On the right the bottom $g$-bit is zero. An incoming $\Pi$-state has positive output count $-m$, forcing $m=0$ and that count zero. The ordinary right column on this subspace is the projected right column, as required for all observers in the proof. \end{proof} The proof applies with a translated vacuum: if its occupied sites are $u\leq s$ and the parameter cut is translated to $s$, the same statements hold on that left region. It also applies to any initial state in $\im(\Pi E_k)$; reachability from an earlier vertical boundary was used only to establish membership in this image. \section{The right ramp and all dominance inequalities}\label{vert:ramp-section} Let $h(u)$ denote the total horizontal occupation of the chosen $k$ pairs immediately after site $u$. We prove the observable bound \begin{equation}\label{vert:ramp} h(u)\leq r_k(u):=\max\{k-\max(u,0),0\}. \end{equation} Here a bound means that the corresponding complete coefficient, with fixed outside data, is zero when the inequality fails. For $u\leq0$ this is Theorem~\ref{vert:left-capacity} and Proposition~\ref{vert:postponed-projection}. \subsection{Successive disappearance on the right} \begin{lemma}\label{vert:peeling} On the right tail, the innermost pair is observably vacuum after the first column and is the vertical identity thereafter. The next pair has this property after the second column, and pair $i$ after column $i$. In particular, all nonempty horizontal states beyond column $k$ are annihilated by the right boundary. The assertion includes the zero-height restriction. \end{lemma} \begin{proof} At the first right site, the adjacent inner pair has parameters $(x_1,-x_1)$. Equation~\eqref{vert:RAW-ALL} puts every incoming raw state on its fused output lines. Their positive-degree span is the invariant capacity-zero submodule and is killed by the empty right boundary. Thus its observable output is vacuum. With both input and output vacuum the pair has coefficient $\ind_{A=C}$, so it is the vertical identity at each subsequent site. At site two the next pair is consequently adjacent. Repeat the same argument. For the zero-height color, a right positive row has input vertical bit zero and horizontal bit zero; both its outputs are therefore zero. In the negative row an incoming horizontal bit can only be deposited vertically, giving $C_g=J_g$ and horizontal output zero. The last-color contribution to $\phi(D,C-J)$ vanishes. The same argument applies on all other colors and at each successive pair. Since every step is an identity for each vertical matrix element, surrounding rows are retained throughout. \end{proof} \subsection{A last auxiliary color when no zero plateau is present} \begin{proposition}\label{vert:ghost-ramp} If there is no zero-height plateau, the bound \eqref{vert:ramp} holds at every positive site. \end{proposition} \begin{proof} Fix $1\leq s\leq k$. Adjoin one last path color $g$. Before the first right site its horizontal state and all its vertical bits are zero. Thus extending the earlier left network has no effect, and its state lies in $\im E_k$ with no $g$-occupation. Encode the first $s$ original right columns as extended left columns. At each encoded column give $g$ bottom vertical bit one and top bit zero, and require its positive horizontal occupations to vanish. The projection can be postponed until the end by \eqref{vert:Pi-column}. The projected left-capacity argument applies to this initial state in $\im(\Pi E_k)$. We give a right observer that selects the required auxiliary path history. Number the negative rows $1,\ldots,k$ from bottom to top. At a negative vertex its single-color equations are \[ A_g+J_g=C_g+L_g,\qquad J_g\leq C_g. \] For a column with bottom bit one and top bit zero, let $J=\{j_1<\cdots0$, translate the positive vacuum of this one last color to charge $s$, with occupied sites $u\leq s$. At the top of the $k$ negative rows add the occupied sites $k+1,\ldots,k+s$. They are disjoint from the old state. After subtracting the charge $s$, the transposed target partition is $(k^s,\nu'_0,\nu'_1,\ldots)$; its untransposed partition is \begin{equation}\label{vert:rectangle-shape} \widetilde\nu=\nu+(s^k). \end{equation} The first $s$ full-height columns of a semistandard tableau of this shape in $k$ letters are forced to contain $1,2,\ldots,k$ from top to bottom. Deleting them gives a tableau of $\nu$. Conversely every entry in row $i$ of a tableau of $\nu$ is at least $i$, so prefixing these forced columns respects the row inequalities. This is a bijection, including every partial shape and every original label. We now track its physical edges. Transposition gives partial shapes $(i^s,\nu(i)')$. The charge translation cancels the row-index shift of every old transposed box. Thus every old box remains at the same physical Maya site. After negative row $i$, the extra vertical sites are \begin{equation}\label{vert:rectangle-sites} \{i+1,\ldots,i+s\}; \end{equation} the extra outgoing horizontal sites in that row are $\{i,\ldots,i+s-1\}$. These sets are disjoint from the old vertical and horizontal sites, respectively. Their incoming horizontal sites are shifted one step to the right, so at every negative vertex the added occupations satisfy $C_g=J_g$. Merging them with the old zero-height paths therefore leaves every edge binary. A meeting of the two paths can give the allowed full state $A_g=J_g=C_g=L_g=1$. Use auxiliary positive parameters $x_i/\tau$ through site $s$, and $x_i$ afterward; keep negative parameters $b_i=-x_i$. The last color is projected to its charge-$s$ positive vacuum. This is the one-projected-color left-capacity model with its cut translated to $s$. At sites $1,\ldots,s$ the extra positive vertical bit is one and its horizontal bits vanish. Its contribution to $\phi(K,A+I)$ on old colors is $|K_{\rm old}|$, converting $x_i/\tau$ into the original right parameter $x_i$. At every negative vertex, the equality $C_g=J_g$ leaves the second argument of $\phi(L,C-J)$ unchanged. The new horizontal bit is last in color order, so it contributes no further term to $\phi$. Each negative row has exactly $s$ added boxes. The auxiliary coefficient is therefore exactly \begin{equation}\label{vert:rectangle-weight} \left(\prod_{i=1}^k b_i^s\right) \times\text{the original coefficient}. \end{equation} This is a bijection of path configurations and a pointwise equality of their weights. At the section after site $s$, the added horizontal arrows occupy the rows $i$ for which $i\leq s\leq i+s-1$, namely $1,\ldots,\min(k,s)$. The translated left capacity gives \[ h(s)+\min(k,s)\leq k. \] This is \eqref{vert:ramp} for every $s>0$. The multiplier in \eqref{vert:rectangle-weight} is nonzero over the generic field, so division is legitimate. If further variables surround the selected block, hold its boundary $\nu$ fixed, apply this inner coefficient identity, and then multiply and sum the original outside coefficients. The changed auxiliary boundary is used only inside that identity. \end{proof} \subsection{Every prefix and dominance support} \begin{theorem}[All-prefix capacity]\label{vert:all-prefix} For every partition $\lambda$, every nonnegative $p\in\RPP(\lambda)$, and every choice of $k$ alphabet variables, with all other variables retained, \begin{equation}\label{vert:all-prefix-equation} \deg_{\{x_{i_1},\ldots,x_{i_k}\}} K_{\lambda,p}[X;T] \leq\lambda_0+\cdots+\lambda_{k-1}. \end{equation} Equivalently, $K_{\lambda,p}$ is supported on $m_\mu$ with $\mu\unlhd\lambda$. \end{theorem} \begin{proof} Proposition~\ref{vert:fixed-p-symmetry} places the selected variables in the innermost $k$ pairs. Fix the intermediate shapes on the boundary of this block for each plateau. They determine which skew subsets of the original diagram carry selected letters. Conservation fixes their total horizontal flux $h(u)$ on each Maya diagonal before any internal signs are summed. Thus the preceding coefficientwise vanishings apply to each such contribution. Let $d_\lambda(u)$ be the number of boxes of $\lambda$ with $i-j=u$. Necessarily $h(u)\leq d_\lambda(u)$. The boxes on this diagonal start in row $\max(u,0)$; hence its intersection with the first $k$ rows has size \[ \min\{d_\lambda(u),\max(k-\max(u,0),0)\}. \] Every box carrying a selected letter contributes one to precisely one such horizontal flux. The left capacity and Propositions~\ref{vert:ghost-ramp}--\ref{vert:rectangle-ramp} therefore give \begin{align*} \deg_{\rm selected} &=\sum_u h(u)\\ &\leq\sum_u\min\{d_\lambda(u),r_k(u)\} =\lambda_0+\cdots+\lambda_{k-1}. \end{align*} This proves \eqref{vert:all-prefix-equation} with every outside variable retained. For a monomial sort its exponents in decreasing order and apply these inequalities to its $k$ largest exponents. Fixed-$p$ symmetry then gives all dominance inequalities $\sum_{i=1}^k\mu_i\leq\sum_{i=1}^k\lambda_i$, where the latter display uses the conventional one-based notation for parts. This is exactly $\mu\unlhd\lambda$. \end{proof} \begin{corollary}[Complete one-sided triangularity] \label{vert:triangularity} Over $\Q(q,t)$, \begin{equation}\label{vert:triangle} F_\lambda[(t-1)X;q,t]\in \operatorname{span}\{m_\mu[X]:\mu\unlhd\lambda\}. \end{equation} The same statement holds with Schur functions in place of monomial symmetric functions. \end{corollary} \begin{proof} Apply Theorem~\ref{vert:all-prefix} to each finite polynomial in \eqref{vert:paired-identity-equation}. Its scalar $M_\lambda(p;q,t)$ and the factor $D_\lambda$ change no alphabet exponents. The height sum is valid coefficientwise in the completion and is rational by Lemma~\ref{vert:rationality}; hence the vanishing holds over $\Q(q,t)$. The Kostka matrix is unitriangular for dominance, so a downward dominance ideal has the same span in the monomial and Schur bases. \end{proof} \section{Identification with the modified Macdonald polynomial} \label{vert:identification-section} We combine the full two-alphabet Cauchy identity of Theorem~\ref{root:cauchy} with the complete triangle \eqref{vert:triangle}. \begin{theorem}\label{vert:macdonald} For every partition $\lambda$, \begin{equation}\label{vert:macdonald-equation} D_\lambda(q,t)G_\lambda[X;q,t] =\widetilde H_\lambda[X;q,t]. \end{equation} Consequently the root vertex is \begin{equation}\label{vert:root-vertex-final} V_{\lambda,s}(x,\TT;X_s) =x^{d_0(\lambda,s)} \frac{\widetilde H_\lambda[X_s;x^{-1},x\TT]} {\displaystyle\prod_{b\in\lambda} (1-x^{a_b+l_b+1}\TT^{l_b})}. \end{equation} \end{theorem} \begin{proof} All functions are over $K=\Q(q,t)$. The geometric two-point calculation in Theorem~\ref{root:cauchy}, with the functions of Definition~\ref{vert:definition}, gives \begin{equation}\label{vert:cauchy-input} \sum_\lambda R^{|\lambda|}c_\lambda F_\lambda[X]F_\lambda[Y] =\Exp\!\left[ \frac{Qp_1[X]p_1[Y]}{(q-1)(1-t)}\right], \end{equation} where \begin{equation}\label{vert:cauchy-norm} c_\lambda^{-1} =\prod_{b\in\lambda} (q^{a_b+1}-t^{l_b})(q^{a_b}-t^{l_b+1}). \end{equation} The plethystic exponential on the right is the reproducing kernel for the nondegenerate pairing \begin{equation}\label{vert:kernel-pairing} \langle p_\rho,p_\sigma\rangle_{\mathcal K} =\delta_{\rho\sigma}z_\rho \prod_{r\in\rho}(q^r-1)(1-t^r). \end{equation} Indeed its logarithm is $\sum_{r\geq1}R^rp_r[X]p_r[Y]/ \bigl(r(q^r-1)(1-t^r)\bigr)$. Fix degree $n$. Let $A$ be the square matrix whose columns are the power-sum coordinates of the $F_\lambda$, and let $B$ be the diagonal matrix of \eqref{vert:kernel-pairing}. If $C=\operatorname{diag}(c_\lambda)$, comparison of coefficients in \eqref{vert:cauchy-input} gives \begin{equation}\label{vert:cauchy-matrices} ACA^{\mathsf t}=B^{-1}. \end{equation} Both $B$ and $C$ are invertible. Taking determinants proves $A$ invertible, so the $F_\lambda$ form a basis. Inversion of \eqref{vert:cauchy-matrices} then gives $A^{\mathsf t}BA=C^{-1}$. Thus they are orthogonal, with nonzero norms $c_\lambda^{-1}$. Let $U$ be the invertible plethystic map $Uf[X]=f[(t-1)X]$. Transport the pairing by $U^{-1}$. Since $Up_r=(t^r-1)p_r$, the new pairing in degree $n$ is \begin{align}\label{vert:transported-pairing} \langle p_\rho,p_\sigma\rangle_U &=\delta_{\rho\sigma}z_\rho \prod_{r\in\rho}\frac{-(1-q^r)}{1-t^r}\nonumber\\ &=t^{-n}\delta_{\rho\sigma}z_\rho \prod_{r\in\rho}\frac{1-q^r}{1-t^{-r}}. \end{align} It is a nonzero degreewise scalar multiple of the ordinary Macdonald pairing with parameters $(q,t^{-1})$. The vectors $UF_\lambda$ are orthogonal and, by \eqref{vert:triangle}, triangular in dominance. To verify their leading terms, order the partitions of $n$ by any total order extending dominance, from smaller to larger. Their monomial coefficient matrix is triangular and invertible. Every diagonal coefficient $[m_\lambda]UF_\lambda$ is consequently nonzero. Divide each vector by that coefficient to obtain a monic family $f_\lambda$. Compare it with the monic Macdonald family $P_\lambda[X;q,t^{-1}]$ in \eqref{main:macdonald-convention}. Induct in the chosen total order. The difference $f_\lambda-P_\lambda$ lies in the span of earlier monomials and therefore of earlier $f_\mu$, whose triangular block is invertible. By induction those earlier vectors are the corresponding $P_\mu$. Both terms in the difference are orthogonal to them. Their span has nonzero orthogonal norms and is nondegenerate, so the difference is zero. This argument includes incomparable partitions, since the total order only enlarges the collection of earlier terms. Under the modified convention of the statement, homogeneity gives \begin{equation}\label{vert:modified-conversion} U\widetilde H_\lambda[X;q,t] =t^{n+n(\lambda)}J_\lambda[X;q,t^{-1}]. \end{equation} The integral form $J_\lambda$ is a nonzero scalar multiple of $P_\lambda$. Therefore $UF_\lambda$ and $U\widetilde H_\lambda$ span the same line. Inverting $U$ gives $F_\lambda=a_\lambda\widetilde H_\lambda$. The one-letter calculation \eqref{vert:one-letter} gives $F_\lambda[1]=1$, and the modified normalization gives $\widetilde H_\lambda[1]=1$. Hence $a_\lambda=1$. This proves \eqref{vert:macdonald-equation}; substituting \eqref{vert:physical-G} proves \eqref{vert:root-vertex-final}. \end{proof} \providecommand{\Loc}{\operatorname{Loc}} \providecommand{\ParEnd}{\operatorname{ParEnd}} \providecommand{\SParEnd}{\operatorname{SParEnd}} \providecommand{\ch}{\operatorname{ch}} \providecommand{\Td}{\operatorname{Td}} \providecommand{\Alt}{\operatorname{Alt}} \providecommand{\Flag}{\operatorname{Flag}} \providecommand{\End}{\operatorname{End}} \section{The Chern and perverse filtrations for arbitrary flags} \label{betti:pc-section} We establish the comparison needed for the Betti realization. Throughout this section, $C$ is a smooth projective complex curve of genus $g\geq0$, $D$ is a nonempty reduced divisor, and $\mu^p=(\mu^p_1,\ldots,\mu^p_{\ell_p})$ is a composition of $n$ at each $p\in D$. Ordering the parts amounts to ordering the parabolic weights; the final statements depend only on the associated partitions. Let $M_{\mathrm{Dol},\mu}$ be the moduli space of stable strongly parabolic Higgs bundles of rank $n$, fixed degree $d_0$, these flag types, and numerically generic rational weights. Thus semistability coincides with stability. We work with rational cohomology. Put \begin{equation}\label{betti:Nmu} N_\mu=(g-1)n^2+1+\frac12\sum_{p\in D} \left(n^2-\sum_a(\mu^p_a)^2\right). \end{equation} When the moduli space is nonempty, it is smooth of dimension $2N_\mu$. Its Hitchin morphism is projective. All Hitchin bases below can be embedded in the same space of characteristic coefficients. A closed embedding of bases does not change the centered perverse filtration, because its direct image is perverse exact. \subsection{Normalizations and the statement} If $h:Y\to A$ is projective and $Y$ is smooth of dimension $d_Y$, our centered convention is \begin{equation}\label{betti:centered-P} P_jH^i(Y)=\im\left[ H^{i-d_Y}\left(A,{}^p\tau_{\leq j}Rh_*\Q_Y[d_Y]\right) \longrightarrow H^i(Y)\right]. \end{equation} For a Hitchin system of relative dimension $N$, the unit has centered perverse degree $-N$. This convention is useful because pullback and Gysin maps over a common base have dimension shifts that cancel in the flag correspondences below. There is a universal bundle $\mathcal E$ on the moduli stack times $C$, and there are universal graded block bundles $\mathcal V_{p,a}$. Choose $x\in C\setminus D$ and set \begin{equation}\label{betti:normalization} u=\frac1n c_1(\mathcal E_x),\qquad \ch^{\mathrm{nor}}(\mathcal E)=e^{-u}\ch(\mathcal E),\qquad \ch^{\mathrm{nor}}(\mathcal V_{p,a})=e^{-u}\ch(\mathcal V_{p,a}). \end{equation} These expressions are unchanged when the universal bundle is tensored by a line bundle from the moduli stack. They therefore define rational classes on the coarse moduli space. Equivalently, use the scalar gerbe and take the coefficient independent of its degree-two generator. The normalized Chern classes of a block are obtained from its normalized Chern character by the Newton identities. Give \[ \int_C\ch^{\mathrm{nor}}_k(\mathcal E)\,\gamma, \quad \gamma\in H^*(C), \qquad c_k^{\mathrm{nor}}(\mathcal V_{p,a}) \] Chern index $k$. Define $C_bH^*(M_{\mathrm{Dol},\mu})$ to be the span of products of these classes whose indices sum to at most $b$, including the empty product in index zero. Negative steps are zero. Numerical constants, even if they occur as degree-zero components of a Chern character, are assigned index zero as scalar multiples of the unit. \begin{theorem}\label{betti:pc} For every $g\geq0$ and every collection of partial flags, the above normalized classes generate the cohomology, and \begin{equation}\label{betti:pc-formula} P_jH^*(M_{\mathrm{Dol},\mu}) =C_{j+N_\mu}H^*(M_{\mathrm{Dol},\mu}). \end{equation} The assertion holds in every degree $d_0$ for which the stability parameter is numerically generic, including noncoprime degrees. \end{theorem} We use the surface Hecke algebra and its polynomial reduction in the precise form of \cite[\S\S4--7]{HMMS}. Those constructions allow every genus. Their geometric hypotheses, tautological generation, passage to zero residues, and passage to partial flags are proved below; these are needed separately from the operator construction. \subsection{Auxiliary markings and stability} When $M_{\mathrm{Dol},\mu}$ is empty, both sides of \eqref{betti:pc-formula} vanish. We henceforth take it to be nonempty. A marking with the single block $(n)$ and weight zero is harmless: strong parabolicity there says that the residue is zero, which says exactly that the Higgs field is regular at that point. Add such markings until $|D|\geq3$ if $g=0$, and $|D|\geq1$ otherwise. Then \begin{equation}\label{betti:positive-L} L=K_C(D),\qquad \deg L>0. \end{equation} The moduli problem and its Hitchin map have not changed. Refine each block to a complete flag. Replace its repeated weight by nearby strictly ordered rational weights, chosen sufficiently close to the original ones. This can be done uniformly for all Higgs bundles of the prescribed type. Indeed, for a subrank $00$ and set \[ A=C_\infty+\kappa F+\epsilon\sum_{p\in D}F_p' =C_\infty+(\kappa+|D|\epsilon)F -\epsilon\sum_{p,i}E_{p,i}. \] The ruled-surface divisor $C_\infty+\kappa F$ is ample for $\kappa\gg0$. After pullback to its blowup, subtracting a sufficiently small positive multiple of the exceptional divisors is ample; this proves the assertion for $A$. Both $C_\infty$ and the $F_p'$ are removed from $S_\xi$, so $A|_{S_\xi}=\kappa F|_{S_\xi}$. Keep $A$ fixed. If an interior spectral subsheaf has rank $r$ over $C$ and profile $\nu$, then \[ \Sigma'=r\pi^*C_0-\sum_{p,i}\nu_{p,i}E_{p,i},\qquad A\cdot\Sigma'=\kappa r. \] For a rational twisting class $\beta$, twisted Gieseker stability for one-dimensional sheaves compares the slopes \[ \frac{\chi(F')+\beta\cdot\Sigma'}{A\cdot\Sigma'} =\frac1\kappa\left( \frac{\deg E'}r+1-g+\beta\cdot\pi^*C_0 +\frac1r\sum_{p,i}\alpha'_{p,i}\nu_{p,i}\right), \qquad \alpha'_{p,i}=-\beta\cdot E_{p,i}. \] The common additive term and the positive factor cancel in every comparison. Initially take $\beta_0=\sum\alpha_{p,i}E_{p,i}$. Small changes of $\beta$ preserve the entire interior stability chamber by the preceding discrete-slope argument. There are only finitely many stability walls in a bounded neighborhood of $\beta_0$. To see this, a destabilizing effective subcycle satisfies $00$; for $g\geq2$ it is greater than $2g-2$. For $i=1$, the residue theorem gives precisely the displayed total-trace relation. Choose linear splittings of the evaluation maps for the characteristic coefficients. Subtract from the $i$th coefficient the chosen lift of the degree-$i$ elementary symmetric polynomial in the residues. These homogeneous polynomial changes of coordinates give the stated equivariant product. The weak and fixed-residue deformation complexes, in degrees zero and one, are \begin{align*} K_{\rm weak}&=[\ParEnd(E)\xrightarrow{[\theta,\,\cdot\,]} \ParEnd(E)\otimes K_C(D)],\\ K_{\rm fix}&=[\ParEnd(E)\xrightarrow{[\theta,\,\cdot\,]} \SParEnd(E)\otimes K_C(D)]. \end{align*} Parabolic Serre duality identifies $\mathbb H^2(K_{\rm weak})^\vee$ with the commuting strongly parabolic endomorphisms. Stability makes a commuting parabolic endomorphism scalar; a strictly lowering scalar at a nonempty marked set is zero. Thus $\mathbb H^2(K_{\rm weak})=0$ and $X$ is smooth. The same duality gives $\mathbb H^2(K_{\rm fix})=\C$, dual to the scalar identity. The residue map gives an exact sequence of complexes \[ 0\longrightarrow K_{\rm fix}\longrightarrow K_{\rm weak} \longrightarrow\bigoplus_{p,i}\C[-1]\longrightarrow0. \] Its connecting map to $\mathbb H^2(K_{\rm fix})$ is the sum of the residue traces, since its dual includes the scalar identity into all diagonal endomorphisms. Hence the end of the long exact sequence is \[ \mathbb H^1(K_{\rm weak})\longrightarrow\C^{n|D|} \xrightarrow{\ \sum\ }\C\longrightarrow0. \] The image of $dr$ is therefore exactly the tangent space of $A_D$, so $r$ is smooth. If $X_0$ is nonempty, the image of this smooth map is an open subset of $A_D$ containing zero and stable under positive scaling. Its complement is closed and scaling invariant. Every nonempty such complement would contain zero, by taking the limit under scaling. Thus $r$ is surjective. The projective Hitchin morphism and its positive-weight base make $X$ semiprojective. We use smooth-fiber semiprojective contraction \cite[Corollary 1.3.3]{HRVPurity}, whose target is an affine line. Here all ordered residues have weight one. For a nonzero $\xi$, take the line $\ell_\xi$ through $0$ and $\xi$ in $A_D$. Its inverse image $X_{\ell_\xi}$ is smooth, since $r$ is a submersion, and is a closed invariant semiprojective subvariety of $X$. It has the same core as $X$ and $X_0$: the core is the zero Hitchin fiber, by properness and positive weights on the Hitchin base. The inclusion of the core in a smooth semiprojective variety is a homotopy equivalence. Therefore $H^*(X)\to H^*(X_{\ell_\xi})$ is an isomorphism. The affine-line fiber theorem now gives isomorphisms from this latter cohomology to both $H^*(X_\xi)$ and $H^*(X_0)$. If $A_D$ is a point, the assertion is immediate. All restriction maps are algebraic, so the resulting isomorphisms respect mixed Hodge structures. Thus \begin{equation}\label{betti:residue-cohomology} H^*(X)\xrightarrow{\sim}H^*(X_\xi),\qquad H^*(X)\xrightarrow{\sim}H^*(X_0). \end{equation} These groups are pure, and the restrictions carry the universal curve and flag classes to the corresponding classes. Lemma \ref{betti:compactification} proves that all three are generated by the normalized tautological classes. Let $U\subset X$ be the open locus of integral spectral curves with distinct marked residues. Choose $\xi$ outside all hyperplanes \begin{equation}\label{betti:residue-generic} \sum_{p,i}\nu_{p,i}\xi_{p,i}=0, \qquad \sum_i\nu_{p,i}=r,\quad 00), & [y_{p,i},T_0(1)]&=X_{p,i}.\nonumber \end{align} For the last equality, equality of the moving Hecke point and the marked branch is one transverse Cartier equation on the smooth universal spectral curve near that branch. Comparing the two eigenlines gives this divisor with multiplicity one; the virtual pullback is its ordinary divisor class. This establishes the coefficient and excludes an extra multiplicity. Let $H_{n,D}$ be the extension of $H^*(C,\Q)$ by degree-two symbols $\tau_{p,i}$, with relations \[ \sum_{i=1}^n\tau_{p,i}=\omega,\qquad \tau_{p,i}H^{>0}(C)=0,\qquad \tau_{p,i}\tau_{q,j}=0. \] Before reduction define \[ \psi_k(\tau_{p,i})=y_{p,i}^{k},\qquad T_k(\tau_{p,i})=y_{p,i}^{k}X_{p,i},\qquad q_1(\gamma)=T_0(\gamma). \] These assignments respect the relations: the sum of the eigenline powers is the point component of the universal Chern character, and the analogous identity for $T_k$ follows by decomposing the Hecke correspondence over $p$ into its ordered branches. This is the parabolic algebra of \cite[Definition 7.8 and Theorem 7.9]{HMMS}. For the degree comparison, let $U_d$ denote the ordered elliptic space in degree $d$, over the same ordered Hitchin base, and put $R_d=H^*_{\mathrm{taut}}(U_d)$ with the normalized Chern filtration. Thus the space previously denoted by $U$ is $U_{d_0}$. On the moduli stack over $U_d$, a scalar gerbe, write $u_d=c_1(\mathcal E_{d,x})/n$ and $\ell^{(d)}_{p,i}=c_1(\mathcal L^{(d)}_{p,i})-u_d$. \begin{lemma}\label{betti:degree-filtered} The degree operator $Q_d:R_{d-1}\to R_d$ induced by $T_0(\omega)$ is an isomorphism, and \[ Q_d C_bR_{d-1}=C_bR_d,\qquad Q_d P_jR_{d-1}=P_jR_d. \] Moreover, $R_d=H^*_{\mathrm{pure}}(U_d)$ for every $d$. The normalized maps $Q_d/n$ and their inverses preserve the unit. \end{lemma} \begin{proof} Fix $p\in D$. The ordered branch $\Gamma_{p,i}$ is a relative Cartier divisor in the smooth locus of the universal spectral curve. Tensoring by $\mathcal O(-\Gamma_{p,i})$ gives an isomorphism $\operatorname{Mod}^{(d)}_{p,i}:U_d\to U_{d-1}$ over the Hitchin base, with inverse tensoring by $\mathcal O(\Gamma_{p,i})$. Integral spectral support makes all these sheaves stable. Consequently $X^{(d)}_{p,i}=(\operatorname{Mod}^{(d)}_{p,i})^*$ is a unital cohomology-ring isomorphism in the opposite direction. The modification is supported at $p$, so $X^{(d)}_{p,i}u_{d-1}=u_d$. On its own branch the eigenline changes by the fixed line $T_p^*C$, and the other eigenlines are unchanged. Hence every normalized flag root is preserved. Pushing forward \eqref{betti:branch-modification} to $C$ gives \[ X^{(d)}_{p,i}\ch^{\mathrm{nor}}(\mathcal E_{d-1}) =\ch^{\mathrm{nor}}(\mathcal E_d) -e^{\ell^{(d)}_{p,i}}\omega. \] Thus, for $k\geq1$ and $a^{(d)}_{k,\gamma}=\int_C \ch^{\mathrm{nor}}_k(\mathcal E_d)\gamma$, \begin{equation}\label{betti:branch-C} X^{(d)}_{p,i}a^{(d-1)}_{k,\gamma} =a^{(d)}_{k,\gamma} -\frac{(\ell^{(d)}_{p,i})^{k-1}}{(k-1)!} \int_C\omega\gamma. \end{equation} The correction has Chern index at most $k-1$, including index zero when $k=1$. Since $X^{(d)}_{p,i}$ is a ring map, this proves the filtration bound for every product of generators. Its inverse has the same formula with a plus sign, and therefore $X^{(d)}_{p,i}C_bR_{d-1}=C_bR_d$. These algebraic isomorphisms also identify the pure cohomologies. The equality $R_{d_0}=H^*_{\mathrm{pure}}(U_{d_0})$ already proved in \eqref{betti:pure-U} consequently holds in every degree. The Hecke correspondence over $p$ is the disjoint union of its $n$ ordered branches, each with multiplicity one. Hence \[ Q_d=\sum_{i=1}^n X^{(d)}_{p,i},\qquad Q_d(1)=n\,1. \] Fix a branch $i_0$. By \eqref{betti:branch-C}, each ring automorphism $(X^{(d)}_{p,i_0})^{-1}X^{(d)}_{p,i}$ is the identity modulo one lower Chern step. Thus \[ (X^{(d)}_{p,i_0})^{-1}Q_d=n\operatorname{id}+N_d, \qquad N_d C_bR_{d-1}\subset C_{b-1}R_{d-1}. \] The finite-dimensionality and exhaustiveness of the Chern filtration make $N_d$ nilpotent. The finite geometric series for $(n\operatorname{id}+N_d)^{-1}$ shows that $Q_d$ and $Q_d^{-1}$ both preserve $C$. Each branch modification is an isomorphism over the same base, so it strictly preserves the centered perverse filtration, with no dimension shift. Therefore $Q_dP_j\subset P_j$, while any single branch isomorphism gives $\dim P_jR_{d-1}=\dim P_jR_d$. The injectivity of $Q_d$ makes this inclusion an equality. The assertion about the unit follows from $Q_d(1)=n\,1$. \end{proof} Here is the precise algebraic reduction input we shall use from \cite[\S\S6--7]{HMMS}. For the above pure one-dimensional surface-sheaf correspondence, assume that colength-one subsheaves remain in the elliptic moduli, that a weight-one tautological line splits the scalar gerbe, and that cohomological positive-degree classes on the coarse space are nilpotent. The degree identifications used in the polynomiality construction are given on the stack by $T_0(\omega)/n$. On normalized tautological cohomology they induce the maps $Q_d/n$ of Lemma~\ref{betti:degree-filtered}. Weyl reduction with \begin{equation}\label{betti:y-reduction} y=\psi_1(\omega)/n=c_1(\mathcal E_x)/n \end{equation} produces operators $D_{a,b}(\gamma)$ on the reduced tautological cohomology. We use $D_{a,b}(\gamma)$ for the operators denoted $\widetilde D_{a,b}(\gamma)_{\mathrm{red}}$ in \cite[Section 6.6 and Proposition 7.5]{HMMS}; $q_1(\omega)=T_0(\omega)$ retains its unreduced meaning as an operator between successive degrees. Reduction replaces every curve and flag Chern series by $e^{-y}$ times that series. The difference $y-v$ descends to the coarse space and is nilpotent, so the splitting by \eqref{betti:eigenline} gives the same reduced operators. The stack operator satisfies $T_0(\omega)(u_{d-1}a)=u_dT_0(\omega)(a)$ for every tautological class $a$ on the source stack. Hence its quotient by $y=u$ induces the operator $Q_d$ of Lemma~\ref{betti:degree-filtered}. The normalized maps $Q_d/n$ fix the unit. This is the quotient identification of reduced cohomology in \cite[Section 6.6]{HMMS}. For the operator argument now fix \[ d_*=n(g-1),\qquad U_*=U_{d_*},\qquad R_*=H^*_{\mathrm{pure}}(U_*)=R_{d_*}. \] The branch isomorphisms identify $U_*$ with $U_{d_0}$ over the ordered Hitchin base. In this degree $\psi_1(1)=d_*+n(1-g)=0$, and the three relations needed below are \begin{align} [D_{2,0}(1),D_{a,b}(\gamma)]&=-2bD_{a+1,b-1}(\gamma),\label{betti:quadratic-relations}\\ [D_{1,1}(1),D_{a,b}(\gamma)]&=(a-b)D_{a,b}(\gamma),\nonumber\\ [D_{0,2}(1),D_{a,b}(\gamma)]&=2aD_{a-1,b+1}(\gamma).\nonumber \end{align} The extended relations \eqref{betti:branch-relations} add the flag generators to this reduction. These are statements about the raw operator construction and its algebraic reduction. All the stated hypotheses now hold. An integral spectral curve has no proper positive-rank spectral subcurve, so its rank-one torsion-free sheaves and their colength-one modifications are stable. The weight-one line was constructed above. The tautological cohomology of the coarse space is finite dimensional and its positive-degree ideal is nilpotent. Hence \begin{equation}\label{betti:sl2} e=\tfrac12D_{0,2}(1),\qquad h=-D_{1,1}(1),\qquad f=-\tfrac12D_{2,0}(1) \end{equation} is an $\mathfrak{sl}_2$-triple, and \begin{equation}\label{betti:chern-commutator} [h,\psi_k(\gamma)]=k\psi_k(\gamma),\qquad [h,y_{p,i}]=y_{p,i} \end{equation} after the indicated normalization. All these operators and exact commutators are now taken on $R_*$. We will establish the filtration comparison there, and then use Lemma~\ref{betti:degree-filtered} to return to degree $d_0$. \subsection{Relative Lefschetz theory and low genus} We give the filtration comparison explicitly. We use two elementary consequences of relative hard Lefschetz \cite[Theorem 3.6 and Propositions 3.7--3.9]{HMMS}. First, a relatively ample class and the centered filtration form a Lefschetz structure, and restriction to an open base is a morphism of such structures. Second, if $a:Y\to Z$ is over a common base, both structure maps are projective and both spaces are smooth, then on pure cohomology \begin{equation}\label{betti:perverse-map-bound} a^*P_j\subset P_{j+\dim Z-\dim Y},\qquad a_*P_j\subset P_{j+\dim Z-\dim Y}. \end{equation} The Gysin map retains its usual cohomological degree $2(\dim Z-\dim Y)$. These inequalities follow either from the perverse truncations of the corresponding cohomological correspondences or from their projective compactifications and the nilpotent filtration for a pulled-back ample base class. Every Hecke correspondence above preserves the spectral curve. Applying \eqref{betti:perverse-map-bound}, with the virtual-class version of \cite[Proposition 3.9]{HMMS} for the Hecke projections, and using the degree of the inserted class gives, for $\gamma\in H^i(C)$, \[ D_{a,b}(\gamma)P_j\subset P_{j+2b-2+i}. \] The degree maps $Q_d$ and their inverses strictly preserve the centered perverse filtration by Lemma~\ref{betti:degree-filtered}. Polynomial extraction and Weyl reduction therefore preserve the displayed bounds. It follows that the operators $(e,h,f)$ on $R_*$ in \eqref{betti:sl2} have perverse degrees $(2,0,-2)$. A relatively ample class $\alpha$ on $U_*$ is pure and, by Lemma~\ref{betti:degree-filtered}, tautological. The following linear-algebra argument takes place on $R_*$ and permits the same comparison in genus zero and one. Write $y(\gamma)=\psi_1(\gamma)$ and $x(\gamma)=D_{1,0}(\gamma)$ for positive-degree curve or flag classes. Put $\langle\gamma,\eta\rangle=\int_C\gamma\eta$ for curve classes. The reduced relations give \begin{equation}\label{betti:weyl-linear} [y(\gamma),x(\eta)]=n\langle\gamma,\eta\rangle \quad(\gamma,\eta\in H^1(C)),\qquad [e,x(\gamma)]=y(\gamma). \end{equation} The first bracket vanishes if one argument is a flag class. Operators $x$ have $h$-degree $-1$, and operators $y$ have degree $1$. Exponential conjugation by the nilpotent flag operators adds arbitrary flag terms to $e$ while preserving the $h$-filtration. For the odd curve generators, the operators $x(\gamma)y(\eta)$ generate the ordinary $\mathfrak{gl}(H^1(C))$ action. They preserve the $h$-filtration. The flag operators $x(\tau_{p,i})$ have perverse degree zero. For odd curve classes, $x(\gamma)y(\eta)$ has perverse degree $-1+1=0$. Their conjugations and inverse conjugations therefore also preserve $P$. In our normalization $\psi_2(1)=2e$. Choose a symplectic basis $\gamma_1,\ldots,\gamma_{2g}$, and put \[ y_i=y(\gamma_i),\qquad \Omega_y=\sum_{i=1}^g y_i y_{i+g},\qquad J=\begin{pmatrix}0&I_g\\-I_g&0\end{pmatrix}. \] The odd generators satisfy \[ [\Omega_y,x(\gamma)]=-n y(\gamma),\qquad [\Omega_y,x(\gamma)y(\eta)]=-n y(\gamma)y(\eta), \qquad [e,x(\gamma)y(\eta)]=y(\gamma)y(\eta). \] For example, the first identity follows by applying the graded product rule to each $y_i y_{i+g}$ and using \eqref{betti:weyl-linear}; the two nonzero pairings in a symplectic pair give $-n$ times the corresponding component of $y(\gamma)$. Thus $K=e+\Omega_y/n$ is fixed by this linear-group action, as are all flag classes. Write the pure degree-two class as \[ \alpha=aK+\sum_{i0$, \begin{equation}\label{betti:alternating-determinant} \det B_\lambda=(2/n)^{2g}\lambda^{2g} +\text{a polynomial of degree at most }2g-1, \end{equation} since $\det J=1$. For $g=0$, the determinant is the empty determinant $1$. Thus $a_\lambda\ne0$ and $B_\lambda$ is nondegenerate outside a finite set of parameters in every genus. Scale the original triple and conjugate by the flag operators to make its raising operator $a_\lambda e+\sum C_{p,i}y_{p,i}$. In the $K$ decomposition its alternating part is $-(a_\lambda/n)J$. The group $\GL(H^1(C))$ carries this nondegenerate alternating form to $B_\lambda$, while fixing $K$ and the flag terms. This gives a triple with raising operator $\alpha+\lambda\psi_2(1)$ and the same $h$-filtration. A sufficiently small permissible rational $\lambda$ preserves relative ampleness. We obtain a triple $(e',h',f')$ with $e'=\alpha+\lambda\psi_2(1)$, the same increasing $h$-filtration, and perverse degrees $(2,0,-2)$. On $\Gr^P R_*$, relative hard Lefschetz provides the centered grading $h_P$ and a triple with the same $e'$. The operator $h'$ has degree zero, so it commutes with $h_P$. We recall the needed finite-dimensional uniqueness fact: two $\mathfrak{sl}_2$-triples with the same raising operator and commuting semisimple grading operators have the same grading operator. One proof decomposes a representation into its raising-operator Jordan strings. A centralizer of the raising operator is the product of the general linear groups on equal-length primitive strings, extended by maps raising the string filtration. Conjugate grading operators differ by the latter maps; commutation with the original grading forces every positive-filtration term to vanish. The remaining product centralizes the grading. Applying this fact gives $h'=h_P$. The unit of $R_*$ has $h$-degree $-N_F$, the negative arithmetic genus of the integral spectral curve. Equation \eqref{betti:chern-commutator} therefore identifies its increasing eigenspace filtration with $C_{j+N_F}R_*$. We have proved \[ P_jR_*=C_{j+N_F}R_*. \] Let $\mathcal T_{d_0,d_*}:R_*\to R_{d_0}$ be the composition of the normalized successive-degree maps $Q_d/n$, using their inverses when necessary. Lemma~\ref{betti:degree-filtered} gives \[ P_jR_{d_0} =\mathcal T_{d_0,d_*}(P_jR_*) =\mathcal T_{d_0,d_*}(C_{j+N_F}R_*) =C_{j+N_F}R_{d_0}. \] Thus, for the original space $U=U_{d_0}$, \begin{equation}\label{betti:pc-U} P_jH^*_{\mathrm{pure}}(U) =C_{j+N_F}H^*_{\mathrm{pure}}(U). \end{equation} Choose a relatively ample class on the original space $X$ and restrict it to $U$. Restriction from $X$ is an isomorphism onto the pure part by \eqref{betti:pure-U}, and is a morphism of the corresponding Lefschetz structures. Such morphisms are strict: kernels and cokernels inherit relative Lefschetz isomorphisms on their associated gradeds, so the two filtrations on the image agree. The same restriction identifies the normalized Chern generators. Hence \eqref{betti:pc-U} proves $P=C$ on the original space $X$ in degree $d_0$. \subsection{The zero-residue fiber and its centered shift} We now check that the second restriction in \eqref{betti:residue-cohomology} and its inverse preserve centered perversity. Let $q=\dim A_D$, $d_X=\dim X$ and $d_0'=\dim X_0=d_X-q$. Use the scaling compactifications \[ \overline X= ((X\times\A^1)\setminus(h^{-1}(0)\times\{0\}))/\Gm, \qquad \overline X_0= ((X_0\times\A^1)\setminus(h_0^{-1}(0)\times\{0\}))/\Gm. \] Away from the excluded zero fiber the scaling stabilizers are finite. These are smooth proper orbifolds; rationally their coarse spaces are smooth for the duality and Lefschetz statements being used. The Hitchin maps extend to weighted projective bases. Write $L_B$ and $L_{B,0}$ for the pulled-back weighted hyperplane classes. The complement of the zero Hitchin fiber in $\overline X$ is the orbifold line bundle over its boundary. Purity and the localization sequence imply \begin{equation}\label{betti:boundary-quotient} H^*(X)=H^*(\overline X)/L_BH^*(\overline X),\qquad H^*(X_0)=H^*(\overline X_0)/L_{B,0}H^*(\overline X_0). \end{equation} Indeed, restriction to the boundary is surjective by purity, and its Gysin map composed with restriction is multiplication by the boundary class $L_B$. The resulting short exact localization sequence gives the first quotient, and gives the second with the identical geometric argument for $X_0$. The inclusion $\iota:\overline X_0\into\overline X$ is the regular zero locus of the $q$ weight-one residue sections. Consequently \begin{equation}\label{betti:zero-selfintersection} \iota_*\iota^*=a L_B^q\quad(a\in\Q^\times). \end{equation} The isomorphism modulo $L_B$ in \eqref{betti:residue-cohomology} shows that $\im\iota_*\subset\im L_B^q$. Write $\beta=\iota^*\gamma+L_{B,0}\beta_1$ and push forward: \[ \iota_*\beta=aL_B^q\gamma+L_B\iota_*\beta_1. \] Repeat this lifting $q$ times. Every term is divisible by $L_B^q$. It follows for every $a'\geq0$ that \begin{equation}\label{betti:kernel-power} (\iota^*)^{-1}(\ker L_{B,0}^{a'}+\im L_{B,0}) \subset\ker L_B^{a'+q}+\im L_B. \end{equation} Indeed, apply \eqref{betti:zero-selfintersection} to a lift of an element of the left side. The error in $\im L_{B,0}$ pushes forward to $\im L_B^{q+1}$ by the preceding inclusion, and subtracting its preimage leaves an element killed by $L_B^{a'+q}$. On the quotient by the image of a nilpotent operator $L_B$, its canonical filtration at index $b\geq0$ is represented by $\ker L_B^{b+1}$; at negative indices it is zero. The projective perverse/nilpotent-filtration formula represents $P_jH^i(X_0)$ by this filtration at index $j+d_0'-i$. Equation \eqref{betti:kernel-power} replaces its kernel exponent by \[ (j+d_0'-i+1)+q=j+d_X-i+1, \] which is precisely the exponent for $P_jH^i(X)$. Thus the inverse of restriction preserves centered $P$. It commutes with a chosen relative ample class, so it is a morphism of Lefschetz structures; strictness gives a filtered isomorphism. We obtain \begin{equation}\label{betti:pc-full-zero} P_jH^*(X_0)=C_{j+N_F}H^*(X_0) \end{equation} in every genus and every numerically generic degree. \subsection{Blockwise Gysin transfer} Return to the original partial flags and put \begin{equation}\label{betti:cflags} c=\sum_{p,a}\binom{\mu^p_a}{2},\qquad N_F=N_\mu+c. \end{equation} Inside $X_0$, impose zero induced residue on each original graded block and call the resulting locus $Z$. It has two maps \[ X_0\xleftarrow{\ i\ }Z\xrightarrow{\ \pi\ }M_{\mathrm{Dol},\mu}. \] The stability perturbation already proved makes $\pi$ the product of complete flag bundles in the graded block bundles; its relative dimension is $c$. The residue condition is the zero of the rank-$c$ bundle of strictly lowering endomorphisms inside those blocks. Both $Z$ and $X_0$ are smooth and their dimensions differ by $c$, so this section is regular. Its normal bundle is $T_\pi^*$. The partial-flag moduli space $M_{\mathrm{Dol},\mu}$ is smooth semiprojective: scaling the Higgs field has a limit by properness of the Hitchin map, and its fixed locus is contained in the proper zero fiber. Its cohomology is pure by \cite[Corollary 1.3.2]{HRVPurity}. The flag-bundle formula gives purity of $H^*(Z)$ as well, so the pure-cohomology bounds \eqref{betti:perverse-map-bound} apply to both maps. Let $y_{p,a,b}$ be the normalized Chern roots of the refined block and set \[ \Delta_\mu=\prod_{p,a}\prod_{b0$, with $C'$ a smooth projective curve. In particular the restriction of a local system defines \begin{equation}\label{betti:restriction-stack} R_f:\Loc_{\GL_n}(\mathcal C)\longrightarrow\Loc_{\GL_n}(C'). \end{equation} This is an algebraic morphism of the entire representation stacks. Choose finite presentations of the two fundamental groups. Every generator of the covering group is a word in the base generators, so restriction is given by regular matrix-word functions on representation schemes, equivariant for simultaneous conjugation. An irreducible representation may become reducible upon restriction; the target of \eqref{betti:restriction-stack} includes those representations. \subsection{The homogeneous weights of orbifold tautological classes} We recall and explain the precise universal-class weight statement \cite{Shende}. For a topological space $Y$ with a finite model, the Betti local-system stack has universal classes \[ \int_\gamma\eta\in H^{2k-j}(\Loc_G(Y)),\qquad \gamma\in H_j(Y,\Q),\quad \eta\in H^{2k}(BG,\Q) \] when $\eta$ is a characteristic class. They lie in \begin{equation}\label{betti:shende-weight} F^k\cap\overline{F^k}\cap W_{2k}. \end{equation} Thus each nonzero rational such class defines a copy of $\Q(-k)$ in the indicated cohomology group. The statement concerns the entire local-system stack, without stability restrictions. Here is the reason for the degree-independent weight in \eqref{betti:shende-weight}. Replace $Y$ by a finite simplicial model and use the constant simplicial algebraic scheme whose components in each simplicial degree are points indexed by that model. Its cochain complex has weight zero. Evaluation of a local system on the vertices and transport on the edges gives an algebraic simplicial evaluation to $BG$. The Chern class on $BG$ has type $(k,k)$ and weight $2k$; its pullback has these same bounds in the simplicial mixed Hodge complex. Pairing with a degree-$j$ topological cycle changes cohomological degree by $j$, while the cochain factor being paired has weight zero. It therefore does not change the weight or the two Hodge bounds. Simplicial descent gives \eqref{betti:shende-weight}. Finally, on $W_{2k-1}$ the two bounds $p,q\geq k$ are incompatible with $p+q\leq2k-1$, so a nonzero rational class satisfying them is a pure Tate class of weight $2k$. Apply this result on $C'$ and pull back algebraically by \eqref{betti:restriction-stack}. Rational homological transfer for \eqref{betti:finite-cover} satisfies \[ f_*f^!=a\id:H_*(\mathcal C,\Q)\longrightarrow H_*(\mathcal C,\Q). \] Naturality of the universal bundle consequently gives \begin{equation}\label{betti:transfer-taut} \int_\gamma\ch_k(\mathcal E_{\mathcal C}) =\frac1a R_f^* \left(\int_{f^!\gamma}\ch_k(\mathcal E_{C'})\right). \end{equation} This proves the homogeneous Tate weight $2k$ for every orbifold universal-curve component. Only the pullback on representation stacks carries mixed Hodge structures in this argument; no transfer of a mixed Hodge structure through an analytic Riemann--Hilbert map is being used. The topological transfer supplies the equality of universal cohomology classes in \eqref{betti:transfer-taut}. The eigenbundle at a marking is algebraic on the fixed-monodromy stack: it is the image of the polynomial spectral projector \[ \prod_{b\ne a}\frac{X_p-\xi_{p,b}\id}{\xi_{p,a}-\xi_{p,b}}. \] Its $k$th Chern class is therefore a Tate class of weight $2k$. Normalization by $e^{-u}$ preserves this statement at total Chern index $k$, since $u^r\ch_{k-r}$ has weight $2r+2(k-r)=2k$. The normalized invariant classes descend via \eqref{betti:gerbe-LH}, which is an identity in mixed Hodge structures. All normalized orbifold and block generators on the coarse Betti space are thus homogeneous of the required weight. \subsection{The coarse/root comparison is triangular} The Dolbeault generators in Theorem \ref{betti:pc} use the coarse curve bundle, while \eqref{betti:transfer-taut} uses the orbifold bundle. We compute their difference. Let $\mathcal E_0=\nu_*\mathcal E$ and let $\mathcal V_{p,c}$ denote the multiplicity bundle of character $c$, with $0\leq c1$ divided $n$, $\chi_0$, and all the $\mu^p_a$, then it would divide $S=n(1-g)-\chi_0$. The proper subprofile $\mu/h$ would have eigenvalue product $e^{-2\pi iS/h}=1$, another contradiction. This proves \eqref{betti:target-primitive}. \end{proof} The auxiliary Euler characteristic one in \eqref{betti:active-wall} chooses chambers for the finite active coefficient construction. For the character variety in Theorem \ref{main:hlrv}, positive-source degree transport is applied to $\chi_0$ from \eqref{betti:target-chi}, followed by its canonical vanishing-cycle realization. Theorem \ref{betti:constancy} carries the resulting weight-graded numbers to every generic parameter in that determinant component. Theorem \ref{betti:torsion} supplies this construction in each component. \section{Recovery of the mixed Hodge polynomial} \label{main:recovery} We now combine the root-curve primitive with the punctual calculation. We retain Hodge classes until the two independent cohomological indices have been recovered. The ordinary-curve argument has already established this procedure in Section~\ref{sec:recovery}; we give the formulas with all puncture data present. \subsection{The full Hodge identity} In each generic determinant component, choose finite-order unitary eigenvalues as in Theorem~\ref{betti:torsion}, avoiding also the finite active-profile walls of Lemma~\ref{root:active-weights}. Write $\xi_{p,a}=\exp(-2\pi\mathrm i\alpha_{p,a})$ with $\alpha_{p,a}\in\Q\cap[0,1)$. Put \[ S=\sum_{p,a}\mu^p_a\alpha_{p,a}\in\Z, \qquad d_0=-S, \qquad\chi_0=n(1-g)-S, \qquad N=\frac12d_{\boldsymbol\mu}. \] Let $M$ be the corresponding stable strongly parabolic Higgs moduli space of coarse degree $d_0$, and let $h:M\to\mathcal A$ be its proper Hitchin morphism. By Theorem~\ref{betti:pw}, its cohomology is identified with that of the Betti variety, with the specified perverse and weight filtrations. Define \begin{equation}\label{main:perverse-objects} P_i={}^{p}\!\mathcal H^i(Rh_*\Q_M[2N]), \qquad A_i=[R\Gamma_c(\mathcal A,P_i)]. \end{equation} All sums over $i$ are finite. For a symmetric series in the $m$ independent alphabets, let $\operatorname{Coeff}_{\boldsymbol\mu}$ mean Hall pairing with $\prod_p h_{\mu^p}[X_p]$. Equivalently, it is the coefficient of the prescribed monomial of multiplicities $\mu^p$ in each alphabet. These two descriptions agree because $h_\rho$ and $m_\rho$ are dual for the Hall pairing. There is no factorial in this coefficient operation. Use $R$ for the formal rank variable in the root stable-pair series $Z_{\mathrm{nil}}$. The Fock formula of Theorem~\ref{root:nil-series}, followed by Theorem~\ref{vert:macdonald}, is \begin{equation}\label{main:nil-macdonald} Z_{\mathrm{nil}}(R,x;X_1,\ldots,X_m) =\sum_\lambda R^{|\lambda|} x^{-(g-1)\|\lambda\|^2} \prod_{b\in\lambda}\zeta_C(x^{h_b}\TT^{l_b}) \prod_{p=1}^m\widetilde H_\lambda[X_p;x^{-1},x\TT]. \end{equation} Here $\zeta_C$ is the full Hodge-class series \eqref{rec:curve-zeta}. The root baseline and removed-point factors have already canceled before this expression is formed. The primitive theorem, with $\chi_0$ as above, gives \begin{equation}\label{main:primitive-identity} [R^n]\operatorname{Coeff}_{\boldsymbol\mu}\Log Z_{\mathrm{nil}} =\frac{x\TT^{-N}}{(1-x)(1-x\TT)}\sum_i(-x)^i A_i. \end{equation} This is Theorem~\ref{root:primitive}, after positive-source transport to $\chi_0$ and application of the canonical critical functor. The coefficientwise finiteness proved for the positive source and the full root-character bounds make both sides meaningful in the same completed Grothendieck group. In any fixed rank and profile, there are finitely many possible lower-rank decompositions and perverse indices. The two nonnegative free indices in the scalar primitive have a prescribed sum in a fixed power of $x$. If $M$ is empty, the right side of \eqref{main:primitive-identity} is zero. The hook substitution below then gives $\mathbb H_{\boldsymbol\mu}=0$, so the formulas hold with zero cohomology. In the separation of degrees that follows, assume $M$ is nonempty. \subsection{Separation of the two degrees} The space $M$ is smooth semiprojective. Its cohomology in degree $D$ is pure of weight $D$, and smooth duality gives the same assertion for its compactly supported cohomology. The proper decomposition theorem identifies \[ H_c^j(\mathcal A,P_i) \quad\text{as a direct summand of}\quad H_c^{2N+i+j}(M). \] It is therefore pure of weight $2N+i+j$. If \[ F_{\boldsymbol\mu}(x) =x^{-1}\TT^N(1-x)(1-x\TT) [R^n]\operatorname{Coeff}_{\boldsymbol\mu}\Log Z_{\mathrm{nil}}, \] then \eqref{main:primitive-identity} yields \begin{equation}\label{main:weight-extraction} [H_c^j(\mathcal A,P_i)] =(-1)^{i+j}\Gr^W_{2N+i+j} \bigl([x^i]F_{\boldsymbol\mu}(x)\bigr). \end{equation} The right side determines the pure Hodge structure, since the category of pure polarizable Hodge structures of a fixed weight is semisimple. Thus this extraction retains both $i$ and $j$. The centered perverse objects in \eqref{main:perverse-objects} correspond to the nonnegative perverse index $k$ by $i=k-N$. Put \begin{equation}\label{main:ordinary-perverse-piece} V_{D,k}=H^{D-N-k}(\mathcal A,P_{k-N}), \qquad \beta_{D,k}=\dim V_{D,k}. \end{equation} Since $\mathbb D P_i=P_{-i}(2N)$, Verdier duality gives \begin{equation}\label{main:duality-index} V_{D,k} \simeq H_c^{N+k-D}(\mathcal A,P_{N-k})^\vee\otimes\TT^{2N}. \end{equation} The compact indices corresponding to $(D,k)$ are consequently \[ i_c=N-k,\qquad j_c=N+k-D, \qquad 2N+i_c+j_c=4N-D. \] By Theorem~\ref{betti:pw}, \begin{equation}\label{main:beta-betti} \beta_{D,k} =\dim\Gr^W_{2k}H^D(\mathcal M_{\boldsymbol\mu},\Q), \qquad H(\mathcal M_{\boldsymbol\mu};Q,u) =\sum_{D,k}\beta_{D,k}Q^ku^D. \end{equation} The Hodge--Tate property follows in the same theorem from the normalized orbifold and flag Chern generators. \subsection{The hook substitution} Let $s$ denote the weight character of the half-Tate object, so $\TT$ has character $s^2$. Then \[ \zeta_C(y;s)=\frac{(1-sy)^{2g}}{(1-y)(1-s^2y)}. \] Take this character only after \eqref{main:weight-extraction}. The substitution \begin{equation}\label{main:substitution} x=z^{-2},\qquad s=zw \end{equation} also sends the two local Macdonald parameters $(x^{-1},xs^2)$ to $(z^2,w^2)$. A box of arm $a$ and leg $l$ contributes \[ z^{-2(g-1)(2a+1)} \frac{(z^{2a+1}-w^{2l+1})^{2g}} {(z^{2a+2}-w^{2l})(z^{2a}-w^{2l+2})}. \] The power outside the fraction cancels the baseline in \eqref{main:nil-macdonald}, because $\sum_{b\in\lambda}(2a_b+1)=\|\lambda\|^2$. Hence the resulting series is the rank-marked version of $\Omega_{g,m}$ in \eqref{main:cauchy-series}. If $E_c(P_i;s)$ is the alternating weight character of $A_i$, multiplying the transformed primitive identity by $(z^2-1)(1-w^2)$ gives \begin{equation}\label{main:Hmu-perverse} \mathbb H_{\boldsymbol\mu}(z,w) =(zw)^{-2N}\sum_i(-z^{-2})^i E_c(P_i;zw). \end{equation} For the compact indices in \eqref{main:duality-index}, the total sign is $(-1)^{i_c+j_c}=(-1)^D$ and the weight is $4N-D$. Consequently \begin{equation}\label{main:finite-Hmu} \mathbb H_{\boldsymbol\mu}(z,w) =\sum_{D,k}\beta_{D,k}(-1)^D z^{2k-D}w^{d_{\boldsymbol\mu}-D}. \end{equation} All signs in the plethystic calculations are additive signs in the lambda ring. An ordinary evaluation of the final rational function at negative $z$ or $w$ is made after these calculations. \subsection{Proof of the main theorem} \begin{proof}[Proof of Theorem~\ref{main:hlrv}] First consider the finite-order unitary parameters fixed above. For each term of \eqref{main:finite-Hmu}, since $d=2N$ is even, \[ (u\sqrt Q)^d(-1)^D(\sqrt Q)^{2k-D} \left(-\frac1{u\sqrt Q}\right)^{d-D} =Q^ku^D. \] Thus \begin{equation}\label{main:ordinary-formula} H(\mathcal M_{\boldsymbol\mu};Q,u) =(u\sqrt Q)^d\mathbb H_{\boldsymbol\mu} \left(\sqrt Q,-\frac1{u\sqrt Q}\right). \end{equation} Smooth Poincare duality then gives \[ \begin{split} H_c(\mathcal M_{\boldsymbol\mu};Q,u) &=Q^du^{2d} H(\mathcal M_{\boldsymbol\mu};Q^{-1},u^{-1})\\ &=(u\sqrt Q)^d \mathbb H_{\boldsymbol\mu}(Q^{-1/2},-u\sqrt Q). \end{split} \] Transposing every summation partition in \eqref{main:cauchy-series} exchanges $z$ and $w$, since the modified Macdonald transpose symmetry and the hook product exchange simultaneously. It follows that $\mathbb H_{\boldsymbol\mu}(z,w) =\mathbb H_{\boldsymbol\mu}(w,z)$. The hook factors and the squared Macdonald parameters also give $\mathbb H_{\boldsymbol\mu}(-z,-w) =\mathbb H_{\boldsymbol\mu}(z,w)$. This proves \eqref{main:hlrv-formula} at the chosen parameters. Theorem~\ref{betti:torsion} supplies such a generic finite-order unitary point in every nonempty generic determinant component. Theorem~\ref{betti:constancy} makes every compactly supported weight-graded rank locally constant on that component. The same argument, or the relative Hodge--Tate stratification used there, keeps the off-diagonal Hodge numbers zero. Applying the formula separately in each component proves it for every generic tuple. The expression obtained is the same function of $g,n,\boldsymbol\mu$ in each case, which proves eigenvalue independence across the components as well. Equation~\eqref{main:finite-Hmu} also proves the polynomiality assertion. The ordinary cohomology of a smooth affine variety of complex dimension $d$ vanishes above degree $d$, and its weights in degree $D$ are at least $D$. Thus $D\leq d$ and $2k\geq D$ whenever $\beta_{D,k}\ne0$. Both exponents in \eqref{main:finite-Hmu} are nonnegative. Replacing $z$ by $-z$ removes its sign, since $(-1)^{D+2k-D}=1$. The coefficients are the nonnegative integers $\beta_{D,k}$. This proves $\mathbb H_{\boldsymbol\mu}(-z,w)\in\Z_{\geq0}[z,w]$. If the variety is nonempty, its nonzero $H^0$ contributes a positive multiple of $w^d$. No term has larger $w$-degree. The symmetry in $z,w$ proves the asserted degree in both variables. For a monomial $z^aw^b$ in $\mathbb H_{\boldsymbol\mu}$, formula \eqref{main:hlrv-formula} has cohomological degree $d+b$ and weight $d-a+b$. Equality of these two degrees holds exactly when $a=0$. Keeping precisely these terms gives \eqref{main:pure-part}. The empty case gives the zero polynomial by the same primitive identity. The displayed ordinary and compact formulas show directly that no choice of square root remains. \end{proof} \begin{corollary}\label{main:hrv-all-genera} The central-twist formula \eqref{eq:hrv-main} holds for every $g\geq0$ and $n\geq1$. \end{corollary} \begin{proof} Take one puncture, $\mu^1=(n)$, and monodromy $\zeta^{-1}I_n$, where $\zeta$ is a primitive $n$th root of unity. This is generic. Pairing a homogeneous symmetric function of degree $n$ with $h_n$ is its one-letter evaluation. Since $\widetilde H_\lambda[1;z^2,w^2]=1$, the specialization of \eqref{main:Hmu} is precisely $\overline H_n$. Formula \eqref{main:ordinary-formula} is therefore \eqref{eq:hrv-main}. For $g=0,n>1$, the defining variety is empty; the formula includes this case with value zero. \end{proof} \begin{thebibliography}{99} \raggedright \bibitem{ABWVertex} A.~Aggarwal, A.~Borodin, and M.~Wheeler, \emph{Colored fermionic vertex models and symmetric functions}, arXiv:2101.01605v2 [math.CO], 2021. \bibitem{bettiBehrendNoohi} K.~Behrend and B.~Noohi, \emph{Uniformization of Deligne--Mumford curves}, arXiv:math/0504309 (2005), Proposition~5.7. \bibitem{BBD} A.~A.~Beilinson, J.~Bernstein and P.~Deligne, \emph{Faisceaux pervers}, Ast\'erisque \textbf{100} (1982), 5--171; \href{https://archive.numdam.org/item/AST_1982__100__1_0/}{Numdam digital edition}. \bibitem{bettiBiquardBoalch} O.~Biquard and P.~Boalch, \emph{Wild non-abelian Hodge theory on curves}, Compos.~Math. \textbf{140} (2004), no.~1, 179--204. \bibitem{rootBGL} I.~Biswas, N.~L.~Gammelgaard, and M.~Logares, \emph{Bohr--Sommerfeld Lagrangians of moduli spaces of Higgs bundles}, J. Geom. Phys. \textbf{94} (2015), 179--184. \href{https://doi.org/10.1016/j.geomphys.2015.04.005} {doi:10.1016/j.geomphys.2015.04.005}. \bibitem{Borri} A.~Borri, \emph{Untwisting the twisted character variety via a stacky curve}, \href{https://arxiv.org/abs/2609.24584v1}{arXiv:2609.24584v1} (2026). \bibitem{Braden} T.~Braden, \emph{Hyperbolic localization of intersection cohomology}, Transform. Groups \textbf{8} (2003), no.~3, 209--216; \href{https://arxiv.org/abs/math/0202251v5}{arXiv:math/0202251v5}. \bibitem{BBJ} C.~Brav, V.~Bussi and D.~Joyce, \emph{A `Darboux theorem' for derived schemes with shifted symplectic structure}, \href{https://arxiv.org/abs/1305.6302v4}{arXiv:1305.6302v4} (2018). \bibitem{BDIKP} C.~Bu, B.~Davison, A.~Ib\'a\~nez N\'u\~nez, T.~Kinjo and T.~P\u{a}durariu, \emph{Cohomology of symmetric stacks}, \href{https://arxiv.org/abs/2502.04253v2}{arXiv:2502.04253v2} (2025). \bibitem{CRV} E.~Carlsson and F.~Rodriguez-Villegas, \emph{Vertex operators and character varieties}, Adv. Math. \textbf{330} (2018), 38--60; \href{https://arxiv.org/abs/1603.09267v1}{arXiv:1603.09267v1}. \bibitem{rootCL} P.-H.~Chaudouard and G.~Laumon, \emph{Un th\'eor\`eme du support pour la fibration de Hitchin}, Ann. Inst. Fourier \textbf{66} (2016), no.~2, 711--727. \href{https://doi.org/10.5802/aif.3023}{doi:10.5802/aif.3023}. \bibitem{rootCLAdd} P.-H.~Chaudouard and G.~Laumon, \emph{Addendum \`a ``Un th\'eor\`eme du support pour la fibration de Hitchin''}, Ann. Inst. Fourier \textbf{67} (2017), no.~3, 1005--1008. \href{https://doi.org/10.5802/aif.3103}{doi:10.5802/aif.3103}. \bibitem{CDDP} W.-Y.~Chuang, D.-E.~Diaconescu, R.~Donagi and T.~Pantev, \emph{Parabolic refined invariants and Macdonald polynomials}, Comm. Math. Phys. \textbf{335} (2015), no.~3, 1323--1379; \href{https://arxiv.org/abs/1311.3624}{arXiv:1311.3624}. \bibitem{CDP} W.-Y.~Chuang, D.-E.~Diaconescu and G.~Pan, \emph{BPS states and the $P=W$ conjecture}, in \emph{Moduli Spaces}, Cambridge University Press, 2014, 132--150; \href{https://arxiv.org/abs/1202.2039v2}{arXiv:1202.2039v2}. \bibitem{DM} B.~Davison and S.~Meinhardt, \emph{Cohomological Donaldson--Thomas theory of a quiver with potential and quantum enveloping algebras}, \href{https://arxiv.org/abs/1601.02479v5}{arXiv:1601.02479v5} (2020). \bibitem{CHM} M.~A.~A. de Cataldo, J.~Heinloth and L.~Migliorini, \emph{A support theorem for the Hitchin fibration: the case of $\GL_n$ and $K_C$}, \href{https://arxiv.org/abs/1906.09582v2}{arXiv:1906.09582v2} (2021). \bibitem{dCHM} M.~A.~A. de Cataldo, T.~Hausel and L.~Migliorini, \emph{Topology of Hitchin systems and Hodge theory of character varieties: the case $A_1$}, Ann. of Math. \textbf{175} (2012), no.~3, 1329--1407, \href{https://doi.org/10.4007/annals.2012.175.3.7}{doi:10.4007/annals.2012.175.3.7}. \bibitem{dCMS} M.~A.~A. de Cataldo, D.~Maulik and J.~Shen, \emph{Hitchin fibrations, abelian surfaces, and the $P=W$ conjecture}, J. Amer. Math. Soc. \textbf{35} (2022), 911--953; \href{https://arxiv.org/abs/1909.11885v2}{arXiv:1909.11885v2}. \bibitem{DJK} F.~D\'eglise, F.~Jin and A.~A.~Khan, \emph{Fundamental classes in motivic homotopy theory}, \href{https://arxiv.org/abs/1805.05920v3}{arXiv:1805.05920v3} (2021). \bibitem{Descombes} P.~Descombes, \emph{Hyperbolic localization in Donaldson--Thomas theory}, \href{https://arxiv.org/abs/2506.22400v1}{arXiv:2506.22400v1} (2025). \bibitem{GKVertex} A.~Gitlin and D.~Keating, \emph{A vertex model for supersymmetric LLT polynomials}, arXiv:2110.10273v1 [math.CO], 2021. \bibitem{HHLVertex} J.~Haglund, M.~Haiman, and N.~Loehr, \emph{A combinatorial formula for Macdonald polynomials}, arXiv:math/0409538v1 [math.CO], 2004. \bibitem{HLRV} T.~Hausel, E.~Letellier and F.~Rodriguez-Villegas, \emph{Arithmetic harmonic analysis on character and quiver varieties}, Duke Math. J. \textbf{160} (2011), no.~2, 323--400; \href{https://arxiv.org/abs/0810.2076v3}{arXiv:0810.2076v3}. \bibitem{HLRVII} T.~Hausel, E.~Letellier and F.~Rodriguez-Villegas, \emph{Arithmetic harmonic analysis on character and quiver varieties II}, Adv. Math. \textbf{234} (2013), 85--128; \href{https://arxiv.org/abs/1109.5202v1}{arXiv:1109.5202v1}. \bibitem{HLRVKac} T.~Hausel, E.~Letellier and F.~Rodriguez-Villegas, \emph{Positivity for Kac polynomials and DT-invariants of quivers}, Ann. of Math. \textbf{177} (2013), 1147--1168; \href{https://arxiv.org/abs/1204.2375v1}{arXiv:1204.2375v1}. \bibitem{HMMS} T.~Hausel, A.~Mellit, A.~Minets and O.~Schiffmann, \emph{$P=W$ via $\mathcal H_2$}, \href{https://arxiv.org/abs/2209.05429v3}{arXiv:2209.05429v3} (2026). \bibitem{HRV} T.~Hausel and F.~Rodriguez-Villegas, \emph{Mixed Hodge polynomials of character varieties}, with an appendix by N.~M.~Katz, Invent. Math. \textbf{174} (2008), 555--624; \href{https://arxiv.org/abs/math/0612668v3}{arXiv:math/0612668v3}. \bibitem{HRVPurity} T.~Hausel and F.~Rodriguez-Villegas, \emph{Cohomology of large semiprojective hyperk\"ahler varieties}, Ast\'erisque \textbf{370} (2015), 113--156; \href{https://arxiv.org/abs/1309.4914v1}{arXiv:1309.4914v1}. \bibitem{HT} T.~Hausel and M.~Thaddeus, \emph{Mirror symmetry, Langlands duality, and the Hitchin system}, Invent. Math. \textbf{153} (2003), no.~1, 197--229, \href{https://doi.org/10.1007/s00222-003-0286-7}{doi:10.1007/s00222-003-0286-7}. \bibitem{HitchinIS} N.~J.~Hitchin, \emph{Stable bundles and integrable systems}, Duke Math. J. \textbf{54} (1987), no.~1, 91--114. \bibitem{HitchinSD} N.~J.~Hitchin, \emph{The self-duality equations on a Riemann surface}, Proc. London Math. Soc. (3) \textbf{55} (1987), no.~1, 59--126, \href{https://doi.org/10.1112/plms/s3-55.1.59}{doi:10.1112/plms/s3-55.1.59}. \bibitem{HSS} M.~Hoyois, S.~Scherotzke and N.~Sibilla, \emph{Higher traces, noncommutative motives, and the categorified Chern character}, Adv. Math. \textbf{309} (2017), 97--154; \href{https://arxiv.org/abs/1511.03589v3}{arXiv:1511.03589v3}. \bibitem{Keller} B.~Keller, \emph{On the cyclic homology of exact categories}, J. Pure Appl. Algebra \textbf{136} (1999), no.~1, 1--56, \href{https://doi.org/10.1016/S0022-4049(97)00152-7}{doi:10.1016/S0022-4049(97)00152-7}. \bibitem{Khan} A.~A.~Khan, \emph{Intersection theory on derived stacks}, \href{https://arxiv.org/abs/1909.01332v2}{arXiv:1909.01332v2} (2026). \bibitem{KPS} T.~Kinjo, H.~Park and P.~Safronov, \emph{Cohomological Hall algebras for $3$-Calabi--Yau categories}, \href{https://arxiv.org/abs/2406.12838v2}{arXiv:2406.12838v2} (2026). \bibitem{Kinjo} T.~Kinjo, \emph{Dimensional reduction in cohomological Donaldson--Thomas theory}, Compos. Math. \textbf{158} (2022), 123--167; \href{https://arxiv.org/abs/2102.01568v2}{arXiv:2102.01568v2}. \bibitem{KK} T.~Kinjo and N.~Koseki, \emph{Cohomological $\chi$-independence for Higgs bundles and Gopakumar--Vafa invariants}, J. Eur. Math. Soc. \textbf{28} (2026), 619--671; \href{https://arxiv.org/abs/2112.10053v2}{arXiv:2112.10053v2}. \bibitem{KM} T.~Kinjo and N.~Masuda, \emph{Global critical chart for local Calabi--Yau threefolds}, Int. Math. Res. Not. \textbf{2024} (2024), no.~5, 4062--4093; \href{https://arxiv.org/abs/2112.10052v2}{arXiv:2112.10052v2}. \bibitem{bettiLeeLee} J.~C.~Lee and S.~Lee, \emph{Generators for the cohomology of the moduli space of irregular parabolic Higgs bundles}, arXiv:2402.05380v2 (2024). \bibitem{rootLin} Y.~Lin, \emph{Moduli spaces of semistable pairs on projective Deligne--Mumford stacks}, Math. Res. Lett. \textbf{30} (2023), no.~4. \href{https://doi.org/10.4310/MRL.2023.v30.n4.a7} {doi:10.4310/MRL.2023.v30.n4.a7}; \href{https://arxiv.org/abs/2006.16530v5}{arXiv:2006.16530v5}. \bibitem{Macdonald} I.~G.~Macdonald, \emph{Symmetric products of an algebraic curve}, Topology \textbf{1} (1962), no.~4, 319--343, \href{https://doi.org/10.1016/0040-9383(62)90019-8}{doi:10.1016/0040-9383(62)90019-8}. \bibitem{Markman} E.~Markman, \emph{Generators of the cohomology ring of moduli spaces of sheaves on symplectic surfaces}, J. Reine Angew. Math. \textbf{544} (2002), 61--82; \href{https://arxiv.org/abs/math/0009109v4}{arXiv:math/0009109v4}. \bibitem{MSY} D.~Maulik, J.~Shen and Q.~Yin, \emph{Perverse filtrations and Fourier transforms}, Acta Math. \textbf{234} (2025), 1--69; \href{https://arxiv.org/abs/2308.13160v3}{arXiv:2308.13160v3}. \bibitem{MSChi} D.~Maulik and J.~Shen, \emph{Cohomological $\chi$-independence for moduli of one-dimensional sheaves and moduli of Higgs bundles}, Geom. Topol. \textbf{27} (2023), 1539--1586; \href{https://arxiv.org/abs/2012.06627v3}{arXiv:2012.06627v3}. \bibitem{MSPW} D.~Maulik and J.~Shen, \emph{The $P=W$ conjecture for $\GL_n$}, Ann. of Math. \textbf{200} (2024), no.~2, 529--556; \href{https://arxiv.org/abs/2209.02568v2}{arXiv:2209.02568v2}. \bibitem{MellitHiggs} A.~Mellit, \emph{Poincar\'e polynomials of moduli spaces of Higgs bundles and character varieties (no punctures)}, Invent. Math. \textbf{221} (2020), 301--327; \href{https://arxiv.org/abs/1707.04214}{arXiv:1707.04214}. \bibitem{MellitIntegral} A.~Mellit, \emph{Integrality of Hausel--Letellier--Villegas kernels}, Duke Math. J. \textbf{167} (2018), no.~17, 3171--3205; \href{https://arxiv.org/abs/1605.01299v2}{arXiv:1605.01299v2}. \bibitem{MellitPoincare} A.~Mellit, \emph{Poincar\'e polynomials of character varieties, Macdonald polynomials and affine Springer fibers}, Ann. of Math. \textbf{192} (2020), no.~1, 165--228; \href{https://arxiv.org/abs/1710.04513}{arXiv:1710.04513}. \bibitem{MellitSurvey} A.~Mellit, \emph{Cohomology rings of character varieties}, \href{https://arxiv.org/abs/2507.12454v1}{arXiv:2507.12454v1} (2025). \bibitem{MellitToric} A.~Mellit, \emph{Toric stratifications of character varieties}, Publ. Math. Inst. Hautes \'Etudes Sci. \textbf{142} (2025), 153--240, \href{https://doi.org/10.1007/s10240-025-00158-0}{doi:10.1007/s10240-025-00158-0}. \bibitem{MSV} L.~Migliorini, V.~Shende and F.~Viviani, \emph{A support theorem for Hilbert schemes of planar curves, II}, Compos. Math. \textbf{157} (2021), 835--882; \href{https://arxiv.org/abs/1508.07602v2}{arXiv:1508.07602v2}. \bibitem{MonavariNested} S.~Monavari, \emph{Double nested Hilbert schemes and the local stable pairs theory of curves}, Compos. Math. \textbf{158} (2022), no.~9, 1799--1849, \href{https://doi.org/10.1112/S0010437X22007606}{doi:10.1112/S0010437X22007606}. \bibitem{MonavariRefined} S.~Monavari, \emph{The refined local Donaldson--Thomas theory of curves}, Geom. Topol. \textbf{30} (2026), no.~9, 3423--3484; \href{https://arxiv.org/abs/2506.14359v3}{arXiv:2506.14359v3}. \bibitem{MozSchiff} S.~Mozgovoy and O.~Schiffmann, \emph{Counting Higgs bundles and type $A$ quiver bundles}, Compos. Math. \textbf{156} (2020), 744--769; \href{https://arxiv.org/abs/1705.04849}{arXiv:1705.04849}. \bibitem{NayakSastry} S.~Nayak and P.~Sastry, \emph{Grothendieck duality and transitivity II: traces and residues via Verdier's isomorphism}, \href{https://arxiv.org/abs/1903.01783v2}{arXiv:1903.01783v2} (2019). \bibitem{PToda} T.~P\u{a}durariu and Y.~Toda, \emph{Topological K-theory of quasi-BPS categories for Higgs bundles}, \href{https://arxiv.org/abs/2409.10800v1}{arXiv:2409.10800v1} (2024). \bibitem{PTVV} T.~Pantev, B.~To\"en, M.~Vaqui\'e and G.~Vezzosi, \emph{Shifted symplectic structures}, \href{https://arxiv.org/abs/1111.3209v4}{arXiv:1111.3209v4} (2013). \bibitem{SaitoMHM} M.~Saito, \emph{Mixed Hodge modules}, Publ. Res. Inst. Math. Sci. \textbf{26} (1990), no.~2, 221--333, \href{https://doi.org/10.2977/prims/1195171082}{doi:10.2977/prims/1195171082}. \bibitem{SaitoTS} M.~Saito, \emph{Thom--Sebastiani theorem for Hodge modules}, \href{https://arxiv.org/abs/2605.05782v1}{arXiv:2605.05782v1} (2026). \bibitem{Schiffmann} O.~Schiffmann, \emph{Indecomposable vector bundles and stable Higgs bundles over smooth projective curves}, Ann. of Math. \textbf{183} (2016), no.~1, 297--362, \href{https://doi.org/10.4007/annals.2016.183.1.6}{doi:10.4007/annals.2016.183.1.6}. \bibitem{Shende} V.~Shende, \emph{The weights of the tautological classes of character varieties}, Int. Math. Res. Not. IMRN (2017), no.~22, 6832--6840; \href{https://arxiv.org/abs/1411.4975v2}{arXiv:1411.4975v2}. \bibitem{Simpson} C.~T.~Simpson, \emph{Higgs bundles and local systems}, Publ. Math. Inst. Hautes \'Etudes Sci. \textbf{75} (1992), 5--95, \href{https://doi.org/10.1007/BF02699491}{doi:10.1007/BF02699491}. \bibitem{bettiSimpson} C.~T.~Simpson, \emph{Harmonic bundles on noncompact curves}, J.~Amer.~Math.~Soc. \textbf{3} (1990), no.~3, 713--770. \bibitem{Spaide} T.~Spaide, \emph{Shifted symplectic and Poisson structures on spaces of framed maps}, \href{https://arxiv.org/abs/1607.03807v1}{arXiv:1607.03807v1} (2016). \bibitem{Stacks} The Stacks Project Authors, \emph{The Stacks Project}, \href{https://stacks.math.columbia.edu}{stacks.math.columbia.edu}; Tags 08GS, 0A1G, 0G99 and 0656. \bibitem{ST} J.~Stoppa and R.~P.~Thomas, \emph{Hilbert schemes and stable pairs: GIT and derived category wall crossings}, \href{https://arxiv.org/abs/0903.1444v3}{arXiv:0903.1444v3} (2009). \bibitem{Su} T.~Su, \emph{Cell decomposition and dual boundary complexes of character varieties}, Adv. Math. \textbf{498} (2026), 111043; \href{https://arxiv.org/abs/2307.16657v5}{arXiv:2307.16657v5}. \bibitem{Sulzgruber} R.~Sulzgruber, \emph{Inserting rim-hooks into reverse plane partitions}, \href{https://arxiv.org/abs/1710.09695v2}{arXiv:1710.09695v2} (2018). \bibitem{TodaCat} Y.~Toda, \emph{Categorical Donaldson--Thomas theory for local surfaces}, \href{https://arxiv.org/abs/1907.09076v4}{arXiv:1907.09076v4} (2021). \bibitem{TodaPT} Y.~Toda, \emph{Curve counting theories via stable objects I. DT/PT correspondence}, \href{https://arxiv.org/abs/0902.4371v3}{arXiv:0902.4371v3} (2009). \bibitem{Tubach} S.~Tubach, \emph{Mixed Hodge modules on stacks}, Forum Math. Sigma \textbf{13} (2025), e175; \href{https://arxiv.org/abs/2407.02256v3}{arXiv:2407.02256v3}. \bibitem{Wildeshaus} J.~Wildeshaus, \emph{Modules de Hodge mixtes \`a croisements normaux}, C. R. Math. Acad. Sci. Paris \textbf{337} (2003), no.~7, 467--472, \href{https://doi.org/10.1016/j.crma.2003.08.001}{doi:10.1016/j.crma.2003.08.001}. \end{thebibliography} \end{document}