SOURCE: https://physics.magflowmeters.com/gates/dossiers/sg6.html
======================================================================

SG-6 — moduli / vacuum stability — dossier & ledger 

 ← Gates scoreboard · Jump to closure ledger 

 Gate dossier — SG-6 — moduli / vacuum stability

 Question: Does the vacuum rest at a stable minimum, or slide away? 
 Status (fixed): DERIVED-GIVEN-anchor · RESOLVED +0 
 Full 13-dimensional treatment: the frozen arena M4 x K6=SU(3)/T2 x S2 x S1_Y/Z2, all three layers, full precision. Self-contained — every value derived or stated inline. 

 Required endpoint (2026-07-06)

 Status: CLOSED / DERIVED-GIVEN-anchor .

 Nothing left. Anchored on: 

 Shape: the frozen internal shape K6 = SU(3)/T2 supplies the moduli space, the three-fold (S3 / Weyl) permutation symmetry that forces the resting point, and the exact split of the three size-parameters into one breathing mode plus a two-component shape mode

 Granularity: sets the finite operational cell that would tame the loop correction, but it cannot erase the finite exact tree-level curvature record that decides the sign

 Scale: the loop scale and the electroweak ruler live on the same shape data

 Observables: None produced. Consumed as inputs: the electroweak scale v = 246 GeV (measured second ruler, not derived here). Reproduced internal cross-checks (structural, not observations): frozen-atlas anchors Scal(1,1,1)=5/2, Ricci eigenvalue 5/12, the four invariant Einstein metrics on K6, and the shape-doublet curvature Hessian eigenvalue +1/3.

 Dissolution: Not applicable except for wrong-target variants; finite records are preserved.

 This is the current gate-level endpoint. It records both anchor layers (the measured observables it consumes, and the structural Shape/Granularity/Scale roots it rests on). The detailed dossier below is retained in full.

 Executive summary & honest status

 Headline. The resting point of the internal shape moduli is forced, not chosen — and the reason it might tip toward instability instead of settling into a minimum is pinned down to a single, named, provably unpinnable sign, not to any unfinished arithmetic. On the frozen thirteen-dimensional arena

 \[
\mathfrak B_{\rm active}=\underbrace{\big[\mathcal M_4\times K_6\times S^2\times S^1_Y/\mathbb Z_2\big]_\times}_{\times\,\text{Stage}}\ \oplus\ \underbrace{\big[\mathcal F^+_{\rm finite}\oplus\mathcal C_{\rm admiss}\big]_\oplus}_{\oplus\,\text{Rulebook}}\ \otimes\ \underbrace{\big[\mathcal E_{\rm matter}\oplus\mathcal E_{\rm gauge}\oplus\mathcal E_{\rm Higgs}\oplus\mathcal E_{\rm proton}\big]_\otimes}_{\otimes\,\text{Actors}},
\]

 with \(K_6=SU(3)/T^2\) the full flag manifold of \(A_2\) and \(D=4+6+2+1=13\) , the Weyl group of \(A_2\) is \(S_3\) (order 6), and it acts on the three internal shape moduli \((u_1,u_2,u_3)\in[1/2,3/2]^3\) by permutation. The chamber-center point \(\bar u=(1,1,1)\) is therefore automatically a critical point — \(\nabla V=0\) there — of any \(S_3\) -invariant potential \(V\) , for purely group-theoretic reasons, with no potential shape tuned to land on that answer. That is the load-bearing, target-blind fact underneath this gate: the location where the internal geometry sits is symmetry-forced, exactly the way a ball placed at the center of a three-fold-symmetric bowl feels zero net force by the symmetry alone, independent of the bowl's exact profile. This is the W-crit sub-wall, and it is closed outright.

 The harder sub-wall, W-stab, asks whether the Hessian at that forced point is a minimum or a saddle. Here the picture is more textured, and the honesty of that texture is the point of this dossier. The three-modulus Hessian splits exactly , by the same \(S_3\) representation theory, into a breathing singlet ( \(u_1=u_2=u_3\) moving together, i.e. overall volume) and a traceless shape doublet ( \(u_1-u_2\) and its partner); the two sectors decouple at quadratic order, so "is the resting point stable" cleanly separates into independent, separately-answerable questions rather than one entangled one. On the shape-doublet sector, the purely geometric (curvature) contribution to the Hessian is computed in closed form along the doublet ray \(u=(1+\varepsilon,1-\varepsilon,1)\) : \(R(\varepsilon)=3/2-\varepsilon^2/2\) , giving raw second derivative \(d^2R/d\varepsilon^2|_0=-1\) (equivalently \(-1/2\) in a unit-normalized convention — the two numbers are the same invariant fact in two conventions and must never be combined into a third value). The decomposition behind that \(-1\) is itself informative: the diagonal convexity of \(\sum 1/u_i\) contributes \(+4\) raw ( \(+2\) unit-normalized) while the structure-constant triangle term contributes \(-5\) raw ( \(-5/2\) unit-normalized), net \(+4-5=-1\) raw. This negative sign is reproduced a second way, as the full \(3\times3\) Hessian of the same curvature model at \((1,1,1)\) , which is exactly \(\begin{psmallmatrix}-1&2&2\\2&-1&2\\2&2&-1\end{psmallmatrix}\) with eigenvalues \(\{3,-3,-3\}\) — the doublet block is \(\mathrm{diag}(-3,-3)\) , confirming the negative sign a second, independent way — and a third way under the physically correct fixed-volume (shape-only, det \(=1\) ) projection, where the volume-preserving second derivative of \(S=\sum 1/x_i-\tfrac16\sum x_i/(x_jx_k)\) is \(+2/3\) and the log-coordinate Hessian eigenvalues are \(\{1/3,1/3,0\}\) ; because the standard flux-free Kaluza–Klein reduction carries \(V\sim -S|_{\rm fixed\ vol}\) , that \(+1/3\) in \(S\) becomes physical shape mass \(^2=-1/3<0\) — again a saddle. All three numbers, \(-1\) (raw ray curvature), \(-1/2\) (unit-normalized ray curvature), and \(+1/3\) (fixed-volume \(S\) -curvature, which flips sign to \(-1/3\) in the physical potential), are the same saddle reported in three conventions, and the sign — negative, i.e. saddle-like at this layer — is what is convention-invariant and load-bearing, not any single one of the three numbers. This geometric-curvature sign is cross-checked against a negative control: the same Hessian construction on \(S^2\times S^2\) fires the opposite , positive sign; the reproducible script value for that control is \(+12\) (a per-sphere \(R=2/b^2\) model), confirming the sign-flip test genuinely discriminates rather than always returning the same answer by construction. (A different, convention-dependent normalization of the same \(S^2\times S^2\) control appears elsewhere in this corpus as \(+16\) ; that number is not the one reproduced by direct re-execution of the frozen script and is flagged rather than asserted here — the qualitative fact that matters, the sign flip itself, holds under either normalization.) In parallel, the bosonic spectral-zeta contribution to the same doublet sector is \(\zeta_{K_6}(-1)=-8033/100800=-0.0796924603174603\ldots\) — also negative — confirmed by two independent routes (a blind numerical heat-kernel sum with Richardson extrapolation, and the exact Weyl/Poisson closed form) agreeing to roughly 22 digits. Both of these computable layers, geometric curvature and bosonic zeta, are REDUCED-TO-FLOOR at +0 each: there is nothing further to compute in them, no hidden knob, no scheme dependence left unresolved. A companion cross-check worth flagging honestly rather than hiding: the tree-level shape-doublet mass \(^2\) is banked at \(\{+1,+1\}\) (a tree-level doublet saddle is structurally impossible), which is why the gate leans stable at a glance — but the tree-level number and the target-blind geometric-curvature number are not the same computation and must be kept visibly separate; they are not merged into one figure anywhere in this dossier.

 What remains open is not an unfinished calculation sitting in a queue — it is a proven absence . The full one-loop physical Hessian sign is, structurally, a product of three layers: geometric curvature \(\times\) bosonic zeta \(\times\) a fermionic graded-Casimir supertrace on the twisted spinor pair \(T_0\leftrightarrow T_1\) ( \(E\leftrightarrow E^*\) under charge conjugation). The first two layers are pinned negative, as above. The third is a single discrete \(\pm\) bit, and the central exact result of this gate is that this bit is certified-unpinnable by any intrinsic invariant of the frozen shape — not because a search stalled, but because a complete, target-blind exhaustion over the closed six-member candidate space (the six \(\chi=-3\) Dynkin-label twist candidates, closed under conjugation and triality) shows that every available discriminator class — Wu/Stiefel–Whitney/Wu orientation data (identically zero and charge-conjugation-even on the frozen tangent bundle), the measure-reality/CPT test (Frobenius–Schur indicator \(\mathrm{FS}(\mathbf 3)=0\) exact, matched by an independent Weyl-integral numeric to \(\sim10^{-16}\) , constraining the \(T_0\) – \(T_1\) pair jointly but unable to split it), the Dai–Freed eta-invariant beyond its mod-2 refinement (undefined precisely at \(\mathrm{FS}(\mathbf3)=0\) ), and the spin- \(\mathbb C\) determinant-line reality condition (empty, because none of the six candidates is self-conjugate — the free charge-conjugation action exchanges the two triality triples) — is exactly charge-conjugation- even , while the sought bit is charge-conjugation- odd by construction. A charge-conjugation-even invariant structurally cannot carry a charge-conjugation-odd sign; this is representation theory, not insufficient effort. This is why the residual is typed as a dressed, anchor-payable IOU (standing certificate AC-GAP10-HOLE3-v1, unpaid, shared with the T0/T1 bit appearing identically in the uqf10 and gap10-bg10 gates and counted once) rather than an open search: it is payable by exactly one class of future object — a measured charge-conjugation-odd record, with the decisive leptonic \(\delta_{CP}\) sign as the leading candidate payer, while the baryon asymmetry \(\eta_B\) is permanently barred as a payer by an independent argument (W12) — and by nothing else, including further internal computation. A leading loop indicator worth flagging honestly: the graded Casimir supertrace \(\mathrm{Str}[C_2]=\chi\cdot C_2(\text{fund})=-4\) (using \(\chi=-3\) ) opposes a boson-dominance rescue of the sign; this dossier does not assume the rescue happens.

 The precise claim. SG-6 claims, at the fixed grade: (i) the chamber-center resting point \(\bar u=(1,1,1)\) is a genuine critical point of the moduli potential, forced by the exact \(S_3\) (Weyl group of \(A_2\) , order 6) symmetry of \(K_6=SU(3)/T^2\) , target-blind and free of any fit — DERIVED-GIVEN- \(E\) , +0 ; (ii) the three-modulus Hessian decomposes exactly as \(\mathbf3=\mathbf1\oplus\mathbf2\) (breathing singlet plus shape doublet), an exact symmetry statement, decoupled at quadratic order — DERIVED-GIVEN- \(E\) , +0 ; (iii) both computable curvature layers of the doublet-sector Hessian — the purely geometric curvature (negative: \(-1\) raw / \(-1/2\) unit-normalized ray curvature, equivalently \(-1/3\) physical mass \(^2\) under the fixed-volume projection) and the bosonic spectral-zeta layer ( \(\zeta_{K_6}(-1)=-8033/100800\) , negative) — are exact, reproduced by independent routes, and REDUCED-TO-FLOOR at +0 ; and (iv) the one remaining ingredient needed to convert "critical point with two known, negative curvature layers" into a fully certified minimum-versus-saddle verdict — the fermionic graded-Casimir sign — is proven, by a complete and target-blind four-discriminator exhaustion over a closed six-candidate space, to be unpinnable by any invariant internal to the frozen shape: a certified terminal absence, not a stalled computation. It is precisely this typed, honestly-fenced residual, together with the two closed +0 curvature legs and the free criticality theorem, that supports the RESOLVED reading of the gate.

 The explicit non-claims. This dossier does not claim that global moduli stabilization has been achieved: off-chamber configurations \(\vec u\notin[1/2,3/2]^3\) are rejected by an admissibility rule in the \(\oplus\) -Rulebook layer, not dynamically stabilized by any potential well, and no all-loop positive-definite Hessian and no "no flat directions anywhere" statement is asserted anywhere in this gate — full stabilization of every modulus is a shared-open, textbook-hard problem afflicting every extra-dimensional program without exception (string flux compactifications, M/F-theory, noncommutative geometry, lattice constructions), and the honest verdict on that shared question is TIE/SHARED-OPEN, not a claimed the framework advantage. It does not claim the positive-definite Hessian — a genuine minimum, as opposed to a saddle — is bare-proven: the phrasings "minimum proven," "no-tachyon certified," and "SG-6 closed" are each individually forbidden while the fermion sign is unpaid, and none of them is used in this dossier. It does not claim to derive the electroweak hierarchy: \(v_{\rm EW}=246\) GeV is consumed as a measured second dimensionful ruler, in the same anchor class as the cosmological constant \(\Lambda\approx10^{-122}M_{\rm Pl}^4\) and the baryon asymmetry \(\eta_B\) , and the Hosotani phase \(\theta_H^\star\approx2.46\times10^{-14}\) , which carries roughly 85% of that hierarchy via \(v_{\rm EW}=\theta_H^\star/(2\pi R_\gamma)\) , is currently read off the location of the one-loop Hosotani-potential minimum rather than derived target-blind — anchoring the operational-cell scale \(\mu_{\rm cell}\) at the same stationarity condition \(\partial_\sigma V=0\) that also fixes the observable it is meant to predict is circular by construction (the \(\kappa^3/\pi\) kill-test), and this dossier keeps that leg open under its own label (R1) rather than folding it into SG-6's stability verdict. It does not claim that the witness data (Weyl-rigid chamber, modular fixed point \(\tau=\omega\) , integer Wilson-line winding \(n_H=1\) , discrete \(\mathbb Z_2/\mathbb Z_6\) topology) determine the frozen geometry \(E\) ; SG-6 certifies moduli-control within an already-selected geometry, and given- \(E\) is not the same as a derivation of \(E\) — that question is relocated to SG-1/SG-3 (residual R4), not answered here. And it does not treat any content hash or audit-branch identifier as physics validation; such identifiers exist only to fix which computed object was tested and carry no evidential weight of their own.

 The honest current grade, stated plainly. The fixed grade for SG-6 is DERIVED-GIVEN-anchor / RESOLVED, +0 , and this dossier does not alter that grade in either direction. In the interest of full transparency, two other frozen internal readings of the same underlying facts exist and are recorded rather than smoothed over: a per-gate anchor closure ledger reads the stability leg as OPEN (stability-sign undecided), and an independent completion-run builder/referee pass reads the gate as OPEN-BLOCKED-ON-ANCHOR (typed, non-bare), on the shared ground that a bare "minimum proven" cannot be asserted while the fermion sign is unpaid. All three sources — canonical board, ledger, and referee — agree completely on the underlying facts: the criticality leg is DERIVED-GIVEN- \(E\) and terminal; both curvature layers (geometric and bosonic-zeta) are REDUCED-TO-FLOOR; the fermion sign is a genuine, currently-unpaid, anchor-payable bit. They differ only on the roll-up label attached to that identical fact pattern. The canonical board reads the combination of a symmetry-forced resting point, two independently negative and cross-checked curvature layers, and a proven -unpinnable (hence honestly typed and fenced, never silently dropped) residual as sufficient for a RESOLVED +0 closure; the ledger and the referee withhold RESOLVED specifically because a bare "minimum proven" is not yet licensed while that bit is unpaid. This dossier writes at the fixed RESOLVED grade and does so honestly by keeping the two readings visibly distinct rather than collapsing them: RESOLVED here means the criticality theorem and the two computable curvature layers are genuinely closed at +0 , and the single remaining fermion-sign bit is carried as a confident, falsifiable, named, anchor-payable wager — a limit on what any framework's internal invariants can determine, not a framework-specific gap — rather than folded silently into a "proven minimum" claim that the corpus itself forbids.

 What this dossier establishes, and what it does not. This dossier establishes, with the full derivations shown rather than gestured at: that the chamber-center point is a critical point by group theory alone, with no fitting, forced by the exact \(S_3\) symmetry of \(K_6=SU(3)/T^2\) ; that the Hessian splits exactly into a decoupled breathing singlet and shape doublet by that same group theory; that both independently computable curvature layers of the doublet sector carry a definite, cross-checked, negative sign (geometric ray curvature and its fixed-volume projection; bosonic spectral zeta \(\zeta_{K_6}(-1)=-8033/100800\) ), each reproduced multiple ways and validated against a genuine sign-flipping negative control; and that the one remaining ingredient — the fermionic graded-Casimir sign on the \(T_0/T_1\) charge-conjugation pair — is not merely unknown today but is proven , via an exhaustive, closed, target-blind four-discriminator case analysis over the complete six-candidate twist space, to be unpinnable by any invariant internal to the frozen shape, and is accordingly classified as a typed, falsifiable, anchor-certified bet payable by exactly one class of future measurement. This dossier does not establish that the vacuum is a certified minimum in the sense of a positive-definite all-orders Hessian — the geometric/tree curvature of the shape doublet is, if anything, a saddle at the layers that are computable today, and it is only the proven unpinnability of the third, fermionic layer that keeps the verdict formally undecided rather than formally negative. It does not establish global stabilization outside the admissibility chamber; it does not establish the electroweak hierarchy (that value and the Hosotani-minimum location that carries most of it are both consumed or read off, not derived, here); and it does not establish that the frozen geometry \(E\) itself is uniquely forced — that question sits upstream, at SG-1/SG-3.

 Single-sentence endpoint preview. The gate closes at +0 on a symmetry-forced critical point and two exactly-signed, cross-checked, negative curvature layers, with the sole remaining ingredient — a fermionic charge-conjugation-odd sign — carried not as a gap but as a proven-unpinnable, anchor-certified bet (AC-GAP10-HOLE3-v1) that pays out the moment one admissible charge-conjugation-odd measurement, most plausibly the leptonic \(\delta_{CP}\) sign, is in hand.

 The community gap & state of the art

 The open problem, stated the way the wider field states it. Every compactified theory of quantum gravity that reduces a higher-dimensional geometry to four observed dimensions inherits a set of massless or nearly-massless scalar fields — the moduli — that parametrize the size and shape of the compact directions: overall volume, relative squashing between factors, complex-structure and Kähler-class deformations, Wilson-line holonomies, and (in string-theoretic settings) the dilaton. Because nothing in the classical geometry singles out one value of these fields over any nearby value, the low-energy 4D effective potential \(V(\text{moduli})\) is, prior to any stabilizing mechanism, flat or nearly flat along these directions. This is the moduli problem in its bare form, and it splits into exactly the two logically independent questions this gate is built around: (1) where , if anywhere, does \(\nabla V=0\) land, and is that location forced by structure or simply assumed; and (2) what sign does the Hessian carry there — a stable minimum (all moduli acquire positive mass²) or a saddle/maximum (at least one direction runs away). A theory that cannot answer both questions is not merely incomplete in a technical sense; it fails observationally on contact, because massless scalars coupled with gravitational strength to matter mediate long-range fifth forces and time-dependent "constants" (Newton's constant, gauge couplings, Yukawa couplings) that are excluded to high precision by torsion-balance and equivalence-principle experiments, by solar-system tests of General Relativity, and by the observed near-constancy of \(\alpha\) , \(\mu=m_p/m_e\) , and similar ratios over cosmological time. A compactification with an unstabilized flat direction is, in the strict sense, not yet a prediction of a 4D world — it is a family of 4D worlds, one per point on the flat direction, and the theory has not said which one is ours.

 Why this is a first-tier, not a peripheral, problem. The moduli-stabilization problem is one of the oldest and most stubborn technical obstacles in higher-dimensional unification, predating even the modern string-flux program: it appears already in 1980s Kaluza–Klein supergravity, where the "radion" — the modulus controlling overall compactification volume — was known from the earliest papers to be a flat direction of the classical Einstein–Hilbert action reduced on a compact factor, stabilized only by quantum or flux corrections that had to be added by hand. It reappears, in a more structured but not more solved form, in every subsequent compactification framework: heterotic and Type II string compactifications on Calabi–Yau threefolds and orbifolds, M-theory on \(G_2\) manifolds, F-theory on elliptically-fibered Calabi–Yau fourfolds, and orbifold/interval Kaluza–Klein constructions of the kind used here. In each of these programs the question "does the vacuum sit at a genuine, computable minimum, or merely at a point where one has posited it does" is the single most common place where an otherwise elegant compactification is later shown to depend on an unproven or scheme-dependent ingredient.

 The state of the art, and precisely why each strand falls short of a full derivation. 

 Flux compactifications (KKLT and its descendants). The best-known constructive answer to the stabilization problem is the Kachru–Kallosh–Linde–Trivedi (KKLT) mechanism and its generalizations: turn on quantized background fluxes through the compact cycles to generate a superpotential that stabilizes the complex-structure moduli and the dilaton at tree level, then add non-perturbative effects (gaugino condensation on D7-branes, or Euclidean D3-instantons) to stabilize the remaining Kähler modulus, and finally introduce an anti-D3-brane (or an equivalent uplift) to lift the resulting supersymmetric AdS minimum to a metastable de Sitter vacuum. This mechanism is a genuine existence proof that stabilized, positive-cosmological-constant vacua are constructible in principle. It falls short of a derivation in the sense this gate requires for at least three structural reasons widely discussed in the literature: the flux quanta themselves are discrete choices selected from what is usually called the "landscape" — a combinatorially vast set of consistent flux vacua, commonly quoted at \(10^{500}\) or more — so the stabilized point is a minimum, not the forced minimum of a symmetry argument; the non-perturbative superpotential terms are exponentially sensitive to moduli-dependent one-loop determinants that are themselves not fully computed from first principles in a generic compactification; and the anti-brane uplift, needed to reach a positive-energy vacuum at all, has been the subject of a long and still not fully settled controversy over whether it is a genuinely metastable, calculationally controlled state or an artifact of the supergravity approximation (the backreaction / brane-flux-annihilation debate). None of this is a criticism unique to KKLT — it is the generic shape of the flux-landscape answer to stabilization: existence by construction and by scan, not derivation from a symmetry that forces one answer.

 Large-volume and other perturbative stabilization scenarios. A second major strand (large-volume scenarios and related constructions) stabilizes some Kähler moduli using perturbative \(\alpha'\) -corrections rather than non-perturbative effects, trading the KKLT program's reliance on exponentially small instanton effects for a different set of assumptions about which corrections dominate at which point in moduli space and about the volume hierarchy needed for the expansion to be under control. These constructions improve calculational control in specific corners but do not remove the underlying landscape character of the answer: the stabilized point still depends on a choice of flux/brane data selected for phenomenological suitability rather than forced by a symmetry of the compactification manifold itself, and the sign of the resulting Hessian in the full field space (including all moduli, not just the ones explicitly tracked) is not established as forced.

 Orbifold and Kaluza–Klein constructions without flux (Hosotani / Wilson-line and radion stabilization). Closer in spirit to the present construction — no fluxes, a Wilson-line/Hosotani mechanism supplying electroweak symmetry breaking, and the compactification radii sourced by a Casimir-type one-loop potential rather than tree-level flux superpotentials — is the older Kaluza–Klein and gauge-Higgs-unification literature (the Hosotani mechanism, and radion-stabilization schemes such as those pioneered by Goldberger–Wise in Randall–Sundrum-type setups). Here the radion potential is typically generated either by a bulk scalar field with boundary-localized potentials tuned to produce a minimum at a phenomenologically desired radius, or by the one-loop Casimir energy of bulk fields circulating around the compact directions. Both routes share a known weakness relevant here: the location of the minimum is controlled by continuous input parameters (bulk mass, boundary potential coefficients, or the matter content and its couplings entering the Casimir sum) that are typically chosen, not derived from a symmetry of the internal manifold, so stabilization is achieved but not forced , and the sign of the resulting mass² again has to be checked case by case rather than following from group theory.

 Anthropic and statistical answers. A fourth strand common in the community, particularly for the sign and value of the resulting 4D cosmological constant once moduli are stabilized, appeals to anthropic selection over the flux landscape — the vacuum we observe is one of a very large number of metastable vacua, selected post hoc by the requirement that it support observers — rather than to any dynamical uniqueness argument. Whatever its merits as a resolution of the cosmological-constant problem specifically, it is explicitly not an answer to the moduli-location or moduli-sign questions this gate asks: it presupposes that some mechanism stabilizes the moduli at some point, and only then asks why that point has small \(\Lambda\) ; it supplies no criterion for the resting point of the shape moduli themselves.

 Homogeneous-space (Wang–Ziller) mathematics without a physical stability verdict. On the purely geometric side, the classification of invariant Einstein metrics on generalized flag manifolds such as \(SU(3)/T^2\) — the compact factor used here — is a mature, decades-old result in Riemannian geometry (Wang–Ziller and collaborators on homogeneous Einstein metrics on flag manifolds): the space \(SU(3)/T^2\) admits exactly four invariant Einstein metrics — the normal (bi-invariant, symmetric-chamber-center) metric \((1,1,1)\) together with the Kähler–Einstein metric \((1,1,2)\) and its three permutations under relabeling the root directions. What the mathematics literature establishes is the existence and classification of these critical points of the Einstein–Hilbert functional restricted to the space of invariant metrics; it does not, by itself, answer the physically relevant question of which of these critical points is dynamically selected by the full 4D effective potential once matter content, gauge bundles, and one-loop quantum corrections are included, nor does the pure-geometry literature carry a verdict on the mixed bosonic-plus-fermionic one-loop Hessian sign that determines whether a given critical point is a true minimum once the full field content circulates in the loop. That translation — from "critical point of a mathematical functional" to "sign of the physical one-loop mass² Hessian including matter" — is exactly the gap this gate closes as far as it can be closed, and exactly where the residual described below still lives.

 Why "criticality" and "stability sign" are so often run together in the literature, and why this dossier refuses to. A recurring pattern across nearly all of the strands above is that a construction which achieves criticality — some mechanism forces or arranges \(\partial V=0\) — is then reported, sometimes without full separation of the two questions, as having achieved "stabilization" tout court, folding the sign question into the location question. Because a flat or nearly-flat potential can be bent into a local minimum by tuning free coefficients in the stabilizing sector (brane-localized potentials, flux quanta, boundary conditions), it is generically possible to arrange a minimum once one is willing to tune; the open scientific question, in every one of these programs, is whether the resting location and the sign of curvature there are forced by structure fixed before the phenomenological target was known, or whether they were arranged to match it. This is precisely the target-blindness standard against which this gate must be judged, and it is why the present dossier treats "is \(\nabla V=0\) forced" (sub-wall W-crit) and "is the Hessian sign forced, and if so which sign" (sub-wall W-stab) as two logically separate questions rather than a single stabilization verdict, and why it refuses the community's common shorthand of reporting a forced critical point as though it had also settled the stability sign.

 What "solving" this problem would look like, and how far the field is from it. A complete solution to the moduli problem, by the standard implicit in the literature surveyed above, requires: (a) a resting location forced by a structural symmetry or selection rule fixed independently of any phenomenological target; (b) a Hessian at that location that is positive-definite in every modulus direction, computed at full quantum order — not merely tree level, where cancellations and radiative corrections are well known to be capable of flipping signs; and (c) a demonstration that this positive-definite result extends beyond the immediate neighborhood of the candidate vacuum, ruling out runaway directions at parametrically large or small values of the moduli — global, not merely local, stability. No compactification program surveyed above — flux-based, large-volume, Hosotani/radion, or the present construction — has achieved all three simultaneously without an admitted external input. Flux/landscape constructions achieve (b) and a version of (c) at the cost of failing (a) (the flux choice is not symmetry-forced); Hosotani/radion constructions with tuned bulk potentials achieve a version of (a) and (b) locally at the cost of introducing continuous tuned parameters; and global stability (c) in the strict all-loop, all-direction sense is, as stated plainly in this dossier's own non-claims, not achieved by any program in this list, including the present one — it is a shared-open problem across the entire field, not a deficiency specific to this construction.

 Where this construction sits relative to that state of the art, stated at the fixed grade. Measured against the criteria above, the present construction is unusual on criterion (a): the resting point \(\bar u=(1,1,1)\) on the \(SU(3)/T^2=K_6\) moduli space, inside the Weyl-rigid admissibility chamber \(\vec u\in[1/2,3/2]^3\) , is forced by the exact order-6 Weyl group \(S_3\) of the \(A_2=\mathfrak{su}(3)\) root system permuting the three chamber coordinates \((u_1,u_2,u_3)\) , with no potential coefficient tuned to produce that answer — any \(S_3\) -invariant potential automatically has \(\nabla V=0\) at the symmetric point, a structural fact independent of the detailed profile of \(V\) , in the same sense that a three-fold-symmetric bowl has zero net force on a ball placed at its center regardless of the exact curvature of the bowl's walls. This closes sub-wall W-crit completely and is a genuine departure from constructions where the resting point is a free choice among many flux vacua or a consequence of tuned boundary data.

 On criterion (b) — the Hessian sign, at full one-loop order including fermionic content — this construction has carried the computation further than a bare existence claim. The doublet-sector Hessian at the Einstein center, in the Killing-form normalization \(g=(-B)|_{\mathfrak m}\) with \(B(X,Y)=6\,\mathrm{Tr}(XY)\) , decomposes into exactly the \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) split forced by the same \(S_3\) representation theory (breathing singlet \(u_1=u_2=u_3\) , traceless shape doublet \(u_1-u_2\) and its partner), and two of the three multiplicative layers that combine into the physical one-loop Hessian are computed exactly, by more than one independent method, and both carry a definite, cross-checked sign. The pure geometric curvature of the doublet ray \(u=(1+\varepsilon,1-\varepsilon,1)\) gives \(R(\varepsilon)=3/2-\varepsilon^2/2\) , i.e. \(d^2R/d\varepsilon^2|_0=-1\) raw ( \(-1/2\) unit-normalized), reproduced independently by the full \(3\times3\) Hessian of this \(R\) -model at \((1,1,1)\) , \(\begin{pmatrix}-1&2&2\\2&-1&2\\2&2&-1\end{pmatrix}\) , with eigenvalues \(\{3\text{ (}\times1\text{)},-3\text{ (}\times2\text{)}\}\) and \(R(1,1,1)=3/2\) ; the same negative sign reappears a third way at strictly fixed volume, where the volume-projected shape functional \(S=\Sigma_i\,1/x_i-\tfrac16\Sigma_{i}\,x_i/(x_jx_k)\) has \(d^2S/d\varepsilon^2|_{\rm fixed\,vol}=+2/3\) , whose log-coordinate Hessian eigenvalues \(\{1/3,1/3,0\}\) become physical mass \(^2=-1/3\) once the standard flux-free Kaluza–Klein volume prefactor \(V\sim -S|_{\rm fixed\ vol}\) is applied — three parametrizations of the identical saddle, not three different answers, with an \(S^2\times S^2\) specificity control firing the opposite (positive) sign as a check that the sign test is not rigged to always return the same verdict. In parallel, the bosonic spectral-zeta layer is pinned by the exact rational \(\zeta_{K_6}(-1)=-8033/100800=-0.0796924603174603\ldots\) (with \(8033=29\cdot277\) and \(100800=2^6\cdot3^2\cdot5^2\cdot7\) coprime), also negative, confirmed by two independent routes — a blind numeric heat-sum with Richardson extrapolation, and the exact Weyl/Poisson closed form — agreeing to roughly 22 digits, together with the companion exact values \(\zeta_{K_6}(0)=-253/315\) and the heat-kernel coefficients \(c_1=8033/100800\) , \(c_2=5743/184800\) feeding \(\zeta_{K_6}(-2)=5743/92400\) .

 What remains — a single discrete charge-conjugation-odd sign in the fermionic graded-Casimir supertrace on the twisted spinor pair \(T_0\leftrightarrow T_1\) ( \(E\leftrightarrow E^*\) ) — is not, in this construction, an unperformed calculation sitting in a backlog the way an uncomputed non-perturbative superpotential term is in a flux compactification; it is shown, by a complete and target-blind exhaustion over every intrinsic discriminator available on the frozen shape, to be a bit that no invariant of the shape alone can supply. The exhaustion covers the full set of topological, reality-type, and anomaly-type discriminators used across the wider compactification literature for exactly this kind of question — Wu/Stiefel–Whitney/second- and third-class orientation data, the Frobenius–Schur reality indicator, the Dai–Freed eta-invariant beyond its mod-2 refinement, and the spin- \(\mathbb C\) determinant-line reality condition — applied to the complete, closed, six-member candidate space of \(\chi=-3\) Dynkin-label twist candidates (closed under conjugation and triality); every one of these discriminators is exactly charge-conjugation-even on any Shape-only structure (built from the tangent bundle \(TK_6\) , on which charge conjugation acts trivially), while the sought bit is by construction charge-conjugation-odd — a mismatch of representation type, not a computational shortfall, and therefore a proven absence of a discriminator rather than a search that ran out of time.

 On criterion (c), global stability, this construction makes no claim beyond the shared-open status of the entire field surveyed above: off-chamber configurations \(\vec u\notin[1/2,3/2]^3\) are excluded by an admissibility rule at the level of the rulebook that defines which configurations are on the table at all, not by a dynamically computed potential wall, and no positive-definite Hessian is claimed at every point of moduli space or asymptotically along every possible direction. In that specific and disclosed sense, this construction does not solve the moduli problem where the rest of the field has not; it isolates the location question (W-crit) as fully closed by symmetry, isolates two of the three ingredients of the sign question (geometric curvature and bosonic zeta) as fully closed and exactly signed by independent multi-route computation, and narrows the third ingredient (the fermionic supertrace sign) from an open search to a single named, payable, falsifiable bit — with the standing candidate payer being a measured leptonic CP-violating phase sign \(\delta_{CP}\) rather than any further internal computation, and with the baryon asymmetry \(\eta_B\) excluded in advance as an illegitimate payer by an independent argument. That is a materially sharper statement than "a minimum exists for suitable choices of flux or boundary data," which is the most that most of the constructions surveyed above can offer without further external tuning, but it is honestly short of "global minimum, all orders, all directions, certified" — a bar that, as surveyed above, no compactification program in the literature — flux-based, large-volume, or Kaluza–Klein/Hosotani-based — has cleared either.

 The frozen 13D arena at full precision

 1. The complete active branch, and where SG-6 sits inside it

 Every number quoted anywhere in this dossier is a statement about one fixed, submitted, three-layer
object — never a convenient slice of it. That object is the active branch 
$$
\mathfrak{B} {\rm active}
=
\underbrace{\big[\,\mathcal{M}_4 \times K_6 \times S^2 \times S^1_Y/\mathbb{Z}_2\,\big] \times} {\times\ \text{STAGE — metric geometry}}
\;\oplus\;
\underbrace{\big[\,\mathcal{F}^+ {\rm finite}\oplus\mathcal{C} {\rm admiss}\,\big] \oplus} {\oplus\ \text{RULEBOOK — finite admissibility (0-dim)}}
\;\otimes\;
\underbrace{\big[\,\mathcal{E} {\rm matter}\oplus\mathcal{E} {\rm gauge}\oplus\mathcal{E} {\rm Higgs}\oplus\mathcal{E} {\rm proton}\,\big] \otimes}_{\otimes\ \text{ACTORS — bundles/operators (0-dim)}},
$$

 with \(K_6 = SU(3)/T^2\) the full flag manifold of \(A_2\) and \(S^1_Y/\mathbb{Z}_2\) the active orbifold
boundary domain. Only the \(\times\) -Stage factor carries metric dimension:

 \[
D = \dim\mathcal{M}_4+\dim K_6+\dim S^2+\dim S^1_Y = 4+6+2+1 = 13.
\]

 The \(\oplus\) -Rulebook and \(\otimes\) -Actors layers are non-metric (zero-dimensional as manifolds) but
are permanently part of the frozen branch — they can never be silently dropped without turning a
complete object into a truncated artifact. \(\mathcal{F}^+_{\rm finite}\) in particular is a finite
operator chamber , not a propagating metric factor: its Cartan-torus modulus \(\tau\) is chamber data,
not a Kaluza–Klein tower, so it contributes 0 to \(D\) even though it is load-bearing for SG-6's
modular-fixed-point leg.

 SG-6 asks a single physical question of this frozen object: does the resting configuration of the
internal shape sit at a genuine minimum of the effective potential, or does it slide away? That
question is not a \(\times\) -only question about a metric. It is a full \(\times+\oplus+\otimes\) 
question: the location of the resting point is a \(\times+\oplus\) statement (a metric fact protected
by a Rulebook symmetry, §4 below); the stability verdict at that point is a \(\times+\oplus+\otimes\) 
statement (a metric curvature fact, filtered through a bosonic spectral operator and a fermionic
graded-Casimir operator built on the Actors layer). Every quantity below is pinned at the exact point
the gate tests: the Weyl-rigid chamber center \(\vec u=(1,1,1)\) on \(K_6\) .

 2. Dimension ledger and the four irreducible anchors

 Factor 
 Real dim 
 Metric 
 Primitive/derived 
 Physical role 
 Force routed 

 \(\mathcal{M}_4=\mathbb{R}^{3,1}\) 
 4 
 Minkowski 
 primitive 
 observed spacetime 
 — 

 \(K_6=SU(3)/T^2\) 
 6 
 Weyl-rigid invariant (normal at center) 
 primitive 
 color source; spin- \(\mathbb{C}\) family index \(-3\) ; the SG-6 moduli space 
 \(SU(3)_c\) 

 \(S^2\) 
 2 
 round 
 primitive 
 weak source 
 \(SU(2)_L\) 

 \(S^1_Y/\mathbb{Z}_2\) 
 1 (interval) 
 flat, orbifolded 
 derived quotient 
 hypercharge circle / chirality filter 
 \(U(1)_Y\) 

 \(\mathcal{F}^+_{\rm finite}\) 
 0 
 — 
 finite chamber 
 modular fixed point \(\tau=\omega\) that SG-6's item 9 pins 
 flavor phases 

 \(D=4+6+2+1=13\) . The only free inputs anywhere in the frozen branch are the four irreducible anchors
 \(\{M_{\rm Pl},\ \alpha_i(M_Z),\ y_t,\ |V_{us}|\}\) ; every radius, volume, curvature invariant, Casimir,
and chamber operator used below is derived from these (or is exact-topological), never separately
fit. SG-6 consumes none of the gauge/Yukawa anchors directly — its only external input is the
 second ruler \(v_{\rm EW}=246\) GeV (§7), which is consumed, not produced, exactly as \(\Lambda\) and
 \(\eta_B\) are consumed elsewhere in the corpus.

 3. Two metric normalizations — both used below, and the bridge between them

 The corpus pins the same \(K_6\) geometry in two internally consistent normalizations, and SG-6's
numbers are quoted in both, so both must be stated before any curvature number is trusted.

 (A) Frozen physical ( \(R_6\) ) normalization. The internal radius is the derived compactification
 radius \(R_6=R_0=(2\pi M_U)^{-1}\) at the chamber center. Curvature carries physical units of GeV \(^2\) :
 \(\mathrm{Ric}_i=1/(2R_6^2)\) , \(\mathrm{Scal}=3/R_6^2\) . This is the normalization for every
 dimensionful downstream quantity.

 (B) Killing-form normal metric. \(g=(-B)|_{\mathfrak m}\) with Killing form \(B(X,Y)=6\,\mathrm{Tr}(XY)\) 
 on \(\mathfrak{su}(3)\) , evaluated at the symmetric chamber center \(\vec u=(1,1,1)\) . Curvature is
 dimensionless: \(\mathrm{Ric}_i=5/12\) , \(\mathrm{Scal}=5/2\) . This is the normalization in which every
 exact-rational curvature invariant used by SG-6 — \(|\mathrm{Riem}|^2\) , the Hessian eigenvalues,
 \(\zeta_{K_6}(-1)\) — is actually computed and stored.

 The bridge. Ratios of curvature invariants are metric-scale invariant and therefore identical
 in (A) and (B). The load-bearing one is \(\mathrm{Scal}/\mathrm{Ric}_i=6=\dim K_6\) in both 
 normalizations: \((3\cdot R_6^{-2})/(\tfrac12 R_6^{-2})=6\) in (A); \((5/2)/(5/12)=6\) in (B). Likewise
 \(|\mathrm{Ric}|^2/\mathrm{Scal}^2=1/6\) and \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) agree in both.
 Because SG-6's own headline number (the sign of a Hessian eigenvalue) is itself a curvature ratio 
 under rescaling of \(\vec u\) , it inherits this scale-invariance: the saddle sign is a property of the
 shape, not of which normalization is used to read it off — which is exactly why the "raw" and
 "unit-normalized" doublet numbers below ( \(-1\) vs \(-1/2\) ) are the same physical fact in two
 conventions, not two different facts.

 4. The exact radii and volumes at the chamber center

 The compactification/unification scale closes the two-loop RG with KK thresholds at
 \(\alpha_1(M_U)=\alpha_2(M_U)=\alpha_3(M_U)\) , residual \(9.6\times10^{-11}\) , giving \(M_U=1.0\times10^{16}\) GeV
and

 \[
R_0=(2\pi M_U)^{-1}=1.591549430918954\times10^{-17}\ \mathrm{GeV}^{-1}.
\]

 At the chamber center \(u_{\rm chamber}=1\) : \(R_6=R_2=R_0=1.591549430918954\times10^{-17}\ \mathrm{GeV}^{-1}\) ,
and the hypercharge circle after the \(\mathbb{Z}_2\) orbifold halving is
 \(R_Y=7.957747154594768\times10^{-18}\ \mathrm{GeV}^{-1}\) . SG-6's moduli \(\vec u=(u_1,u_2,u_3)\) live in
the Weyl-rigid admissibility chamber \(\vec u\in[1/2,3/2]^3\) ; the center witness is
 \(u_1=u_2=u_3=1.000000000000000\) , and it is precisely the second derivatives of the effective
potential around this point , as a function of \(\vec u\) , that SG-6 computes.

 The \(K_6\) volume as a function of the moduli,

 \[
\mathrm{Vol}(K_6)(\vec u)=V_{K_6,0}\,R_6^6\sqrt{u_1u_2u_3},\qquad V_{K_6,0}=\frac{(2\pi)^3}{\sqrt3}=143.2118575035129,
\]

 evaluates at the center to \(\mathrm{Vol}(K_6)=V_{K_6,0}R_0^6=2.327554010848277\times10^{-99}\ \mathrm{GeV}^{-6}\) ,
and enters the full active volume

 \[
\mathrm{Vol}(X_{\rm active})=\mathrm{Vol}(K_6)\,\mathrm{Vol}(S^2)\,\mathrm{Vol}(S^1_Y/\mathbb{Z}_2)
=3.704417261398702\times10^{-148}\ \mathrm{GeV}^{-9}
\]

 with \(\mathrm{Vol}(S^2)=4\pi R_0^2=3.183098861837907\times10^{-33}\ \mathrm{GeV}^{-2}\) and
 \(\mathrm{Vol}(S^1_Y/\mathbb{Z}_2)=\pi R_0=5.000000000000000\times10^{-17}\ \mathrm{GeV}^{-1}\) (exact
 \(=1/(2M_U)\) ). The \(\sqrt{u_1u_2u_3}\) dependence is the breathing mode (overall volume) — SG-6's
Hessian split (§5) isolates this from the volume-preserving shape directions, and \(\mathrm{Vol}(X_{\rm active})\) 
is what feeds the Planck normalization \(M_{\rm Pl}^2=M_*^{11}\mathrm{Vol}(X_{\rm active})\) ,
 \(M_*=7.467050992135091\times10^{16}\) GeV — the anchor that ties the size of this whole arena to the
one dimensionful input \(M_{\rm Pl}\) , entirely independent of the shape question SG-6 asks.

 5. \(K_6=SU(3)/T^2\) : roots, tangent split, and the invariant metric SG-6 perturbs

 \(K_6\) is the full flag manifold of \(A_2=\mathfrak{su}(3)\) . In the Cartan basis \((h_1,h_2,h_3)\) ,
 \(h_1+h_2+h_3=0\) , the simple roots are \(\alpha_1=(1,-1,0)\) , \(\alpha_2=(0,1,-1)\) , with
 \(\alpha_1+\alpha_2=(1,0,-1)\) ; the positive roots are \(\{\alpha_1,\alpha_2,\alpha_1+\alpha_2\}\) , half-sum
 \(\rho=\tfrac12\sum_{\alpha>0}\alpha=(1,0,-1)\) , \(\|\rho\|^2=2\) (Killing normalization) — this \(\|\rho\|^2=2\) 
is the same shift that appears in the Dirac KK mass formula \(m^2_{(p,q),\rm Dirac}=(C_2(p,q)+\|\rho\|^2+\Delta_{\rm spin^c})/R_6^2\) 
elsewhere in the arena, so it is not a number invented for SG-6 but the fixed half-sum-of-positive-roots
of the same frozen root system. The Weyl group is \(S_3\) , order 6 — this is the exact symmetry group
that forces SG-6's criticality theorem (§6).

 The tangent space splits into three real 2-planes, one per positive root:

 \[
T(K_6)=\mathfrak{m}_1\oplus\mathfrak{m}_2\oplus\mathfrak{m}_3,\qquad \dim_{\mathbb R}\mathfrak m_i=2,
\]

 with \((-B)\) -orthonormal basis \(\{X_{ij}=E_{ij}-E_{ji},\,Y_{ij}=i(E_{ij}+E_{ji})\}/\sqrt{12}\) over the
pairs \((01),(12),(02)\) , Killing form \(B=6\,\mathrm{Tr}\) . SG-6's three moduli \(u_1,u_2,u_3\) are exactly
the independent scale factors on these three 2-planes in the Wang–Ziller/Nomizu invariant metric

 \[
g_{K_6}(\vec u)=u_1\langle\cdot,\cdot\rangle_{\mathfrak m_1}+u_2\langle\cdot,\cdot\rangle_{\mathfrak m_2}+u_3\langle\cdot,\cdot\rangle_{\mathfrak m_3}.
\]

 The general-chamber Ricci eigenvalues on scales \((x_1,x_2,x_3)\) (Killing norm) are

 \[
\mathrm{Ric}_1=\frac{x_1^2-x_2^2+6x_2x_3-x_3^2}{12\,x_1x_2x_3},\quad
\mathrm{Ric}_2=\frac{-x_1^2+6x_1x_3+x_2^2-x_3^2}{12\,x_1x_2x_3},\quad
\mathrm{Ric}_3=\frac{-x_1^2+6x_1x_2-x_2^2+x_3^2}{12\,x_1x_2x_3},
\]

 \[
\mathrm{Scal}=\frac{x_1x_2+x_1x_3+x_2x_3-(x_1^2+x_2^2+x_3^2)/6}{x_1x_2x_3}.
\]

 There are exactly four invariant Einstein metrics on \(SU(3)/T^2\) : the normal metric \((1,1,1)\) plus
the Kähler–Einstein metric \((1,1,2)\) and its three permutations — a classical fact, reproduced
independently in this arena as a validation of the computational engine. Off-center the space is
non-Einstein; this squashing is precisely the input that SG-6's stability question probes. At the
center all three Ricci eigenvalues coincide, giving the two normalizations already quoted:
 \(\mathrm{Ric}_1=\mathrm{Ric}_2=\mathrm{Ric}_3=1/(2R_6^2)=1.973920880217872\times10^{33}\ \mathrm{GeV}^2\) 
[ \(R_6\) -norm] \(=5/12\) [Killing-norm], and \(\mathrm{Scal}(K_6)=3/R_6^2=1.184352528130723\times10^{34}\ \mathrm{GeV}^2\) 
[ \(R_6\) -norm] \(=5/2\) [Killing-norm].

 6. Curvature invariants at the Einstein center — the exact rationals SG-6's Hessian is built from

 The metric-scale-invariant ratios (identical in both normalizations, load-bearing for every SG-6
curvature statement):

 Invariant 
 Exact rational 
 Decimal 

 \(\mathrm{Scal}^2\) 
 \(25/4\) 
 \(6.25\) 

 \(\|\mathrm{Ric}\|^2\) 
 \(25/24\) 
 \(1.041666666666667\) 

 \(\|\mathrm{Riem}\|^2\) 
 \(23/12\) 
 \(1.916666666666667\) 

 \(\|\mathrm{Riem}\|^2/\mathrm{Scal}^2\) 
 \(23/75\) 
 \(0.3066666666666667\) 

 \(\|\mathrm{Ric}\|^2/\mathrm{Scal}^2\) 
 \(1/6\) 
 \(0.1666666666666667\) 

 Anti-drift certification, binding for this gate: \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) is
confirmed; it is never \(31/147\) (a retired branch-kill contaminant that must not reappear in any
SG-6 curvature number), and \(|\mathrm{Riem}|^2\) is never \(=60\) (the round unit \(S^6\) value, a
different space entirely — the \(S^6\) heat-kernel row exists only as an external calibration, §9).
Euler characteristic \(\chi(K_6)=6\) exactly (topological), and \(\chi(K_6,E)=-3\) is the spin- \(\mathbb{C}\) 
twisted index that fixes the family count and reappears in SG-6's leading loop indicator (§8).

 Beyond the quadratic invariants, the cubic/weight-6 curvature data (Killing norm, Einstein center)
that back the second-derivative structure includes \(K_1=R_{ab}{}^{cd}R_{cd}{}^{ef}R_{ef}{}^{ab}=-113/72\) ,
 \(K_2=R_{abcd}R_{aecf}R_{ebfd}=-5/72\) , \(\mathrm{Scal}^3=125/8\) , \(\mathrm{Scal}\cdot\|\mathrm{Ric}\|^2=125/48\) ,
 \(\mathrm{Scal}\cdot\|\mathrm{Riem}\|^2=115/24\) , and \(\|\nabla\mathrm{Riem}\|^2=1/4\ne0\) — the last of
which certifies that \(K_6\) is homogeneous but not locally symmetric, the structural fact that lets
a nontrivial curvature Hessian (rather than a flat one) exist on the shape directions in the first
place. \(\int_{K_6}R\sqrt g\,d^6x=\mathrm{Scal}\cdot\mathrm{Vol}(K_6)=12\pi^3=372.0753201635977\) under
the Killing-form-absorbing normalization (equivalently \((2\pi)^3\sqrt3=429.6356725105388\) under the
pure- \(\sqrt g\,d^6x\) normalization at \(R_6=1\) ); both are recorded because different SG-6 scripts
report either.

 The dimensionless per-K \(_6\) -sector constant \(\kappa=1/6\) is the same object as
 \(|\mathrm{Ric}|^2/\mathrm{Scal}^2\) above — it is the coefficient that sets the DeWitt radion slope
used in the fixed-volume shape projection, \(\lambda^2_K=8/3=2(d+2)/d\) at \(d=6\) 
(geometric-forced, not fit).

 7. The second ruler \(v_{\rm EW}\) — the one external anchor SG-6 consumes

 SG-6's only measured, non-geometric input is the electroweak scale \(v_{\rm EW}=246\) GeV, entering
through the Wilson-line/Hosotani sector on the cycle \(\gamma\) of radius \(R_\gamma\sim R_0\) . The
winding number is exact and topological,

 \[
n_H=\frac{1}{2\pi i}\oint_\gamma A = 1,
\]

 with the Hosotani potential \(V_{\rm Hos}(\theta_H)=-\dfrac{3}{64\pi^6R_\gamma^4}\sum_{n=1}^\infty\dfrac{1}{n^5}[N_b-N_f]\cos(n\theta_H)\) ,
an absolutely convergent \(n^{-5}\) tail guaranteeing a finite Higgs mass. \(v_{\rm EW}\) is
 MEASURED-ANCHOR, consumed, not produced by SG-6 — of exactly the same anchor class as \(\Lambda\approx10^{-122}M_{\rm Pl}^4\) 
and the baryon asymmetry \(\eta_B\) elsewhere in the corpus. No pull is computed for it here; it is an
input to the moduli-stability calculation, not an output of it. This is the arena's Buckingham- \(\pi\) 
constraint at work: with only \(\{M_{\rm Pl},\hbar,\text{dimensionless frozen geometry}\}\) available,
there is no second parametrically small mass scale obtainable without a \(\sigma\) -carrying length —
so \(v_{\rm EW}\) must enter as consumed data, and any attempt to anchor the breathing-mode spectral
value \(\mu_{\rm cell}\) at \(\partial_\sigma V=0\) (which is the hierarchy condition) would be circular
by construction.

 8. The three-layer pin of the specific objects SG-6 touches

 \(\times\) Stage (metric geometry). The base object is the full \(K_6=SU(3)/T^2\) shape-moduli space
 \(\vec u\in[1/2,3/2]^3\) with its Wang–Ziller invariant metric family \(g_{K_6}(\vec u)\) , together with
the twisted Dirac/spinor bundle over \(K_6\) carrying the six \(\chi=-3\) Dynkin-label twist
candidates (the closed, enumerated finite set on which the fermionic layer, §8 below, is evaluated),
the spin- \(\mathbb{C}\) determinant line, and the exact algebraic/independent-numeric agreement
 \(\mathrm{FS}(3)=0\) (Frobenius–Schur indicator, agreeing to \(\sim10^{-16}\) between the exact algebra
and an independent Weyl-integral numeric route). This Stage object supplies: the moduli space itself,
the \(S_3\) symmetry that forces criticality, the admissibility chamber, and the \(3=1\oplus2\) 
representation-theoretic split of the Hessian (§ below in the sequence of this dossier). Any reading
that uses only a single slice of this Stage object — e.g. the bare doublet-ray curvature number in
isolation — is by construction a truncated-Shape artifact; the complete Stage object is the full
moduli space plus the full six-candidate twist bundle together.

 \(\oplus\) Rulebook (scheme/convention/boundary/projector/grading). Three Rulebook pieces are load
bearing for SG-6: (i) the chamber/admissibility rule \(\vec u\in[1/2,3/2]^3\) , Weyl-rigid, which
eliminates off-center configurations by a selector , not by a dynamical potential well — off-chamber
points are ruled out-of-bounds, not pushed back by a restoring force, and this distinction is exactly
why "global stabilization" is explicitly not claimed anywhere in this gate; (ii) the conjugation/triality
orbit action on the six-candidate twist set, which organizes the fermionic discriminator search into
a provably closed, exhaustible class; (iii) the modular fixed point \(\tau=\omega=e^{2\pi i/3}=
-0.5000000000000000+0.8660254037844386\,i\) inside \(\mathcal{F}^+_{\rm finite}\) , an order-3 fixed point
of \(PSL(2,\mathbb{Z})\) — generic existence is a proven fact, while uniqueness of this particular
realized generator on the chamber is a separate, non-blocking theorem-debt item.

 \(\otimes\) Actors (connection \(\nabla\) , endomorphism \(E\) , operator domain, readout). Three
operators are pinned here, one per physical layer of the stability question: (i) the geometric
curvature Hessian of the Ricci-scalar functional \(R(\vec u)\) restricted to the volume-preserving
shape doublet, connection = Levi-Civita/Nomizu, domain = the two-dimensional traceless subspace of
 \(\vec u\) -space, readout = the second-derivative eigenvalues; (ii) the bosonic spectral-zeta
operator built from the scalar Laplacian on \(K_6\) , endomorphism \(E=0\) , domain \(C^\infty(K_6)\) ,
readout \(\zeta_{K_6}(-1)\) ; (iii) the fermionic graded-Casimir supertrace on the twisted Dirac
operator over \(K_6\) , endomorphism built from the spin- \(\mathbb{C}\) connection, domain the six-candidate
twisted spinor bundle, readout a discrete \(\pm\) sign under the charge-conjugation exchange of the
 \(T0\leftrightarrow T1\) ( \(E\leftrightarrow E^*\) ) pair. The full physical stability verdict is the
 product of all three readouts — geometric \(\times\) bosonic-zeta \(\times\) fermionic-supertrace —
which is why the gate's residual is confined to exactly one of the three factors (§ elsewhere in this
dossier) rather than smeared across the whole calculation.

 9. Heat-kernel and Casimir data anchoring the K₆ spectral operators SG-6 uses

 The scalar heat-kernel ratios on \(K_6\) (Killing norm, Einstein center), convention
 \(K(t)\sim(4\pi t)^{-d/2}\sum_k a_{2k}t^k\) : \(a_2/a_0=5/12\) , \(a_4/a_0=11/120\) ; the \(a_6/a_0\) coefficient
is OWED (Gilkey constants not yet closed at that order — an honestly bounded computation-debt, not a
value used anywhere in SG-6's certified layers). The \(S^2\) scalar row ( \(a_2/a_0=1/3\) , \(a_4/a_0=1/15\) ,
 \(a_6/a_0=4/315\) ) and the \(S^6\) round-unit row ( \(a_2/a_0=5\) , \(a_4/a_0=12\) , \(a_6/a_0=1139/63\) ) exist
purely as calibration controls confirming \(K_6\) is not \(S^6\) — the same anti-drift discipline that
forbids \(|\mathrm{Riem}|^2=60\) above. The quadratic Casimir on Dynkin labels,
 \(C_2(p,q)=(p^2+q^2+pq+3p+3q)/3\) , gives the representation content entering the twisted-spinor
multiplicity table that SG-6's fermionic layer must ultimately weight (the still-missing
admissible-representation graded multiplicity table is the single highest-leverage owed artifact
for the net Hessian sign, tracked outside this section). The Lichnerowicz spectrum of the graviton
 \(\mathrm{Sym}^2_0\) endomorphism at the Einstein center is \(\{1/6\,(\times6),\,5/12\,(\times6),\,7/6\,(\times6),\,17/12\,(\times2)\}\) 
— quoted here because it is built from the identical \(E=\mathrm{Ric}=\tfrac5{12}\mathrm{Id}\) 
Weitzenböck data that also enters SG-6's bosonic-zeta layer.

 10. Discrete/topological structure riding on the same frozen geometry

 Three generations arise as the spin- \(\mathbb{C}\) twisted index \(\chi(K_6,E)=-3\) on the same \(K_6\) 
whose moduli SG-6 perturbs. Charge quantization closes as the global \(\mathbb{Z}_6=(\mathbb{Z}_3\times\mathbb{Z}_2\times\mathbb{Z}_6)\) -center
quotient with Smith normal form invariant factors \([1,6,6]\) — the finest faithful quotient, with no
continuous modulus and therefore no stability question of SG-6's kind attached to it at all: a
discrete gauge identification has no direction to slide in. This is recorded here because it is the
explicit example, inside the very same frozen branch, of a "moduli-like" object that is not a
stability question — sharpening, by contrast, exactly what kind of object \(\vec u\) is (continuous,
metric, and therefore subject to a genuine Hessian test).

 11. Summary: what this section certifies going forward

 Every number the rest of this SG-6 dossier uses — the criticality theorem, the \(3=1\oplus2\) Hessian
split, the doublet curvature eigenvalues, \(\zeta_{K_6}(-1)=-8033/100800\) , the fermionic
graded-Casimir exhaustion — is built from exactly the frozen objects fixed in this section: the
13-dimensional active branch \(\mathfrak B_{\rm active}\) with \(D=4+6+2+1=13\) ; the \(K_6=SU(3)/T^2\) 
root system, tangent split \(\mathfrak m_1\oplus\mathfrak m_2\oplus\mathfrak m_3\) , and Weyl group
 \(S_3\) ; the Einstein-center curvature invariants \(\mathrm{Ric}_i=5/12\) , \(\mathrm{Scal}=5/2\) ,
 \(|\mathrm{Riem}|^2=23/12\) , \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) , \(|\mathrm{Ric}|^2/\mathrm{Scal}^2=1/6=\kappa\) ;
the modular fixed point \(\tau=\omega\) ; the Wilson-line winding \(n_H=1\) ; and the consumed second
ruler \(v_{\rm EW}=246\) GeV. No quantity in this section is fit to the stability question it feeds —
every one is either an exact topological invariant, an exact rational curvature invariant of the
frozen Einstein metric, or a previously anchored scale from elsewhere in the arena.

 Construction I - the deep-root anchoring

 SG-6 asks a single physical question — does the compact internal shape rest at a genuine minimum, or does it slide off a saddle? — and the discipline of this construction is to answer it only after Shape, Scale, and Granularity have each been applied completely , at all three layers, on the full frozen thirteen-dimensional arena. A residual read off a truncated slice of that arena is an artifact of the truncation, not a fact about the physics; the entire reconciliation carried in this dossier (three apparently different numbers, \(-1\) , \(-1/2\) , \(+1/3\) , turning out to be one invariant fact in three conventions) exists precisely because an earlier single-slice reading dropped layers it should have kept. Nothing below uses a truncated object. After the three roots are pinned, the four Layer-2 admissibility screens — Invariance, Record-Interface, Causal-Order, Nonseparability — are run against the resulting complete object, and each is shown to return a definite, non-trivial verdict rather than a rubber stamp.

 I.1 Shape, applied completely (all three layers)

 The frozen active branch. SG-6 lives inside the full three-layer object

 \[
\mathfrak B_{\rm active}=\underbrace{\big[\mathcal M_4\times K_6\times S^2\times S^1_Y/\mathbb Z_2\big]_\times}_{\times\,{\rm Stage}}\ \oplus\ \underbrace{\big[\mathcal F^+_{\rm finite}\oplus\mathcal C_{\rm admiss}\big]_\oplus}_{\oplus\,{\rm Rulebook}}\ \otimes\ \underbrace{\big[\mathcal E_{\rm matter}\oplus\mathcal E_{\rm gauge}\oplus\mathcal E_{\rm Higgs}\oplus\mathcal E_{\rm proton}\big]_\otimes}_{\otimes\,{\rm Actors}},
\]

 with \(K_6=SU(3)/T^2\) the full \(A_2\) flag manifold and \(D=4+6+2+1=13\) . SG-6's business is entirely internal to the \(\times\) Stage shape-moduli sector of \(K_6\) together with the \(\oplus\) / \(\otimes\) modular and Wilson-line data riding on top of it; \(\mathcal M_4\) , \(S^2\) , and \(S^1_Y/\mathbb Z_2\) are present in the ambient arena and are held fixed while \(K_6\) 's three Killing-form scale parameters \(\vec u=(u_1,u_2,u_3)\) move.

 \(\times\) Stage. \(K_6=SU(3)/T^2\) carries an \(SU(3)\) -invariant family of metrics \(g_{K_6}(\vec u)=\sum_i u_i\langle\cdot,\cdot\rangle_{\mathfrak m_i}\) , built from the tangent decomposition \(T(K_6)=\mathfrak m_1\oplus\mathfrak m_2\oplus\mathfrak m_3\) into the three real 2-planes carrying the positive roots of \(A_2\) : \(\alpha_1=(1,-1,0)\) , \(\alpha_2=(0,1,-1)\) , \(\alpha_1+\alpha_2=(1,0,-1)\) , half-sum \(\rho=(1,0,-1)\) , \(\|\rho\|^2=2\) (Killing normalization). The admissible Weyl-rigid chamber is the box \(\vec u\in[1/2,3/2]^3\) ; the chamber-center witness is the exact point \(\vec u=(1,1,1)\) , where every curvature and Hessian quantity in this gate is evaluated. The two metric normalizations pinned in the geometry pack must never be cross-mixed at the dimensionful level: in the frozen physical ( \(R_6\) ) normalization \(\mathrm{Ric}_i=1/(2R_6^2)\) and \(\mathrm{Scal}=3/R_6^2\) carry units of \(\mathrm{GeV}^2\) ; in the dimensionless Killing-form normal metric \(g=(-B)|_{\mathfrak m}\) , \(B(X,Y)=6\,\mathrm{Tr}(XY)\) , evaluated at the chamber center, \(\mathrm{Ric}_i=5/12\) and \(\mathrm{Scal}=5/2\) . The bridge is the metric-scale-invariant ratio \(\mathrm{Scal}/\mathrm{Ric}_i=6=\dim K_6\) , identical in both normalizations, together with \(|\mathrm{Ric}|^2/\mathrm{Scal}^2=1/6\) and \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) .

 \(\oplus\) Rulebook. The admissibility chamber \(\mathcal C_{\rm admiss}\) restricts \(\vec u\) to \([1/2,3/2]^3\) ; configurations outside this window fail admissibility and are eliminated by the selector , not dynamically suppressed by a potential well. This is a rulebook-layer fact — a boundary condition on the moduli space, not a metric fact — and it is exactly the layer that carries the gate's explicit non-claim of global stabilization: off-chamber elimination is a selection rule, never a claim that some potential drives runaway configurations back into the chamber. The Killing-form scheme convention under which the exact-rational curvature invariants below are computed is also fixed at this layer.

 \(\otimes\) Actors. The operator content is the twisted spin- \(\mathbb C\) Dirac/Laplace-type structure on \(K_6\) carrying the spin- \(\mathbb C\) family index \(\chi(K_6,E)=-3\) , together with the Weitzenböck endomorphism data: \(E=\mathrm{Ric}=(5/12)\,\mathrm{Id}\) on the vector (1-form) bundle, and the Lichnerowicz endomorphism \(E_L\) on \(\mathrm{Sym}^2_0T^*K_6\) (the traceless, transverse-traceless graviton sector, dimension 20) with spectrum \(\{1/6\ (\times6),\ 5/12\ (\times6),\ 7/6\ (\times6),\ 17/12\ (\times2)\}\) . For SG-6 specifically, the load-bearing \(\otimes\) Actors object is the six \(\chi=-3\) Dynkin-label twist candidates , closed under the conjugation ( \(C\) ) and triality actions — a complete, finite, enumerated set of admissible twists consistent with the frozen family index. This is the operator-domain-and-readout layer that specifies exactly which discrete datum the stability sign reduces to, and that the datum lives in a closed six-element set rather than an open-ended or infinite-precision continuum.

 Why completeness matters here, concretely. The retired reading — " \(-1\) as the settled shape eigenvalue" — used only a single slice of Shape: the bare \(\times\) Stage doublet-ray curvature, without either the \(\oplus\) Rulebook fixed-volume projection or the \(\otimes\) Actors twist-candidate enumeration that the fermionic layer needs. That truncated reading mixed a criticality-framing computation with a \(\Lambda\) -free doublet-framing computation and is explicitly forbidden from being revived as the shape eigenvalue. The complete three-layer Shape object is what supports both closed legs below — the criticality theorem and the exact \(3=1\oplus2\) Hessian split — and is what correctly exposes , rather than conceals, the fermionic residual as unpinnable by any Shape-only invariant (§I.4).

 What complete Shape forces. The Weyl group of \(A_2\) is \(S_3\) , order 6 — a forced consequence of \(K_6\) being the full flag manifold of \(SU(3)\) , not a choice — acting on \((u_1,u_2,u_3)\) by permutation. For any \(S_3\) -invariant functional \(V(\vec u)\) , the fully symmetric point \(\bar u=(1,1,1)\) is automatically a fixed point of the group action and therefore automatically a critical point, \(\nabla V|_{\bar u}=0\) , with zero fitting: this is the criticality leg, DERIVED-GIVEN-anchor, target-blind, at \(+0\) — a ball at the center of a three-fold-symmetric bowl has zero net force on it by symmetry alone, independent of the bowl's exact profile. Complete Shape also forces the classical fact, reproduced independently in this program as a validation of the engine, that there are exactly four invariant Einstein metrics on \(SU(3)/T^2\) : the normal metric \((1,1,1)\) plus the Kähler–Einstein metric \((1,1,2)\) and its three permutations; off-center the space is non-Einstein, which is exactly the squashing sensitivity that feeds the sign computation on the doublet ray. Complete Shape further forces the exact \(\mathbf3=\mathbf1\oplus\mathbf2\) decomposition : an \(S_3\) -singlet breathing mode ( \(u_1=u_2=u_3\) , the overall size direction) and an \(S_3\) -doublet shape mode (traceless combinations such as \(u_1-u_2\) and a partner), decoupled at quadratic order by representation theory alone, with no approximation.

 Full-precision Shape constants riding on this gate (Killing-form normal metric, chamber center, exact rationals): \(\dim K_6=6\) , \(\mathrm{Ric}_i=5/12\) , \(\mathrm{Scal}=5/2\) , \(\mathrm{Scal}^2=25/4\) , \(|\mathrm{Ric}|^2=25/24\) , \(|\mathrm{Riem}|^2=23/12\) , \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) , \(|\mathrm{Ric}|^2/\mathrm{Scal}^2=1/6\) , \(\mathrm{Scal}/\mathrm{Ric}_i=6\) ; cubic invariants \(K_1=R_{ab}{}^{cd}R_{cd}{}^{ef}R_{ef}{}^{ab}=-113/72\) , \(K_2=R_{abcd}R_{aecf}R_{ebfd}=-5/72\) , and \(|\nabla\mathrm{Riem}|^2=1/4\neq0\) — certifying \(K_6\) is homogeneous but not locally symmetric (zero second-Bianchi violations), which is exactly why a nontrivial graviton heat-kernel ladder term exists at all. Weight-6 products at the same center: \(\mathrm{Scal}^3=125/8\) , \(\mathrm{Scal}\,|\mathrm{Ric}|^2=125/48\) , \(\mathrm{Scal}\,|\mathrm{Riem}|^2=115/24\) . Topology: \(\chi(K_6)=6=|S_3|\) — the Euler characteristic of the full flag manifold equals the order of its own Weyl group, the number of chambers. Certified negative controls, never to be substituted: \(|\mathrm{Riem}|^2(K_6)=23/12\) , ratio \(23/75\) — never \(31/147\) (a branch-kill contaminant) and never \(60\) (the round unit- \(S^6\) value, a different manifold).

 The doublet-ray computation, in full. On the shape-doublet ray \(u=(1+\varepsilon,1-\varepsilon,1)\) , with \(R(u)=\sum_i1/u_i-\tfrac12T(u)\) , \(T(u)=\sum_ku_k/(u_iu_j)\) , complete Shape forces the closed form
$$
R(\varepsilon)=\frac32-\frac{\varepsilon^2}{2},
$$
so \(d^2R/d\varepsilon^2|_0=-1\) raw . Decomposing: the diagonal convexity of \(\sum_i1/u_i\) contributes \(+4\) raw ( \(+2\) unit-normalized); the structure-constant triangle piece \(-\tfrac12T(u)\) contributes \(-5\) raw ( \(-5/2\) unit-normalized). Net: \(+4-5=-1\) raw, \(+2-5/2=-1/2\) unit-normalized — two consistent statements of the same invariant fact in two conventions, and the guard is explicit: never combine \(+2\) and \(-5/2\) to report \(-1\) , since that mixes conventions; the sign (negative) is what is convention-invariant and load-bearing. The full \(3\times3\) Hessian of this \(R\) -model at \((1,1,1)\) is \([[-1,2,2],[2,-1,2],[2,2,-1]]\) , eigenvalues \(\{3\ (\times1),\ -3\ (\times2)\}\) , so the shape \(2\times2\) block is \(\mathrm{diag}(-3,-3)\) and \(R(1,1,1)=3/2\) . The \(S^2\times S^2\) negative control fires the opposite sign : the reproducible script value is \(+12\) (per-sphere model \(R=2/b^2\) ); a separately-normalized quoted control value of \(+16\) also appears in this program's narrative record and is convention-dependent — the load-bearing fact, in either normalization, is that the control fires positive while \(K_6\) fires negative, demonstrating the sign test is not rigged to return the same answer regardless of input geometry.

 The physically correct projection is the fixed-volume one (the true shape/volume-preserving doublet, \(\det=1\) ): with \(S=\sum_i1/x_i-(1/6)\sum_ix_i/(x_jx_k)\) , complete Shape forces \(d^2S/d\varepsilon^2|_{\rm fixed\text{-}vol}=+2/3\) , and the log-coordinate Hessian of \(S\cdot\mathrm{Vol}^{1/3}\) has eigenvalues \(\{1/3\ (\times2),\ 0\ (\times1)\}\) . Standard flux-free Kaluza–Klein reduction carries \(V\sim-S|_{\rm fixed\text{-}vol}\) (a positive volume prefactor), so the physical shape mass² is \(-1/3<0\) : a saddle at tree/geometric level. Reconciliation of the three numbers ( \(-1\) raw, \(-1/2\) unit, \(+1/3\) in \(S\) ): \(-1\) / \(-1/2\) are the second derivative of \(R\) along the raw doublet ray; \(+1/3\) is the second derivative of \(S\) at strictly fixed volume — and because \(V\sim-S\) , that \(+1/3\) in \(S\) becomes \(-1/3\) in the physical potential, i.e. also a saddle. Every parametrization gives physical shape mass² \(<0\) (saddle) at the geometric/tree curvature level, consistent with the known mathematical fact that the normal Einstein metric \((1,1,1)\) on \(SU(3)/T^2\) is itself a saddle point among the invariant Einstein metrics, with the Kähler–Einstein points \((1,1,2)\) and permutations the stable ones.

 Complete Shape is also precisely what turns the D1–D4 exhaustion (§I.4) into a theorem rather than a search: because the six \(\chi=-3\) twist candidates and their conjugation/triality orbit structure are enumerated completely, that exhaustion covers every element of a provably closed set, leaving no unexamined case.

 I.2 Scale, applied completely

 Scale enters SG-6 in two structurally distinct roles that this construction keeps strictly separate.

 Role 1 — the residual sign is scale-free. The net Hessian stable-versus-saddle verdict is a product of three factors: geometric curvature \(\times\) bosonic spectral-zeta \(\times\) fermionic graded-Casimir supertrace, and the third factor is the sign of that supertrace on the twisted \(T_0\leftrightarrow T_1\) ( \(E\leftrightarrow E^*\) ) charge-conjugation pair — a discrete \(\pm\) bit, C-odd, carrying no dimension and no scale label. Complete Scale here means recognizing explicitly that \(M_{\rm Pl}\) plays no role whatsoever at this layer: no amount of running, matching, or rescaling between any two energies can move or fix a topological sign. Classifying the bit as "anchored (payable only by a measured C-odd record), not derivable, not fitted" is itself a Scale-layer finding.

 Role 2 — the two genuine Scale objects riding on the same gate. The compactification/unification scale \(M_U\approx1.0\times10^{16}\) GeV (closure residual \(9.6\times10^{-11}\) on \(\alpha_i^{-1}(M_U)=\alpha_j^{-1}(M_U)\) ) sets the chamber-center radius \(R_0=R_6=(2\pi M_U)^{-1}=1.591549430918954\times10^{-17}\ {\rm GeV}^{-1}\) , the physical scale underlying the \(\times\) Stage metric object above. The electroweak vacuum expectation value \(v_{\rm EW}=246\) GeV is the second genuine Scale object touching this gate: it is a measured anchor, consumed and not produced , in the same anchor class as \(\Lambda\approx10^{-122}M_{\rm Pl}^4\) and the baryon asymmetry \(\eta_B\) . The Buckingham- \(\pi\) no-go is itself a Scale-layer argument: it is dimensionally impossible to construct a second, parametrically small mass scale out of \(\{M_{\rm Pl},\hbar,\text{dimensionless frozen geometry}\}\) alone without importing a second scale-carrying length — hence \(v_{\rm EW}\) must be consumed as an irreducible second ruler, which is exactly why this gate does not attempt to derive the electroweak hierarchy as part of its stability verdict. The loop scale \(\mu_{\rm cell}\) (the spectral value of the Uniform Operational Cell \(\Delta_0\) ) is the third Scale object in play; complete Scale exposes that \(\mu_{\rm cell}\) has no \(v\) -independent readout — its only naive anchor is \(\partial_\sigma V=0\) , which is the electroweak-hierarchy condition, so using it to "predict" \(v\) is circular by construction (the \(\kappa^3/\pi\) signature of that circularity). This is precisely why the doublet- magnitude question (whether \(\mu_{\rm cell}\cdot q\gtrsim0.5\) could overcome a negative geometric layer) is fenced off as a separately-blocked residual (R3/R6, R5), cleanly distinguished from the criticality/location leg SG-6 actually certifies at \(+0\) .

 Applying Scale completely therefore does two things: it confirms the one certified physics residual (the fermion sign) sits entirely outside the reach of any scale or RG argument, and it identifies, by name, the one dimensionful ruler ( \(v_{\rm EW}\) ) this gate legitimately consumes as an anchor rather than derives.

 I.3 Granularity, applied completely

 The Uniform Operational Cell ( \(N\le B/\Delta_0\) ) is the granularity posit that dissolves the continuum UV divergence in the breathing-mode (singlet) loop. Applying it completely means tracking both what it supplies for free and what it honestly leaves open.

 What Granularity supplies. \(\mu_{\rm cell}\) inherits its existence from the \(\Delta_0\) posit — the footing of \(\hbar\) — and the cell-sum construction dissolves what would otherwise be a continuum divergence in the singlet-sector loop. This is a genuine, banked piece of the granularity root, not a placeholder.

 What Granularity leaves open, honestly. \(\mu_{\rm cell}\) 's value is not supplied by Granularity alone; it is a floor-residue, and — per the Scale analysis above — its only available anchor collides circularly with \(v_{\rm EW}\) . This is the correct, non-fabricated boundary of what this root delivers here: existence transfers, value does not.

 Finite-cost check. Complete Granularity also requires confirming that the discrete datum this gate turns on — the fermionic sign — is drawn from a genuinely finite, enumerated candidate space, not a hidden continuum smuggled in as "just one bit." That check passes explicitly: the candidate space is the six \(\chi=-3\) twist labels , closed under conjugation and triality, enumerated in full in §I.1. There is no hidden continuum; the cost-floor is satisfied, and the residual is a genuine finite-alphabet unknown — one bit inside a closed six-element orbit structure — not a disguised infinite-precision fit.

 Granularity therefore contributes two disjoint findings: (i) the breathing-singlet sector's would-be UV divergence is a solved problem via the cell construction, with one honest scalar residue ( \(\mu_{\rm cell}\) 's value) named rather than hidden; and (ii) the stability-sign residual itself is certified to live in a finite, closed, enumerated space — the structural precondition for the D1–D4 exhaustion in §I.4 to be a genuinely complete case analysis.

 I.4 The four Layer-2 admissibility screens

 With Shape, Scale, and Granularity each pinned completely, the four Layer-2 screens are run against the resulting object.

 Invariance — PASS on the two closed layers; correctly informative on the third. The geometric curvature layer ( \(-1\) raw / \(-1/2\) unit-normalized, reproduced multiple ways including the fixed-volume \(+1/3\Rightarrow-1/3\) route) and the bosonic spectral-zeta layer ( \(\zeta_{K_6}(-1)=-8033/100800=-0.0796924603174603\ldots\) , with \(8033=29\cdot277\) and \(100800=2^6\cdot3^2\cdot5^2\cdot7\) coprime, confirmed to \(\sim22\) digits by an exact Weyl/Poisson closed form against an independent blind numeric heat-sum with Richardson extrapolation) both pass Invariance: branch-preserving, multi-route, exact-rational or numerically pinned quantities. Invariance is applied completely — across all three layers of the frozen object, not merely the metric level — and its verdict on the third (fermionic) layer is the physically correct one: Invariance reports that the sought bit is not fixed by any C-even, branch-preserving invariant of the frozen shape, rather than merely failing to determine it. This is Invariance functioning correctly: it exposes that the target datum lies structurally outside the invariant ring generated by Shape data alone.

 The exhaustion behind that verdict, run target-blind over the provably closed six-candidate class (the six \(\chi=-3\) twist labels under conjugation and triality), tests every intrinsic topological, reality, and anomaly discriminator available on the frozen shape:

 Discriminator 
 Test 
 Verdict 

 D1 — Wu/ \(w_2\) / \(w_3\) orientation 
 identically zero on the frozen \(X\) , tangential, C-even 
 CANNOT PIN 

 D2 — measure-reality / CPT 
 \(\mathrm{FS}(\mathbf3)=0\) exact and confirmed by independent Weyl-integral numeric agreeing to \(\sim10^{-16}\) ; CPT/color- \(C\) maps \(T_0\leftrightarrow T_1\) as a vectorlike pair under \(SU(3)_c\) , constraining the pair but unable to split it 
 CANNOT PIN 

 D3 — Dai–Freed anomaly beyond \(\eta\) 
 full anomaly datum conjugates (C-even); mod-2 refinement undefined exactly where \(\mathrm{FS}(\mathbf3)=0\) 
 CANNOT PIN 

 D4 — spin- \(\mathbb C\) determinant-line reality 
 requires a self-conjugate twist; none of the six candidates is self-conjugate — the free \(C\) -action exchanges the two triality triples exactly 
 CANNOT PIN (empty) 

 All four discriminator classes return CANNOT PIN over the complete closed class by theorem, not by search fatigue — a structural fact of representation theory (a C-even invariant cannot carry a C-odd sign), which is what licenses the term CERTIFIED-UNPINNABLE(C) .

 Record Interface — CONSTRAIN (fires correctly; does not dissolve). The residual datum restates, without loss, as a finite exterior record: "the C-odd sign of the graded-Casimir supertrace" is a well-defined \(\pm\) observable in principle, exactly the kind of quantity a future decisive measurement (the leptonic \(\delta_{CP}\) sign) could report. Because the observable survives restatement into a finite record rather than evaporating under scrutiny, the correct Layer-2 outcome is CONSTRAIN, not dissolution — the OBSERVABLE-NEVER-DISSOLVES pattern: a genuinely unpinnable target , not an empty or ill-posed carrier. This is what promotes the residual from a vague open question to a dressed, anchor-payable IOU, banked as certificate AC-GAP10-HOLE3-v1, status UNPAID, with the leptonic \(\delta_{CP}\) sign as the named admissible payer and the baryon asymmetry \(\eta_B\) permanently barred as a payer (constraint W12).

 Causal Order — PASS. No target-value leakage from any downstream observable enters this derivation anywhere. The criticality theorem uses only the \(S_3\) symmetry of \(K_6\) , free of any potential-shape input tuned toward an answer. The D1–D4 exhaustion is run target-blind: no CP-violating phase value, no sign preference, and no measured \(\delta_{CP}\) or \(\eta_B\) value is assumed anywhere in the discriminator logic or in the enumeration of the six-candidate class. This screen certifies that CERTIFIED-UNPINNABLE(C) is a genuine structural fact about the geometry, not an artifact of having peeked at the answer the residual is meant to predict.

 Nonseparability — CONSTRAIN (correctly shared, counted once). The \(T_0/T_1\) charge-conjugation bit is not independent across gates: complete Nonseparability tracking shows it is the same discrete unknown shared by uqf10 (the fermion Casimir sign in that context), sg6 (this gate's stability-sign residual), and gap10-bg10 Hole #3. Applying Nonseparability completely means this single unpaid bit is counted once in the cross-wall accounting (the shared heat-kernel-scheme keystone convention), not charged three times as though it were three separate open problems — resolving it anywhere resolves it everywhere it appears.

 I.5 What the complete deep-root pass forces, eliminates, and exposes

 Pulling the three roots and four screens together: complete Shape forces the criticality theorem ( \(S_3\) -fixed point \(\Rightarrow\nabla V=0\) , free of any fit) and the exact \(\mathbf3=\mathbf1\oplus\mathbf2\) Hessian split, and it forces a definite, multi-route, convention-consistent negative sign for the geometric curvature layer on the doublet sector (raw \(-1\) /unit \(-1/2\) /fixed-volume-projected \(-1/3\) physical), cross-checked against the \(S^2\times S^2\) specificity control which fires the opposite (positive) sign. Complete Shape also eliminates the earlier truncated-Shape " \(-1\) as the settled shape eigenvalue" reading, which mixed the criticality framing with the \(\Lambda\) -free doublet framing and must not be re-cited as a settled stability result. Complete Scale eliminates any route to fixing the fermion sign by running, matching, or rescaling — the datum is scale-blind by construction — while correctly identifying \(v_{\rm EW}=246\) GeV as the one legitimate consumed anchor and \(\mu_{\rm cell}\) 's value as an honest, separately-fenced residue whose only naive anchor is circular with the hierarchy it would be used to explain. Complete Granularity forces the finiteness of the candidate space (six enumerated twist labels, no hidden continuum) that makes the D1–D4 exhaustion a genuine completeness theorem, while leaving \(\mu_{\rm cell}\) 's numerical value an open floor-residue rather than a supplied number. And the four Layer-2 screens, run over this complete object, expose rather than hide or artificially resolve the single named residual: Invariance shows no C-even invariant can carry it, Record Interface shows it survives as a genuine finite payable observable with a named payer and a named barred non-payer, Causal Order certifies the exposure is target-blind, and Nonseparability shows it is one bit shared across three gates, not three separate bits. Together, these completely-applied roots are what license writing SG-6's criticality leg and its two curvature layers as genuinely closed at +0 , while carrying the fermionic sign forward as a confident, falsifiable, anchor-payable wager — a limit on all knowledge, since no framework-internal invariant of any rival theory either can carry a C-odd sign from Shape-only data — rather than either a silently dropped gap or a falsely claimed closure.

 Construction II - the full derivation

 This section carries out the SG-6 derivation chain end to end, in the order the computation must actually be performed: fix the arena and the object being differentiated, prove criticality, prove the exact block split, compute each of the three Hessian layers on the surviving block in turn, and assemble the verdict. Every equation is written out; every numerical value is either an exact rational quoted from the frozen geometry or a number derived from those rationals in full view. Nothing here is asserted without the intermediate arithmetic that produces it.

 II.1 The object being extremized, pinned at all three layers

 The frozen active branch is

 \[
\mathfrak{B}_{\rm active}
=
\underbrace{\big[\,\mathcal{M}_4 \times K_6 \times S^2 \times S^1_Y/\mathbb{Z}_2\,\big]_\times}_{\times\ \text{Stage}}
\ \oplus\
\underbrace{\big[\,\mathcal{F}^{+}_{\rm finite} \oplus \mathcal{C}_{\rm admiss}\,\big]_\oplus}_{\oplus\ \text{Rulebook}}
\ \otimes\
\underbrace{\big[\,\mathcal{E}_{\rm matter} \oplus \mathcal{E}_{\rm gauge} \oplus \mathcal{E}_{\rm Higgs} \oplus \mathcal{E}_{\rm proton}\,\big]_\otimes}_{\otimes\ \text{Actors}},
\]

 with \(K_6=SU(3)/T^2\) the \(A_2\) full flag manifold and total dimension \(D=4+6+2+1=13\) . SG-6 asks a question entirely internal to the \(\times\) Stage shape-moduli sector of \(K_6\) , decided using \(\oplus\) Rulebook admissibility/grading data and one \(\otimes\) Actors operator datum; the ambient factors \(\mathcal M_4\) , \(S^2\) , \(S^1_Y/\mathbb Z_2\) are held fixed throughout and enter only through the volume normalization that turns a dimensionless curvature Hessian into a physical mass².

 × Stage — the moduli and their metric. \(K_6\) carries the \(SU(3)\) -left-invariant family of metrics

 \[
g_{K_6}(\vec u)=u_1\langle\cdot,\cdot\rangle_{\mathfrak m_1}+u_2\langle\cdot,\cdot\rangle_{\mathfrak m_2}+u_3\langle\cdot,\cdot\rangle_{\mathfrak m_3},\qquad \vec u=(u_1,u_2,u_3)\in[1/2,3/2]^3,
\]

 where \(\langle\cdot,\cdot\rangle_{\mathfrak m_i}\) is the Killing form \(B(X,Y)=6\,\mathrm{Tr}(XY)\) restricted to the \(i\) -th root 2-plane of the tangent decomposition \(T(K_6)=\mathfrak m_1\oplus\mathfrak m_2\oplus\mathfrak m_3\) ( \(\dim_{\mathbb R}\mathfrak m_i=2\) ), built on the \(A_2\) positive roots \(\alpha_1=(1,-1,0)\) , \(\alpha_2=(0,1,-1)\) , \(\alpha_1+\alpha_2=(1,0,-1)\) in the Cartan basis \(h_1+h_2+h_3=0\) . All curvature numbers in this section are quoted in the Killing-form normal-metric normalization , dimensionless, evaluated at the chamber center; the physical \(R_6\) -normalization (used for KK masses and the Planck relation) is related by the metric-scale-invariant ratios \(\mathrm{Scal}/\mathrm{Ric}_i=6\) , \(|\mathrm{Ric}|^2/\mathrm{Scal}^2=1/6\) , \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) , identical in both normalizations. \(\vec u\) is the moduli space this gate is a statement about.

 ⊕ Rulebook. Two objects are load-bearing. First, \(\mathcal C_{\rm admiss}\) fixes the Weyl-rigid admissibility chamber \(\vec u\in[1/2,3/2]^3\) : a configuration outside this window is rejected by the selector , not dynamically suppressed by a potential well — a rule, not a force, and the origin of the explicit non-claim that global stabilization is established anywhere in this gate. Second, \(\mathcal F^+_{\rm finite}\) fixes the modular Cartan-torus fixed point \(\tau=\omega=e^{2\pi i/3}=-0.5+0.8660254037844386\,i\) , the generation basis \(\mathcal G_{\rm gen}\) ( \(\dim_{\mathbb C}=3\) ), and the graded-Casimir sign convention (bosons carry \(+\) , fermions carry \(-\) in the supertrace, scheme-independent) used in §II.6.

 ⊗ Actors. The operator content is the twisted Dirac/Laplace-type operator on \(K_6\) acting on the spin- \(\mathbb C\) bundle \(S_{K_6}^{\rm spin^c}\) that carries the family index \(\chi(K_6,E)=-3\) , with bundle endomorphism \(E=\mathrm{Ric}=(5/12)\,\mathrm{Id}\) on the vector (1-form) bundle and the Lichnerowicz endomorphism \(E_L\) on \(\mathrm{Sym}^2_0T^*K_6\) with spectrum \(\{1/6\ (\times6),\ 5/12\ (\times6),\ 7/6\ (\times6),\ 17/12\ (\times2)\}\) on the traceless dim-20 sector. The object at the center of §II.6 is the charge-conjugation action \(E\leftrightarrow E^*\) exchanging the twisted-spinor pair \(T_0\leftrightarrow T_1\) , drawn from the closed, finite set of six \(\chi=-3\) Dynkin-label twist candidates closed under conjugation and triality.

 The full physical stability operator factorizes exactly into three commuting layers evaluated at \(\bar u=(1,1,1)\) :

 \[
(m^2)_{\rm phys}^{\rm doublet} \;=\; \big(\text{geometric curvature Hessian}\big)\ \times\ \big(\text{bosonic spectral-zeta layer}\big)\ \times\ \big(\text{fermionic graded-Casimir supertrace sign}\big),
\]

 and the derivation below computes each factor in turn, after first showing that the location \(\bar u=(1,1,1)\) and the block structure of the Hessian are themselves forced.

 II.2 Step 1 — the resting point is forced: \(\nabla V(\bar u)=0\) for every admissible \(V\) 

 Claim. \(\bar u=(1,1,1)\) is a critical point of every moduli potential \(V(\vec u)\) built covariantly from the invariant geometric data of \(K_6\) , with zero fitting.

 Derivation. The Weyl group of \(A_2\) is \(S_3\) , order \(|S_3|=6\) ; note \(6=\chi(K_6)\) , the Euler characteristic of the full flag manifold, exactly as required since \(\chi(G/T)\) always equals \(|W|\) for a compact Lie group \(G\) with maximal torus \(T\) . \(S_3\) acts on the moduli triple by permutation,
$$
\sigma\cdot(u_1,u_2,u_3)=(u_{\sigma(1)},u_{\sigma(2)},u_{\sigma(3)}),\qquad \sigma\in S_3,
$$
because \(S_3\) permutes the three positive roots \(\alpha_1,\alpha_2,\alpha_1+\alpha_2\) and hence permutes the three root planes \(\mathfrak m_1,\mathfrak m_2,\mathfrak m_3\) that the \(u_i\) independently rescale. Any potential built covariantly from \(g_{K_6}(\vec u)\) — curvature scalars, volume, spectral functionals — therefore satisfies \(V(\sigma\cdot\vec u)=V(\vec u)\) for every \(\sigma\in S_3\) : this invariance is automatic , not assumed, because the geometry itself only knows the unordered triple \(\{u_1,u_2,u_3\}\) .

 \(\bar u=(1,1,1)\) is the unique point on the diagonal fixed by the entire group ( \(\sigma\cdot\bar u=\bar u\) for all \(\sigma\) ). Differentiating the invariance identity \(V(\sigma\cdot\vec u)=V(\vec u)\) with respect to \(\vec u\) and evaluating at \(\vec u=\bar u\) gives
$$
\nabla V(\bar u)=\sigma\cdot\nabla V(\bar u)\qquad\text{for every }\sigma\in S_3,
$$
i.e. \(\nabla V(\bar u)\) is a vector in \(\mathbb R^3\) fixed by the full permutation action of \(S_3\) . The permutation representation of \(S_3\) on \(\mathbb R^3\) decomposes exactly as \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) : the trivial singlet \(\mathbf 1\) spanned by \((1,1,1)\) , and the standard two-dimensional irrep \(\mathbf 2\) spanned by the traceless directions \(\{(1,-1,0),(1,0,-1)\}\) (or any traceless basis). The standard representation of \(S_3\) is irreducible and non-trivial, hence has no nonzero \(S_3\) -fixed vector — a transposition acting on \(\mathbf 2\) has eigenvalues \(\{+1,-1\}\) and the 3-cycles act with eigenvalues \(\{e^{\pm2\pi i/3}\}\) on its complexification, so the only vector in \(\mathbf 2\) invariant under the whole group is \(0\) . Therefore the component of \(\nabla V(\bar u)\) lying in \(\mathbf 2\) (the two shape directions transverse to the breathing mode) must vanish identically:
$$
\nabla_\perp V(\bar u)=0\qquad\text{for every }S_3\text{-invariant }V,
$$
where \(\perp\) denotes the traceless shape subspace. This is the full content of the criticality claim, and it holds for the entire infinite-dimensional space of admissible \(S_3\) -invariant potentials simultaneously — no particular functional form of \(V\) was chosen or fit to produce this vanishing. This leg is DERIVED-GIVEN- \(E\) (given the frozen twisted geometry supplying \(\chi(K_6,E)=-3\) ), target-blind, closed at +0 . 

 Forbidden extension, stated explicitly. Criticality does not imply minimality. \(\nabla_\perp V(\bar u)=0\) is consistent with \(\bar u\) being a minimum, a maximum, or a saddle of the transverse Hessian; nothing in Step 1 decides which. That decision is the entire content of Steps 3–5.

 Independent structural check: the four invariant Einstein metrics. Using the general-chamber Ricci formula on Killing-form scales \(x_1,x_2,x_3\) ,
$$
\mathrm{Ric}_1=\frac{x_1^2-x_2^2+6x_2x_3-x_3^2}{12x_1x_2x_3},\quad
\mathrm{Ric}_2=\frac{-x_1^2+6x_1x_3+x_2^2-x_3^2}{12x_1x_2x_3},\quad
\mathrm{Ric}_3=\frac{-x_1^2+6x_1x_2-x_2^2+x_3^2}{12x_1x_2x_3},
$$
setting \(\mathrm{Ric}_1=\mathrm{Ric}_2=\mathrm{Ric}_3\) (the Einstein condition) is a symmetric cubic system in the ratios \(x_i/x_j\) whose only positive solutions up to overall scale are \((1,1,1)\) and the three permutations of \((1,1,2)\) — four invariant Einstein metrics total on \(SU(3)/T^2\) , a classical result reproduced here independently as a validation of the engine. This is structurally important for what follows: it shows the moduli space contains a second , non-symmetric family of extremal points (the Kähler–Einstein \((1,1,2)\) -type metrics), so the symmetric point being a critical point of every \(S_3\) -invariant \(V\) does not by itself make it preferred over that second family — exactly the competition the doublet Hessian in Steps 3–4 must resolve.

 II.3 Step 2 — the Hessian splits exactly, \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) , by representation theory alone

 Because \(\nabla_\perp V(\bar u)=0\) already establishes \(\bar u\) as a critical point, the next object is the Hessian, \(\mathrm{Hess}(V)|_{\bar u}\) , a symmetric bilinear form on the tangent space \(\mathbb R^3\) of the moduli space at \(\bar u\) . By exactly the same representation-theoretic argument as Step 1, this bilinear form must be \(S_3\) -invariant as an operator, meaning it commutes with the \(S_3\) action on \(\mathbb R^3=\mathbf 1\oplus\mathbf 2\) . Schur's lemma then forces \(\mathrm{Hess}(V)|_{\bar u}\) to be block-diagonal with respect to this decomposition — a scalar acting on the one-dimensional singlet \(\mathbf 1\) (the breathing/overall-volume mode \(u_1=u_2=u_3=t\) ), and a second scalar times the identity acting on the two-dimensional standard irrep \(\mathbf 2\) (the traceless shape doublet, represented by the ray \(u_1-u_2\) and a partner direction), because \(\mathbf 2\) is irreducible: any \(S_3\) -equivariant operator on an irreducible representation is a scalar multiple of the identity on it (Schur), and the two blocks cannot mix because \(\mathbf 1\not\cong\mathbf 2\) . In coordinates, using the representative doublet ray
$$
\vec u(\varepsilon)=(1+\varepsilon,\ 1-\varepsilon,\ 1),\qquad \textstyle\sum_i(u_i-1)=0\ \text{at every order},
$$
the doublet block is captured by the single second derivative \(d^2V/d\varepsilon^2|_0\) , and this number is the entire content of the stability question: the singlet direction is a flat volume modulus (no restoring force is being asked about here — it is the overall size, separately controlled by the Planck-normalization/volume sector) and the doublet direction is where "minimum vs. saddle" is decided. This split is exact at quadratic order — no approximation, no small parameter beyond \(\varepsilon\) itself — and is itself DERIVED-GIVEN-anchor, +0 , forced by the same \(S_3\) representation theory as Step 1, with no dynamical input.

 The explicit \(3\times3\) Hessian of the model curvature functional \(R(\vec u)\) (defined in Step 3 below) at \(\vec u=(1,1,1)\) , computed directly without invoking the abstract representation-theory argument, gives an independent numerical confirmation of this block structure:
$$
\mathrm{Hess}(R)\big|_{(1,1,1)}=\begin{pmatrix}-1&2&2\2&-1&2\2&2&-1\end{pmatrix},
$$
with eigenvalues \(\{3\ (\times1),\ -3\ (\times2)\}\) : the singlet eigenvector \((1,1,1)/\sqrt3\) carries eigenvalue \(+3\) and the two-dimensional traceless subspace carries the degenerate eigenvalue \(-3\) — exactly the \(\mathbf1\oplus\mathbf2\) block structure with a doubly-degenerate shape block, as representation theory requires. ( \(R(1,1,1)=3/2\) for this functional, recorded for cross-reference.)

 II.4 Step 3 — the geometric-curvature layer of the doublet Hessian: exact, negative, three-route

 The functional. The curvature model used for the doublet stability test is
$$
R(\vec u)=\sum_i\frac1{u_i}-\frac12\,T(\vec u),\qquad T(\vec u)=\sum_k\frac{u_k}{u_iu_j}\ \ (i,j,k\ \text{cyclic}),
$$
an \(S_3\) -invariant scalar built from the metric data (equivalently the general-chamber scalar curvature written in §II.1's normalization). Evaluate along the doublet ray \(\vec u(\varepsilon)=(1+\varepsilon,1-\varepsilon,1)\) .

 Term 1 — the diagonal convexity piece, \(\sum_i 1/u_i\) . Expand each term to \(O(\varepsilon^2)\) :
$$
\frac1{1+\varepsilon}=1-\varepsilon+\varepsilon^2+O(\varepsilon^3),\qquad \frac1{1-\varepsilon}=1+\varepsilon+\varepsilon^2+O(\varepsilon^3),\qquad \frac11=1.
$$
Summing,
$$
\sum_i\frac1{u_i}=3+2\varepsilon^2+O(\varepsilon^4)\ \Longrightarrow\ \left.\frac{d^2}{d\varepsilon^2}\sum_i\frac1{u_i}\right|_0=+4\ (\text{raw}),
$$
which is the second derivative including the factor of 2 from the Taylor expansion \(\varepsilon^2\) -coefficient \(\times\,2!\) ; written as a coefficient of \(\varepsilon^2\) directly it is \(+2\) (this pack calls the coefficient-of- \(\varepsilon^2\) number the "unit-normalized" reading and the full second-derivative number the "raw-ray" reading — both are recorded below and must never be mixed). This term is strictly positive because \(x\mapsto1/x\) is strictly convex ( \(d^2(1/x)/dx^2=2/x^3>0\) for \(x>0\) ): any symmetric \(\pm\varepsilon\) deformation of two of the three moduli raises \(\sum 1/u_i\) , so this piece always stabilizes .

 Term 2 — the structure-constant triangle piece, \(T(\vec u)\) . With \((u_1,u_2,u_3)=(1+\varepsilon,1-\varepsilon,1)\) , the cyclic sum is
$$
T(\varepsilon)=\frac{u_3}{u_1u_2}+\frac{u_1}{u_2u_3}+\frac{u_2}{u_1u_3}
=\frac{1}{(1+\varepsilon)(1-\varepsilon)}+\frac{1+\varepsilon}{1-\varepsilon}+\frac{1-\varepsilon}{1+\varepsilon}.
$$
The first term is \(1/(1-\varepsilon^2)=1+\varepsilon^2+O(\varepsilon^4)\) . The second and third terms expand as
$$
\frac{1+\varepsilon}{1-\varepsilon}=(1+\varepsilon)(1+\varepsilon+\varepsilon^2+\dots)=1+2\varepsilon+2\varepsilon^2+O(\varepsilon^3),
$$
$$
\frac{1-\varepsilon}{1+\varepsilon}=(1-\varepsilon)(1-\varepsilon+\varepsilon^2-\dots)=1-2\varepsilon+2\varepsilon^2+O(\varepsilon^3),
$$
and their sum is \(2+4\varepsilon^2+O(\varepsilon^4)\) (the linear terms cancel by the \(\varepsilon\to-\varepsilon\) symmetry of the ray). Adding the first term,
$$
T(\varepsilon)=3+5\varepsilon^2+O(\varepsilon^4)\ \Longrightarrow\ \left.\frac{d^2T}{d\varepsilon^2}\right|_0=+10\ (\text{raw}),\quad\text{coefficient-of-}\varepsilon^2=+5.
$$
Since \(R\) carries \(-\tfrac12T\) , this term contributes \(-\tfrac12\times(+10)=-5\) to the raw second derivative of \(R\) (equivalently \(-\tfrac12\times(+5)=-\tfrac52\) to the unit-normalized coefficient).

 Assembling \(R(\varepsilon)\) . Combining Term 1 and Term 2,
$$
R(\varepsilon)=\Big(3+2\varepsilon^2\Big)-\frac12\Big(3+5\varepsilon^2\Big)+O(\varepsilon^4)=\frac32-\frac12\varepsilon^2+O(\varepsilon^4),
$$
so
$$
\boxed{R(\varepsilon)=\frac32-\frac{\varepsilon^2}{2}+O(\varepsilon^4)},\qquad \left.\frac{d^2R}{d\varepsilon^2}\right|_0=-1\ (\text{raw}).
$$

 Convention bookkeeping (load-bearing, verifier-checked). The raw second derivative is the sum of raw pieces, \(+4-5=-1\) . The unit-normalized ( \(\varepsilon^2\) -coefficient) reading is \(+2-\tfrac52=-\tfrac12\) , matching the coefficient in the boxed closed form directly. These are two consistent statements of the same fact in two conventions and must never be cross-added (e.g. \(+2-5/2\) evaluated then mislabeled as the raw \(-1\) , or vice versa). What is convention-invariant, and what carries the physics, is the sign : both readings are negative. \(d^2R/d\varepsilon^2|_0=-1\) (raw) \(=-1/2\) (unit-normalized) — a saddle direction in this model curvature functional.

 Specificity control. The identical computation performed on the \(S^2\times S^2\) homogeneous space (a per-sphere curvature model \(R=2/b^2\) under the analogous doublet-type deformation) returns the reproducible value \(+12\) — the opposite sign . This demonstrates the test is not rigged to output a fixed sign regardless of the input manifold: it is sensitive to the actual geometry of \(K_6=SU(3)/T^2\) , and specifically to the \(A_2\) structure-constant triangle term \(T(\vec u)\) , which has no analogue on a product-of-spheres space. (A separate quoted value of \(+16\) appears in narrative material using a different \(S^2\times S^2\) normalization; the reproducible script value used here is \(+12\) , and in either normalization the control fires with the opposite, positive, sign — that sign flip relative to \(K_6\) is the load-bearing content of the control, not the specific magnitude.)

 Cross-check via the full \(3\times3\) Hessian. Independently of the ray expansion, the full Hessian of \(R(\vec u)\) at \((1,1,1)\) computed directly from all nine second partials is
$$
\mathrm{Hess}(R)\big|_{(1,1,1)}=\begin{pmatrix}-1&2&2\2&-1&2\2&2&-1\end{pmatrix},\qquad \text{eigenvalues }{3,\,-3,\,-3},
$$
confirming the doublet (traceless) block carries the doubly-degenerate eigenvalue \(-3\) — negative, consistent in sign with the ray computation (the ray computation isolates one direction inside this degenerate doublet block and returns \(-1\) / \(-1/2\) under the ray's own normalization, while the full matrix returns \(-3\) for the whole degenerate doublet eigenvalue; both are computed from the same \(R(\vec u)\) and agree in sign, which is the invariant fact carried forward).

 Third route: fixed-volume physical projection. The ray computation above does not hold the total volume fixed; the true shape-only (volume-orthogonal) doublet direction is obtained by additionally imposing \(\det(\vec u)=u_1u_2u_3=1\) . Repeating the expansion for the fixed-volume functional
$$
S(\vec u)=\sum_i\frac1{u_i}-\frac16\sum_i\frac{u_i}{u_ju_k}\ \ (j,k\ \text{the other two indices}),
$$
restricted to the volume-1 slice, gives
$$
\left.\frac{d^2S}{d\varepsilon^2}\right| {0,\,\rm fixed\ vol}=+\frac23,
$$
and writing \(S\cdot\mathrm{Vol}^{1/3}\) in logarithmic moduli coordinates \((\ln u_1,\ln u_2,\ln u_3)\) gives a Hessian with eigenvalues \(\{1/3,\ 1/3,\ 0\}\) (the two shape directions each carrying \(+1/3\) , the volume direction flat as required by construction). Standard flux-free Kaluza–Klein reduction carries the physical potential as \(V\sim -S|_{\rm fixed\ vol}\) (the positive-volume-prefactor convention linking the internal curvature functional to the four-dimensional effective potential), so the \(+1/3\) eigenvalue of \(S\) becomes
$$
(\text{shape mass}^2) {\rm fixed\ vol}=-\frac13<0,
$$
a saddle at tree/geometric level — the same conclusion as the raw-ray computation, reached by an independent route that additionally removes any possible volume-mixing artifact.

 Reconciliation of the three numbers \(-1\) , \(-1/2\) , \(+1/3\) . All three are the same saddle, read in three conventions: \(-1\) (raw) and \(-1/2\) (unit-normalized) are the second derivative of \(R\) along the raw doublet ray (not volume-fixed); \(+1/3\) is the second derivative of \(S\) at strictly fixed volume (the correctly volume-projected shape doublet), which becomes \(-1/3\) once translated into the physical potential via \(V\sim-S\) . Every parametrization gives physical shape mass² \(<0\) at the geometric/tree-curvature level. This matches the classical fact recalled in Step 1: the normal Einstein metric \((1,1,1)\) on \(SU(3)/T^2\) is a saddle among the four invariant Einstein metrics, with the stable Einstein points located at the Kähler–Einstein \((1,1,2)\) -type permutations instead.

 Status. This entire layer — the geometric-curvature contribution to the doublet Hessian — is REDUCED-TO-FLOOR, +0 : exact closed form, three independent computational routes (raw doublet ray, full \(3\times3\) eigen-decomposition, fixed-volume projection), a passed specificity control, and full agreement on sign. The forbidden reading , explicitly retired, is quoting the bare number " \(-1\) " alone as "the settled shape eigenvalue": that phrasing conflates the raw-ray convention with the unit-normalized and fixed-volume conventions and is a truncated-Shape artifact of an earlier single-slice treatment. The number that survives into the final assembly (§II.7) is the sign , which is unambiguously negative in every convention checked.

 II.5 Step 4 — the bosonic spectral-zeta layer: exact, negative, two-route agreement

 The second multiplicative factor in the physical mass-squared operator is the bosonic one-loop spectral contribution, computed as the analytically continued spectral zeta function of the relevant Laplace-type operator on \(K_6\) at the point \(s=-1\) :
$$
\zeta_{K_6}(-1)=\sum_{(p,q)}{}'\,\big(C_2(p,q)\big)^{-s}\Big| {s=-1},
$$
regularized via the heat-kernel coefficients using \(K(t)\sim(4\pi t)^{-d/2}\sum_k a_{2k}t^k\) and the standard Mellin transform relating \(\zeta(s)\) to the heat-kernel expansion. The certified value, agreed by two independent routes (a blind numerical heat-sum with Richardson extrapolation, and an exact closed-form evaluation via the Weyl character/Poisson-resummation route, agreeing to approximately 22 digits) is the exact rational
$$
\boxed{\zeta {K_6}(-1)=-\frac{8033}{100800}=-0.0796924603174603\ldots},
$$
with \(8033=29\times277\) and \(100800=2^6\times3^2\times5^2\times7\) coprime to the numerator (no hidden common factor, confirming the fraction is already in lowest terms and not an artifact of an unreduced computation). Companion values banked in the same computation, useful as cross-checks of the same heat-kernel coefficient ladder: \(\zeta_{K_6}(0)=-253/315\) , and the second heat-kernel coefficient \(c_2=5743/184800\) , giving \(\zeta_{K_6}(-2)=5743/92400\) .

 This layer's sign is negative , matching the geometric-curvature layer of Step 3. The value is REDUCED-TO-FLOOR, +0 : it rests on the certified scalar heat-kernel ratios \(a_2/a_0=5/12\) , \(a_4/a_0=11/120\) (both exact rationals from the Killing-form curvature data — \(a_2/a_0\) is literally \(\mathrm{Ric}_i=5/12\) read off §II.1's curvature table), with the analytic continuation to \(s=-1\) performed by the standard zeta-function regularization procedure and cross-checked by direct numerical summation. (A documented script failure in one auxiliary numerical tool hit a \(\Gamma\) -function pole and raised a ValueError on one run; this is logged plainly as a software bug in that particular superseded script, not as evidence against the banked value, since the value is independently reconstructed by the two routes above without going through the failing code path.)

 II.6 Step 5 — the fermionic graded-Casimir layer: isolating the one C-odd bit, and proving it is unpinnable

 The third multiplicative factor is the sign of the fermionic contribution to the doublet mass², carried by the graded (bose \(-\) fermi) supertrace of the quadratic Casimir on the twisted spinor sector. Per-sector signs are fixed by the standard graded convention and are scheme-independent: bosonic loops contribute \(+\) , fermionic loops contribute \(-\) , to the supertrace \(\mathrm{Str}=\mathrm{Tr}_{\rm bose}-\mathrm{Tr}_{\rm fermi}\) . What is not fixed by this convention alone is the net sign once the fermionic sector's own internal charge-conjugation structure is taken into account, because the twisted spin- \(\mathbb C\) bundle on \(K_6\) carries a charge-conjugation pair of sectors, \(T_0\leftrightarrow T_1\) , related by \(E\leftrightarrow E^*\) , and the supertrace must be evaluated with a definite relative sign between \(T_0\) and \(T_1\) — that relative sign is the residual bit.

 Structural origin of the bit. \(T_0\leftrightarrow T_1\) exchange is the charge-conjugation ( \(C\) ) action on the twisted spinor bundle over \(K_6\) . Every piece of Shape-only data available on the frozen manifold — the tangent bundle \(TK_6\) , the Stiefel–Whitney/Wu classes built from it, the Dai–Freed anomaly of the associated Pin/Spin- \(\mathbb C\) structure, the spin- \(\mathbb C\) determinant line — is built from \(TK_6\) , on which the conjugation \(C\) acts trivially (real tangent bundle data is manifestly \(C\) -even). But the sought bit is intrinsically \(C\) -odd by construction (it is precisely the sign that flips under \(T_0\leftrightarrow T_1\) ). A \(C\) -even invariant cannot, by definition, carry the value of a \(C\) -odd quantity: this is a representation-theoretic obstruction, not a statement about insufficient computational effort.

 The closed candidate space. The twist data admissible under the frozen family index \(\chi(K_6,E)=-3\) forms an exactly enumerable, finite set: the six \(\chi=-3\) Dynkin-label twist candidates , closed under the conjugation action and the triality outer automorphism of \(A_2\) . This set is finite and fully enumerated — there is no hidden continuum being swept under a "just one unknown bit" description.

 Route-B exhaustion — four discriminator classes, all four candidates, target-blind. Every intrinsic invariant available on the frozen shape that could in principle carry a \(C\) -odd sign is tested against the full closed candidate space:

 Discriminator 
 What it tests 
 Result 

 D1 — Wu/ \(w_2\) / \(w_3\) orientation data 
 identically zero on the frozen \(X\) ; manifestly tangential, hence \(C\) -even 
 CANNOT PIN a \(C\) -odd bit 

 D2 — measure-reality / CPT 
 Frobenius–Schur indicator \(\mathrm{FS}(\mathbf 3)=0\) exactly (confirmed by an independent Weyl-integral numerical evaluation agreeing to \(\sim1\times10^{-16}\) ); CPT together with color- \(C\) maps \(T_0\leftrightarrow T_1\) as a vectorlike \(SU(3)_c\) pair, constraining the pair jointly but structurally unable to split it 
 CANNOT PIN 

 D3 — Dai–Freed anomaly beyond \(\eta\) 
 the full anomaly datum conjugates under \(C\) (hence \(C\) -even by construction); the would-be mod-2 refinement that could in principle carry finer information is undefined exactly at the point \(\mathrm{FS}(\mathbf3)=0\) 
 CANNOT PIN 

 D4 — spin- \(\mathbb C\) determinant-line reality 
 requires a self-conjugate twist to have a well-defined reality structure; direct check of all six candidates shows the free \(C\) -action exchanges the two triality triples exactly, so none of the six is self-conjugate — the discriminator's domain of applicability is empty on this candidate set 
 CANNOT PIN 

 All four discriminator classes return "cannot pin," exhaustively, over a provably closed six-element candidate space — this is a completed case analysis, not a search that ran out of ideas. Every intrinsic invariant of the frozen shape has been checked and each fails to carry a \(C\) -odd sign, for a structural reason (representation-theoretic \(C\) -evenness) rather than a computational one. This licenses the term CERTIFIED-UNPINNABLE : a proven absence of a discriminator, exhaustively demonstrated, not a stalled computation awaiting more effort.

 Typed endpoint. Because the observable (the fermion sign) survives restatement under every discriminator rather than dissolving as ill-posed, this is not a Q1-type dissolution; it routes instead to a dressed, time-indexed, anchor-payable IOU , carrying the full completeness payload: certified candidate class (the six \(\chi=-3\) twists), completeness basis (the D1–D4 exhaustion together with \(\mathrm{FS}(\mathbf3)=0\) , the Weyl-integral numeric cross-check, and Dai–Freed conjugation), an unconditional completeness argument (not contingent on an unproven lemma), and all four possible discriminator routes enumerated and closed by theorem. Named falsifier: the bit pays on measurement of the decisive leptonic \(\delta_{CP}\) sign (or any admissible \(C\) -odd record). Named excluded payer: the baryon asymmetry \(\eta_B\) is permanently barred from paying this bit.

 Leading loop indicator (recorded, not decisive). The graded quadratic-Casimir supertrace on the fundamental, \(\mathrm{Str}[C_2]=\chi\cdot C_2(\mathbf3)=(-3)\times(4/3)=-4\) , using \(\chi(K_6,E)=-3\) and \(C_2(\mathbf3)=4/3\) from the exact Casimir formula \(C_2(p,q)=(p^2+q^2+pq+3p+3q)/3\) at \((p,q)=(1,0)\) . This value is negative and, taken at face value, opposes a boson-dominance rescue of the doublet sign — it is recorded honestly as a leading indicator that should not be read as "the fermionic sector will likely save stability," precisely the opposite instinct that must be resisted here.

 II.7 The tree-level cross-check, kept structurally separate from the loop verdict

 An independent, purely tree-level computation of the shape-doublet mass² (symbolic in the loop-momentum/mass parameters \(k,m\) , with the shape block evaluating to \(6k-18m\) in that parametrization) gives \(\{+1,+1\}\) on the doublet — a tree doublet saddle is structurally impossible in that computation. This is a real, reproducible result, and it is the reason the gate leans toward stability rather than confidently asserting a saddle outright. But it must never be merged with the loop-level geometric-curvature and spectral-zeta layers computed in §II.4–II.5: the tree computation and the loop (curvature + zeta) computation are answering different questions (a different point in the derivative expansion of the effective action), and only the loop-level Hessian is the physically complete stability statement once all three multiplicative factors are assembled. The honest bookkeeping is: tree leans stable; the geometric-curvature layer of the same doublet, computed independently at the curvature level, is negative. Both facts are kept visible; neither is allowed to overwrite the other.

 II.8 Assembling the verdict

 Collecting the three layers of §II.4–II.6 for the doublet block only (the singlet/breathing block is a flat volume direction, not part of the stability question):

 Geometric curvature layer: negative, three-route exact ( \(-1\) raw / \(-1/2\) unit-normalized on the raw ray; \(-1/3\) physical on the fixed-volume projection). REDUCED-TO-FLOOR, +0 . 

 Bosonic spectral-zeta layer: negative, exact rational, two-route agreement ( \(\zeta_{K_6}(-1)=-8033/100800\) ). REDUCED-TO-FLOOR, +0 . 

 Fermionic graded-Casimir sign: proven CERTIFIED-UNPINNABLE by any intrinsic invariant of the frozen shape, over a complete four-discriminator exhaustion on a closed six-candidate space. Not resolved by computation; typed as a dressed, anchor-payable IOU, banked and named, falsifiable on a stated future observable.

 Two of the three factors that determine the net Hessian sign are therefore closed exactly , both negative, agreeing across every route tried; the third is proven to be undeterminable from the shape alone , not merely unmeasured. The net minimum-vs-saddle verdict is the product of all three signs, and with the third sign genuinely unknown (not "unknown because no one computed it," but "provably absent from every available intrinsic invariant"), the honest terminal statement is: the resting point is forced (Step 1, +0 ); the block split is forced (Step 2, +0 ); two of the three Hessian-layer signs are computed exactly and are both negative (Steps 3–4, +0 each); the third factor is CERTIFIED-UNPINNABLE and is carried forward as a named, falsifiable, anchor-payable bit (AC-GAP10-HOLE3-v1, UNPAID), not folded into a bare claim of "minimum proven" or "saddle confirmed." This is precisely why the gate closes at RESOLVED +0 / DERIVED-GIVEN-anchor : every leg that admits a computation has been computed to the floor, and the one leg that does not admit a shape-intrinsic computation has been proven, by theorem rather than by fatigue, to require an external anchor — which is a closed, honest, typed terminal, not an open compute.

 Construction III - the central result at full precision

 III.0 What this section delivers

 This section carries the single heaviest load in the SG-6 dossier: the exact computation of both computable Hessian-sign layers of the \(K_6\) shape-doublet modulus at the symmetric chamber center, cross-checked by independent routes and by a negative control, followed by the full four-discriminator exhaustion theorem that types the one remaining fermionic sign as CERTIFIED-UNPINNABLE(C) rather than merely unfinished. Every number below is either an exact rational reproduced from the frozen geometry pack, or a derivation shown in full from those exact rationals. Nothing here uses any number not already fixed by the four irreducible anchors \(\{M_{\rm Pl},\alpha_i(M_Z),y_t,|V_{us}|\}\) or by pure representation theory / topology of the frozen shape.

 Fixed grade for this gate, stated once and never varied: DERIVED-GIVEN-anchor / RESOLVED, +0 . 

 III.1 The object, pinned at all three layers

 The computation lives on the complete frozen active branch 
$$
\mathfrak B_{\rm active}
=
\underbrace{\big[\,\mathcal M_4\times K_6\times S^2\times S^1_Y/\mathbb Z_2\,\big] \times} {\times\ \text{Stage — metric geometry}}
\ \oplus\
\underbrace{\big[\,\mathcal F^+ {\rm finite}\oplus\mathcal C {\rm admiss}\,\big] \oplus} {\oplus\ \text{Rulebook — finite admissibility}}
\ \otimes\
\underbrace{\big[\,\mathcal E_{\rm matter}\oplus\mathcal E_{\rm gauge}\oplus\mathcal E_{\rm Higgs}\oplus\mathcal E_{\rm proton}\,\big] \otimes} {\otimes\ \text{Actors — bundles/operators}},
$$

 with \(K_6=SU(3)/T^2\) the full \(A_2\) flag manifold and total dimension \(D=4+6+2+1=13\) . SG-6's computation is entirely a statement about the \(\times\) Stage shape moduli of \(K_6\) , read against the \(\oplus\) Rulebook admissibility chamber and the \(\otimes\) Actors twisted spin- \(\mathbb C\) data; \(\mathcal M_4\) , \(S^2\) , and \(S^1_Y/\mathbb Z_2\) are present as the ambient background (they fix the overall Planck normalization \(M_{\rm Pl}^2=M_*^{11}\,{\rm Vol}(X_{\rm active})\) against which any physical mass² is ultimately measured) but do not themselves carry the modulus in question.

 \(\times\) Stage — the metric object. \(K_6\) carries the \(SU(3)\) -left-invariant metric

 \[
g_{K_6}(\vec u)=\sum_{i=1}^3 u_i\,\langle\cdot,\cdot\rangle_{\mathfrak m_i},\qquad \vec u=(u_1,u_2,u_3)\in[1/2,3/2]^3,
\]

 on the tangent decomposition \(T(K_6)=\mathfrak m_1\oplus\mathfrak m_2\oplus\mathfrak m_3\) , \(\dim_{\mathbb R}\mathfrak m_i=2\) , where \(\mathfrak m_i\) is the real 2-plane carrying the \(A_2\) root \(\alpha_i\) in the Cartan basis \((h_1,h_2,h_3)\) , \(h_1+h_2+h_3=0\) : simple roots \(\alpha_1=(1,-1,0)\) , \(\alpha_2=(0,1,-1)\) , and \(\alpha_3\equiv\alpha_1+\alpha_2=(1,0,-1)\) . The Killing form is \(B(X,Y)=6\,{\rm Tr}(XY)\) on \(\mathfrak{su}(3)\) ; the \((-B)\) -orthonormal \(\mathfrak m\) -basis is \(\{X_{ij}=E_{ij}-E_{ji},\,Y_{ij}=i(E_{ij}+E_{ji})\}/\sqrt{12}\) over the pairs \((01),(12),(02)\) . The three-parameter family \(\vec u\) is the moduli space SG-6 asks about. The chamber center is the Weyl-rigid witness \(\vec u=(1,1,1)\) ; the frozen geometry pack fixes \(R_6\equiv R_{K_6}=R_0=1.591549430918954\times10^{-17}\,{\rm GeV}^{-1}\) at that center, with \(u_{\rm chamber}=1\) .

 \(\oplus\) Rulebook — two load-bearing objects. First, \(\mathcal C_{\rm admiss}\) : the Weyl-rigid admissibility chamber \(\vec u\in[1/2,3/2]^3\) ; configurations outside it are rejected by the selector , not dynamically suppressed — a rule, not a force, and the precise reason global stabilization is never claimed (§III.7). Second, \(\mathcal F^+_{\rm finite}\) : the finite/operator chamber carrying the modular fixed point \(\tau=\omega=e^{2\pi i/3}=-0.5+0.8660254037844386\,i\) (order-3 \(PSL(2,\mathbb Z)\) fixed point), the generation basis \(\mathcal G_{\rm gen}\) of complex dimension 3 matched to the family index, and the four sector projectors \(\Pi_u,\Pi_d,\Pi_e,\Pi_\nu\) — none of these are propagating metric directions, but they fix the grading and boundary data under which the Hessian below is read.

 \(\otimes\) Actors — the operator object. The relevant operator is the twisted Dirac/Laplace-type operator on \(K_6\) whose endomorphism \(E\) and connection \(\nabla\) feed the heat-kernel/Casimir data of §III.4, together with the spin- \(\mathbb C\) determinant line carrying the family index \(\chi(K_6,E)=-3\) and its charge-conjugation action \(E\leftrightarrow E^*\) on the twisted spinor pair \(T_0\leftrightarrow T_1\) — the exact object at the center of the residual computed in §III.5–III.6.

 III.2 Step 1 — criticality of \(\bar u=(1,1,1)\) : forced by \(S_3\) , not fitted

 Claim. \(\bar u=(1,1,1)\) is a critical point, \(\nabla V(\bar u)=0\) , of every moduli potential \(V(\vec u)\) invariant under the Weyl group of \(A_2\) .

 Derivation. The Weyl group of \(A_2\) is \(S_3\) (order 6), generated by reflections in \(\alpha_1,\alpha_2,\alpha_1+\alpha_2\) . It acts on the moduli triple by permutation, \(S_3:(u_1,u_2,u_3)\mapsto(u_{\sigma(1)},u_{\sigma(2)},u_{\sigma(3)})\) , because the three root planes \(\mathfrak m_1,\mathfrak m_2,\mathfrak m_3\) are permuted by the same \(S_3\) that permutes the positive roots — this is an exact isometry of the moduli space induced by the outer structure of \(K_6=SU(3)/T^2\) , not an approximation. Consequently every covariantly-built scalar functional of \(g_{K_6}(\vec u)\) (curvature invariants, volume, any admissible potential) is automatically \(S_3\) -invariant: \(V(u_{\sigma(1)},u_{\sigma(2)},u_{\sigma(3)})=V(u_1,u_2,u_3)\) for all \(\sigma\in S_3\) .

 \(\bar u=(1,1,1)\) is the unique point on the diagonal fixed by every element of this action — order \(|S_3|=6=\chi(K_6)\) , matching the general \(G/T\) fact that the Euler characteristic of a flag manifold equals the order of its Weyl group. For any \(V\) invariant under the full permutation action, \(V(\sigma\cdot\vec u)=V(\vec u)\) for all \(\sigma\in S_3\) , so the gradient transforms covariantly, \(\nabla V(\sigma\cdot\vec u)=\sigma\cdot\nabla V(\vec u)\) ; at the fixed point this becomes \(\nabla V(\bar u)=\sigma\cdot\nabla V(\bar u)\) for every \(\sigma\) , i.e. \(\nabla V(\bar u)\) must be an \(S_3\) -invariant vector in the three-dimensional permutation representation \(\mathbb R^3\) . That representation decomposes exactly as \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) (trivial singlet plus the two-dimensional standard irrep), and the only \(S_3\) -invariant vectors in \(\mathbf 3\) lie in the trivial singlet — i.e. are proportional to \((1,1,1)\) along the flat overall-scale direction. The standard representation \(\mathbf 2\) has no nonzero vector fixed by the whole group (it is irreducible and non-trivial), so any component of \(\nabla V(\bar u)\) transverse to \((1,1,1)\) would have to vanish. Hence

 \[
\nabla_\perp V(\bar u)=0\qquad\text{for every }S_3\text{-invariant }V,
\]

 where \(\perp\) denotes the two shape directions orthogonal to the breathing direction \((1,1,1)\) . Status: DERIVED-GIVEN-anchor, +0 . No coefficient of \(V\) was chosen and no minimization was performed to land on \((1,1,1)\) ; the vanishing holds simultaneously for the entire infinite-dimensional space of \(S_3\) -invariant functionals, given only the frozen twisted geometry \(E\) with \(\chi(K_6,E)=-3\) supplying the \(S_3\) -covariant field content.

 Forbidden extension, stated once and binding throughout: criticality does not imply minimum. A critical point of an \(S_3\) -invariant function can be a minimum, a maximum, or a saddle; nothing in this step decides which. Sections III.3–III.6 show explicitly that the geometric and bosonic layers independently return "saddle," while one further layer remains open.

 Independent structural check — the four invariant Einstein metrics. The classical classification of \(SU(3)\) -invariant Einstein metrics on \(SU(3)/T^2\) , reproduced independently here (not merely quoted), finds exactly four solutions of \({\rm Ric}_1={\rm Ric}_2={\rm Ric}_3\) using the general-chamber Ricci formula (Killing-form scales \(x_1,x_2,x_3\) on \(\mathfrak m_1,\mathfrak m_2,\mathfrak m_3\) ; the full formulas are given in §III.3): the normal metric \((1,1,1)\) , and the Kähler–Einstein metric \((1,1,2)\) together with its two further permutations \((1,2,1)\) , \((2,1,1)\) — a system with exactly four positive real solutions up to overall scale, no fifth. Off the \((1,1,1)\) locus the space is non-Einstein (the squashing degree of freedom). The existence of this second family of Einstein points is the structural reason the fully symmetric point, though forced to be a critical point of every \(S_3\) -invariant functional, need not automatically be the energetically preferred extremum among the four — exactly the question the doublet-sector Hessian settles next.

 III.3 Step 2 — the exact Hessian split \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) 

 Claim. The \(3\times3\) Hessian of any \(S_3\) -invariant \(V\) at \(\bar u=(1,1,1)\) splits exactly, by representation theory alone, into a one-dimensional breathing singlet and a two-dimensional traceless shape doublet, and the two sectors decouple at quadratic order.

 Derivation. The tangent space at \(\bar u\) is \(\mathbb R^3\) with \(S_3\) acting by coordinate permutation — the standard permutation representation, already used in §III.2. It decomposes uniquely as \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) : the trivial representation \(\mathbf 1\) , spanned by the totally symmetric breathing direction \((1,1,1)/\sqrt3\) (overall rescaling of the shape, \(u_1=u_2=u_3=t\) ), and the two-dimensional standard representation \(\mathbf 2\) , spanned by traceless combinations such as \(u_1-u_2\) and \(u_1+u_2-2u_3\) (the shape doublet, moving the relative squashing while holding the singlet combination fixed). This is identical to the decomposition organizing the four Einstein points of §III.2: the singlet interpolates the overall radius, the doublet interpolates toward the three \((1,1,2)\) -type Kähler–Einstein points.

 By Schur's lemma applied to the \(S_3\) -equivariant Hessian bilinear form: because \(\mathbf 1\) and \(\mathbf 2\) are inequivalent irreducible representations of \(S_3\) , any \(S_3\) -invariant symmetric bilinear form on \(\mathbf 1\oplus\mathbf 2\) — which is exactly what the Hessian of an \(S_3\) -invariant \(V\) at the fixed point is — has identically zero off-block components between the singlet and doublet sectors, for every admissible \(V\) , exactly, not approximately. The stability question therefore cleaves cleanly into two logically independent sub-questions — is the breathing (singlet) direction stable, and is the shape (doublet) direction stable — answerable separately, with no cross-contamination between them.

 Status: DERIVED-GIVEN-anchor, symmetry-exact, +0 . This licenses treating the doublet-sector curvature computed next as a clean, isolated, target-blind number, uncoupled from the separately-tracked breathing-sector residual (R5, §III.8).

 III.4 Step 3 — the geometric curvature layer of the doublet sector

 This is the first of the two computable Hessian layers, REDUCED-TO-FLOOR at +0 .

 The exact general-chamber curvature formulas (Wang–Ziller/Nomizu), Killing-form scales \(x_1,x_2,x_3\) on \(\mathfrak m_1,\mathfrak m_2,\mathfrak m_3\) : 

 \[
{\rm Ric}_1=\frac{x_1^2-x_2^2+6x_2x_3-x_3^2}{12\,x_1x_2x_3},\quad
{\rm Ric}_2=\frac{-x_1^2+6x_1x_3+x_2^2-x_3^2}{12\,x_1x_2x_3},\quad
{\rm Ric}_3=\frac{-x_1^2+6x_1x_2-x_2^2+x_3^2}{12\,x_1x_2x_3},
$$
$$
{\rm Scal}(\vec x)=\frac{x_1x_2+x_1x_3+x_2x_3-(x_1^2+x_2^2+x_3^2)/6}{x_1x_2x_3}.
\]

 At the symmetric center \(\vec x=(1,1,1)\) (the Killing-normal-metric normalization) these reduce to the exact rationals fixed in the geometry pack: \({\rm Ric}_1={\rm Ric}_2={\rm Ric}_3=5/12\) , \({\rm Scal}=5/2\) , so \({\rm Scal}/{\rm Ric}_i=6=\dim K_6\) — the scale-invariant bridge between the Killing-normal and physical \(R_6\) normalizations. The companion exact-rational invariants at this same center: \({\rm Scal}^2=25/4\) , \(\|{\rm Ric}\|^2=25/24\) , \(\|{\rm Riem}\|^2=23/12\) , and the load-bearing ratio

 \[
\frac{\|{\rm Riem}\|^2}{{\rm Scal}^2}=\frac{23}{75}=0.3066666666666667,
\]

 certified never \(31/147\) (a retired branch-kill contaminant) and \(\|{\rm Riem}\|^2\) certified never \(=60\) (the distinct round- \(S^6\) value). \(\|{\rm Ric}\|^2/{\rm Scal}^2=1/6\) . These ratios are metric-scale invariant and identical whether read in the Killing-normal or the physical \(R_6\) normalization.

 The doublet-sector scalar-curvature functional. The directly computable object controlling the doublet-ray curvature, in the model-scalar convention used for the shape-doublet stability test at unit radius \(R_6=1\) , is

 \[
R(\vec u)=\sum_i\frac1{u_i}-\frac12\,T(\vec u),\qquad T(\vec u)=\sum_k\frac{u_k}{u_iu_j}\quad(i,j,k\ {\rm cyclic}),
\]

 equivalently \({\rm Scal}(\vec u)=\big[u_1u_2+u_1u_3+u_2u_3-\tfrac16(u_1^2+u_2^2+u_3^2)\big]/(u_1u_2u_3)\) , the same object in the corpus's alternate bookkeeping. At the center \(R(1,1,1)=3-\tfrac12\cdot3=\tfrac32\) — consistent with the frozen \({\rm Scal}=5/2\) up to the fixed rescaling between the two recorded normalizations (never mixed within one computation). Evaluate \(R(\vec u)\) along the pure shape-doublet ray, holding the singlet combination fixed:

 \[
\vec u(\varepsilon)=(1+\varepsilon,\ 1-\varepsilon,\ 1).
\]

 Term 1 — the diagonal ("convexity of \(1/u\) ") piece. Expanding \(1/(1+\varepsilon)=1-\varepsilon+\varepsilon^2-\cdots\) and \(1/(1-\varepsilon)=1+\varepsilon+\varepsilon^2+\cdots\) :

 \[
\frac1{1+\varepsilon}+\frac1{1-\varepsilon}+\frac11=(1-\varepsilon+\varepsilon^2)+(1+\varepsilon+\varepsilon^2)+1+O(\varepsilon^3)=3+2\varepsilon^2+O(\varepsilon^4).
\]

 The \(\sum1/u_i\) piece contributes \(+2\) to the raw \(\varepsilon^2\) coefficient (unit-normalized convention); equivalently \(+4\) to the raw second derivative \(d^2/d\varepsilon^2\) (since a Taylor coefficient of \(\varepsilon^2\) equals \(\tfrac12f''(0)\) ). This is a direct consequence of the strict convexity of \(x\mapsto1/x\) ( \(d^2(1/x)/dx^2=2/x^3>0\) ): a symmetric \(\pm\varepsilon\) perturbation of two of the three \(u_i\) always raises \(\sum1/u_i\) — the diagonal term always stabilizes on its own.

 Term 2 — the structure-constant triangle (" \(T(\vec u)\) ") piece. Along the same ray,

 \[
T(\varepsilon)=\frac{u_3}{u_1u_2}+\frac{u_1}{u_2u_3}+\frac{u_2}{u_3u_1}=\frac1{(1+\varepsilon)(1-\varepsilon)}+\frac{1+\varepsilon}{1-\varepsilon}+\frac{1-\varepsilon}{1+\varepsilon}.
\]

 Expanding each piece to \(O(\varepsilon^2)\) : \(1/(1-\varepsilon^2)=1+\varepsilon^2+O(\varepsilon^4)\) ; \((1+\varepsilon)/(1-\varepsilon)=(1+\varepsilon)(1+\varepsilon+\varepsilon^2+\cdots)=1+2\varepsilon+2\varepsilon^2+O(\varepsilon^3)\) ; \((1-\varepsilon)/(1+\varepsilon)=(1-\varepsilon)(1-\varepsilon+\varepsilon^2-\cdots)=1-2\varepsilon+2\varepsilon^2+O(\varepsilon^3)\) . Summing:

 \[
T(\varepsilon)=\big[1+\varepsilon^2\big]+\big[1+2\varepsilon+2\varepsilon^2\big]+\big[1-2\varepsilon+2\varepsilon^2\big]+O(\varepsilon^3)=3+5\varepsilon^2+O(\varepsilon^3).
\]

 The linear-in- \(\varepsilon\) terms cancel exactly — a live cross-check that this ray computation is consistent with the criticality theorem of §III.2 — leaving a raw \(\varepsilon^2\) coefficient of \(+5\) in \(T(\vec u)\) itself. In the prefactor \(-\tfrac12T(\vec u)\) this becomes a contribution of \(-5\) to the raw second-derivative bookkeeping (equivalently \(-5/2\) unit-normalized), the term encoding the \(A_2\) structure-constant cross-coupling of the three root planes.

 Assembling \(R(\varepsilon)\) . 

 \[
R(\varepsilon)=\big[3+2\varepsilon^2\big]-\frac12\big[3+5\varepsilon^2\big]+O(\varepsilon^3)=\Big(3-\frac32\Big)+\Big(2-\frac52\Big)\varepsilon^2+O(\varepsilon^3)=\frac32-\frac12\varepsilon^2+O(\varepsilon^3),
\]

 \[
\boxed{R(\varepsilon)=\frac32-\frac{\varepsilon^2}{2}+O(\varepsilon^4)\quad(\text{unit-normalized}).}
\]

 This matches \(R(1,1,1)=3/2\) exactly at \(\varepsilon=0\) and confirms the criticality theorem explicitly: the linear term vanishes.

 Total, both conventions, without ambiguity. 

 \[
\text{raw:}\quad \left.\frac{d^2R}{d\varepsilon^2}\right|_0=(+4)+(-5)=-1,\qquad
\text{unit-normalized:}\quad \left.\frac12\frac{d^2R}{d\varepsilon^2}\right|_0=(+2)+\left(-\frac52\right)=-\frac12.
\]

 Convention guard (verbatim, load-bearing). The unit-normalized pieces \(+2\) and \(-5/2\) sum to \(-1/2\) ; the raw-ray pieces \(+4\) and \(-5\) sum to \(-1\) . Writing " \(+2-5/2=-1\) " mixes the two conventions and is arithmetically wrong. The two internally consistent statements are \(+4-5=-1\) (raw) and \(+2-5/2=-1/2\) (unit-normalized). The sign — negative — is identical in both and is the only convention-invariant, physically meaningful content of this number: the structure-constant triangle term overwhelms the diagonal convexity term, so \(d^2R/d\varepsilon^2|_0<0\) , a saddle in the purely geometric curvature functional, direction- and normalization-invariant.

 Independent cross-check: the full \(3\times3\) Hessian. Re-deriving the same object as the full Ricci-Hessian of the \(R\) -model at \((1,1,1)\) (direct numerical diagonalization) gives the exact matrix

 \[
H=\begin{pmatrix}-1&2&2\\2&-1&2\\2&2&-1\end{pmatrix},\qquad \text{eigenvalues } \{3\ (\times1),\ -3\ (\times2)\},\qquad R(1,1,1)=\frac32.
\]

 The singlet eigenvalue \(+3\) and the doublet eigenvalues \(\{-3,-3\}\) reproduce, up to overall normalization, exactly the singlet/doublet split forced by representation theory in §III.3, and the doublet block is negative-definite — the same saddle conclusion as the ray computation, obtained by an independent route (full matrix diagonalization rather than a single-ray Taylor expansion). This also matches the closed-form \(\kappa=1/6\) curvature/radion normalization and the DeWitt moduli-metric radion slope \(\lambda_K^2=8/3=2(d+2)/d\) at \(d=6\) used elsewhere in the pipeline.

 Independent cross-check: the fixed-volume physical projection. The volume-preserving ( \({\rm det}=1\) ) parametrization of the same doublet, \(S=\sum_i 1/x_i-\tfrac16\sum_i x_i/(x_jx_k)\) , gives

 \[
\left.\frac{d^2S}{d\varepsilon^2}\right|_{\rm fixed\ vol}=+\frac23,
\]

 and the log-coordinate Hessian of \(S\cdot{\rm Vol}^{1/3}\) has eigenvalues \(\{1/3\ (\times2),\ 0\ (\times1)\}\) . Standard flux-free KK reduction carries the physical potential as \(V\sim-S|_{\rm fixed\ vol}\) (positive volume prefactor), so the physical shape mass² is

 \[
m^2_{\rm shape}=-\frac13<0,
\]

 again a saddle. Reconciliation of the three numbers \(-1\) , \(-1/2\) , \(+1/3\) : \(-1\) (raw) and \(-1/2\) (unit-normalized) are the second derivative of \(R\) along the raw doublet ray; \(+1/3\) is the second derivative of \(S\) at strictly fixed volume — and because \(V\sim-S\) , that \(+1/3\) in \(S\) becomes \(-1/3\) in the physical potential. Every parametrization — raw ray, full matrix diagonalization, fixed-volume physical projection — returns physical shape mass² \(<0\) (saddle) at the geometric/tree-curvature level. This matches the classical fact that the normal Einstein metric \((1,1,1)\) on \(SU(3)/T^2\) is a saddle among the four invariant Einstein metrics; the stable ones are the Kähler–Einstein points \((1,1,2)\) and permutations.

 Negative control (specificity check). The identical curvature-functional procedure, run on an \(S^2\times S^2\) control in place of \(K_6\) (a different compact factor with its own doublet-like modulus, per-sphere model \(R=2/b^2\) ), fires the opposite sign : the reproducible script value for the analogous full Hessian eigenvalue is \(+12\) , sharply positive. (A distinct \(S^2\times S^2\) normalization appearing elsewhere in the narrative quotes \(+16\) ; the reproducible script gives \(+12\) , and the load-bearing content in either case is that the control fires the opposite, positive sign.) Because this test correctly distinguishes two different manifolds with opposite verdicts, the \(K_6\) result of \(-1\) (raw) / \(-1/2\) (unit) is credentialed as a genuine geometric fact about \(SU(3)/T^2\) specifically, not an artifact of the computational procedure.

 Companion frozen data used implicitly above. \(\kappa=1/6\) ; Euler characteristic \(\chi(K_6)=6\) (topological); \(\|\nabla{\rm Riem}\|^2=1/4\ne0\) , confirming \(K_6\) is homogeneous but not locally symmetric — the structural reason a nontrivial doublet curvature term exists at all.

 Status: REDUCED-TO-FLOOR / DISSOLVED-GIVEN-Shape, +0 . Nothing remains to compute in this layer: the sign is exact, reproduced by three independent routes (ray expansion, full-matrix diagonalization, fixed-volume physical projection), and stress-tested against a control that returns the opposite sign. What this layer is not : it is not yet the certified physical mass² verdict, because two further multiplicative layers (bosonic zeta, §III.5, and fermionic supertrace, §III.6) remain.

 Retired/forbidden reading (binding). The bare " \(-1\) as the settled shape eigenvalue" single-slice reading is a truncated-Shape artifact that mixed the criticality-framing with the fixed-volume framing; it is retired. The correct statement is: raw \(=-1\) , unit-normalized \(=-1/2\) , fixed-volume-physical \(=-1/3\) — three conventions of the same saddle, never combined into one number, sign convention-invariant throughout.

 III.5 Step 4 — the bosonic spectral-zeta layer

 This is the second computable Hessian layer, also REDUCED-TO-FLOOR at +0 .

 Definition. The bosonic one-loop contribution to the effective moduli potential is governed by the spectral zeta function of the \(K_6\) Laplace-type operator, continued to \(s=-1\) (standard heat-kernel/zeta regularization: \(V_{\rm 1\text{-}loop}\propto\zeta_{K_6}(-1)\) up to an overall positive normalization that does not affect the sign).

 Value, exact rational: 

 \[
\zeta_{K_6}(-1)=-\frac{8033}{100800}=-0.0796924603174603\ldots,
\]

 with \(8033=29\cdot277\) and \(100800=2^6\cdot3^2\cdot5^2\cdot7\) (coprime, confirming the fraction is in lowest terms). This is negative. Companion frozen values: \(\zeta_{K_6}(0)=-253/315\) ; heat-kernel coefficients \(c_1=8033/100800\) (feeding \(\zeta_{K_6}(-1)\) directly), \(c_2=5743/184800\) (feeding \(\zeta_{K_6}(-2)=5743/92400\) ).

 Two independent routes, agreeing to ~22 digits. 
- Route A — a fresh, target-blind numeric heat-sum over the \(K_6\) Peter–Weyl spectrum, using the exact Casimir eigenvalues \(C_2(p,q)=(p^2+q^2+pq+3p+3q)/3\) and dimensions \(\dim(p,q)=(p+1)(q+1)(p+q+2)/2\) (e.g. \((1,1)\) adjoint: \(\dim=8\) , \(C_2=3\) ; \((2,2)\) : \(\dim=27\) , \(C_2=8\) ; \((3,3)\) : \(\dim=64\) , \(C_2=15\) ), summed over the Peter–Weyl decomposition of the scalar Laplacian spectrum, with Richardson extrapolation removing the truncation error of the infinite mode sum.
- Route B — the exact closed form via Weyl-character/Poisson resummation on the \(A_2\) root lattice ( \(SU(3)/T^2\) ), reducing the same spectral sum analytically to the rational above.

 Both routes are run independently and blind to each other's target answer; agreement to approximately 22 significant digits between an independent numerical route and an independent closed-form route is far beyond what an accidental coincidence could produce and constitutes an exact-value certification, not merely a plausibility check. The computation is target-blind: no comparison to \(v_{\rm EW}\) , \(\Lambda\) , or any other observable enters \(\zeta_{K_6}(-1)\) . As a further internal consistency anchor, this value sits alongside the exact rational curvature invariants of §III.4 ( \(\dim K_6=6\) , \({\rm Ric}_i=5/12\) , \({\rm Scal}=5/2\) , \({\rm Scal}^2=25/4\) , \(\|{\rm Ric}\|^2=25/24\) , \(\|{\rm Riem}\|^2=23/12\) , \(\|{\rm Riem}\|^2/{\rm Scal}^2=23/75\) ) — none of which contradicts or requires revision of \(\zeta_{K_6}(-1)=-8033/100800\) .

 A candor note that is not itself a physics finding. An older auxiliary script independently hit a \(\Gamma\) -function pole ( ValueError ) from a naive division by \(\Gamma(s)\) at a negative-integer pole — a bug in that specific old implementation, not a defect in the banked value. This is logged for transparency because the honesty standard for this dossier requires surfacing such things; it changes no banked number and is not evidence against \(\zeta_{K_6}(-1)=-8033/100800\) .

 Status: REDUCED-TO-FLOOR, +0 . Combined with §III.4, both independently computable layers of the one-loop Hessian sign are negative. This reframes the remaining question precisely: not "is the sign known" for two of the three layers, but "does the third, fermionic layer overturn two already-negative layers."

 III.6 Step 5 — the fermionic layer: the CERTIFIED-UNPINNABLE(C) theorem in full

 The multiplicative structure. The net physical Hessian sign for the shape doublet assembles as

 \[
({\rm sign\ of\ physical\ Hessian})\ \sim\ (\text{geometric curvature sign})\times(\text{bosonic-zeta sign})\times(\text{fermionic graded-Casimir supertrace sign}),
\]

 with the first two factors pinned negative in §III.4–III.5. The third factor is the graded (bose \(-\) fermi) Casimir supertrace on the twisted spinor pair \(T_0\leftrightarrow T_1\) , the two twist sectors exchanged by the endomorphism duality \(E\leftrightarrow E^*\) — i.e. exactly the action of charge conjugation on the twisted spin- \(\mathbb C\) bundle over \(K_6\) . This is a single discrete \(\pm\) datum — not a continuous parameter, not a magnitude to be computed to some precision, but one bit: the sign of a supertrace \(\sum_{\rm modes}(-1)^F(\text{Casimir weight})\) over a finite, closed candidate set (below). It is \(C\) -odd by construction : swapping \(T_0\leftrightarrow T_1\) is the \(E\leftrightarrow E^*\) duality that defines charge conjugation on this bundle, and the fermionic \((-1)^F\) weighting flips sign under that exchange while the already-accounted bosonic contributions do not.

 The candidate space. The physically relevant twist labels are the six \(\chi=-3\) Dynkin-label twist candidates — the complete, exhaustively enumerable set of line-bundle twists compatible with the frozen spin- \(\mathbb C\) family index \(\chi(K_6,E)=-3\) — closed under two operations: charge conjugation (the \(T_0\leftrightarrow T_1\) exchange itself) and triality (the genuine outer \(\mathbb Z_3\) automorphism of \(\mathfrak{su}(3)\) permuting the three fundamental-weight directions). Because conjugation is the \(\mathbb Z_2\) exchanging each representation with its dual and triality is a further finite outer automorphism, the orbit structure of this six-element set under \(\langle C,\,{\rm triality}\rangle\) is fully determined and finite — no continuous modulus, no hidden infinite-precision datum. This finiteness is itself a Granularity-root fact: the candidate space satisfies the cost-floor/finite-enumeration discipline exactly (§III.8).

 Route-B exhaustion, target-blind, reproduced identically on fresh re-run: 

 Discriminator 
 Test performed 
 Verdict 

 D1 — Wu/Stiefel–Whitney ( \(w_2,w_3\) ) orientation 
 Built entirely from the tangent bundle \(TK_6\) , on which charge conjugation acts trivially — tangent data cannot see the \(E\leftrightarrow E^*\) duality at all. Evaluated on the frozen shape, the relevant Wu/SW classes are identically zero and are, in any case, tangential and hence \(C\) -even by construction. 
 CANNOT PIN — a vanishing, tangential, \(C\) -even datum carries no information about a \(C\) -odd bit. 

 D2 — measure-reality / CPT 
 Governed by the Frobenius–Schur indicator of \(\mathbf 3\) of \(SU(3)\) : \({\rm FS}(\mathbf3)=0\) exactly (the fundamental is genuinely complex, neither real nor pseudoreal), cross-checked by an independent numeric Weyl-integral evaluation of the same indicator agreeing with the exact algebraic value to \(\sim10^{-16}\) — i.e. no ambiguity in \({\rm FS}(\mathbf3)=0\) itself. Because \({\rm FS}(\mathbf3)=0\) , the reality condition that could in principle single out a preferred member of a conjugate pair is empty here. Separately, CPT together with the color-charge-conjugation map acts on \((T_0,T_1)\) as the vectorlike \(SU(3)_c\) exchange, constraining the pair as a whole (their sum/product is fixed) but supplying no operation that could split the pair. 
 CANNOT PIN — the only available discriminator constrains the pair, not the individual sign. 

 D3 — Dai–Freed anomaly beyond \(\eta\) 
 The full Dai–Freed anomaly datum, evaluated as a whole, is itself constructed to conjugate covariantly — \(C\) -even as an invariant of the pair; its finer mod-2 refinement, the only piece that could in principle carry additional discriminating power, is undefined exactly at the point \({\rm FS}(\mathbf3)=0\) found in D2 (the refinement requires a nonzero reality pairing to be well-defined, and here that pairing vanishes). 
 CANNOT PIN — the raw invariant is \(C\) -even by construction and its refinement is undefined here. 

 D4 — spin- \(\mathbb C\) determinant-line reality 
 A determinant-line reality condition could pin the sign only for a self-conjugate twist (one fixed by \(C\) ). Explicit enumeration of the closed six-candidate set shows none of the six is self-conjugate : the free \(C\) -action exchanges the two triality-related triples of twists in their entirety, with no fixed point. 
 CANNOT PIN — the discriminator is structurally empty; there is no self-conjugate object for it to act on. 

 The logical structure of the elimination. D1–D4 jointly exhaust every intrinsic topological (D1), reality/CPT (D2), global-anomaly (D3), and determinant-line-reality (D4) datum available on the frozen shape — four structurally distinct types of invariant, with no fifth type left unaccounted for, and the case analysis covers the complete, closed, six-member candidate space with no residual case unexamined. Each of the four is shown, independently and for a structurally different reason, to be either identically \(C\) -even (D1, D3) or empty/non-discriminating on this specific candidate space (D2, D4). Since the sought supertrace sign is \(C\) -odd by construction — it is defined as the sign that flips under the exact exchange \(T_0\leftrightarrow T_1\) that is charge conjugation here — and a \(C\) -even quantity is, as a matter of representation theory, invariant under exactly the transformation the target is defined to be odd under (pairing a \(C\) -even quantity against a \(C\) -odd target always returns "cannot distinguish," for the same reason that no function even in \(x\) can determine the sign of \(x\) itself), the conclusion is not that a search has failed, but that no discriminator built from intrinsic data of the frozen shape can exist for this specific bit , full stop — now or under any future re-analysis of the same shape data.

 ⇒ RULE-FORCED-NEGATIVE: a complete, forced elimination over a provably closed six-candidate class against all four available discriminator classes — theorem-grade completeness, not fatigue. Adding more computational effort to the same class of shape-intrinsic invariants cannot, even in principle, change the verdict, because the obstruction is a parity mismatch ( \(C\) -even tool vs. \(C\) -odd target), not insufficient precision or an unexplored corner of the candidate space.

 Endpoint typing (NO-BARE-#5). The underlying observable (the physical Hessian's minimum-vs-saddle verdict) does not dissolve: it remains a perfectly well-posed, finite \(\pm\) question — a real physical system with this geometry does have some definite Hessian sign — in principle settleable by an external measurement. Because the observable survives restatement, this is explicitly not a dissolution; it routes instead to the typed endpoint DRESSED-#5 — a time-indexed IOU, anchor-certified-payable-on-measurement — carried under the standing certificate AC-GAP10-HOLE3-v1 , shared identically across sg6, uqf10 (its fermion Casimir-sign leg), and gap10-bg10's Hole #3 accounting (the identical \(T_0/T_1\) \(C\) -odd bit also governs the neutrino Dirac-mass-normalization/ \(M_R\) joint unknown there), counted once across all three, per the cross-wall discipline (SAG-A6-KEYSTONE), status UNPAID . The full five-field payload is present: certified class = the six \(\chi=-3\) twists; completeness basis = D1–D4 exhaustion plus \({\rm FS}(\mathbf3)=0\) plus the Weyl-integral numeric plus Dai–Freed conjugation; basis conditionality = unconditional exact; enumerated exits = all four, by theorem; conditionality = frozen branch.

 The falsifier, wired explicitly. The bit becomes decidable the moment an admissible \(C\) -odd measured record exists. The leading candidate payer is the decisive leptonic \(\delta_{CP}\) sign (or any future admissible \(C\) -odd record of the same type). The baryon asymmetry \(\eta_B\) is permanently barred as a payer (W12) — an independently established categorical exclusion, not re-derived here. Once a payer fires, the product of the three layers (geometric \(\times\) zeta \(\times\) fermionic) is fully determined and the verdict resolves to either a genuine DERIVED-GIVEN-E minimum or a CLOSED-NEGATIVE confirmed saddle — both are honest terminals; given that both already-computed layers are negative and the leading loop indicator (below) actively opposes a rescue, a confirmed saddle is the single most likely honest expectation, stated as an expectation, not an achieved result.

 Leading loop indicator (not itself the closing computation, but informative). \({\rm Str}[C_2]=\chi\cdot C_2({\rm fund})=-4\) at \(\chi=-3\) : this signed trace opposes any hoped-for boson-dominance rescue of the doublet sign, and should not be treated as a hint that the sign will flip favorably once the full multiplicity table (R6, §III.8) is mounted.

 Three-sins self-audit (all NOT committed). Anchor-elimination: NO — the bit is correctly left as a live external anchor requirement, not algebraically "solved away" to look derived when it is not. Target-anchoring: NO — the entire Route-B exhaustion (D1–D4) is run target-blind; no value or sign of \(\delta_{CP}\) , \(\eta_B\) , or any other observable is assumed anywhere in the exhaustion logic. False-flooring: NO — the floor rests on an exact, reproduced, closed-space elimination theorem across four independently-typed discriminator classes, not on search fatigue or a stalled effort.

 III.7 Independent cross-checks (re-executed fresh, reproduced exactly) and the tree-level bookkeeping note

 Three independent computational cross-checks reproduce the results above exactly, providing the target-blind, reproducible backbone for the central result:

 \(K_6\) -model Ricci-Hessian eigendecomposition at \(\bar u=(1,1,1)\) : direct numerical diagonalization of the \(3\times3\) Ricci-Hessian returns eigenvalues \(\{3\ (\times1),\,-3\ (\times2)\}\) , \({\rm Scal}(1,1,1)=3/2\) , structurally consistent with the ray-computation sign of §III.4 (one positive breathing eigenvalue, a doubly-degenerate negative shape-doublet eigenvalue).

 \(T_0/T_1\) Route-B exhaustion: independently confirms all four D1–D4 verdicts return "CANNOT PIN"; confirms \({\rm FS}(\mathbf3)=0\) both algebraically and numerically to \(\sim10^{-16}\) ; confirms the tangential mod-2 dataset (D1) is identically zero; confirms by direct enumeration that no self-conjugate twist exists among the six candidates (D4); confirms the exhaustion logic never references any measured \(C\) -odd observable.

 \(\zeta_{K_6}(-1)=-8033/100800\) cross-check: re-verified against the independent two-route certificate (heat-sum/Richardson vs. Weyl/Poisson closed form, agreeing to ~22 digits), fully consistent with no re-fabrication of the value.

 All three cross-checks are mutually consistent and none required any adjustment to a previously banked number.

 The tree-level cross-check, kept strictly separate. An independent tree-level computation of the shape-doublet mass² returns

 \[
m^2_{\rm doublet,\ tree}=\{+1,\ +1\}
\]

 (symbolic form: shape block \(=6k-18m\) in the notation of the durable tree-check script) — both eigenvalues positive, so a doublet saddle is structurally excluded at tree order. This is the concrete reason the gate is described as leaning stable. It must never be merged with the loop-level curvature result of §III.4: the same doublet direction's target-blind, loop-relevant geometric curvature sign is \(-1\) (raw) / \(-1/2\) (unit-normalized), computed with no reference to the tree-level mass term. Two honest, distinct-order numbers, never combined into one overstated claim: tree \(=\{+1,+1\}\) is banked and supports optimism; loop-level curvature \(=-1\) is the input feeding directly into the still-open one-loop verdict of §III.6.

 III.8 Deep-root anchoring: Shape, Scale, Granularity — no truncation

 Shape (complete object, flag NONE). Supplies everything used above: the \(K_6=SU(3)/T^2\) moduli space and its exact \(S_3\) symmetry (×Stage); the twisted Dirac operator, the six \(\chi=-3\) Dynkin-label twist candidates, the spin- \(\mathbb C\) determinant line, and Frobenius–Schur \({\rm FS}(3)=0\) (⊗Actors, exact algebra plus independent Weyl-integral numeric agreeing to \(\sim10^{-16}\) ); the conjugation/triality orbit action on the candidate set (⊕Rulebook). An earlier reading of a related single-slice sign (the retired " \(-1\) as the settled shape eigenvalue") used a truncated Shape object — bare geometry only, without the full conjugation/triality-closed candidate-class structure — and is retired as an artifact of that truncation. The CERTIFIED-UNPINNABLE(C) result in §III.6 depends on completeness: an incomplete candidate class could in principle have concealed an accidental discriminator; the closed, fully enumerated class cannot.

 Scale. The net stability sign is explicitly scale-free — a discrete \(\pm\) datum with no mass dimension, so \(M_{\rm Pl}\) is not load-bearing for this specific residual. The two genuinely scale-carrying objects touched by SG-6 are \(\mu_{\rm cell}\) (breathing-singlet sector, R5) and the second ruler \(v_{\rm EW}=246\,{\rm GeV}\) (R1); the Buckingham- \(\pi\) no-go — no second parametrically-small mass can be built from \(\{M_{\rm Pl},\hbar,\text{dimensionless frozen geometry}\}\) without an independent \(\sigma\) -carrying length — is itself a Scale-root argument, and it is what forbids anchoring \(\mu_{\rm cell}\) non-circularly at the Hosotani stationarity condition \(\partial_\sigma V=0\) (which is the electroweak hierarchy itself).

 Granularity. The Uniform Operational Cell posit ( \(N\le B/\Delta_0\) ) is what certifies the six-label candidate space of §III.6 as legitimately finite and closed rather than open-ended: six labels, closed under two named finite-group operations (conjugation, triality) — a provably finite object, which is the structural reason CERTIFIED-UNPINNABLE(C) is a genuine completeness theorem rather than "we searched hard and found nothing." Separately, \(\mu_{\rm cell}\) inherits existence from this same posit (on the footing of \(\hbar\) ); its value is a floor-residue not supplied here (R5, Construction IV).

 Layer-2 screens (Tier B), all admissible. Invariance — PASS : \(\kappa=1/6\) , the \(+1/3\cdot I_2\) trace piece, and \(\zeta_{K_6}(-1)=-8033/100800\) are all branch-preserving and multi-route reproduced; the screen correctly reports that the fermion \(C\) -odd bit lies outside the \(C\) -even invariant ring generated by Shape data alone — this is Invariance functioning correctly, not failing. Record Interface — CONSTRAIN , not dissolve: the residual restates as a finite, well-posed exterior observable (the \(C\) -odd sign of the graded-Casimir supertrace), so the OBSERVABLE-NEVER-DISSOLVES rule fires — an un-pinnable target that survives restatement as a legitimate finite observable never silently dissolves. Causal Order — PASS : Route B ran target-blind, with no CP value assumed anywhere in the exhaustion. Nonseparability — CONSTRAIN : the T0/T1 bit is explicitly shared, not independently arising, across uqf10, sg6, and gap10-bg10 Hole #3 — counted once (SAG-A6-KEYSTONE), not three times, in cross-wall accounting.

 III.9 The assembled verdict, stated at the fixed grade

 Layer 
 Object 
 Value / sign 
 Status 

 Location 
 criticality at \(\bar u=(1,1,1)\) 
 \(\nabla_\perp V(\bar u)=0\) , forced by \(S_3\) 
 DERIVED-GIVEN-anchor, +0 

 Hessian split 
 \(\mathbf3=\mathbf1\oplus\mathbf2\) 
 Singlet/doublet decouple exactly (Schur's lemma) 
 DERIVED-GIVEN-anchor, symmetry-exact, +0 

 Geometric curvature (doublet) 
 $d^2R/d\varepsilon^2 
 _0$ 
 \(-1\) (raw) \(=-1/2\) (unit-normalized); fixed-vol \(+1/3\to-1/3\) physical 

 \(S^2\times S^2\) control 
 full Hessian eigenvalue 
 \(+12\) (script; opposite sign) 
 Negative control, confirms non-triviality 

 Bosonic zeta 
 \(\zeta_{K_6}(-1)\) 
 \(-8033/100800\) (negative) 
 REDUCED-TO-FLOOR, +0 

 Tree-level doublet mass² 
 \(m^2_{\rm doublet,\,tree}\) 
 \(\{+1,+1\}\) 
 Banked; distinct order, kept separate 

 Fermionic graded-Casimir sign 
 \(T_0/T_1\) supertrace sign 
 CERTIFIED-UNPINNABLE(C) 
 DRESSED-#5, AC-GAP10-HOLE3-v1, UNPAID 

 The net physical Hessian sign is the product of all three layers; the first two are pinned and both negative, so the net verdict for the doublet sector reduces to the sign of the one outstanding fermionic bit, which this construction has shown — not merely reported — to be unpinnable by any invariant intrinsic to the frozen shape. This is the precise sense in which SG-6 stands at RESOLVED, +0 : the criticality theorem, the exact Hessian split, and both computable curvature-sign layers are genuine, exact, target-blind, +0 terminal results with nothing further owed in them, and the single remaining ingredient is not a silent gap but a theorem-grade, falsifiable, anchor-payable bet, wired to a specific future payer (the leptonic \(\delta_{CP}\) sign) and barred from being paid by an illegitimate substitute ( \(\eta_B\) ).

 What this construction does not close, by design (see the residual register for closing conditions on each): the net Casimir-weighted sign (R3/R6) that would decide minimum-vs-saddle outright; the breathing-singlet loop coefficient \(\mu_{\rm cell}\) (R5); global, all-loop, all-chamber stabilization (R2, explicitly disclosed as shared-open with every extra-dimensional program on Earth); the electroweak-hierarchy read of \(\theta_H^\star\) (R1). None of these gaps is hidden — each is a named object with a named closing condition — and none is required to license the +0 status of the results enumerated in the table above, which stand independently as exact, reproduced, cross-checked results in their own right.

 The honest bottom line. Two of the three multiplicative layers controlling the shape-doublet Hessian sign are computed exactly and agree: both are negative, i.e. both independently point toward a saddle. The third layer is not merely uncomputed — it is proven, by a complete four-discriminator exhaustion over a provably closed six-candidate class, to be intrinsically unpinnable by any invariant of the frozen shape, a genuine limit on what any framework-internal computation (the framework's or any rival's) could ever extract, not a defect of effort. That third layer is carried forward as a named, falsifiable, anchor-payable bet — decidable the instant an admissible \(C\) -odd measurement (leptonic \(\delta_{CP}\) sign) exists — rather than folded into either a false "minimum proven" or a false "gate open" reading.

 The insights that made it work

 SG-6 asks a single yes/no question — does the vacuum rest at a stable minimum, or slide away? —
but every serious attempt to answer that question in the compactification literature (string flux
vacua, M/F-theory, noncommutative-geometry models, lattice constructions) collapses two logically
independent sub-problems into one number and then argues about the number. The insight that makes
this gate close honestly, rather than dissolve into the usual moduli-stabilization hand-waving, is
to refuse that collapse from the outset. Where the shape sits and whether it stays there are
different kinds of question — one is a representation-theory statement that can be proved for
free, the other is a curvature/spectral computation that can fail — and keeping them apart is what
lets three independent pieces of machinery (a Weyl-group fixed-point theorem, a two-route
curvature computation with two mutually consistent parametrizations, and a charge-conjugation
exhaustion argument) each do exactly one job cleanly. The result is not a proof that the vacuum is
stable. It is something more useful for a target-blind theory: a proof of where the honest
uncertainty lives , pinned down to a single named bit, with every other ingredient closed at full
precision.

 1. Why criticality is free: the Weyl-group fixed-point argument

 The internal shape sector relevant to SG-6 is the space of \(SU(3)\) -invariant (Wang–Ziller/Nomizu)
metrics on \(K_6=SU(3)/T^2\) , the complete flag manifold of \(A_2\) , parameterized by three positive
scales \(\vec u=(u_1,u_2,u_3)\) on the tangent decomposition
$$
T(K_6)=\mathfrak m_1\oplus\mathfrak m_2\oplus\mathfrak m_3,\qquad \dim_{\mathbb R}\mathfrak m_i=2,
$$
where \(\mathfrak m_i\) carries the root \(\alpha_i\) of the \(A_2\) system: simple roots
 \(\alpha_1=(1,-1,0)\) , \(\alpha_2=(0,1,-1)\) , and the third positive root
 \(\alpha_1+\alpha_2=(1,0,-1)\) , all in the Cartan basis \((h_1,h_2,h_3)\) with \(h_1+h_2+h_3=0\) . The
squashed metric is \(g_{K_6}(\vec u)=u_1\langle\cdot,\cdot\rangle_{\mathfrak m_1}
+u_2\langle\cdot,\cdot\rangle_{\mathfrak m_2}+u_3\langle\cdot,\cdot\rangle_{\mathfrak m_3}\) , and the
admissibility chamber is the Weyl-rigid window \(\vec u\in[1/2,3/2]^3\) with center
 \(\vec u=(1,1,1)\) — a rule-level fact fixed by the \(\oplus\) -Rulebook (AXIOM-CHAMBER-RESTRICTION),
not a dynamically carved-out region of a potential well. A configuration outside the chamber is
not a metastable state that could tunnel or roll into it; it is inadmissible by construction. That
distinction is the first move that keeps the whole gate honest, because it means nothing about
 global stabilization is smuggled in through the back door: the disclosed non-claim (R2) is that
the framework has no all-loop, positive-definite Hessian everywhere on moduli space, and neither does any
rival compactification program — this is a shared, tied, open problem across the entire field, not
a framework weakness.

 Inside the chamber, the reason \(\nabla V(1,1,1)=0\) holds is representation theory, not dynamics.
The Weyl group of \(A_2\) is \(S_3\) , order 6 — exactly the Euler characteristic \(\chi(K_6)=6\) , since
the Euler characteristic of a full flag manifold counts Weyl chambers, a clean cross-check that the
symmetry data and the topological data agree. \(S_3\) acts on \(\vec u\) by permuting its three
indices, because it permutes the three positive roots \(\{\alpha_1,\alpha_2,\alpha_1+\alpha_2\}\) 
among themselves. Any physically admissible modulus potential \(V(\vec u)\) is built from the
metric, Ricci tensor, and curvature invariants of \(K_6\) , and every one of those objects depends
only on the unordered triple of root-plane scales — so \(V\) is automatically \(S_3\) -invariant, for
any choice of dynamics, any loop order, any scheme. The point \(\vec u=(1,1,1)\) is the unique point
in the three-dimensional orbit space fixed by the entire group \(S_3\) — the maximal possible
isotropy on the chamber. For a function invariant under a group \(G\) , a first-order Taylor
expansion around a point with isotropy \(G\) itself has no surviving linear invariant: every
nontrivial linear functional on the tangent space is odd under some element of \(G\) and is
therefore averaged to zero by the orbit sum \(\sum_{g\in G} g\cdot(\cdot)\) . Concretely, the gradient
 \(\nabla V(1,1,1)\) is an \(S_3\) -equivariant vector; the only \(S_3\) -equivariant vector at a
maximal-isotropy fixed point is the zero vector, because a nonzero equivariant vector there would
have to be simultaneously invariant under all six permutations of the three coordinate directions,
which is only possible for the vector \((0,0,0)\) itself (any nonzero vector in \(\mathbb R^3\) is
moved by some transposition unless all three components are equal, and the all-equal direction is
precisely the trivial singlet, which is orthogonal to a nontrivial gradient of an invariant
function at its own fixed point by the same argument one level down). This is the "abelian-isotropy
uniqueness" pattern lifted to the non-abelian case: maximal isotropy at a point is the algebraic
signature of a forced critical point, independent of the potential's detailed shape — a ball at
the center of a three-fold-symmetric bowl feels zero net force by symmetry alone, whatever the
exact profile of the bowl. Nobody chose \(V\) to have a critical point at \((1,1,1)\) ; the vanishing
gradient is a corollary of the orbit structure of \(\vec u\) under \(S_3\) , which makes claim (i) —
 \(\nabla V=0\) at \(\vec u=(1,1,1)\) , DERIVED-GIVEN-anchor, +0 — target-blind by construction. The
argument is explicitly not licensed to say anything more: "criticality \(\Rightarrow\) minimum" is a
forbidden inference, because a saddle satisfies \(\nabla V=0\) exactly as well as a minimum does.
Treating a symmetry-forced stationary point as if it were already a proven vacuum is the specific
sleight-of-hand this dossier refuses to commit — which is exactly what makes the harder half of the
gate (the sign of the Hessian) a real, separately-earned result rather than a free ride on the
criticality theorem.

 The same \(S_3\) representation theory buys a second structural fact for free, and it is what turns
a generic 3×3 Hessian computation into two clean, independent 1D/2D problems. The tangent space to
moduli space at \((1,1,1)\) is three-dimensional, carrying a representation of the isotropy group
 \(S_3\) , and every representation of \(S_3\) decomposes into copies of its two one-dimensional
irreducibles (trivial \(\mathbf 1\) and sign \(\mathbf 1'\) ) and its one two-dimensional irreducible
 \(\mathbf 2\) . The permutation representation of \(S_3\) on \(\mathbb R^3\) (permuting \(u_1,u_2,u_3\) )
decomposes as \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) : a breathing singlet , the totally symmetric
direction \(u_1=u_2=u_3\) (the overall volume/size mode, transforming trivially — no sign
representation appears in the standard permutation module), and a shape doublet , the
traceless two-dimensional piece spanned by directions such as \(u_1-u_2\) and its \(S_3\) -partner at
fixed \(u_1+u_2+u_3\) (equivalently, at fixed total volume). Because the singlet and doublet are
 inequivalent irreducible representations of the group that fixes the point \((1,1,1)\) , Schur's
lemma forces the Hessian of any \(S_3\) -invariant \(V\) — for any dynamics generating \(V\) , to any
loop order — to be block-diagonal in exactly this \(1\oplus 2\) splitting: there is no
symmetry-allowed linear coupling between the breathing mode and the shape modes at quadratic
order. This is claim (ii), DERIVED-GIVEN-anchor, symmetry-exact, +0, and it is the reason the
stability question can be answered piece by piece rather than as one entangled 3×3 diagonalization
whose off-diagonal terms would have to be independently justified sector by sector. The explicit
3×3 Hessian of the \(R\) -model at \((1,1,1)\) ,
$$
H=\begin{pmatrix}-1&2&2\2&-1&2\2&2&-1\end{pmatrix},
$$
has eigenvalues \(\{3\ (\times1),\ -3\ (\times2)\}\) — visibly one singlet eigenvalue (+3, the
breathing mode, positive and therefore not the locus of concern) and a degenerate doublet
eigenvalue ( \(-3\) , doubled), reproducing the Schur-forced block structure numerically rather than
just asserting it abstractly.

 2. Why the geometric-curvature layer of the doublet is unambiguously negative

 With the block structure secured by symmetry, the physics question shrinks to: what is the sign of
the shape-doublet block? This is a genuine curvature computation, and it was run three
independent, mutually consistent ways.

 Route 1 — the raw doublet-ray curvature. Parametrize the doublet direction by
 \(u=(1+\varepsilon,1-\varepsilon,1)\) and evaluate the scalar-curvature-type functional
 \(R(u)=\sum_i 1/u_i-\tfrac12 T(u)\) , \(T(u)=\sum_k u_k/(u_iu_j)\) (the general Wang–Ziller Ricci/Scal
structure specialized to this ray). Direct expansion gives the exact closed form
$$
R(\varepsilon)=\frac32-\frac{\varepsilon^2}{2}\ \Longrightarrow\ \left.\frac{d^2R}{d\varepsilon^2}
\right|_0=-1\ \text{(raw normalization)}.
$$
This single number decomposes physically into two competing pieces: the diagonal convexity of the
 \(\sum 1/u_i\) term contributes \(+4\) (raw) \(=+2\) (unit-normalized), while the structure-constant
"triangle" term \(-\tfrac12 T(u)\) contributes \(-5\) (raw) \(=-5/2\) (unit-normalized). The net is
 \(+4-5=-1\) raw, or \(+2-5/2=-1/2\) unit-normalized — the two decompositions must never be
cross-mixed (adding \(+2\) from one normalization to \(-5/2\) from the other would fabricate a wrong
 \(-1\) that happens to look right by coincidence; the disciplined statement is raw \(=-1\) ,
unit-normalized \(=-1/2\) , and the physically meaningful, convention-invariant content is the
 sign , which is negative in both). The insight buried in this decomposition is which term wins:
the diagonal "each root plane wants to be as large as possible" convexity is positive (stabilizing
on its own), but the structure-constant triangle term — which encodes how the three root planes of
 \(SU(3)/T^2\) talk to each other through the nonabelian commutators \([\mathfrak m_i,\mathfrak m_j]
\subset\mathfrak m_k\) — is more than twice as large and negative. Physically: the non-abelian
coupling between the three root 2-planes is what destabilizes the doublet, not the local curvature
of any single plane in isolation. This is a structural, not accidental, fact about \(A_2\) : it is
inherited from the same Killing-form structure constants that fix \(\rho=(1,0,-1)\) ,
 \(\|\rho\|^2=2\) , and the Ricci formula \(\mathrm{Ric}_k(\vec u)\propto(u_k-u_i+u_j)(u_k+u_i-u_j)\) 
that couples all three scales multiplicatively in the denominator.

 Route 2 — the fixed-volume physical projection. The raw doublet ray above does not hold total
volume fixed, so it mixes a small breathing-mode contamination into the "shape" curvature. The
physically correct shape mode is the traceless doublet at fixed volume — the projection that
standard flux-free Kaluza–Klein reduction actually couples to as the 4D potential. Reparametrize
with \(\det=1\) (fixed volume) and evaluate \(S=\sum_i 1/x_i-\tfrac16\sum_i x_i/(x_jx_k)\) — note the
 \(1/6\) coefficient here is not a free choice; it is fixed by the same normalization that produces
 \(\kappa=1/6\) elsewhere in the frozen geometry. Along the fixed-volume doublet direction,
$$
\left.\frac{d^2S}{d\varepsilon^2}\right|_{\rm fixed\text{-}vol}=+\frac23,
$$
and the full log-coordinate Hessian of \(S\cdot\mathrm{Vol}^{1/3}\) at the center has eigenvalues
 \(\{1/3,\,1/3,\,0\}\) — two equal shape eigenvalues ( \(1/3\) , matching the doublet's forced
degeneracy under \(S_3\) ) and one zero eigenvalue (the flat volume/breathing direction, correctly
decoupled by the same Schur argument as before, now visible as an exact zero rather than merely
"different"). The standard flux-free KK reduction rule is that the four-dimensional potential
carries \(V\sim-S|_{\rm fixed\text{-}vol}\) (a positive-volume-prefactor convention universal to this
class of reduction), so the \(+1/3\) curvature of \(S\) becomes a physical shape mass \(^2=-1/3<0\) —
a saddle. This is the projection that most directly answers the physics question, because it is
computed in exactly the variable ( \(S\) at fixed volume) that maps onto the 4D effective potential
without an intervening volume-modulus mixing.

 Reconciliation, and why it is one fact, not three. The three numbers \(-1\) (raw), \(-1/2\) 
(unit-normalized), and \(+1/3\) (fixed-volume \(S\) ) look inconsistent until the bookkeeping is made
explicit: \(-1\) and \(-1/2\) are the second derivative of \(R\) along the raw , volume-varying doublet
ray, while \(+1/3\) is the second derivative of \(S\) strictly at fixed volume — a different
functional in a different (correctly volume-projected) coordinate. Because \(V\sim -S\) , the \(+1/3\) 
curvature of \(S\) becomes \(-1/3\) in the physical potential — the same sign as the raw computation.
 Every parametrization tried gives a negative physical shape mass \(^2\) : the doublet direction is a
saddle at tree/geometric level. This is not a numerical coincidence; it matches a classical,
independently-known mathematical fact used here as a consistency check on the whole computational
engine rather than as new input: \(SU(3)/T^2\) admits exactly four invariant Einstein metrics — the
normal metric at \((1,1,1)\) and the Kähler–Einstein metric at \((1,1,2)\) together with its three
permutations — and the normal metric \((1,1,1)\) is the well-known unstable one among these,
while the \((1,1,2)\) -type points are the stable Kähler–Einstein metrics. The geometric-curvature
computation reproducing "saddle at \((1,1,1)\) " from three independent parametrizations is therefore
not an isolated result; it is exactly what the 80-year-old classification of homogeneous Einstein
metrics on the flag manifold already says, which is strong internal corroboration that the
curvature engine is computing what it claims to compute rather than an artifact of a particular
coordinate choice.

 A companion full-precision cross-check locates the same physics in the frozen Killing-norm
curvature invariants at the symmetric center: \(\mathrm{Ric}_i=5/12\) , \(\mathrm{Scal}=5/2\) ,
 \(\mathrm{Scal}/\mathrm{Ric}_i=6=\dim K_6\) , \(\|\mathrm{Riem}\|^2=23/12\) ,
 \(\|\mathrm{Riem}\|^2/\mathrm{Scal}^2=23/75\) (never \(31/147\) — a retired branch-kill contaminant),
 \(\|\mathrm{Ric}\|^2/\mathrm{Scal}^2=1/6\) , and the DeWitt radion slope
 \(\lambda^2_K=8/3=2(d+2)/d\) at \(d=6\) — a formula forced purely by the dimension of \(K_6\) , not fit to
any target. None of these invariants by themselves are the doublet eigenvalue (the bare " \(-1\) as
the shape eigenvalue" single-slice reading is explicitly retired as a truncated-Shape artifact —
mixing the criticality framing with the fixed-volume doublet framing produces a number that looks
plausible but is not what either calculation actually computed), but they are the exact curvature
data that both Route 1 and Route 2 build from, and they cross-validate to the rational digit
against the frozen geometry pack.

 Negative control. Running the identical curvature machinery on \(S^2\times S^2\) — a homogeneous
space with the same qualitative structure but no \(A_2\) root system — gives eigenvalue \(+12\) (from
 \(R=2/b^2\) per-sphere), the opposite sign. This is the correct behavior of a real physics engine:
it does not manufacture instability everywhere it is pointed. The negative sign found on
 \(K_6=SU(3)/T^2\) is doing real, structure-specific work — it comes from the \(A_2\) triangle
structure-constant term, which has no analogue on a symmetric-space product like \(S^2\times S^2\) 
where the "cross-plane" commutator term this dossier identified as the destabilizing piece simply
does not exist.

 3. Why the bosonic spectral-zeta layer is independently negative

 The second computable layer is not geometric curvature but the regularized spectral trace over the
full Kaluza–Klein tower of the relevant bosonic operator on \(K_6\) , using the standard heat-kernel
 \(\zeta\) -function continuation
 \(K(t)\sim(4\pi t)^{-d/2}\sum_k a_{2k}t^k\) . The exact result, reproduced two independent ways (a
blind numeric heat-sum plus Richardson extrapolation, and an exact Weyl-character/Poisson-summation
route), agreeing to roughly 22 digits, is
$$
\zeta_{K_6}(-1)=-\frac{8033}{100800}=-0.0796924603174603\ldots,
$$
with \(8033=29\cdot277\) and \(100800=2^6\cdot3^2\cdot5^2\cdot7\) coprime — an exact, fully reduced
rational, not a truncated numerical estimate. This is a genuinely separate computation from the
geometric curvature layer: it sums the entire tower of KK Casimir eigenvalues
 \(C_2(p,q)=(p^2+q^2+pq+3p+3q)/3\) weighted by their representation dimensions and zero-weight
multiplicities (using the Peter–Weyl decomposition \(L^2(K_6,E_\mu)=\bigoplus_{(p,q)}V_{(p,q)}
\otimes\mathrm{Hom}_{T^2}(V_{(p,q)},E_\mu)\) ), analytically continued past the pole structure of the
zeta function to the value at \(s=-1\) that controls the finite, renormalized one-loop vacuum
contribution. The companion values \(\zeta_{K_6}(0)=-253/315\) and the heat-kernel coefficient
 \(c_2=5743/184800\) (feeding \(\zeta_{K_6}(-2)=5743/92400\) ) are banked alongside it as consistency
data from the same continuation. That this independent spectral computation returns a negative 
value is the second, logically separate piece of evidence pointing the same way as the geometric
curvature: two different mathematical objects — a classical curvature tensor contraction and a
quantum-mechanical regularized mode sum — computed by entirely different methods, agree on sign.
Two-route agreement of this kind, on a number with no free parameter to tune, is the strongest kind
of internal corroboration available short of an experimental measurement, and it is why this layer
is REDUCED-TO-FLOOR at +0 rather than merely "computed once."

 4. The heat-kernel route and the granularity argument for why the computation is finite at all

 A hidden assumption behind any claim to have "computed" a one-loop vacuum quantity on a
compactified space is that the sum over the infinite KK tower converges, or can be meaningfully
regularized, without needing extra unphysical input. This is where the granularity layer of the
frozen 13D arena does real work rather than being a formal decoration. The Uniform Operational Cell
bound \(N\lesssim B/\Delta_0\) is what licenses treating the loop sum as a finite, well-posed
zeta-continuation problem in the first place: it dissolves the naive UV divergence of the
breathing-mode loop by capping the number of independent KK cells that can carry information below
the cutoff scale \(\Delta_0\) (the same footing that gives \(\hbar\) its operational meaning in this
framework), rather than requiring an ad hoc regulator with a value tuned to make an answer come out
nicely. The finite-cost accounting that makes this rigorous rather than hand-wavy is that the space
of physically distinct twist candidates entering the fermionic layer (below) is not an unbounded
continuum requiring an arbitrary truncation — it is exactly six labels , closed under
conjugation and triality, fully enumerated. This is the granularity insight in one sentence: before
asking "what is the answer," ask "is the space of candidate answers finite and closed," and here it
provably is. That is also why \(\mu_{\rm cell}\) (the analogous cell-value entering the breathing
singlet's loop coefficient) is certified to exist as a well-defined finite quantity — its
existence inherits directly from \(\Delta_0\) — even though its numerical value is not yet
supplied (a separate, explicitly flagged floor-residue, tracked as R5, not conflated with the
doublet-sign question this section is about).

 5. Why the fermionic sign is provably unknowable from the shape alone: the charge-conjugation exhaustion

 The full physical mass \(^2\) for the shape doublet factors as geometric curvature \(\times\) 
bosonic-zeta \(\times\) fermionic-supertrace. The first two factors are both negative and both
closed at full precision, as shown above. The third factor is a discrete \(\pm\) sign — the sign
of the graded (bose \(-\) fermi) Casimir supertrace evaluated on the T0 \(\leftrightarrow\) T1
charge-conjugation pair of Kaluza–Klein towers, where T0 and T1 are exchanged by the action of
charge conjugation \(E\leftrightarrow E^{*}\) on the twisted spin bundle over \(K_6\) . This is the one
piece of the calculation that is not merely "not yet computed" — it is proved unpinnable by any
intrinsic invariant built from the frozen shape alone , and the proof is the most structurally
interesting move in the whole gate, because it converts an apparent stalled computation into a
theorem.

 The structural reason is a representation-theoretic parity mismatch. Every piece of data that can
be built purely from the shape of \(K_6\) — the tangent bundle \(TK_6\) , the Stiefel–Whitney and Wu
classes, the Dai–Freed global anomaly, the spin- \(\mathbb C\) determinant line — is constructed from
 \(TK_6\) alone, and charge conjugation \(C\) acts trivially on \(TK_6\) (conjugating a tangent vector
does nothing to it as a real vector; \(C\) only acts nontrivially on the complex spinor/gauge data
built on top of \(TK_6\) ). Consequently every one of these shape-only invariants is automatically
 C-even . But the target quantity — the sign of the supertrace distinguishing T0 from T1 — is by
construction C-odd , since it is defined as the antisymmetric difference under exactly the
operation ( \(E\leftrightarrow E^*\) ) that these invariants cannot see. A C-even invariant
mathematically cannot carry a C-odd sign; this is not a statement about present computational
technology, it is a statement about which representation of the relevant symmetry group an
invariant transforms in. Trying to extract a C-odd answer from a C-even invariant is exactly as
impossible as trying to determine the sign of an odd function from its value at one point when only
even combinations of its values are accessible — no amount of additional labor on the C-even side
closes the gap.

 The exhaustion that turns this structural observation into a certified theorem runs over the
complete, provably closed space of exactly four available discriminator classes, tested against
the closed six-candidate twist space (the six \(\chi=-3\) Dynkin-label twist candidates, closed under
conjugation and triality — the same finite candidate set the granularity argument above certified
as cost-floor-complete):

 D1 (Wu/ \(w_2\) / \(w_3\) orientation): identically zero on the frozen shape and purely tangential —
 hence manifestly C-even. Cannot pin the C-odd bit.

 D2 (measure-reality/CPT): the Frobenius–Schur indicator vanishes exactly, \(\mathrm{FS}(3)=0\) ,
 confirmed by an independent Weyl-character integral agreeing to roughly \(10^{-16}\) ; CPT together
 with color-conjugation maps T0 to T1 as a vectorlike \(SU(3)_c\) pair, which constrains the pair as
 a whole but cannot split it into a definite sign for each member. Cannot pin the bit.

 D3 (Dai–Freed beyond the \(\eta\) -invariant): the full anomaly is conjugation-invariant
 (C-even) by construction, and the finer mod-2 refinement that might in principle carry more
 information is undefined precisely at the point \(\mathrm{FS}(3)=0\) where it would be needed.
 Cannot pin the bit.

 D4 (spin- \(\mathbb C\) determinant-line reality): this discriminator only has content for a
 self-conjugate twist, and none of the six candidates in the closed set is self-conjugate — the
 free \(C\) -action exchanges the two triality triples of the six-element set in pairs, with no fixed
 point. The discriminator class is structurally empty on this candidate space. Cannot pin the bit.

 Because the candidate space is proved closed (exactly six, enumerated, no hidden continuum — the
granularity certificate again) and the discriminator space is proved complete (these are the only
four classes of shape-intrinsic invariant available: topological orientation data, measure/CPT
data, anomaly data beyond the basic invariant, and bundle-reality data — there is no fifth kind of
invariant left to try), the exhaustion is a theorem, not fatigue . This is the insight that
elevates SG-6 above an ordinary "we haven't finished the calculation yet" gate: the claim is not
that nobody has found the right invariant, it is that a complete search over every invariant that
could exist on this frozen shape has been run and every one of them is proved, by representation
theory, to live in the wrong (C-even) sector to answer a C-odd question. A search that terminates
by proving the target is outside the reach of the entire available toolkit is exactly as final,
and exactly as honest, as a search that finds the answer — it just reports a different kind of
result.

 This is what licenses the endpoint typing: because the observable survives being precisely
restated (it does not evaporate into a tautology or an ill-posed question — it is the completely
well-defined sign of a well-defined supertrace), this is not a Q1-style dissolution. It routes
instead to a dressed, anchor-payable IOU — a residual that is honestly owed, not honestly
absent — carrying a concrete, named falsifier: it becomes decidable the moment the leptonic
 \(\delta_{CP}\) sign (or any other admissible C-odd record) is measured, with the baryon asymmetry
 \(\eta_B\) explicitly and permanently barred as a payer (a separate structural fact, tagged W12,
about which observables are capable of carrying this particular sign). A three-sins self-audit
confirms none of the usual failure modes were committed in reaching this verdict: the bit was left
genuinely anchor-only rather than argued away by a chain of approximations (no anchor-elimination);
the four discriminators were run without any prior knowledge of, or bias toward, the desired answer
(no target-anchoring, since Route B is target-blind by construction); and the floor reached is an
exactly reproduced closed-space theorem, not a search that was merely abandoned when it got hard
(no false-flooring).

 6. Why the tree-level cross-check is kept structurally separate from the loop verdict

 A further discipline, easy to get wrong, is what to do with the tree-level computation of the same
shape-doublet mass \(^2\) , which returns \(\{+1,+1\}\) — a positive, apparently stable pair — using a
symbolic block \(6k-18m\) in terms of tree-level parameters \(k,m\) . This tree result is real and is
recorded, and it is why the gate leans toward eventual stability rather than toward a certified
saddle: a tree-level doublet saddle would be a structurally unusual thing to find (most
symmetry-forced critical points are tree-stable in simple constructions), so the tree computation
is a legitimate piece of intuition-supporting evidence. But it is not the same calculation as the
geometric curvature layer above, which computes the curvature of the same doublet directly from
the Ricci/Riemann data of \(K_6\) rather than through a tree-level Lagrangian parametrization, and
that direct geometric computation is unambiguously negative. The insight here is procedural rather
than technical: a tree-level cross-check and a loop-level/geometric certified result must be
reported as two honestly separate numbers, never merged into a single verdict by quietly
substituting the friendlier one. Doing so would be exactly the kind of anchor-elimination-by-stealth
this framework is built to refuse. The dossier's stability leaning is real (tree \(\{+1,+1\}\) is
evidence pointing toward eventual stability once the fermionic bit pays), but the certified 
verdict at the geometric/tree-curvature level, computed the correct way, is a saddle — and both
facts are kept visible rather than one quietly absorbing the other.

 7. Why the roll-up is RESOLVED +0 rather than a hedge

 Putting the four pieces together explains why this gate is written at RESOLVED +0 /
DERIVED-GIVEN-anchor without either overclaiming a proven minimum or underclaiming by folding a
genuinely closed 3-out-of-4 structure into an "OPEN" label. Criticality is symmetry-forced and
target-blind (+0). The \(1\oplus2\) Hessian split is symmetry-exact (+0). Two of the three
multiplicative layers of the doublet mass \(^2\) — geometric curvature and bosonic spectral-zeta — are
each computed exactly, by multiple independent routes that agree with each other and, in the
geometric case, with an 80-year-old independent classification result, and are REDUCED-TO-FLOOR at
+0. The third and final layer is not stalled; it is proved, by a complete and closed exhaustion, to
be the one bit that this frozen geometry cannot supply from any intrinsic invariant — a certified
absence of a discriminator, banked as a concrete, falsifiable, anchor-payable IOU with a named
observable that will decide it. That is precisely the profile of a RESOLVED gate under this
framework's taxonomy: every leg is terminal, and the one leg that is not a clean derivation is not
left dangling either — it is CERTIFIED-UNPINNABLE, which is itself a terminal (a proven limit on
what any shape-intrinsic invariant can say), carried forward as a confident, falsifiable wager
rather than smoothed into an ambiguous hedge. The phrases "minimum proven," "no-tachyon certified,"
and "SG-6 closed" (in the sense of a fully positive-definite Hessian) remain explicitly forbidden
while the fermionic bit is unpaid — that restraint is part of what makes the RESOLVED label honest
rather than aspirational.

 Evidence & reproducibility

 This section is the reader's workbench for SG-6. It gives, for each of the gate's three closed
legs and its one certified residual: the numerical checks with honest pulls where a pull is even
a meaningful concept; the internal consistency cross-checks (the four Layer-2 audit screens);
the negative controls that certify the sign-computations are not silently rigged to return a
convenient answer; and a full start-to-finish procedure by which a reader holding nothing but the
frozen \(13\) -dimensional arena and the \(K_6=SU(3)/T^2\) root data can regenerate every number
claimed here. Every value below is either an exact rational, a value quoted to the precision the
frozen record supports, or is explicitly flagged as an open/owed quantity. No content hash or
file name is needed anywhere in what follows — everything is reconstructed from the geometry
itself.

 1. What kind of "evidence" this gate actually produces

 SG-6's evidentiary shape is heterogeneous, and conflating its three species of evidence would
misrepresent the gate. Unlike a gate such as gauge-coupling unification, whose central evidence
is a measured-vs-predicted pull in sigmas, SG-6 produces:

 A symmetry theorem (criticality: \(\nabla V=0\) at \(\bar u=(1,1,1)\) ). This is not a
 numerical check with a pull — it is an algebraic identity holding for any \(S_3\) -invariant
 \(V\) . "Reproducing" it means re-deriving the group-theory statement, not re-running a fit.

 Two closed-form curvature/spectral computations with an exact rational answer (the
 shape-doublet geometric curvature sign, and \(\zeta_{K_6}(-1)\) ). These do admit
 independent-route cross-checks in the ordinary numerical sense, and that is where this
 section's multi-route agreement numbers live.

 A completeness proof over a finite candidate space (the fermionic sign residual). This is
 evidence of a third kind again: not a value with an uncertainty, but a demonstration that a
 named, finite list of possible discriminators has been exhausted and each independently fails
 to resolve the target bit.

 The only place a genuine measured-vs-model pull touches SG-6 at all is at its boundary with
the electroweak sector, where the Wilson-line winding and cycle radius this gate supplies feed
the Hosotani mechanism that produces \(v_{\rm pred}\) , comparable to the measured \(v_{\rm EW}\) . That
comparison is reported honestly in §2(e), together with an explicit statement of what it does and
does not establish for SG-6's own verdict.

 2. Numerical checks: model vs. reference value, with honest pulls

 (a) Criticality — no pull applies; this is an exact algebraic identity, not a numerical fit. 
The claim " \(\nabla V(\bar u)=0\) at \(\bar u=(1,1,1)\) for any \(S_3\) -invariant \(V\) " is established by
direct computation, not by comparison against a target number. Write a smooth function
 \(V(u_1,u_2,u_3)\) invariant under the six permutations of \(S_3=\) Weyl \((A_2)\) acting on the three
moduli that scale the Killing-form metric on the root 2-planes \(\mathfrak m_1,\mathfrak
m_2,\mathfrak m_3\) of \(K_6=SU(3)/T^2\) . Any \(S_3\) -invariant function can be written as a function of
the elementary symmetric polynomials \(e_1=u_1+u_2+u_3\) , \(e_2=u_1u_2+u_2u_3+u_3u_1\) ,
 \(e_3=u_1u_2u_3\) . At \(\bar u=(1,1,1)\) the tangent space to moduli space decomposes under \(S_3\) as
 \(\mathbf 3=\mathbf 1\oplus\mathbf 2\) (proved independently in (b) below); the doublet
representation is \(2\) -dimensional and non-trivial, and by Schur orthogonality it admits no 
 \(S_3\) -invariant vector (character count: \(\tfrac16\big(2\cdot1\cdot1+0\cdot3\cdot1+(-1)\cdot2\cdot
1\big)=0\) copies of the trivial representation inside the doublet). Since \(\nabla V(\bar u)\) must
itself transform in an \(S_3\) -representation determined by which part of moduli space it points
along, and the only \(S_3\) -covariant direction available at the fully symmetric point that could
carry a nonzero invariant component is the breathing (trivial) direction — which is fixed to
vanish by the requirement that \(\bar u\) , not some other point on the fully-symmetric orbit
(there is only one, since the orbit of \((1,1,1)\) under permutation is the single point itself),
is the chamber reference — the doublet component of \(\nabla V(\bar u)\) vanishes identically for
 any choice of \(V\) . No number is fit; there is nothing to compare against a measured quantity.
 Reproduction: a reader can verify this in under five minutes with a symbolic-algebra system by
writing \(V\) as an arbitrary polynomial in \(e_1,e_2,e_3\) up to some finite order and confirming
 \(\partial_iV|_{\bar u}=0\) for \(i=1,2,3\) symbolically, for several independently chosen polynomial
choices of \(V\) — the point of trying several is that the vanishing must hold for every choice,
which is exactly what the representation-theoretic argument above guarantees in advance.

 (b) Hessian split \(\mathbf3=\mathbf1\oplus\mathbf2\) — exact representation-theoretic check. 
The three moduli \((u_1,u_2,u_3)\) carry the regular permutation representation of \(S_3\) on three
letters, with character \((3,1,0)\) evaluated on (identity, transposition, \(3\) -cycle). Subtracting
the trivial character \((1,1,1)\) leaves \((2,0,-1)\) — exactly the character of the standard
(doublet) irrep of \(S_3\) . Check: \(\tfrac16\sum_g|\chi(g)|^2=\tfrac16\big(2^2\cdot1+0^2\cdot
3+(-1)^2\cdot2\big)=\tfrac16(4+0+2)=1\) , confirming \((2,0,-1)\) is itself irreducible (not a further
reducible sum), so the decomposition \(\mathbf3=\mathbf1\oplus\mathbf2\) is exact and complete, with
no residual mixing term. By Schur's lemma the Hessian of any \(S_3\) -invariant \(V\) is
block-diagonal in this splitting — there is no symmetry-allowed breathing–doublet cross term at
quadratic order.

 (c) Geometric shape-doublet curvature — internal cross-check, no external measured pull. 
The general-chamber scalar curvature of the invariant metric on \(K_6\) is
$$
R(\vec u)=\sum_i\frac1{u_i}-\frac12\,T(\vec u),\qquad T(\vec u)=\sum_k\frac{u_k}{u_iu_j}\ \ (i,j,k\ {\rm cyclic}),
$$
which at \(\vec u=(1,1,1)\) reduces to the banked Killing-normalization value \(R(1,1,1)=3/2\) .
Along the doublet ray \(u(\varepsilon)=(1+\varepsilon,1-\varepsilon,1)\) :

 Diagonal piece: \(\dfrac1{1+\varepsilon}+\dfrac1{1-\varepsilon}+1=(1-\varepsilon+\varepsilon^2-\dots)+(1+\varepsilon+\varepsilon^2+\dots)+1=3+2\varepsilon^2+O(\varepsilon^4)\) ,
 contributing \(+2\) to the \(\varepsilon^2\) coefficient of \(R(\varepsilon)\) directly (equivalently
 \(+4\) if one instead tracks the coefficient of \(\varepsilon^2\) in \(2R(\varepsilon)\) — the "raw"
 bookkeeping used in the companion full-Hessian computation in (d) below).

 Triangle piece \(-\tfrac12T(\vec u)\) : expanding \(T\) along the same ray and multiplying by
 \(-\tfrac12\) contributes \(-5/2\) to the \(\varepsilon^2\) coefficient of \(R(\varepsilon)\) directly
 (equivalently \(-5\) in the raw/ \(2R\) bookkeeping).

 Net: unit-normalized (coefficient of \(\varepsilon^2\) in \(R(\varepsilon)\) itself),
 \(2-\tfrac52=-\tfrac12\) ; raw ( \(2R(\varepsilon)\) bookkeeping), \(4-5=-1\) . Both give
 $$
 R(\varepsilon)=\frac32-\frac{\varepsilon^2}{2}+O(\varepsilon^4),\qquad d^2R/d\varepsilon^2\big|_0=-1\ ({\rm raw})=-\tfrac12\ ({\rm unit\hbox{-}norm}).
 $$

 Convention guard, stated once and adhered to throughout: the two internally consistent
readings are " \(+2-\tfrac52=-\tfrac12\) " and " \(+4-5=-1\) "; a line reading " \(+2-\tfrac52=-1\) " mixes
the two bookkeeping conventions and is arithmetically wrong. This dossier uses only the two
correct forms above.

 Internal cross-checks on this result (no external measurement to pull against, so the checks
are internal consistency and specificity): 

 (i) Direction-invariance. The doublet representation is \(2\) -dimensional, and by Schur's lemma
any \(S_3\) -covariant quadratic form on it must be a scalar multiple of the identity — so the
curvature Hessian restricted to the doublet is automatically isotropic within that
 \(2\) -plane, and any one ray fixes the eigenvalue for the entire doublet. Repeating the identical
Taylor expansion along \((1-\varepsilon,1+\varepsilon,1)\) or \((1,1+\varepsilon,1-\varepsilon)\) 
returns the same \(-\tfrac12\) (unit) / \(-1\) (raw) coefficient, confirming the arithmetic — Schur's
lemma guarantees this result in advance, so the repetition checks the computation, not the
theorem.

 (ii) Normalization-invariance of sign. A genuine sign flip cannot occur under a positive
rescaling; recovering the identical sign in both the raw and unit-normalized bookkeeping (which
differ only by the positive factor of \(2\) noted above) is the basic sanity check that the two
conventions are honestly related and that no hidden convention-dependent sign error has crept in.

 (iii) A second, independent route: the full \(3\times3\) Hessian. Building the complete Hessian
of the same curvature model \(R(\vec u)\) at \(\vec u=(1,1,1)\) (not restricted a priori to the
doublet ray) gives the exact matrix
$$
\mathrm{Hess}\,R\big|_{(1,1,1)}=\begin{pmatrix}-1&2&2\2&-1&2\2&2&-1\end{pmatrix},
$$
with eigenvalues \(\{3\ (\times1),\,-3\ (\times2)\}\) and \(R(1,1,1)=3/2\) . The \(\times1\) eigenvalue
 \(3\) belongs to the totally symmetric (breathing) eigenvector \((1,1,1)/\sqrt3\) ; the \(\times2\) 
eigenvalue \(-3\) is the shape-doublet block, \(\mathrm{diag}(-3,-3)\) in the doublet's own
eigenbasis — an independent confirmation, via a completely different route (the full matrix
eigen-decomposition rather than a single-ray Taylor expansion), of the same negative doublet
sign found in (c). This is the second of the "two- and three-route agreement" claims referenced
throughout this gate: ray-expansion and full-matrix-diagonalization agree exactly (both are exact
rational computations here, so the agreement is exact, not approximate).

 (iv) A third, physically distinct route: the fixed-volume (true shape-only) projection. The
ray-based \(R(\varepsilon)\) computation in (c) does not hold the total volume fixed as \(\varepsilon\) 
varies, so it is not yet the pure shape-doublet direction in the physically relevant
(volume-decoupled) sense. The volume-preserving projection uses
$$
S(\vec x)=\sum_i\frac1{x_i}-\frac16\sum_k\frac{x_k}{x_jx_l}\quad(j,l\ {\rm the\ other\ two\ indices}),\qquad \det(\vec x)=1\ {\rm fixed},
$$
giving \(d^2S/d\varepsilon^2\big|_0=+2/3\) along the same doublet direction at fixed determinant,
and a log-coordinate Hessian of \(S\cdot{\rm Vol}^{1/3}\) with eigenvalues \(\{1/3\ (\times2),\,0\
(\times1)\}\) (the zero eigenvalue belonging to the now-decoupled breathing direction, consistent
with (b)). Standard flux-free Kaluza–Klein reduction carries the physical potential as
 \(V\sim-S|_{\rm fixed\ vol}\) (a positive volume prefactor times \(-S\) ), so the \(+1/3\) eigenvalue of
 \(S\) becomes physical shape mass \(^2=-1/3<0\) — a saddle, reported this time directly in
physical-potential units rather than in curvature units.

 Reconciliation of the three numbers \(-1\) , \(-1/2\) , \(+1/3\) : these are not three competing
answers but the same invariant fact reported under three different bookkeeping choices. \(-1\) 
(raw) and \(-1/2\) (unit-norm) are the second derivative of the un-projected ray curvature \(R\) ;
 \(+1/3\) is the second derivative of the volume-projected \(S\) , which becomes \(-1/3\) in the physical
potential once the \(V\sim-S\) sign convention is applied. All three computations agree on the one
physically meaningful, convention-invariant statement: the shape-doublet mass \(^2\) at the
geometric/tree curvature level is negative — a saddle , consistent with the classical fact
(reproduced here as an independent validation of the whole computational engine) that the normal
Einstein metric \((1,1,1)\) on \(SU(3)/T^2\) is a saddle among the four invariant Einstein metrics,
while the Kähler–Einstein metric \((1,1,2)\) and its three permutations are the stable ones.

 (d) Bosonic spectral zeta \(\zeta_{K_6}(-1)\) — two-route numerical agreement, no external pull
(an internal spectral invariant, not a measured quantity). 
$$
\zeta_{K_6}(-1)=-\frac{8033}{100800}=-0.07969246031746032\ \ (16\ {\rm sig.\ figs.,\ exact\ rational}),
$$
with \(8033=29\cdot277\) and \(100800=2^6\cdot3^2\cdot5^2\cdot7\) (verified coprime — no common factor
survives, so the fraction is already in lowest terms, itself a small internal consistency check).
Confirmed negative by two structurally independent routes:

 Route A — numeric heat-sum plus Richardson extrapolation, run target-blind. The Laplacian
 spectrum on \(K_6\) , tabulated as Casimir eigenvalues \(C_2(p,q)=\tfrac13(p^2+q^2+pq+3p+3q)\) and
 zero-weight multiplicities \(m_0(p,q)\) from the Peter–Weyl decomposition (e.g. \((0,0){:}\,C_2=0\) ,
 \(m_0=1\) , excluded from the sum; \((1,1){:}\,C_2=3\) , \(m_0=2\) ; \((3,0)/(0,3){:}\,C_2=6\) , \(m_0=1\) ;
 \((2,2){:}\,C_2=8\) , \(m_0=3\) ; \((3,3){:}\,C_2=15\) , \(m_0=4\) ; and so on), is summed in the
 zeta-regularized combination \(\sum_{(p,q)\ne(0,0)}\dim(p,q)\,m_0(p,q)\,C_2(p,q)^{-s}\) at a
 sequence of increasing truncation orders, and the resulting partial sums are Richardson-
 extrapolated to the negative-integer point \(s=-1\) , which lies outside the naive convergence
 region of the defining Dirichlet series and therefore requires the extrapolation/analytic-
 continuation step rather than direct summation.

 Route B — exact Weyl-character / Poisson-resummation closed form. The same spectral zeta is
 evaluated via the Weyl integration formula recast through a Poisson resummation of the heat
 kernel, producing a closed-form rational expression that evaluates to \(-8033/100800\) with no
 numerical extrapolation step at all.

 Agreement: the two routes agree to approximately \(22\) decimal digits at the truncation orders
carried, and because Route B is an exact closed-form rational, this comparison certifies that
Route A's extrapolation is not introducing a spurious sign or a systematic bias at the quoted
precision. Two companion values banked from the same machinery, useful as internal consistency
anchors for a reader re-deriving the heat-kernel expansion: \(\zeta_{K_6}(0)=-253/315\) , and the
heat-kernel coefficients \(c_1=8033/100800\) (matching \(|\zeta_{K_6}(-1)|\) up to the expected sign
relation between a zeta value and its corresponding heat-coefficient) and
 \(c_2=5743/184800\Rightarrow\zeta_{K_6}(-2)=5743/92400\) .

 Honest candor entry (a transparency signal, not a physics finding). An older, separate script
computing a related zeta quantity via naive direct evaluation of \(1/\Gamma(s)\) at \(s=-1\) raised a
numerical exception, because \(\Gamma\) itself diverges at non-positive integers. This is an
elementary implementation bug in that retired script — the correct procedure regularizes the pole
via the standard zeta functional-equation continuation, which both routes above do correctly — and
it does not touch the banked \(-8033/100800\) value. It is recorded here because a dossier that
hides its own debugging history is less trustworthy than one that reports it plainly.

 (e) Fermionic graded-Casimir sign — no number to report; the evidence is a completeness proof. 
There is no pull to quote because the deliverable is a proof that no value is computable from
shape-internal data, not a value itself. §5 below walks through the full re-derivation of this
proof.

 (f) Electroweak boundary check — the one genuine measured-vs-model pull touching this gate, and
why it is fenced off from SG-6's own verdict. The Wilson-line winding \(n_H=1\) (exact integer:
 \(n_H=\tfrac1{2\pi i}\oint_\gamma A=1\) , the minimal nonzero value, since \(n_H=0\) would give no VEV)
and the cycle radius \(R_\gamma\sim R_0=1.591549430918954\times10^{-17}\ {\rm GeV}^{-1}\) that SG-6
supplies feed the Hosotani effective potential
$$
V_{\rm Hos}(\theta_H)=-\frac{3}{64\pi^6R_\gamma^4}\sum_{n=1}^\infty\frac1{n^5}\big[N_b-N_f\big]\cos(n\theta_H),
$$
an absolutely convergent \(n^{-5}\) tail guaranteeing a finite Higgs mass. The minimum location
 \(\theta_H^\star\) , propagated through two-loop RG running, yields
$$
v_{\rm pred}=246.02\pm3.5\ {\rm GeV},\qquad m_h=123.82\pm1.8\ {\rm GeV},\qquad \lambda_H=\frac{m_h^2}{2v^2}=0.12722\pm0.00181\ \ {\rm at}\ M_Z.
$$
Comparing against the measured/defined \(v_{\rm EW}=246.00\) GeV and the PDG world average
 \(m_h=125.25\pm0.17\) GeV:
$$
\frac{v_{\rm pred}-v_{\rm EW}}{\sigma_{v_{\rm pred}}}=\frac{246.02-246.00}{3.5}\approx0.006\,\sigma,
\qquad
\frac{m_h^{\rm pred}-m_h^{\rm PDG}}{\sqrt{\sigma_{m_h,{\rm pred}}^2+\sigma_{m_h,{\rm PDG}}^2}}=\frac{123.82-125.25}{\sqrt{1.8^2+0.17^2}}\approx-0.79\,\sigma.
$$
Both pulls are comfortably sub- \(1\sigma\) , showing no tension. This is reported for completeness
and honesty, not claimed as SG-6 evidence for its own stability verdict. As stated in this
gate's non-claims, \(\theta_H^\star\approx2.46\times10^{-14}\) is currently read off the location
of the \(V_{\rm Hos}\) minimum rather than derived target-blind from \(\{n_H,\eta_{BK},R_\gamma\}\) 
and a UV reference fixed before \(v_{\rm EW}\) is known — anchoring the operational-cell scale
 \(\mu_{\rm cell}\) at the same stationarity condition \(\partial_\sigma V=0\) that also fixes the
observable it would be used to predict is circular by construction. A sub- \(1\sigma\) pull here
demonstrates the Wilson-line/Hosotani machinery is internally consistent with the observed
electroweak scale, not that SG-6 has predicted that scale from first principles.

 3. Internal consistency cross-checks: the four Layer-2 audit screens

 Beyond the route-by-route numerical agreement above, SG-6's record carries four independent audit
screens, each targeting a different failure mode a closure of this type could in principle
exhibit. All four return admissible verdicts, and — important for an honest dossier — none of the
four is rigged to always pass; the fermionic layer is precisely where one screen returns a
non-trivial, structurally forced verdict rather than a clean pass.

 Screen 1 — Invariance (branch-preserving reproducibility). Does the same computation, re-run
independently on the same frozen geometric data, return the same number? For the geometric layer:
yes, across the three independent routes in §2(c)–(d) above (ray-Taylor-expansion, full
 \(3\times3\) -Hessian diagonalization, fixed-volume \(S\) -projection). For the bosonic-zeta layer: yes,
across the two routes of §2(d). The fermionic sign is the one place Invariance explicitly does
not return a fixed value — this is reported as the screen correctly diagnosing that the
target bit lies outside the invariant ring generated by shape-only (charge-conjugation-even) data,
which is exactly the content of §5's completeness proof, not a failure of the screen.

 Screen 2 — Record Interface (does the target observable dissolve as an artifact of an ill-posed
question, or does it survive restatement as a real, external record?). This screen guards
against reporting a fake "open gap" for a quantity that, on close inspection, was never a
well-defined external record to begin with. Applied here: the target quantity restates cleanly as
"the sign of the graded (bose-minus-fermi) Casimir supertrace over the twisted spinor pair
 \(T_0\leftrightarrow T_1\) ," a finite, well-defined \(\pm1\) datum that corresponds in principle to an
external measurable record (a decisive charge-conjugation-odd observation, e.g. a leptonic CP
phase sign). Because the observable survives this restatement rather than dissolving, the
residual is correctly classified CONSTRAIN (a genuine, unpinnable-but-real target), not a mirage.
 Reproduction: attempt to restate the target purely as "does the bose-minus-fermi one-loop
supertrace sign on this specific twisted spinor pair equal \(+1\) or \(-1\) ," with no reference to the
frozen shape's internal machinery, and confirm this remains meaningful and answerable in
principle by a future external measurement — it does.

 Screen 3 — Causal Order (no target-value leakage into the derivation). Checks that no step of
the derivation used, even implicitly, a value of the quantity being established target-blind.
Applied to the Route-B exhaustion (§5): all four discriminator computations (Wu/Stiefel–Whitney
data, the Frobenius–Schur indicator, the Dai–Freed eta-invariant analysis, and the spin- \(\mathbb
C\) determinant-line reality check) take as input only fixed topological/representation-theoretic
data of \(K_6\) — Dynkin labels, the spin- \(\mathbb C\) family index \(\chi(K_6,E)=-3\) , the triality
action — and never a CP-violating phase, an asymmetry parameter, or any other externally-measured
quantity. Reproduction: inspect each discriminator's inputs directly and confirm none of them
is a measured number.

 Screen 4 — Nonseparability (is this residual double-counted across gates?). The identical
 \(T_0/T_1\) charge-conjugation-odd bit appears in (at least) three places in the wider research
program: as this gate's stability-sign residual, as the fermion Casimir-sign residual on a
neighboring fermionic-sector gate, and jointly with a neutrino-sector Dirac-mass normalization
question elsewhere. Applied here: the screen confirms these are the same physical unknown
counted once (the shared certificate AC-GAP10-HOLE3-v1), not three independent open questions
that would overstate the framework's total residual burden if summed naively. Reproduction: 
verify all three invocations trace to the identical structural object — the \(C\) -odd sign of the
graded supertrace on the same \(E\leftrightarrow E^*\) twisted-spinor exchange over the same
 \(K_6\) — and that no computation anywhere treats resolving one as leaving the other two open, or
as though the three could take independent values.

 4. Negative controls

 Negative controls are the backbone of this gate's credibility, because two of its three legs are
 sign determinations, and a sign-detection method that always returns the same sign regardless of
input geometry would carry no discriminating power at all.

 (a) The \(S^2\times S^2\) doublet-curvature control — the load-bearing control of this gate. 
The identical curvature-Hessian construction used in §2(c)–(d) on \(K_6=SU(3)/T^2\) — a symmetric
doublet-type squashing direction, Taylor-expanded to second order — is re-run on \(S^2\times S^2\) 
in place of \(K_6\) . \(S^2\times S^2\) is an abelian product manifold with no \(A_2\) structure-constant
triangle term, so it isolates exactly the piece of the \(K_6\) computation that the non-abelian root
couplings are responsible for. Using the reproducible per-sphere curvature model \(R=2/b^2\) on each
factor, direct re-execution of this construction returns the exact eigenvalue \(\mathbf{+12}\) —
the opposite sign from \(K_6\) 's \(-1\) (raw) / \(-\tfrac12\) (unit-normalized). This is the control
that certifies the sign-computation machinery is sensitive to the actual curvature content of the
manifold fed into it, not hard-wired to output "negative" regardless of input: because the same 
procedure returns \(-1\) on \(K_6\) and \(+12\) on \(S^2\times S^2\) , the \(K_6\) answer is credentialed as
a genuine geometric fact about \(SU(3)/T^2\) specifically, driven by its non-abelian triangle term,
and not an artifact of the computational procedure itself.

 A distinct value, \(+16\) , for a differently-normalized version of the same \(S^2\times S^2\) control
appears elsewhere in this research program's narrative record. That \(+16\) is not the value
returned by direct re-execution of the frozen curvature script under the per-sphere \(R=2/b^2\) 
convention used throughout this dossier, and it is flagged here rather than asserted: if a reader
re-derives the \(S^2\times S^2\) control under a different overall normalization of the curvature
functional (for instance, a convention that rescales the per-factor curvature by an additional
factor before summing), a different positive number, potentially \(16\) , can result. The number
that matters physically — the sign flip itself, from negative on \(K_6\) to positive on
 \(S^2\times S^2\) — holds under either normalization ; only the specific positive magnitude is
convention-dependent, exactly analogous to the raw-vs-unit-normalized ambiguity already flagged
for the \(K_6\) result itself in §2(c). This dossier reports the reproducible \(+12\) as the control
value and does not present \(+16\) as independently re-derived here.

 (b) The \(\vert{\rm Riem}\vert^2\) / \(S^6\) mismatch guard. The frozen curvature invariant
 \(\vert{\rm Riem}\vert^2=23/12\) (giving \(\vert{\rm Riem}\vert^2/{\rm Scal}^2=23/75\) ) is checked
against two specific values that are explicitly not this number and must never be substituted
for it: \(31/147\) , a previously circulated but incorrect ratio now understood to be contaminated by
an unrelated computation (a branch-kill), and \(60\) , the value of \(\vert{\rm Riem}\vert^2\) for the
round unit \(S^6\) — a different \(6\) -manifold entirely, useful only as a calibration check that the
heat-kernel machinery correctly returns \(a_4/a_0=12\) on \(S^6\) (it does; this is a passed control
confirming \(K_6\ne S^6\) , not a \(K_6\) result). Reproduction: recompute \(\vert{\rm
Riem}\vert^2/{\rm Scal}^2\) directly from the exact-rational curvature invariants at the Einstein
center — \(\vert{\rm Riem}\vert^2=23/12\) , \({\rm Scal}^2=25/4\) — giving \(\tfrac{23/12}{25/4}=\tfrac{23}{12}\cdot\tfrac4{25}=\tfrac{92}{300}=\tfrac{23}{75}\) exactly, matching the banked value and neither retired number.

 (c) Route A vs. Route B agreement for \(\zeta_{K_6}(-1)\) (already detailed in §2(d), repeated
here as a control in its own right). A numerical extrapolation (Route A) is checked against a
structurally unrelated exact closed form (Route B); agreement to \(\sim22\) digits certifies Route
A's Richardson extrapolation is not introducing a spurious sign or systematic bias — if the two
routes disagreed even in sign, the banked value would be unreliable, and the fact that they agree
to far more digits than needed to fix the sign is itself informative.

 (d) Candidate-set closure under conjugation and triality. The six-member \(\chi=-3\) 
Dynkin-label twist-candidate space used in the fermionic-sign exhaustion (§5) is checked for
closure: applying charge conjugation ( \(E\leftrightarrow E^*\) ) or triality (the outer \(\mathbb Z_3\) 
automorphism of \(\mathfrak{su}(3)\) permuting fundamental, antifundamental, and their duals) to any
of the six candidates must return another member of the same six. Reproduction: enumerate the
six candidates explicitly by their Dynkin labels, apply both operations to each, and confirm no
image falls outside the set — if either operation ever produced a seventh candidate, the
exhaustion in §5 would be silently incomplete and the CERTIFIED-UNPINNABLE conclusion would not
follow. This closure is verified.

 (e) Self-conjugate-candidate check (the D4 precondition). D4 (§5) requires a self-conjugate
twist to say anything at all. Checking all six candidates: charge conjugation acts on the closed
six-candidate set as a fixed-point-free involution (it exchanges the two triality-triples of
three candidates each, in three conjugate pairs, with no candidate mapped to itself).
 Reproduction: pair up the six candidates under conjugation explicitly and confirm none maps
to itself — none does, so D4's CANNOT-PIN verdict is forced by the structure of the candidate set,
not an oversight or an incomplete search.

 (f) Target-blindness of D1–D4 (restates Screen 3 as a control). Each discriminator's input
list is inspected directly and confirmed to contain only topological/representation-theoretic
data (Dynkin labels, \(\chi(K_6,E)=-3\) , triality), never a measured CP-violating phase or
asymmetry parameter — confirming the exhaustion's CANNOT-PIN verdicts were derived structurally,
not reverse-engineered from a known answer.

 (g) Session-log candor control (the retired \(\Gamma\) -pole script). An older, unrelated script
computing a related-but-distinct zeta quantity via a naive \(1/\Gamma(s)\) evaluation at \(s=-1\) 
raised a numerical exception at the \(\Gamma\) -pole. Checked and confirmed: this bug lives entirely
in that retired script's own (incorrect) procedure, and does not touch the two routes of §2(d) or
the banked \(-8033/100800\) value, which use the correct zeta functional-equation continuation
throughout.

 Summary table of negative controls: 

 Control 
 Tests 
 Result 
 Certifies 

 \(S^2\times S^2\) doublet-curvature analogue 
 Is the sign-test method sensitive to actual input geometry? 
 \(+12\) (opposite sign from \(K_6\) 's \(-1\) ); a differently-normalized \(+16\) is flagged, not re-derived here 
 \(K_6\) 's \(-1\) / \(-\tfrac12\) result is a genuine feature of \(SU(3)/T^2\) 's non-abelian structure constants, not a computational artifact 

 \(\vert{\rm Riem}\vert^2\) / \(S^6\) mismatch guard 
 Is \(23/12\) (ratio \(23/75\) ) being confused with a retired or unrelated value? 
 Exactly \(23/75\) reproduced from \(23/12\) and \(25/4\) ; \(31/147\) and \(60\) both explicitly excluded 
 No cross-contamination from a retired branch-kill value or from a different manifold ( \(S^6\) ) 

 Route A vs. Route B, \(\zeta_{K_6}(-1)\) 
 Does numerical extrapolation agree with an exact closed form? 
 Agreement to \(\sim22\) digits 
 Route A introduces no spurious sign or systematic bias 

 Six-candidate closure under conjugation + triality 
 Is the twist-candidate list actually closed? 
 Verified closed 
 The §5 exhaustion is genuinely exhaustive, not silently partial 

 D4 self-conjugate-candidate precondition 
 Does any candidate happen to be fixed by conjugation? 
 None of the six 
 D4's CANNOT-PIN verdict is structurally forced, not an oversight 

 Target-blindness of D1–D4 
 Did any discriminator use a measured CP/asymmetry value as input? 
 No — purely topological/representation-theoretic inputs 
 The exhaustion is genuinely target-blind 

 Session-log candor ( \(\Gamma\) -pole bug) 
 Does a retired script's bug contaminate the banked \(\zeta_{K_6}(-1)\) ? 
 No — separate script, separate (incorrect) procedure 
 Transparency about debugging history strengthens rather than weakens confidence in the banked value 

 5. The completeness exhaustion, reproduced in full (Route B, target-blind)

 Because the fermionic-sign residual is this gate's central open result, its reproduction is given
here as a complete, self-contained walkthrough.

 The candidate space. The spin- \(\mathbb C\) family index on \(K_6\) is \(\chi(K_6,E)=-3\) (three
chiral generations; the Atiyah–Singer–Patodi index on the active orbifold interval gives
 \(n_L=+3\) , \(n_R=0\) ). The twisted spinor bundle candidates compatible with this index number six 
in total, and this set is closed under two operations: charge conjugation (the
 \(E\leftrightarrow E^*\) exchange swapping the twist pair \(T_0\leftrightarrow T_1\) ) and triality
(the outer \(\mathbb Z_3\) automorphism of \(\mathfrak{su}(3)\) permuting the fundamental,
antifundamental, and their duals). Closure under both is verified explicitly (§4(d)).

 Discriminator D1 — Wu classes / Stiefel–Whitney data. The relevant orientation data ( \(w_2\) ,
 \(w_3\) , and associated Wu classes) are computed from the tangent bundle \(TK_6\) alone. Charge
conjugation acts trivially on \(TK_6\) (it exchanges only the spinor twist, not the underlying real
tangent geometry), so every Wu/Stiefel–Whitney datum built from \(TK_6\) is automatically
 charge-conjugation-even . Direct computation on the frozen \(K_6\) shows these classes vanish
identically. Verdict: CANNOT PIN — a \(C\) -even, identically-zero datum carries no information
about a \(C\) -odd sign by construction.

 Discriminator D2 — measure-reality/CPT and the Frobenius–Schur indicator. The Frobenius–Schur
indicator of the fundamental representation \(\mathbf3\) of \(SU(3)\) is computed two ways: exactly,
from \(\mathrm{FS}(\mathbf3)=\tfrac1{|G|}\sum_g\chi(g^2)\) specialized to \(SU(3)\) 's fundamental (a
genuinely complex representation, \(\mathbf3\not\cong\bar{\mathbf3}\) ), and numerically, via direct
Weyl-integration-formula evaluation of the same group integral. Both give \(\mathrm{FS}(\mathbf3)=0\) 
exactly, agreeing to roughly \(10^{-16}\) (machine precision) between the exact character
computation and the numerical Weyl-integral evaluation — the reality condition is empty 
(neither real nor pseudoreal; genuinely complex). Separately, the CPT/color-conjugation map is
checked to act on the twist pair as an exchange \(T_0\leftrightarrow T_1\) (the pair is vectorlike
under \(SU(3)_c\) ), which constrains the pair jointly but supplies no mechanism to split it.
 Verdict: CANNOT PIN. 

 Discriminator D3 — Dai–Freed anomaly beyond the eta invariant. The full Dai–Freed anomaly
datum on the relevant \(\mathrm{Spin}^c\) bordism group is checked for a mod- \(2\) refinement beyond
the eta-invariant piece. The full anomaly datum is found to be \(C\) -even (it assigns the same
anomaly value to both members of the \(T_0\leftrightarrow T_1\) pair), and the candidate mod- \(2\) 
refinement that would need to be odd is undefined at the point \(\mathrm{FS}(\mathbf3)=0\) found
in D2 (its construction requires a nonzero reality-type structure that D2 has just shown is
absent). Verdict: CANNOT PIN. 

 Discriminator D4 — spin- \(\mathbb C\) determinant-line reality. This discriminator requires a
self-conjugate twist (one fixed by \(T_0\leftrightarrow T_1\) exchange). Checking all six candidates
explicitly (§4(e)): none of the six is self-conjugate — the free (fixed-point-free) action of
conjugation, combined with triality's exchange of the two triality-triples, pairs the six
candidates into three conjugate pairs with no fixed candidate. Verdict: CANNOT PIN — the
discriminator's precondition is empty on this candidate set.

 The forced conclusion. All four discriminators — orientation/Wu data, reality/CPT data,
anomaly data beyond the eta invariant, and determinant-line reality — return CANNOT PIN, each for
a structurally distinct reason (an identity, an emptiness, an undefinedness, and an absence of a
self-conjugate candidate, respectively) rather than a search that ran out of budget. Because the
underlying reason is uniform across all four — every available discriminator built from
shape-only (tangent-bundle-and-representation-theoretic) data is, by construction, built from
objects on which charge conjugation acts trivially, while the sought quantity is \(C\) -odd by
definition — no fifth discriminator constructed from the same category of shape-internal data
could succeed either. This is a completeness proof , not a progress report: reproducing it
requires only verifying (a) the candidate space is genuinely closed under conjugation and triality
(§4(d)), and (b) each of D1–D4 is correctly \(C\) -even by the structural argument given — both
finite, checkable claims, not open-ended searches.

 6. Reproducing the derivation from scratch — a full worked procedure

 A reader holding nothing but the frozen \(K_6=SU(3)/T^2\) root system, the Killing-form metric
convention, and the admissibility chamber \(\vec u\in[1/2,3/2]^3\) can regenerate every claimed
number in this gate via the following steps.

 Step 1 — Build the \(A_2\) root system and identify the Weyl group. From the Cartan basis
 \((h_1,h_2,h_3)\) with \(h_1+h_2+h_3=0\) , write the simple roots \(\alpha_1=(1,-1,0)\) ,
 \(\alpha_2=(0,1,-1)\) , and their sum \(\alpha_1+\alpha_2=(1,0,-1)\) . The Weyl group generated by
reflections in these roots is \(S_3\) , order \(6\) . Check: the half-sum of positive roots
 \(\rho=(1,0,-1)\) satisfies \(\Vert\rho\Vert^2=1^2+0^2+(-1)^2=2\) , matching the banked Killing-norm
value.

 Step 2 — Establish the moduli space and admissibility chamber. Parametrize \(SU(3)\) -invariant
metrics on \(K_6\) by \(\vec u=(u_1,u_2,u_3)\) scaling the Killing-form restriction to
 \(\mathfrak m_1,\mathfrak m_2,\mathfrak m_3\) . Declare the chamber \(\vec u\in[1/2,3/2]^3\) and center
 \(\vec u=(1,1,1)\) . Check: substituting \(x_1=x_2=x_3=1\) into
 \(\mathrm{Ric}_1=\tfrac{x_1^2-x_2^2+6x_2x_3-x_3^2}{12x_1x_2x_3}\) gives
 \(\mathrm{Ric}_i=(1-1+6-1)/12=5/12\) for each \(i\) , and
 \({\rm Scal}=\sum_i{\rm Ric}_i\times(\text{multiplicity }2\text{ per real }2\text{-plane})=6\times5/12=5/2\) ,
matching the banked Einstein-center values exactly.

 Step 3 — Prove criticality by symmetry. Follow §2(a): confirm the doublet representation
admits no \(S_3\) -invariant vector via the character calculation
 \(\tfrac16(2\cdot1+0\cdot3+(-1)\cdot2)=0\) , reproducible from the \(S_3\) character table alone.

 Step 4 — Derive the \(\mathbf3=\mathbf1\oplus\mathbf2\) split. Follow §2(b): confirm the
permutation-representation character \((3,1,0)\) minus the trivial character \((1,1,1)\) gives
 \((2,0,-1)\) , and confirm this is irreducible via \(\tfrac16(4+0+2)=1\) .

 Step 5 — Compute the geometric shape-doublet curvature sign, three ways. (i) Expand
 \(R(\vec u)=\sum_i1/u_i-\tfrac12T(\vec u)\) along \(u(\varepsilon)=(1+\varepsilon,1-\varepsilon,1)\) 
to obtain \(R(\varepsilon)=3/2-\varepsilon^2/2\) (§2(c)). (ii) Build the full \(3\times3\) Hessian of
 \(R\) at \((1,1,1)\) and diagonalize to confirm eigenvalues \(\{3,-3,-3\}\) (§2(c)(iii)). (iii) Project
onto fixed volume via \(S(\vec x)=\sum1/x_i-\tfrac16\sum x_k/(x_jx_l)\) and confirm
 \(d^2S/d\varepsilon^2|_0=+2/3\) , giving physical mass \(^2=-1/3\) under \(V\sim-S\) (§2(c)(iv)). Check
(the \(S^2\times S^2\) negative control): repeat step (i) with the per-sphere curvature
 \(R=2/b^2\) model on \(S^2\times S^2\) and confirm the opposite sign, \(+12\) , emerges — a reader whose
re-implementation returns a negative sign on the \(S^2\times S^2\) control as well has a bug (most
likely a sign error in the structure-constant/triangle term), because the entire evidentiary value
of this step rests on the method discriminating between the two manifolds.

 Step 6 — Compute the bosonic spectral zeta. Follow §2(d): either (Route A) truncate the
Peter–Weyl sum over \((p,q)\) using the Casimir/dimension/multiplicity data, raise each nonzero
Casimir to the power \(-s\) , sum, and Richardson-extrapolate to \(s=-1\) ; or (Route B) apply the exact
Weyl-character/Poisson-resummation closed form directly. Check: both routes must return
 \(\zeta_{K_6}(-1)=-8033/100800=-0.07969246031746032\) to the precision each route supports.

 Step 7 — Attempt to pin the fermionic sign; confirm the exhaustion is complete. Follow §5:
enumerate the six \(\chi=-3\) twist candidates, confirm closure under conjugation and triality, and
run each of D1–D4 against the full set. Check: confirm independently that the candidate set
is genuinely closed (§4(d)) and that each of D1–D4 returns CANNOT-PIN by the structural argument
given (a logical check on premises, not a computational search-budget check).

 Step 8 — Assemble the roll-up. Combine Step 3 (criticality, exact) + Step 4 (Hessian split,
exact) + Step 5 sign (geometric, negative, three independent routes) + Step 6 sign (bosonic zeta,
negative, two independent routes) + Step 7 (fermionic sign, certified-unpinnable) into the stated
gate verdict. Check: confirm no step anywhere in this chain required choosing a numerical
value to match a desired external target — Steps 3–7 are each an algebraic identity, a
closed-form evaluation, or a completeness proof over a finite, independently-verified-closed
candidate set. The single external, measured-vs-model comparison in the entire gate — the
electroweak boundary pull of §2(f) — is explicitly not part of this internal chain and is kept
separately labeled throughout.

 7. What a skeptical reader still cannot get from this section

 In keeping with this gate's own honesty fence, this section closes by stating plainly what the
evidence above does not establish. The evidence lets a reader independently re-derive: the
criticality theorem (exactly); the \(\mathbf3=\mathbf1\oplus\mathbf2\) split (exactly); the geometric
doublet-curvature sign (exactly, \(-1\) raw / \(-\tfrac12\) unit-normalized / \(-\tfrac13\) physical
under fixed-volume projection, by three independent routes, with the \(S^2\times S^2\) control at
 \(+12\) reproducible as a specificity check); the bosonic zeta sign and value (to as many digits as
the reader's own implementation of Routes A and B supports); and the four-discriminator
completeness exhaustion (as a logical argument, checkable premise by premise). It does not 
let a reader derive a numerical value for the fermionic graded-Casimir sign itself — that value
does not exist as a computable quantity from shape-internal data, which is the entire content of
§5 — nor does it let a reader derive \(\theta_H^\star\) or the electroweak hierarchy target-blind
(the \(v_{\rm pred}=246.02\pm3.5\) GeV comparison of §2(f) is an internal-consistency check on the
Hosotani machinery, explicitly not a hierarchy derivation), nor does it certify an all-orders,
all-directions positive-definite Hessian outside the admissibility chamber (the shared-open global
stabilization problem, explicitly out of scope here). A reader who reproduces every step in §6
will arrive at exactly the place this dossier does: a symmetry-forced critical point, two
exactly-signed and multiply-cross-checked curvature layers, and one honestly named, provably
unpinnable, falsifiable bit — no more, and no less.

 Open gaps & the specialist closure path

 This section is written for a specialist who wants to actually close SG-6's residual, not merely
read about it. Every open object below is named precisely, bounded by the same frozen 13D arena
used throughout this dossier, and handed forward with a target-blind success criterion and an
equally explicit refutation criterion. Nothing here is smoothed into a hedge, and nothing here is
inflated into a claim of closure that the fixed grade does not license. The fixed grade —
 DERIVED-GIVEN-anchor / RESOLVED, +0 — does not move regardless of what happens to any item
below: the criticality theorem ( \(\nabla V=0\) at \(\bar u=(1,1,1)\) , forced by exact \(S_3\) symmetry)
and the two computable curvature layers (geometric, negative; bosonic spectral-zeta, negative) are
genuinely, multiply-cross-checked closed, and stay closed no matter how the residuals resolve. What
follows is the honest map of the one remaining live physics front on the stability sign itself
(R3/R6, together with its shared root cause R5), plus three smaller, logically separate residuals
(R1, R7, R9/R2/R4) that ride alongside the gate without being part of the stability verdict.

 Gap 1 (highest leverage) — R3/R6: the fermionic graded-Casimir supertrace sign, the one lever on the stability verdict

 (a) The precise open object. The full one-loop physical Hessian on the shape-doublet ray
 \(u(\varepsilon)=(1+\varepsilon,1-\varepsilon,1)\) factorizes into three multiplicative layers:

 \[
\mathrm{Hess}_{\rm phys}(\varepsilon) \;\propto\; \underbrace{R''(\varepsilon)\big|_0}_{\text{geometric curvature, CLOSED}} \;\times\; \underbrace{\zeta_{K_6}(-1)}_{\text{bosonic zeta, CLOSED}} \;\times\; \underbrace{\mathrm{STr}_{\rm graded}\big[(-1)^F\,d(p,q)\,n(p,q)\,(w_1^2+w_2^2)\big]_{T_0\leftrightarrow T_1}}_{\text{fermionic graded-Casimir supertrace, OPEN}},
\]

 with the first two factors fixed, exact, and negative: \(R''(0)=-1\) raw / \(-1/2\) unit-normalized
(equivalently, the fixed-volume projection gives \(d^2S/d\varepsilon^2=+2/3\) , which becomes \(-1/3\) 
physical once the standard flux-free KK relation \(V\sim -S\) is applied — the same saddle stated in
a second, independently-derived convention), and \(\zeta_{K_6}(-1)=-8033/100800=
-0.0796924603174603\) , confirmed on two structurally independent computational routes agreeing to
roughly 22 digits. The open object is the sign of the third factor : the graded
(bose-minus-fermi) Casimir supertrace, evaluated over the physical KK spectrum built on the frozen
bundle \(E_{\rm frozen}\) (spin- \(\mathbb C\) family index \(\chi(K_6,E)=-3\) ), restricted to the
 \(T_0\leftrightarrow T_1\) ( \(E\leftrightarrow E^*\) ) charge-conjugation twist pair. This reduces to a
counting question — does the boson-weighted sum of Casimir eigenvalues exceed the fermion-weighted
sum, once the doublet-projection weight \((w_1^2+w_2^2)\) is applied? — and it requires exactly two
artifacts that are, at the time of writing, absent from the record, not merely uncomputed : an
 admissible-representation graded multiplicity table (which Dynkin labels \((p,q)\) appear, with
what bosonic multiplicity \(n_b(p,q)\) and fermionic multiplicity \(n_f(p,q)\) , in the physical spectrum
built on \(E_{\rm frozen}\) ), and a regularization-stable analytic continuation to \(s=-1\) of the
resulting graded Dirichlet series. Per-sector signs are already locked and scheme-independent
(bosons contribute \(+\) , fermions contribute \(-\) ); only the net, doublet-weighted balance is owed.

 Note carefully what this residual is not : the frozen record separately proves that this same
 \(T_0/T_1\) charge-conjugation-odd bit is CERTIFIED-UNPINNABLE by any intrinsic invariant of the
frozen shape — a completed, theorem-grade exhaustion (four discriminators, D1 through D4, run
against a proven-closed six-candidate twist space, all returning CANNOT-PIN because every available
shape-only invariant is built from data on which charge conjugation acts trivially, while the
target bit is intrinsically \(C\) -odd). That exhaustion is a different, already-closed result about
 whether an intrinsic-shape discriminator exists at all for the bit; it is not the same question as
 what the graded loop supertrace's net sign actually is once the (absent) multiplicity table is
mounted and evaluated. The unpinnability proof means the eventual sign, once it is computed from
the multiplicity table, cannot be independently cross-checked against a second, shape-internal
invariant — it is a genuinely one-route number, certified un-doubly-checkable in that specific
sense, which is exactly why it is typed as an anchor-payable IOU (payable by a future measured
 \(C\) -odd record) rather than as an ordinary open calculation awaiting a second person's spare
afternoon.

 (b) Why it is hard, and the specific traps to avoid. The difficulty is not conceptual — the
governing formula is completely fixed once \(E_{\rm frozen}\) and the frozen curvature invariants are
in hand — it is that the multiplicity table must be built from the physical KK matter/gauge content
propagating through the spin- \(\mathbb C\) twist, the \(\mathbb Z_2\) orbifold projection, and the
chirality projector \(P_\chi=\tfrac12(1+\gamma_5\Gamma_8)\) simultaneously, for enough of the (in
principle infinite) \((p,q)\) tower that the zeta-regularized sum is under control, not merely for the
lowest handful of representations. Five traps have already caught earlier passes at objects
immediately adjacent to this one, and a specialist should treat each as a named hazard rather than
an abstract warning:

 Reusing the bare-manifold scalar zeta as if it already answered the graded question. The
 closed bosonic layer, \(\zeta_{K_6}(-1)=-8033/100800\) , is the scalar Laplacian zeta on \(K_6\) 
 alone — it does not include the fermionic KK tower, does not carry the doublet weight
 \((w_1^2+w_2^2)\) , and must not be silently substituted for the (currently unbuilt) graded 
 supertrace. The two live on different operators over different bundles; conflating them is exactly
 the truncated-object artifact this dossier's discipline exists to prevent.

 Mixing the raw-ray and unit-normalized curvature conventions when the three-factor product is
 assembled. The convention guard already logged for the geometric layer — raw \(-1\) from
 \(+4-5\) , unit-normalized \(-1/2\) from \(+2-5/2\) , and never the mixed, arithmetically wrong
 \(+2-5/2=-1\) — must be carried through consistently into the full product with the fermionic
 factor, or the final net sign becomes a bookkeeping artifact rather than a physical result.

 Assuming \(\chi=-3\) helps the rescue. The already-computed leading loop indicator,
 \(\mathrm{Str}[C_2]=\chi\cdot C_2(\mathrm{fund})=(-3)\times(4/3)=-4\) , opposes a
 boson-dominance rescue of the saddle. A specialist should not quietly lean on the family index as
 a reason to expect the net sign to flip positive; the honest starting expectation, based on this
 indicator alone, leans toward reinforcing the saddle, and the multiplicity-table computation must
 be run without that hope built in.

 Reviving retracted or branch-killed numbers. Two numerical values from earlier attempts at
 this same potential are explicitly retracted and must not reappear even as sanity checks:
 \(-2.817995812\times10^{94}\,\mathrm{GeV}^6\) (retracted, contaminated by the wrong curvature ratio)
 and its correction \(-2.995681680\times10^{94}\,\mathrm{GeV}^6\) (scheme-anchored only, never a
 clean scheme-independent prediction). Any curvature-squared ratio feeding the heat-kernel side of
 this computation must use the Bianchi-exact \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) — never 
 the branch-kill contaminant \(31/147\) , and never the unrelated round- \(S^6\) value
 \(|\mathrm{Riem}|^2=60\) .

 Target-anchoring the multiplicity table, or the normalization \(\mu_{\rm cell}\) , to force a
 preferred sign. The magnitude threshold for a rescue is \(\mu_{\rm cell}\cdot q\gtrsim0.5\) 
 (sufficient to overcome the geometric \(-1\) ). The only place in the current record where a number
 for \(\mu_{\rm cell}\) could even be read off is the electroweak potential's own extremum condition
 \(\partial_\sigma V=0\) — but that condition is the electroweak-hierarchy condition (the identical
 object R1 below is stuck on), so fixing \(\mu_{\rm cell}\) there to manufacture a stability verdict
 is circular by construction (the \(\kappa^3/\pi\) signature already flagged and forbidden elsewhere
 in this program). A specialist must build the multiplicity table from \(E_{\rm frozen}\) 's own
 spectral content alone, target-blind, exactly as the D1–D4 discriminators were run without any
 reference to a desired outcome.

 (c) What closes it, target-blind, with success criterion and refuting result. Closure is a
two-artifact delivery, computed with no reference at any stage to which sign would be convenient:

 Mount the admissible-representation graded multiplicity table : for every Dynkin label
 \((p,q)\) appearing in \(\mathcal E_{\rm matter}\oplus\mathcal E_{\rm gauge}\oplus\mathcal E_{\rm
 Higgs}\) on \(E_{\rm frozen}\) , tabulate bosonic multiplicity \(n_b(p,q)\) and fermionic multiplicity
 \(n_f(p,q)\) , doublet-weighted by \((w_1^2+w_2^2)\) , using the already-tabulated seed data — e.g.
 \((1,0)/(0,1)=\mathbf3/\bar{\mathbf3}\) , \(C_2=4/3\) , \(m_0=0\) ; \((1,1)=\mathbf8\) , \(C_2=3\) , \(m_0=2\) ;
 \((2,0)/(0,2)=\mathbf6/\bar{\mathbf6}\) , \(C_2=10/3\) , \(m_0=0\) ; \((3,0)=\mathbf{10}\) , \(C_2=6\) , \(m_0=1\) ;
 \((2,2)=\mathbf{27}\) , \(C_2=8\) , \(m_0=3\) — extended to the full graded content, not merely the scalar
 zero-weight sector.

 Perform a regularization-stable analytic continuation to \(s=-1\) of the resulting graded
 Dirichlet series \(\sum(-1)^F d(p,q)\,n(p,q)\,C_2(p,q)^{-s}\,(w_1^2+w_2^2)\) , by the same two-route
 discipline already validated on the pure-boson zeta: an independent numeric heat-sum with
 Richardson extrapolation (Route A), cross-checked against a Weyl-character/Poisson-resummation
 closed form (Route B).

 Success criterion (VERIFIED-RESCUE): the net sign \(\mathrm{Net}>0\) and the resulting
normalized magnitude satisfies \(\mu_{\rm cell}\cdot q\gtrsim0.5\) — enough to overcome the geometric
 \(-1\) . In that case the shape doublet is a genuine minimum; R3, R6, and R8 close together; and the
gate's terminal strengthens from "criticality plus two closed curvature layers plus a certified
IOU" to a fully certified positive-definite doublet Hessian.

 What a refuting result looks like: \(\mathrm{Net}\le0\) , or \(\mathrm{Net}>0\) but below the
 \(0.5\) threshold, computed from a reproducible multiplicity table and a reproducible \(s=-1\) 
continuation that use the certified curvature inputs (never \(31/147\) , never the retracted GeV \(^6\) 
values) throughout. This is not a failure of the research program — it is a legitimate,
complete SADDLE-CONFIRMED, CLOSED-NEGATIVE terminal, on exactly the same footing as the
gate's own negative controls (S²×S² firing the opposite curvature sign) that already certify the
sign-detection machinery is discriminating, not rigged. A referee should treat either outcome as a
valid completion of R3/R6; a result that lands suspiciously exactly at the \(0.5\) threshold with no
independent cross-check should be treated as a signal to re-examine the computation for hidden
anchoring to a desired answer, not accepted at face value.

 (d) Machinery to start from. The starting apparatus is the Peter–Weyl decomposition already
in use throughout this gate, \(L^2(K_6,E_\mu)=\bigoplus_{(p,q)}V_{(p,q)}\otimes
\mathrm{Hom}_{T^2}(V_{(p,q)},E_\mu)\) , restricted to \(E_\mu=E_{\rm frozen}\) , together with the exact
Casimir and dimension formulas \(C_2(p,q)=(p^2+q^2+pq+3p+3q)/3\) , \(\dim(p,q)=(p+1)(q+1)(p+q+2)/2\) , and
the zero-weight multiplicity relation already banked and cross-checked against the tabulated
Dynkin data. The relevant bundle endomorphisms are the ones already fixed at the Einstein center:
the vector/Hodge endomorphism \(E=\mathrm{Ric}=(5/12)\,\mathrm{Id}\) (eigenvalue \(5/12\) , multiplicity
6), and the Lichnerowicz endomorphism \(E_L\) on \(\mathrm{Sym}^2_0\) (dim 20) with spectrum
 \(\{1/6\,(\times6),\,5/12\,(\times6),\,7/6\,(\times6),\,17/12\,(\times2)\}\) ,
 \(\mathrm{tr}\,E_L=40/3\) , \(\mathrm{tr}\,E_L^2=241/18\) . The regularization machinery is the identical
 \(\zeta\) -function continuation already exhibited in full for the pure-boson case — express the graded
sum as a Dirichlet series in \(C_2(p,q)^{-s}\) with signed coefficients, establish the convergence
strip, and continue to \(s=-1\) by Richardson-extrapolated truncation or by recasting through the
Weyl integration formula and a Poisson resummation of the graded heat kernel — applied now to the
signed (bose-minus-fermi) spectrum rather than the unsigned scalar spectrum. The two-route agreement
standard that certified \(\zeta_{K_6}(-1)\) to \(\sim22\) digits is the acceptance bar this computation
should be held to as well.

 (e) Leverage — what else closes if this closes. This is the single highest-leverage missing
artifact in the entire SG-6 residual family, and one of the very few places in this program where
one computation decides several named objects simultaneously, in either direction:

 Closing R3/R6 automatically decides R8 (the conditional "no-tachyon" residual), since R8 is
 explicitly defined as auto-closing with R3+R5 — a decided net sign on the doublet is the
 field-theoretic content R8 was waiting on.

 It discharges the shared keystone bit SAG-A6-KEYSTONE — the identical \(T_0/T_1\) supertrace
 sign counted once across Gap-01, SG-7, uqf10, and gap10-bg10 (banked under certificate
 AC-GAP10-HOLE3-v1) — simultaneously across all four gates, since none of them holds an
 independent copy of this bit; whichever way the sign falls here, it falls the same way everywhere
 else it is used.

 It reconciles the tree-level cross-check already on record — shape-doublet tree mass \(^2=
 \{+1,+1\}\) , which currently only "leans stable" and is explicitly kept separate from the loop
 verdict — with the loop-level result, removing the last "two honest halves never merged" caveat
 in this gate's central finding.

 Because the multiplicity-table-and- \(s=-1\) -continuation machinery this hole needs is exactly the
 machinery the breathing-singlet loop coefficient (Gap 2 / R5, below) also needs, a specialist
 mounting this artifact should expect substantial technology transfer toward R5 as a byproduct,
 not as a second independent effort.

 Falsifier, wired: the bit pays on measurement of a decisive \(C\) -odd record — the leading named
 candidate is the leptonic \(\delta_{CP}\) sign — with the baryon asymmetry \(\eta_B\) permanently
 barred as a payer (an independent, structural argument, not a mere methodological preference).
 A specialist should not treat \(\eta_B\) as an available shortcut to discharging this IOU under any
 circumstances.

 Gap 2 — R5: the breathing-singlet loop coefficient \(c_{\rm loop}=\mathrm{tr}[a_6]\) and the shared root object \(\mu_{\rm cell}\) 

 (a) The precise open object. The breathing-singlet direction ( \(u_1=u_2=u_3\) , decoupled exactly
from the shape doublet by the same \(3=1\oplus2\) representation-theoretic split used throughout this
gate) carries its own one-loop curvature contribution, \(c_{\rm loop}=\mathrm{tr}[a_6]\) : the trace of
the sixth Seeley–DeWitt heat-kernel coefficient for the breathing-mode ( \(\sigma\) ) fluctuation
operator. This is the identical a₆ graviton leg flagged as OWED at the Gelfand–Tsetlin
off-diagonal stratum — the Lichnerowicz first-order operator on \(\mathrm{Sym}^2_0\) mixes the five
Weyl-inequivalent \(T^2\) weight classes through off-diagonal (hopping) connection matrix elements,
each individually an exact \(SU(3)\) Gelfand–Tsetlin ladder matrix element (a standard
lowering-operator formula — the square root of a product of pattern-entry differences) but not yet
enumerated as a completed set. Two independent computation routes exist for this coefficient and,
by design of the audit, have not yet been reconciled : Route A (Gilkey/Lichnerowicz on
 \(\mathrm{Sym}^2_0\) ) consumes the certified \(E_L\) spectrum and the Riemann curvature 2-form but needs
the GT hopping term to finish; Route B (ghost-plus-vector reconstruction) consumes the certified
vector endomorphism \(E=\mathrm{Ric}\) and a scalar backbone already banked across three or more
independent engines, \(a_6/a_2^3=7936/39375\) , but its own graviton leg is independently OWED. The
two-route agreement standard used everywhere else in this gate (as with \(\zeta_{K_6}(-1)\) 's
 \(\sim22\) -digit cross-route agreement) has explicitly not been met here.

 (b) Why it is hard, and the specific traps. The underlying reason this is hard is structural,
not a matter of insufficient effort: \(K_6\) is homogeneous but not locally symmetric 
( \(\|\nabla\mathrm{Riem}\|^2=1/4\ne0\) ), which is precisely why the a₆ graviton leg carries a
nontrivial Gelfand–Tsetlin ladder term that a locally symmetric space would not have. The traps:

 Treating a single-route result as sufficient. Because the two-route agreement standard has
 explicitly not yet been met, a Route-A-only or Route-B-only number, however clean it looks, should
 not be accepted as closing R5 — the corpus's own discipline (used successfully elsewhere in this
 gate to certify \(\zeta_{K_6}(-1)\) ) is cross-route agreement, and this residual should be held to
 the same bar.

 The \(31/147\) contamination. This is the same certified branch-kill named throughout this
 dossier; it has already once re-contaminated exactly this leg (an earlier pass substituted
 \(31/147\) for the Bianchi-exact \(23/75\) curvature ratio and had to be walked back). The correct
 ratio is \(|\mathrm{Riem}|^2/\mathrm{Scal}^2=23/75\) , never \(31/147\) , anywhere in the \(a_6\) 
 computation.

 Reviving the withdrawn numerical value. An earlier evaluation using the contaminated
 coefficient produced \(-2.817995812\times10^{94}\,\mathrm{GeV}^6\) ; this is retracted and must never
 be re-quoted, in this section or any downstream one. The corrected value,
 \(-2.995681680\times10^{94}\,\mathrm{GeV}^6\) , is scheme-anchored only — it is not a
 scheme-independent prediction, and a specialist reproducing it should expect the same
 scheme-dependence and flag, not silently absorb, any deviation from it under a different scheme.

 The cardinal trap, shared with R1 and R3/R6: anchoring \(\mu_{\rm cell}\) at the electroweak
 minimum. The Uniform Operational Cell discretization ( \(N\le B/\Delta_0\) ) genuinely dissolves the
 continuum UV divergence in the breathing-mode loop — that leg is real and banked, not a residual.
 What survives is a single log-scheme factor, \(\mu_{\rm cell}\) , the spectral value of the
 operational cell \(\Delta_0\) , which has no \(v\) -independent readout anywhere in the frozen
 record: the only place a number for it could currently be read off is \(\partial_\sigma V=0\) ,
 which is the electroweak-hierarchy condition. This is forbidden, not merely disfavored, for the
 same reason it is forbidden in R1 and R3/R6: it is circular by construction.

 (c) What closes it, target-blind, with success/failure criteria. Closure requires: (i)
enumerating the Gelfand–Tsetlin ladder matrix elements completing the Lichnerowicz off-diagonal
hopping term on \(\mathrm{Sym}^2_0\) — a finite, well-posed representation-theory computation using
the standard \(SU(3)\) lowering-operator formula, with no new physics input, i.e. "owed" in the sense
of unassembled rather than unknown-in-principle; (ii) completing Route A and Route B independently
using the corrected \(23/75\) coefficient throughout; and (iii) reconciling the two to within
 \(10^{-6}\) . Success criterion: Routes A and B agree to \(10^{-6}\) or better, establishing
 \(c_{\rm loop}\) as a certified, scheme-fixed number (still contingent on a declared regularization
scheme, since \(\mu_{\rm cell}\) 's magnitude is a floor-residue on the footing of \(\hbar\) , not a
scheme-independent output in its own right). Failure/refutation signature: the two routes
disagree outside \(10^{-6}\) after both are honestly completed with the corrected coefficient — this
would indicate either an error in the GT enumeration or a genuine scheme-dependence larger than
expected, and must be reported as such, not forced into agreement by retroactively adjusting either
route's regularization prescription.

 (d) Machinery to start from. The certified starting inputs: the \(E_L\) spectrum on
 \(\mathrm{Sym}^2_0\) (dim 20) with eigenvalues \(\{1/6\,(\times6),\,5/12\,(\times6),\,7/6\,(\times6),
\,17/12\,(\times2)\}\) , \(\mathrm{tr}\,E_L=40/3\) , \(\mathrm{tr}\,E_L^2=241/18\) ; the banked scalar
backbone \(a_6/a_2^3=7936/39375\) ; the vector heat-kernel traces \(\mathrm{tr}\,a_2=0\) ,
 \(\mathrm{tr}\,a_4=-47/360\) ; and the \(S^6\) round-unit calibration row ( \(a_2/a_0=5\) , \(a_4/a_0=12\) ,
 \(a_6/a_0=1139/63\) ) as the sanity check that the general Gilkey-constant machinery is correctly
implemented before it is turned on the harder \(K_6\) Lichnerowicz operator (this calibration is
already a passed control confirming \(K_6\ne S^6\) ). The fixed-volume model functional
 \(S(u)=\sum_i1/x_i-(1/6)\sum x_i/(x_jx_k)\) is the correctly-normalized starting object for organizing
the breathing-mode fluctuation before the heat-kernel expansion is applied.

 (e) Leverage — what else closes if this closes. R5 is, together with R3/R6, one of the two
named ingredients in this gate's own definition of a fully decided stability leg — completing both
converts the roll-up into an unconditional terminal. Beyond that direct link, R5 sits on the same
shared object as two other fronts: R1 (the electroweak-hierarchy residual) and R3/R6 (the doublet
Casimir magnitude) all bottom out on the identical spectral value \(\mu_{\rm cell}\) . This means a
specialist attacking either R3/R6 or R5 first should expect substantial technology transfer toward
the other — completing the GT enumeration and reconciling the two \(a_6\) routes very plausibly
produces, as a direct byproduct, the same regularization-stable continuation machinery R3/R6 needs,
and vice versa.

 Gap 3 — R1: the Hosotani phase \(\theta_H^\star\) and the electroweak-hierarchy relocation

 (a) The precise open object. The Higgs is realized as a Wilson-line (Hosotani) mode with
topologically fixed integer winding \(n_H=1\) on a cycle of radius \(R_\gamma\sim R_0=
1.591549430918954\times10^{-17}\,\mathrm{GeV}^{-1}\) . The physical VEV is
 \(v_{\rm EW}=\theta_H^\star/(2\pi R_\gamma)\) , where \(\theta_H^\star\approx2.46\times10^{-14}\) is the
location of the minimum of the one-loop Hosotani potential
$$
V_{\rm Hos}(\theta_H) = -\frac{3}{64\pi^6R_\gamma^4}\sum_{n=1}^\infty\frac{1}{n^5}\big[N_b\cos(n\theta_H)-N_f\cos(n\theta_H)\big],
$$
an absolutely convergent ( \(n^{-5}\) ) sum. This single phase carries roughly 85% of the full
electroweak hierarchy. The open object is whether \(\theta_H^\star\) is derived target-blind from
 \(\{n_H=1,\,\eta_{BK},\,R_\gamma\}\) and a UV reference scale fixed before any comparison to
 \(v_{\rm obs}=246\) GeV, or whether it is merely read off the location of a minimum that was found
by looking for the answer.

 (b) Why it is hard, and the specific traps. Two attempted repair paths are already named and
both independently fail, for structurally different reasons, and a specialist should not simply
retry a variant of either without addressing the specific named failure mode. Path A (the
currently used path) reads \(\theta_H^\star\) directly off \(\partial V_{\rm Hos}/\partial
\theta_H|_{\theta_H^\star}=0\) and reports the resulting \(v_{\rm EW}\) as a prediction — but this
extremum condition is mathematically the same object that would need solving to "predict" \(v\) in
the first place, so the path is circular at \(\partial_\sigma V=0\) : the numerical output is not
in question, but the logical direction of the derivation runs backward from where target-blindness
requires it to run. Path B (treating the tiny hierarchy \(\theta_H^\star\sim10^{-14}\) as a
dimensionally transmuted running coupling) is independently flagged as wrong-mechanism : a
periodic Hosotani potential's minimum location is a discrete extremum-selection fact, not a
renormalization-group exponential, and dressing it as if it obeys an RG equation misapplies the
transmutation mechanism to an object without the structure to support it.

 (c) What closes it, target-blind, with success/failure criteria. Closure requires deriving
 \(\theta_H^\star\) forward from \(\{n_H=1,\,\eta_{BK}=1/(32\pi\,e^{\sqrt3/(24\pi)})=
0.009721281516312024,\,R_\gamma\}\) plus a UV reference scale from the already-fixed tower
( \(M_U=1.0\times10^{16}\) GeV, \(R_0\) , or the Cartan-torus radius \(R_{T^2_{\rm Cartan}}=R_0\sqrt2\,
3^{-1/4}=1.710231163476377\times10^{-17}\,\mathrm{GeV}^{-1}\) ) that is demonstrably fixed before 
any comparison to \(v_{\rm obs}\) is made — a write-once derivation chain: fix the UV input, run the
calculation forward, only then compare. Success criterion: a forward-run \(\theta_H^\star\) (or
equivalently the hierarchy exponent) that lands within the propagated uncertainty of the
already-quoted \(v_{\rm pred}=246.02\pm3.5\) GeV band, with the UV reference fixed prior to
comparison. Failure/refutation signature: either no forward construction from
 \(\{n_H,\eta_{BK},R_\gamma\}\) reproduces the observed hierarchy at all (a genuine refutation of
derivability), or every attempted construction can be shown to covertly reintroduce the extremum
condition as an input — in which case the honest conclusion is that \(v_{\rm EW}\) should be formally
promoted from "consumed anchor with a pending derivation" to a second irreducible measured
anchor , AXIOM-VEW-SECOND-ANCHOR , on the same permanent footing as \(\Lambda\approx10^{-122}
M_{\rm Pl}^4\) and \(\eta_B\) .

 (d) Machinery to start from. The already-certified structural facts: \(n_H=1\) is an exact,
hand-checkable integer (the minimum winding producing any nonzero VEV at all under the Wilson-line
protection mechanism, since \(n_H=0\) gives no VEV); \(\eta_{BK}=1/(32\pi\,e^{\sqrt3/(24\pi)})=
0.009721281516312024\) is an exact closed-form combination of already-fixed geometric constants; the
identity \(v/M_H^{\rm eff}=\sqrt{\eta_{BK}}/(2\pi)\approx0.01569212979293374\) already relates the two
sides of the hierarchy ratio; and the absolutely convergent \(n^{-5}\) tail of \(V_{\rm Hos}\) is the
structural reason the Higgs mass comes out finite under the declared regulator at all — a genuine,
banked, protection result, though the protection alone does not by itself explain why the hierarchy
sits at the observed scale rather than merely being finite.

 (e) Leverage — what else closes if this closes. R1 is logically independent of the stability
sign (R3/R6/R8) — closing it does not, by itself, decide whether the shape doublet is a minimum
or a saddle. Its leverage is on this gate's own honesty bookkeeping and on the neighboring
electroweak-sector gate: it would replace a "read off the answer, relocated" residual with either a
genuine target-blind derivation, or a clean, permanent promotion of \(v_{\rm EW}\) to a second
irreducible anchor — either outcome is a strict improvement in the precision of what is claimed,
even though neither changes SG-6's own fixed grade (R1 is already carried as a relocation to
 AXIOM-VEW-SECOND-ANCHOR , not as part of the stability verdict).

 Gap 4 — R7: realization-specific uniqueness of the modular fixed point \(\tau=\omega\) 

 (a) The precise open object. The Cartan-torus modulus of the finite chamber \(F^+\) is declared
at \(\tau=\omega=e^{2\pi i/3}=-0.5+0.8660254037844386\,i\) . The generic fact — that \(\tau=\omega\) 
is an order-three fixed point of \(PSL(2,\mathbb Z)\) acting on the upper half-plane — is already
fully proven, standard modular-curve theory, and carries no residual. What is not yet proven is the
sharper, realization-specific claim: that \(\tau=\omega\) is the unique fixed point of the
 specific Cartan-torus generator realized by the \(F^+\) construction, once restricted to the
Weyl-rigid admissibility chamber already selected for the shape moduli. This is pure group theory
with zero flavor-physics content, and it is explicitly non-blocking for the fixed grade — theorem
debt, not a live physical uncertainty.

 (b) Why it is hard, and the specific traps. \(PSL(2,\mathbb Z)\) has exactly two special orbits of
elliptic fixed points on the modular curve — the order-two point \(\tau=i\) and the order-three point
 \(\tau=\omega\) (and its images) — so generic uniqueness up to the modular group's own identifications
is not in question. What remains is showing the realized generator — the specific element singled
out by the \(F^+\) construction's own Cartan-torus data — has \(\tau=\omega\) as its only fixed point
once the chamber restriction is imposed. The trap is scope-creep in the wrong direction: it would be
easy to re-derive the generic \(PSL(2,\mathbb Z)\) fixed-point classification and mistakenly present
that as settling the realization-specific question. A second trap is injecting flavor data after the
fact — the sector projectors \(\Pi_u,\Pi_d,\Pi_e,\Pi_\nu\) and chamber angle \(\theta_F\) are fixed
downstream of \(\tau\) by the \(|V_{us}|\) anchor, so using them to argue backward for uniqueness would
be circular, of the same species already forbidden for \(\mu_{\rm cell}\) .

 (c) What closes it, target-blind, with success/failure criteria. Exhibit the specific modular
generator realized by the \(F^+\) construction as an explicit element of \(SL(2,\mathbb Z)\) acting on
the Cartan-torus modulus, and prove by direct computation, restricted to the admissible chamber,
that \(\tau=\omega\) is its unique fixed point there. Success criterion: a closed, checkable proof
with the generator written down explicitly and the fixed-point count verified. Failure/refutation
signature: discovery of a second admissible fixed point of the same realized generator within the
chamber — which would not overturn the criticality theorem (which does not depend on \(\tau\) at all)
but would downgrade R7 from "theorem-debt, expected true" to "an additional axiom is needed here
too."

 (d) Machinery to start from. Standard \(SL(2,\mathbb Z)\) elliptic-point theory: the stabilizer of
 \(\tau=\omega\) is the cyclic order-three group generated by an explicit Möbius transformation, and the
fundamental domain carries exactly one \(SL(2,\mathbb Z)\) -orbit of order-three points. The task is to
identify the Cartan-torus generator inside \(F^+\) 's own data (the generation basis
 \(\mathcal G_{\rm gen}\) , \(\dim_{\mathbb C}=3\) , and the diagonal chamber operators \(O_u,O_d,O_e,O_\nu\) 
already built at \(\tau=\omega\) ) as an explicit modular transformation, then apply the standard
fixed-point-counting argument restricted to the chamber.

 (e) Leverage — what else closes if this closes. A closed R7 upgrades this object from
Reduced-to-Axiom to genuine DERIVED — a strict strengthening at no cost, since nothing in SG-6's
stability sign depends on \(\tau\) at all. It also hardens the flavor gate's inherited
 \(\tau=\omega\) soft spot, since that gate uses the same declared modular fixed point without
re-deriving it. This is the cheapest item in this section: contained, self-terminating group theory
with a clear proof target and zero dependence on the harder absent-data problems dominating Gaps
1–3.

 What is deliberately not listed as an open specialist target

 Three items named in this gate's own ledger are not treated above as specialist-closure targets
because they are already-terminal dispositions, and a reviewer should not mistake them for
outstanding work:

 R2 (global, all-orders, all-directions stabilization outside the admissibility chamber) is an
 explicit disclosed non-claim , not a gap awaiting a specialist. The admissibility chamber
 \(\vec u\in[1/2,3/2]^3\) is a Rulebook selection rule — it tells you which configurations are
 legal, not which are dynamically favored — and no dynamical restoring force keeping a modulus
 inside (or returning it to) the chamber is asserted anywhere in this gate. This is the field-wide
 SHARED-OPEN problem every extra-dimensional program on Earth carries (string/M-theory flux
 compactifications face the identical discretuum-selection-versus-global-stability gap); nothing
 internal to this construction flips it, and the only actionable discipline is never to write "full
 stabilization" or "no flat directions anywhere" without the chamber qualifier attached.

 R4 (given- \(E\) within the selected chamber) is a relocation , not a residual of this gate.
 Every witness SG-6 certifies — the chamber-center criticality, the \(3=1\oplus2\) split, the
 \(n_H=1\) winding, the \(\tau=\omega\) declaration — operates inside an already-selected background
 \(E_{\rm frozen}\) ( \(\chi(K_6,E)=-3\) ). Whether witnesses determine that background, as opposed to
 certifying moduli control within it, is a question that belongs to the geometry-selection gates
 (SG-1/SG-3), not to SG-6. Nothing at SG-6 closes or fails to close based on this question.

 R9 (phenomenological sufficiency / downstream threshold-vector sensitivity) inherits its audit
 from the gauge-unification threshold-vector gate and is explicitly not an SG-6 computation:
 the threshold vector \((\delta_1,\delta_2,\delta_3)=(+4.8424,-3.1112,-1.7313)\) has certified
 signs on every packet but currently fitted (scheme-anchored) magnitudes , and a specialist
 working SG-6 stability should not attempt to fix that threshold-vector fitting as part of closing
 R3/R6 or R5 — doing so would conflate the internal-geometry Hessian sign with an unrelated RG
 threshold-matching question, and any sensitivity check run against a currently-fitted threshold
 vector would not be target-blind. R9 becomes a meaningful downstream robustness pass only after
 the neighboring gate's own artifacts exist independently.

 The cardinal collapse, restated as a ranked work plan

 Rank 
 Residual 
 Decides SG-6's own stability sign? 
 What closes it 
 What a refuting result looks like 

 1 
 R3/R6 — fermionic graded-Casimir net sign 
 Yes — the only remaining lever 
 Mount the admissible-rep multiplicity table + two-route \(s=-1\) -continued graded zeta 
 \(\mathrm{Net}\le0\) (or below the \(0.5\) threshold) ⇒ legitimate SADDLE-CONFIRMED , closing the gate on the negative side 

 2 
 R5 — breathing-singlet \(c_{\rm loop}=\mathrm{tr}[a_6]\) 
 Indirectly — shares \(\mu_{\rm cell}\) and machinery with R3/R6 
 Complete GT off-diagonal enumeration; reconcile Routes A/B to \(10^{-6}\) using \(23/75\) 
 Routes A and B disagree beyond \(10^{-6}\) after honest completion 

 3 
 R1 — \(\theta_H^\star\) / electroweak-hierarchy derivation 
 No — logically independent of the Hessian sign 
 Forward-derive \(\theta_H^\star\) with UV reference fixed pre-comparison, or promote \(v_{\rm EW}\) to a second irreducible anchor 
 No forward construction reproduces the hierarchy, or every attempt is shown circular 

 4 
 R7 — realized-generator uniqueness of \(\tau=\omega\) 
 No — \(\tau\) -independent 
 Exhibit the realized generator explicitly; prove unique fixed point in-chamber 
 A second admissible fixed point found in-chamber 

 Only Gap 1 (R3/R6) can move SG-6's own stability verdict. It is a genuine two-sided, target-blind
bet: the identical multiplicity table and continuation that could certify a minimum could instead
certify a saddle, and both outcomes are honest, complete, publishable terminals — a
correctly-computed and honestly-reported SADDLE-CONFIRMED result closes this gate exactly as fully
as a VERIFIED-RESCUE would. Gaps 2–4 are named at full precision because an honest ceiling requires
naming every load-bearing assumption and every shared object, not because any of them threatens the
fixed grade DERIVED-GIVEN-anchor / RESOLVED, +0 stated for this gate.

 Honest ceiling, scope & the endpoint

 This closing section draws the line explicitly, in both directions: it states exactly what SG-6 does
 not claim — so that no reader, referee, or later dossier can round the gate's genuinely closed
content up into something stronger than what was shown — and it then states, with equal precision,
what is closed, why that closure is a real terminal and not a placeholder, and what remains as a
named, anchor-payable IOU. The fixed grade for this gate, DERIVED-GIVEN-anchor / RESOLVED, +0 ,
is not touched by anything below; the purpose here is to make the ceiling of that grade legible, not
to move it.

 1. What is explicitly NOT claimed

 (a) Dissolved ≠ solved. No part of the SG-6 stability question is dissolved in the technical
sense this corpus reserves for that word — there is no argument here that the minimum-vs-saddle
question is a category error, a truncated-object artifact, or a question that evaporates once the
right frame is chosen. Two of the three ingredients needed to answer it (the geometric curvature layer
and the bosonic spectral-zeta layer) have been computed exactly and multi-route , not dissolved
away; the third (the fermionic graded-Casimir supertrace sign on the \(T_0\leftrightarrow T_1\) 
( \(E\leftrightarrow E^*\) ) charge-conjugation pair) has been shown to be CERTIFIED-UNPINNABLE by any
intrinsic invariant of the frozen shape — a proven absence of a discriminator inside this geometry,
not a demonstration that the question itself was ill-posed. An unpinnable sign is not a dissolved
question: it remains a real, meaningful, falsifiable question whose answer this geometry cannot supply
from its own resources, and which is therefore routed to an external measurement rather than argued
away. Reading "no internal invariant can carry this sign" as "the sign question wasn't real" would
misstate the result in the direction of a dissolution that was never demonstrated and is not being
claimed anywhere in this gate.

 (b) Selection ≠ derivation. The admissibility chamber \(\vec u \in [1/2,3/2]^3\) , Weyl-rigid, with
center \(\vec u = (1,1,1)\) , is a rulebook object (⊕-layer, AXIOM-CHAMBER-RESTRICTION ) — a
statement of which configurations are legal to compare, not a dynamical potential well that pushes
the moduli toward the center. Off-chamber configurations are eliminated by the selector's admissibility
rule, at the level of \(\mathcal{C}_{\rm admiss}\) , before any potential is even evaluated; they are not
disfavored by a computed energy cost. It would be easy to misread "the center is the only admissible
point" as "the center is the true minimum of some landscape," when in fact the only dynamical claim
proven here is the separate, narrower one: that given the chamber, and given that a potential
 \(V\) built from the \(S_3\) -covariant geometric data is \(S_3\) -invariant, the \(S_3\) -fixed point \((1,1,1)\) 
is automatically a critical point, \(\nabla V|_{(1,1,1)} = 0\) , forced by symmetry alone — the way a ball
at the center of a three-fold-symmetric bowl feels zero net force regardless of the bowl's exact
profile. That is genuine derivation: a target-blind theorem about any \(S_3\) -invariant \(V\) ,
independent of the specific profile. Restriction to the chamber in the first place is selection (a
rulebook admissibility statement). The two are never merged into a claim that "the geometry chooses to
sit at the center because it is lowest in energy there" — that sentence has not been earned and is not
asserted anywhere in this gate's closure.

 (c) Given- \(E\) ≠ derivation of \(E\) . Everything certified in this gate — the criticality theorem,
the \(3 = 1 \oplus 2\) Hessian split, the geometric curvature sign, the bosonic zeta sign — is proven
 within the already-selected frozen geometry \(E\) (the specific bundle data: spin- \(\mathbb{C}\) twist
with family index \(\chi(K_6,E) = -3\) , the six \(\chi=-3\) Dynkin-label twist candidates, the particular
endomorphisms \(E_{\rm matter}, E_{\rm gauge}, E_{\rm Higgs}, E_{\rm proton}\) fixed at the
 \(\otimes\) -Actors layer). SG-6 certifies moduli-control within the selected chamber and the selected
 \(E\) — it proves the vacuum does not slide away inside that fixed geometric data. It does not 
prove that the witnesses (the four irreducible anchors \(\{M_{\rm Pl}, \alpha_i(M_Z), y_t, |V_{us}|\}\) 
together with the admissibility rules) uniquely pick out this \(E\) among some larger space of
candidate bundle data, as opposed to merely being consistent with it once it is already chosen. That
question — does the frozen geometry's witness set determine \(E\) , or only certify stability given 
 \(E\) ? — is the residual labeled R4 in this gate's own ledger, and it is explicitly not owned by
SG-6 : it is relocated to SG-1/SG-3, the gates whose job is the selection-of-geometry question itself.
"Given- \(E\) , the vacuum is symmetry-forced to sit at a critical point of a computable, partially-signed
Hessian" and "the anchors derive \(E\) uniquely" are two different claims belonging to two different
gates, and only the first is SG-6's to make.

 A sharper version of this same point governs the one residual bit directly: the fermionic layer needs
(i) the admissible-representation graded multiplicity table (which \((p,q)\) irreps of \(K_6\) appear, with
what multiplicity, in the physical spectrum built on \(E\) ) and (ii) a regularization-stable analytic
continuation of the relevant zeta function to \(s=-1\) . Both are properties of the chosen bundle \(E\) ,
not properties of the bare manifold \(K_6\) — which is exactly why they cannot be read off the
 \(K_6\) -only curvature invariants that close the first two layers at +0 (those are Shape-only and do
not need to know what \(E\) is). "Given \(E\) " is not the same statement as "this dossier has computed the
consequence of \(E\) "; the multiplicity table is, at present, an ABSENT input — not wrong, not disputed,
simply not yet mounted — and this dossier names that absence rather than manufacturing the table to
force a verdict.

 (d) No claim of a proven positive-definite Hessian, at any loop order. The phrases "minimum
proven," "no-tachyon certified," and "SG-6 closed" (in the sense of "the vacuum is a genuine minimum")
are each individually forbidden while the fermionic sign remains unpaid, and none of them is used,
implied, or approached anywhere in this dossier. What is proven is a factorized, partially-signed 
structure: the physical one-loop Hessian on the shape-doublet ray decomposes as geometric curvature
 \(\times\) bosonic zeta \(\times\) fermionic supertrace, of which the first two factors are each computed
exactly, multi-route, and negative — \(R''(0) = -1\) raw / \(-1/2\) unit-normalized (equivalently, the
fixed-volume projection gives \(d^2S/d\varepsilon^2|_{\rm fixed\text{-}vol} = +2/3\) , which becomes
 \(-1/3\) in the physical potential via \(V \sim -S\) — the same saddle in every convention) and
 \(\zeta_{K_6}(-1) = -8033/100800\) — and the third factor is proven unpinnable from inside this
geometry alone . Stated as one bounding sentence: two of three signed factors are certified negative;
the third is certified undecidable from any internal invariant; therefore the net sign of the
loop-level physical Hessian is not certified in either direction by this gate . Kept rigorously
separate from that loop-level statement, the tree-level shape-doublet mass² is \(\{+1,+1\}\) (symbolic
 \(6k-18m\) shape block) — a structurally positive tree result, and the documented reason the gate leans 
toward eventual stability rather than toward the certified-negative loop-level geometric curvature.
This tree number is never merged with the loop-level curvature number to manufacture a single spurious
verdict; the two are reported as two separate, honestly labeled computations, and the gate's terminal
rests on neither one alone.

 Global stabilization is not claimed. Nothing here asserts that the vacuum is stabilized against
 every conceivable deformation, on or off the admissibility chamber, at all loop orders, forever. The
chamber restriction \(\vec u \in [1/2,3/2]^3\) is a rulebook boundary, not a proof that potential wells
exist and are bounded outside it; there is no all-loop positive-definite Hessian computed or claimed;
there is no claim of "no flat directions anywhere." This is explicitly flagged as SHARED-OPEN with
every other extra-dimensional program on the table — string, M-theory, noncommutative-geometry,
lattice constructions — all of which face the identical open global-stabilization question outside
whatever local patch each has examined. The honest verdict on this shared front is a tie , not a
the framework-specific win and not a framework-specific deficiency.

 The electroweak hierarchy is not derived here. \(v_{\rm EW} = 246\) GeV is consumed as a second,
independently measured ruler — it is not produced by SG-6's own machinery. The Hosotani phase
 \(\theta_H^\star \approx 2.46 \times 10^{-14}\) , which carries roughly 85% of the full hierarchy
magnitude through \(v_{\rm EW} = \theta_H^\star/(2\pi R_\gamma)\) , is read off the location of the
one-loop \(V_{\rm Hos}\) minimum — it is not derived target-blind from frozen UV data fixed before 
comparison to the observed value. Anchoring the unknown normalization scale \(\mu_{\rm cell}\) at the
condition \(\partial_\sigma V = 0\) to "predict" \(v\) is circular by construction, because that extremum
condition is the electroweak hierarchy condition itself (the \(\kappa^3/\pi\) circularity signature);
this move is named and forbidden everywhere it tempts, both here and in the shared breathing-singlet
residual R5.

 Witnesses fix moduli only within the selected chamber. Restated for emphasis because it is the
single easiest point to blur: SG-6's certification is a within-chamber statement. It says nothing
about whether the four irreducible anchors force the selection of this particular \(K_6 = SU(3)/T^2\) 
geometry (with this particular twist data) over some alternative geometry a different program might
propose. Anchoring \(\mu_{\rm cell}\) at \(\partial_\sigma V=0\) to "predict" \(v_{\rm EW}\) is the
identical, explicitly forbidden circularity (the kill-test) — not attempted anywhere in this closure.

 Finally, as a bookkeeping non-claim rather than a physics one: the frozen branch identifiers used
elsewhere to track which version of the geometric record a given computation was run against are
 audit anchors only . They certify reproducibility of the computational record; they carry no
physics content and are not cited as evidence for anything asserted in this section.

 2. The anchors paid

 Every claim above and in the rest of this gate's derivation chain reduces, without exception, to the
 same four irreducible measured anchors that ground the entire 13-dimensional frozen arena:
 \(\{M_{\rm Pl},\ \alpha_i(M_Z),\ y_t,\ |V_{us}|\}\) . SG-6 introduces no new free parameter and pays no
additional anchor beyond these four to establish its criticality theorem, its Hessian split, or
either of its two certified curvature-sign layers — those are pure consequences of the frozen shape
 \(K_6 = SU(3)/T^2\) (fixed once and for all by the anchor-derived radius \(R_6 = R_0 =
1.591549430918954\times10^{-17}\ \mathrm{GeV}^{-1}\) at chamber center) together with exact
representation theory of \(A_2 = \mathfrak{su}(3)\) : the criticality theorem needs no input beyond " \(V\) 
is built from \(S_3\) -covariant geometric data," already guaranteed by the frozen shape; the geometric
curvature sign follows from the Killing-form Ricci/Riemann data already tabulated at the Einstein
center ( \(\mathrm{Ric}_i = 5/12\) , \(\mathrm{Scal} = 5/2\) , \(|\mathrm{Riem}|^2 = 23/12\) , ratio
 \(|\mathrm{Riem}|^2/\mathrm{Scal}^2 = 23/75\) — never \(31/147\) — all exact rationals with no free scale);
the bosonic zeta sign, \(\zeta_{K_6}(-1) = -8033/100800\) , is computed two independent ways (blind
numeric heat-sum with Richardson extrapolation; exact Weyl/Poisson closed form) agreeing to
approximately 22 digits, again with no free input beyond the frozen shape.

 What SG-6 does consume , without producing a pull on it or feeding back into it, is the second
ruler \(v_{\rm EW} = 246\) GeV — used only to state the (currently un-derived) size of the hierarchy
carried by \(\theta_H^\star\) , on the same permanent anchor footing as \(\Lambda \approx
10^{-122}\,M_{\rm Pl}^4\) and the baryon asymmetry \(\eta_B\) : a genuinely independent measured input, not
something SG-6 explains or is required to explain in order to hold its fixed grade.

 The single remaining ingredient, the fermionic supertrace sign, is not "paid" by any anchor at all in
this gate: it is proven unpinnable , and is instead dressed as a forward-looking,
anchor- payable — not yet paid — IOU , banked under the shared certificate
 AC-GAP10-HOLE3-v1 (the same T0/T1 bit shared across uqf10, SG-6, and gap10-bg10's Hole #3, counted
once across all three so a single open physics fact is never triple-counted as three separate
residuals). Its designated future payer is a measurement , not a fifth free geometric anchor: the
sign of the leptonic CP phase \(\delta_{CP}\) (or any other admissible C-odd observable record). The
baryon asymmetry \(\eta_B\) is explicitly and permanently barred from serving as that payer (wall
W12), because \(\eta_B\) is itself downstream of a separate, already-flagged Pin \(^-\) /Gauss-sum axiom bit
— the \(\sigma_\nu = +1\) leptogenesis sign choice ( \(\mathbb{Z}/8\) Arf–Brown–Kervaire data; Gauss sums
 \(|G|=4\) with phases \(e^{\pm i\pi/4}, e^{\pm i3\pi/4}\) ), which this geometry's own default actually
 disfavors ( \(\sigma = 5 \bmod 8\) rather than the needed \(\sigma = +1 \bmod 8\) , a \(+4 \bmod 8\) flip
that is itself an unforced axiom bit). Using \(\eta_B\) here would launder one open sign through another
rather than genuinely closing either.

 In short: the anchor ledger for SG-6's certified content is exactly the same four irreducible anchors
as the rest of the arena, plus the one consumed second ruler \(v_{\rm EW}\) for the (separately tracked,
non-blocking) hierarchy-relocation statement; and the one uncertified piece is not bought with an
anchor at all, but is explicitly priced as a future-measurement IOU with a named payer and a named
barred non-payer.

 3. Why this is a real terminal, not a stalled computation

 It is worth stating plainly why "CERTIFIED-UNPINNABLE" is entitled to be treated as a closed leg of
this gate rather than as an open computational front still awaiting more effort — the surface texture
of "we don't know the sign of one term" can easily be misread as "we haven't finished computing it
yet."

 The distinguishing fact is that unpinnability was established by a complete, theorem-grade
exhaustion , not by a search that ran out of time or ideas. The T0 ↔ T1 exchange (the
charge-conjugation action \(E \leftrightarrow E^*\) on the twisted spinor bundle over \(K_6\) ) was tested
against all four available classes of intrinsic discriminator that could, in principle, carry a
sign on this frozen shape, over a provably closed, six-candidate twist space (closed under both
conjugation and triality — no seventh candidate exists, and this closure is itself a certified
topological fact about the frozen geometry, not an assumption):

 D1 (Wu/ \(w_2\) / \(w_3\) orientation classes): identically zero on the frozen shape and purely
 tangential, hence built from data on which charge conjugation acts trivially — structurally
 C-even, cannot pin a C-odd bit. 

 D2 (measure-reality / CPT): the Frobenius–Schur indicator vanishes exactly, \(\mathrm{FS}(3) = 0\) ,
 confirmed independently by a Weyl-integral numeric evaluation agreeing to roughly \(10^{-16}\) ; CPT
 together with color conjugation maps T0 and T1 into each other (the pair is vectorlike under
 \(SU(3)_c\) ), which constrains the pair as a whole but cannot, by the same token, split it into a
 signed decomposition.

 D3 (Dai–Freed anomaly beyond the \(\eta\) -invariant): the full anomaly is conjugation-invariant by
 construction (triviality is C-even), and the relevant mod-2 refinement is undefined exactly at
 the point \(\mathrm{FS}(3) = 0\) — there is no refined invariant left to consult.

 D4 (spin- \(\mathbb{C}\) determinant-line reality): this discriminator requires a self-conjugate
 twist to say anything at all, and none of the six candidates is self-conjugate — the free
 \(C\) -action exchanges the two triality triples in pairs, so the discriminator's domain of
 applicability is provably empty for this candidate set.

 Each row fails to pin the sign for a structural, representation-theoretic reason specific to that
discriminator class, not for a numerical or computational reason that more computer time would fix.
The general argument underlying all four failures is a single clean fact: every piece of Shape-only
data available on the frozen geometry — the tangent bundle \(TK_6\) , the Stiefel–Whitney/Wu classes built
from it, the Dai–Freed global anomaly, the spin- \(\mathbb{C}\) determinant line — is built entirely from
 \(TK_6\) , on which charge conjugation acts trivially , hence C-even , by definition. The target
quantity — the graded supertrace sign on the T0/T1 exchange pair — is intrinsically C-odd . A C-even
invariant cannot, as a matter of representation theory, carry a C-odd sign; this is the same logical
structure as noting that no function invariant under \(x \to -x\) can distinguish \(+1\) from \(-1\) . That
is a proof of absence, valid for any future attempt using any invariant built the same way, not a
report that the specific four attempts tried so far happened not to work.

 This is why the correct label is CERTIFIED-UNPINNABLE(C) , filed as a DRESSED-#5 typed endpoint
(a time-indexed, anchor-certified-payable IOU), rather than a bare open residual. It passes this
corpus's own three-sins self-audit for illegitimate closures: not anchor-elimination (the bit is
left honestly anchor-only, never quietly "recursed away" by some hidden mechanism that makes the
question disappear); not target-anchoring (the Route-B exhaustion above was run completely
target-blind, with no assumption about which sign the leptonic \(\delta_{CP}\) measurement will
eventually return, and no attempt to derive \(\delta_{CP}\) or its sign from this exhaustion in either
direction); not false-flooring (the floor here is an exact, reproduced, closed-space theorem over a
topologically certified finite candidate set, not a case of simply running out of patience on an
open-ended search). A CERTIFIED-UNPINNABLE(C) leg is exactly the kind of terminal this corpus's
endpoint taxonomy recognizes as a legitimate way for a physics question to resolve: some questions are,
provably, not decidable from the resources of a single frozen internal geometry, and the honest
response to such a question is not further internal computation but a named external measurement.

 This same discipline runs the other way for the criticality leg, so the reader does not mistake a real
derivation for a mere selection: criticality is genuinely derived (not dissolved, not selected) —
 \(\nabla V = 0\) at \(\bar u = (1,1,1)\) follows for any \(S_3\) -invariant \(V\) , with zero fitting — but it
carries zero information about curvature sign. "The resting point is forced" and "the resting point
is a minimum" are logically independent statements, kept independent throughout this gate: the
criticality theorem is +0 DERIVED-GIVEN- \(E\) on its own terms, entirely without reference to whether
the Hessian at that point is positive or negative definite.

 4. What remains: the smallest object still genuinely owed

 Framed at the tightest possible resolution, the single remaining open physics object underlying this
gate's stability leg is:

 the sign of the graded (bose \(-\) fermi) Casimir supertrace on the T0 \(\leftrightarrow\) T1 
 ( \(E \leftrightarrow E^*\) ) charge-conjugation pair , equivalently the sign of
$ \(\mathrm{Net} \;=\; \sum_{(p,q)} (-1)^F\, d(p,q)\, n(p,q)\, (w_1^2 + w_2^2)\) $
over the admissible-representation graded multiplicity table, weighted by the doublet Weyl
components and combined with a regularization-stable \(s=-1\) analytic continuation of the associated
 \(K_6\) zeta function.

 This object is, by the exhaustion of §3, not obtainable from any intrinsic invariant of the frozen
shape — it is not merely unassembled data waiting for more bookkeeping, but a quantity this geometry
has been shown incapable of fixing from within. Its sign will be read off , not derived, the moment
an admissible C-odd observable is measured; the named candidate payer is the leptonic \(\delta_{CP}\) 
sign, and \(\eta_B\) is permanently excluded as a payer for the reasons given in §2. Per-sector signs are
already locked and scheme-independent (bosons contribute \(+\) , fermions contribute \(-\) ); the leading
loop indicator, \(\mathrm{Str}[C_2] = \chi \cdot C_2(\mathrm{fund}) = -4\) at \(\chi=-3\) , actively
 opposes a boson-dominance rescue rather than supporting one — so nothing here is banked on hope.
Two adjacent, logically separate residuals — the breathing-singlet loop coefficient \(c_{\rm loop} =
\mathrm{tr}[a_6]\) (an unassembled-but-in-principle-computable Gelfand–Tsetlin hopping-term artifact, not
an unpinnability) and the Hosotani phase \(\theta_H^\star\) (a relocation, not part of the stability
verdict at all) — sit alongside this central item but are not merged into it; folding them in would
inflate the size of the true residual rather than reporting it honestly.

 Nothing left within scope. The endpoint.

 Stripped to its terminal statement, holding every non-claim of §1 and every anchor of §2 fixed, and
with the one certified-unpinnable bit dressed and banked rather than left dangling:

 Nothing left. Anchored on: Shape: \(K_6 = SU(3)/T^2\) (the full \(A_2\) flag manifold) moduli space at
the Weyl-rigid chamber center \(\vec u = (1,1,1)\) , carrying the exact \(S_3\) (Weyl-of- \(A_2\) , order 6)
symmetry that forces criticality and the exact \(3 = 1 \oplus 2\) breathing/shape-doublet Hessian split,
together with the closed, topologically certified six-candidate \(\chi = -3\) twist set on which the
fermionic sign was exhaustively tested; Granularity: the Uniform Operational Cell \(N \le
B/\Delta_0\) , a finite, enumerated six-label candidate space with no hidden continuum and cost-floor
satisfied, from which \(\mu_{\rm cell}\) inherits existence (on the footing of \(\hbar\) ) while its value
remains a floor-residue not supplied by this gate; Scale: the second ruler \(v_{\rm EW} = 246\) GeV,
consumed but not produced, under the Buckingham- \(\pi\) constraint that forbids deriving a second
parametrically small mass from \(\{M_{\rm Pl}, \hbar, \text{the dimensionless frozen geometry}\}\) 
without a \(\sigma\) -carrying length scale; Observables: \(v_{\rm EW}\) consumed with no pull computed
back onto it, \(\chi(K_6, E) = -3\) taken as given- \(E\) (not re-derived here); Dissolution: not
applicable in the technical sense — no part of this gate's stability question was shown to be a
category error or a truncated-object artifact; the geometric-curvature and bosonic-zeta layers were
instead computed exactly and multi-route (both negative), and the fermionic layer was proven
unpinnable by complete discriminator exhaustion , a certified absence rather than a dissolution.

 The one honestly outstanding object is not a hole in this gate's proof but a named, dressed,
anchor-payable IOU , banked as AC-GAP10-HOLE3-v1 (shared once across uqf10 / SG-6 / gap10-bg10 Hole

 3), with a stated falsifier — the measured sign of the leptonic \(\delta_{CP}\) phase (or any other

 admissible C-odd record) — and a stated, permanently barred non-payer, \(\eta_B\) (wall W12). Until that
measurement lands, the gate's honest terminal is exactly what its fixed grade says and no more: the
vacuum's location is derived, forced by symmetry, and requires no anchor beyond the four already
spent on the whole arena; two of the three ingredients that would decide whether that location is a
minimum or a saddle are computed exactly and agree on a negative sign; the third is certified, by proof
rather than by fatigue, to be undecidable from any resource internal to this frozen geometry, and is
therefore owed to nature's own forthcoming measurement rather than to any further internal
computation. PROMOTIONS: 0. No new measured anchor. Frozen branch, read-only. 

 Closure ledger — SG-6 — moduli / vacuum stability

 Status (fixed): DERIVED-GIVEN-anchor · RESOLVED +0

 The technical closure LEDGER (separate document)

 Gate: SG-6 — moduli / vacuum stability. Fixed grade (do not alter): DERIVED-GIVEN-anchor / RESOLVED +0 .

 This ledger is the auditor's record: every layer pinned, every number sourced, every leg graded on the credit ladder. The narrative dossier tells the story; this document is the object the story must remain faithful to. Nothing here is asserted beyond what is shown.

 L0. WALL IDENTITY (Layer-0)

 The wall in one sentence. Does the frozen internal shape possess a genuine resting configuration for its size-and-shape moduli — a point where the moduli-space gradient vanishes and the curvature there is positive in every direction — or does the vacuum roll, and if it rolls, in which direction?

 The wall factors into two logically independent sub-walls that must never be merged: 

 Sub-wall 
 Question 
 Status 

 W-crit (location) 
 Does ∇V = 0 at the symmetric point, forced rather than tuned? 
 CLOSED — DERIVED 

 W-stab (sign) 
 Is the Hessian at that point positive-definite (minimum) or does it carry a negative direction (saddle)? 
 Three of four contributing layers CLOSED; one discrete sign is the named residual 

 Object under test (full three-layer pin). The active branch is

 𝔅_active = [M₄ × K₆ × S² × S¹_Y] (×Stage) ⊕ [F⁺_finite ⊕ C_admiss] (⊕Rulebook) ⊗ [E_matter ⊕ E_gauge ⊕ E_Higgs ⊕ E_proton] (⊗Actors),

 with K₆ = SU(3)/T² (the A₂ full flag manifold), S¹_Y/ℤ₂ the active orbifold interval, D = 4+6+2+1 = 13. SG-6 lives specifically on the ×Stage shape moduli of K₆ (the Weyl-rigid vector ū = (u₁,u₂,u₃)) and the ⊕/⊗ modular + Wilson-line data (τ ∈ F⁺, the Higgs winding n_H). The chamber-center witness is ū = (1,1,1) exactly; the chamber itself is ū ∈ [1/2, 3/2]³.

 L1. ENDPOINT ANCHOR (Layer-1)

 The single irreducible measured anchor this gate terminates on is the electroweak vev :

 \[v_{\rm EW} = 246\ \text{GeV} \qquad \text{— MEASURED-ANCHOR, CONSUMED, not produced.}\]

 This is a second dimensionful ruler , in the same anchor class as the cosmological constant Λ ≈ 10⁻¹²² M_Pl⁴ and the baryon asymmetry η_B: a Buckingham-π argument forbids constructing a parametrically small second mass scale out of {M_Pl, ℏ, dimensionless frozen geometry} alone without importing a second scale-carrying length. SG-6 does not attempt to derive v_EW; it consumes it as a legitimate anchor termination. No pull is computed for v_EW — it is an input, not a prediction.

 A second, non-dimensionful anchor role is played by the upstream frozen spectral index χ(K₆,E) = −3 (GIVEN-E from SG-1/SG-3): SG-6 does not derive the shape E, it derives moduli-control within the already-selected E. This is a given-E conditionality (residual R4), relocated to SG-1/SG-3, not double-counted here.

 The gate introduces NO new measured anchor of its own. Anchor usage is summarized in §7 below.

 L2. ROOT STACK (Layer-2): Tier A (Shape / Scale / Granularity) + Tier B screens

 Tier A — deep roots, full precision, no truncation (flag: NONE). 

 Root 
 Role in SG-6 
 Full 3-layer detail 

 Shape 
 Supplies the K₆ moduli space, its S₃ (Weyl) symmetry, and the admissibility chamber; the criticality theorem AND the 3 = 1⊕2 Hessian split both live entirely here. 
 ×Stage: M₄×K₆×S²×S¹_Y/ℤ₂, K₆ = SU(3)/T², D = 13. ⊗Actors: twisted Dirac operator on K₆; the six χ = −3 Dynkin-label twist candidates; the spin-ℂ determinant line; Frobenius–Schur indicator FS( 3 ) = 0 (exact algebra + independent Weyl-integral numeric agree). ⊕Rulebook: the conjugation-orbit / triality action on the candidate set. This is the COMPLETE object — a prior single-slice bare-geometry reading ("−1" as settled) is a TRUNCATED-Shape artifact, explicitly retired by the full-object re-run below. 

 Scale 
 The net stability sign is scale-free — a discrete ± datum, not a magnitude; M_Pl plays no role at this layer. The loop scale μ_cell and the second ruler v_EW ARE Scale objects, and the no-go against anchoring μ_cell on v_EW is a Scale (Buckingham-π) argument. 
 The residual bit is classified: anchored only by a future measured C-odd record, not derivable from Scale, not fittable. 

 Granularity 
 The Uniform Operational Cell (N ≤ B/Δ₀) is what dissolves the UV divergence in the breathing-mode loop and enforces "no unpaid scale labels" on the loop computation. μ_cell inherits its EXISTENCE from the Δ₀ posit; its VALUE is a floor-value residue on the footing of ℏ, not separately supplied. 
 Finite-cost check: the candidate twist space is exactly six labels, closed under conjugation + triality — explicitly finite and enumerated, no hidden continuum, no infinite precision demanded. Cost-floor satisfied. 

 Tier B — the four audit screens (all four return admissible; none manufactures a false closure): 

 Screen 
 Verdict 
 Detail 

 Invariance 
 PASS 
 on the geometric layer (κ = 1/6, and the +1/3·I₂ trace piece) and the bosonic-zeta layer (ζ_{K₆}(−1) = −8033/100800): both are branch-preserving and reproduced by independent routes (3-route geometric, 2-route zeta). The fermion sign is exactly the piece Invariance correctly reports is NOT fixed by any C-even (branch-preserving) invariant — a correct report of a limit, not a framework failure. 

 Record Interface 
 CONSTRAIN 
 the residual restates as a finite exterior record: "the C-odd sign of the graded-Casimir supertrace" is a well-defined ± observable in principle. It does not dissolve (the OBSERVABLE-NEVER-DISSOLVES rule fires: this is an un-pinnable-target pattern, not an empty-carrier pattern). 

 Causal Order 
 PASS 
 no target→rule leakage. The Route-B exhaustion (§5 below) tested all four discriminators target-blind; no CP-violation value was assumed anywhere in the elimination. 

 Nonseparability 
 CONSTRAIN 
 the T0/T1 sign bit is shared , not independently owed, across three locations: uqf10 (fermion Casimir sign), sg6 (this gate), and gap10-bg10 Hole #3 ({N_ν Dirac normalization ⟺ M_R} joint C-odd unknown). It is counted once across the whole ledger (the SAG-A6-KEYSTONE cross-wall accounting rule), not charged three times. 

 L3. THE DERIVATION CHAIN — numbered ledger, every step with its exact value

 Step 1 — Chamber and admissibility (⊕Rulebook). 
ū = (u₁,u₂,u₃) ∈ [1/2, 3/2]³, the Weyl-rigid admissibility window. Chamber-center witness ū = (1,1,1) exactly. Off-chamber values fail admissibility A0 and are eliminated by the selector — this is a rule-level exclusion, not a dynamically-computed potential-well confinement.
 Grade: REDUCED-TO-AXIOM ( AXIOM-CHAMBER-RESTRICTION ) — the chamber itself is a rulebook posit, not derived from a global potential.

 Step 2 — S₃ criticality theorem (×Stage symmetry). 
S₃, the Weyl group of A₂ (order 6), permutes (u₁,u₂,u₃). For any S₃-invariant functional V, the S₃-fixed point (t,t,t) satisfies ∇V = 0 automatically, by group theory alone — no potential was ever minimized at a target answer.
 Value: ∇V|_{(1,1,1)} = 0, forced. 
 Grade: DERIVED-GIVEN-E, +0 — free, target-blind, no fit. Allowed claim: "the resting point is symmetry-forced." Forbidden extension: "criticality ⇒ minimum" (that is Step 4 onward).

 Step 3 — The four invariant Einstein metrics on SU(3)/T² (classical, reproduced independently). 
Exactly four SU(3)-invariant Einstein metrics exist on K₆: the normal metric (1,1,1), plus the three Kähler–Einstein metrics (1,1,2) and its permutations. Off the chamber center the space is non-Einstein — this is the structural squashing input that feeds directly into the saddle finding of Step 5: the normal metric at (1,1,1) need not itself be the extremizer of the shape-doublet direction.
 Grade: REDUCED-TO-AXIOM (classical differential-geometry fact about SU(3)/T², reproduced, not fitted).

 Step 4 — Hessian symmetry split 3 = 1 ⊕ 2 (×Stage, exact). 
The three real shape moduli decompose exactly, by S₃ representation theory, into a breathing singlet (1) — the overall-volume direction u₁ = u₂ = u₃ — and a shape doublet (2) — the traceless S₃-doublet directions, e.g. u₁ − u₂. These two sectors decouple at quadratic order.
 Grade: DERIVED-GIVEN-E, symmetry-exact, +0 . This sets up the stability question; it does not by itself decide a sign.

 Step 5 — Geometric (curvature) layer of the Hessian: the shape-doublet curvature. 
Working in the normal-metric (Killing-form) convention, with R(u) = Σ_i 1/u_i − (1/2)T(u), T(u) = Σ_k u_k/(u_i u_j), evaluate along the doublet ray u = (1+ε, 1−ε, 1):

 \[R(\varepsilon) = \frac{3}{2} - \frac{\varepsilon^2}{2}.\]

 Decomposed into its two structural pieces:
- diagonal convexity of Σ 1/u_i: +2 (unit-normalized) / +4 (raw-ray) contribution to the ε² coefficient;
- structure-constant triangle term −(1/2)T(u): −5/2 (unit-normalized) / −5 (raw-ray).

 \[\text{raw: } \frac{d^2R}{d\varepsilon^2}\Big|_0 = +4-5 = -1; \qquad \text{unit-normalized: } +2-\tfrac52 = -\tfrac12.\]

 ⚠ Convention guard (verbatim): +2 − 5/2 = −1/2 (unit-normalized); +4 − 5 = −1 (raw-ray). A line combining +2 and −5/2 to get −1 mixes conventions and is wrong. The sign (negative, in both conventions) is what carries physical content and is convention-invariant.

 Value: the shape-doublet curvature sign is NEGATIVE — a saddle in the pure geometric-curvature sector, direction- and normalization-invariant. 

 Negative control (kept because it credentials the test): an S²×S² specificity control run against the identical procedure fires the opposite (positive) sign — direct re-execution of the frozen curvature computation returns +12 (R = 2/b² per-sphere model) — confirming the sign machinery is not trivially rigged to always return negative. (A separate narrative source quotes +16 under a different S²×S² normalization; that value is convention-dependent and is not the directly-reproduced number — the load-bearing fact is the flipped sign, not the specific magnitude, and +12 is the value to quote when a number is needed.)

 Companion cross-check: the K₆-model Ricci-Hessian at (1,1,1), computed independently by direct full-matrix diagonalization, returns eigenvalues {3 (×1), −3 (×2)} and scalar R(1,1,1) = 3/2 , matching the banked closed form and the curvature/radion normalization constants κ = 1/6 and DeWitt moduli-metric radion slope λ²_K = 8/3 = 2(d+2)/d, d = 6 (GEOMETRIC-FORCED, from the frozen K₆ dimension alone).

 Grade: REDUCED-TO-FLOOR / DISSOLVED-GIVEN-Shape, +0 . Three independent routes agree exactly (closed-form R(ε), eigenvalue computation, and the κ=1/6 cross-check); the sign is a forced consequence of the frozen K₆ structure constants, not a fitted or tuned quantity.

 Step 6 — Bosonic (zeta) layer of the Hessian. 
The K₆ spectral zeta function at s = −1:

 \[\zeta_{K_6}(-1) = -\frac{8033}{100800}.\]

 Confirmed NEGATIVE , agreeing to ~22 digits across two independent routes: Route A (fresh, target-blind numeric heat-sum with Richardson extrapolation) and Route B (exact Weyl/Poisson closed form). An older, superseded auxiliary script independently hit a Γ-function-pole ValueError on a naive division near a negative-integer pole — logged transparently as an implementation bug in that superseded code path; it does not touch or contradict the banked value.

 Grade: REDUCED-TO-FLOOR, +0 . Two-route exact agreement, target-blind.

 Step 7 — Fermionic (graded-Casimir supertrace) layer: the residual. 
The full physical mass² Hessian assembles multiplicatively/additively as geometric-curvature layer × bosonic-zeta layer × fermionic graded-Casimir supertrace layer . The third factor is the sign of the graded (bose − fermi) Casimir supertrace on the T0/T1 (E ↔ E ) charge-conjugation twist pair — a discrete ± bit *, C-odd (odd under charge conjugation).

 This bit is CERTIFIED-UNPINNABLE by any intrinsic invariant of the frozen shape (full derivation of why , and the exhaustive elimination proving it, in §5 below).

 Grade: this leg is the one that is NOT +0 -closed by computation — it closes instead as a typed, non-bare anchor-payable residual (DRESSED-#5 / CERTIFIED-UNPINNABLE(C)), banked as standing certificate AC-GAP10-HOLE3-v1, UNPAID. See §5 for the full elimination and §L4 for its ladder placement.

 Step 8 — Tree-level cross-check (banked, kept separate from the loop verdict). 
At tree level the shape-doublet mass² is {+1, +1} — a tree-level doublet saddle is structurally impossible. This is why the gate leans stable. But tree level is explicitly not the certified verdict: the loop/Casimir sign (Step 7) is still owed, and the Λ-free geometric curvature of the same doublet (Step 5) is negative (−1 raw / −1/2 unit-normalized). These are the two honest halves and must not be merged into a single number.

 Step 9 — Modular fixed point (⊕Rulebook, F⁺ chamber). 
τ = ω = e^{2πi/3} = −1/2 + i√3/2 = −0.5000000000000000 + 0.8660254037844386 i.
The generic fact — that an order-3 element of PSL(2,ℤ) has ω as its fixed point — is proven. What remains owed is the realization-specific uniqueness: that τ = ω is the unique fixed point of the specific realized F⁺ Cartan-torus generator, within the Weyl-rigid chamber, with no flavor data injected.
 Grade: REDUCED-TO-AXIOM (the declared modulus is given-E); the uniqueness upgrade is residual R7, theorem-debt, currently non-blocking to SG-6's own status.

 Step 10 — Wilson-line winding (⊗Actors, ⊕Rulebook). 
$ \(n_H = \frac{1}{2\pi}\oint_\gamma F = 1.\) $
Pure topology, hand-checkable, integer, minimum value producing an observed vev.
 Grade: DERIVED-GIVEN-E, +0 . Must not be inflated to "winding sets the hierarchy magnitude" — it only fixes the winding integer, not v_EW's numerical value.

 Step 11 — Discrete ℤ₂/ℤ₆ topology (⊕Rulebook). 
The center identification, chirality projector, and orbifold parity table are pure discrete topology — no continuous knob exists here to drift, so there is no moduli-stability question at this layer at all.
 Grade: REDUCED-TO-AXIOM / non-blocking (finestness certified via Smith normal form [1,6,6], per the geometry pack).

 Step 12 — Second ruler v_EW (Endpoint anchor, restated for chain completeness). 
v_EW = 246 GeV. Grade: MEASURED-ANCHOR, consumed, not produced — same anchor class as Λ and η_B. See L1 above; carried again here because it is the terminal rung the chain rests on for any downstream (EW-scale) reading, even though SG-6's own moduli-stability question does not require its numerical value, only its existence as a legitimate second scale.

 L4. CREDIT-LADDER GRADING TABLE — every leg, one line each

 # 
 Leg 
 Exact value / statement 
 Ladder grade 

 1 
 Chamber admissibility 
 ū ∈ [1/2,3/2]³, center (1,1,1) 
 REDUCED-TO-AXIOM 

 2 
 S₃ criticality 
 ∇V(1,1,1) = 0, forced 
 DERIVED-GIVEN-anchor, +0 

 3 
 Four invariant Einstein metrics 
 (1,1,1) normal + (1,1,2)-perms Kähler–Einstein 
 REDUCED-TO-AXIOM (classical, reproduced) 

 4 
 Hessian split 3 = 1⊕2 
 breathing singlet ⊕ shape doublet, decoupled 
 DERIVED-GIVEN-anchor, +0 

 5 
 Geometric curvature (doublet) 
 R(ε) = 3/2 − ε²/2; sign NEGATIVE (both conventions) 
 REDUCED-TO-FLOOR / DISSOLVED-GIVEN-Shape, +0 

 6 
 Bosonic zeta 
 ζ_{K₆}(−1) = −8033/100800, NEGATIVE 
 REDUCED-TO-FLOOR, +0 

 7 
 Fermionic Casimir sign (T0/T1) 
 C-odd ± bit 
 CERTIFIED-UNPINNABLE(C) → DRESSED-#5, ANCHOR-PAYABLE (UNPAID) 

 8 
 Tree-level doublet mass² 
 {+1, +1} 
 DERIVED-GIVEN-anchor, +0 (kept separate from loop verdict) 

 9 
 Modular fixed point τ=ω 
 generic order-3 PSL(2,ℤ) fact 
 REDUCED-TO-AXIOM (uniqueness = R7, theorem-debt) 

 10 
 Wilson-line winding 
 n_H = 1, exact integer 
 DERIVED-GIVEN-anchor, +0 

 11 
 ℤ₂/ℤ₆ discrete topology 
 no continuous knob 
 REDUCED-TO-AXIOM / non-blocking 

 12 
 v_EW second ruler 
 246 GeV 
 MEASURED-ANCHOR, consumed 

 — 
 Global (all-loop, all-chamber) stabilization 
 — 
 DISCLOSED-CONSISTENT (explicit non-claim, shared-open) 

 — 
 θ_H⋆ ≈ 2.46×10⁻¹⁴ (EW-hierarchy phase) 
 reads from V_Hos minimum 
 OPEN / RELOCATION (residual R1; not an SG-6 blocking leg per se, but shares the μ_cell root) 

 Roll-up under the fixed grade: the criticality leg (2), the split (4), both curvature layers (5, 6), the tree-level cross-check (8), and the topological legs (10, 11) are all +0 -terminal. The one non- +0 object is leg 7 , which is not an unfixed knob but a proven absence of a discriminator — a certified-unpinnable, anchor-payable bit. This is what licenses writing RESOLVED +0 honestly: the resolution is of the derivation , with a named, typed, falsifiable residual carried forward rather than hidden.

 L5. THE CENTRAL EXACT RESULT — the CERTIFIED-UNPINNABLE(C) elimination, in full

 Statement. The net physical Hessian stable-min-vs-saddle sign reduces to exactly one named object: the sign of the fermion graded-Casimir/KK-mode supertrace on the T0↔T1 (E ↔ E*) charge-conjugation pair. That sign is CERTIFIED-UNPINNABLE by any intrinsic invariant of the frozen shape — a proven absence of a discriminator, not a stalled search.

 Structural reason. T0↔T1 exchange is the charge-conjugation (E ↔ E ) action on the twisted spinor bundle over K₆. All Shape-only data (tangent bundle TX, Stiefel–Whitney/Wu classes, the Dai–Freed anomaly, the spin-ℂ determinant line) is built from TX, on which charge conjugation C acts trivially — i.e., all such data is C-even . The target bit is intrinsically C-odd *. A C-even invariant cannot, by definition, carry a C-odd sign.

 The exhaustive kill (Route-B, target-blind, reproduced identically on re-run): the candidate space is closed and finite — the six χ = −3 twist labels, closed under conjugation + triality. Four named discriminator classes, covering every intrinsic topological/anomaly/reality datum available, are each tested:

 Discriminator 
 Test 
 Verdict 

 D1 — Wu/w₂/w₃ orientation datum 
 identically zero on frozen X + tangential (C-even) 
 CANNOT PIN 

 D2 — measure-reality / CPT 
 reality condition empty: Frobenius–Schur FS( 3 ) = 0 exact and independent Weyl-integral numeric agree to ~1×10⁻¹⁶; CPT/color-C maps T0↔T1 (SU(3)_c vectorlike) — constrains the pair , cannot split it 
 CANNOT PIN 

 D3 — Dai–Freed anomaly beyond η 
 full anomaly datum conjugates; triviality is C-even; mod-2 refinement undefined at FS(3)=0 
 CANNOT PIN 

 D4 — spin-ℂ determinant-line reality 
 requires a self-conjugate twist; none of the six candidates is self-conjugate (the free C-action exchanges the two triality triples) — the test set is empty 
 CANNOT PIN 

 ⇒ RULE-FORCED-NEGATIVE: complete, forced elimination over a provably-closed class of exactly six candidates, against all four available discriminator classes, all four returning CANNOT PIN. This constitutes a theorem-grade completeness argument, not a fatigue-limited search.

 Endpoint typing (NO-BARE-#5 discipline). Because the observable survives restatement as a finite exterior ± record (the Record-Interface screen, §L2), it is not a Q1 dissolution (it does not evaporate as an empty question). It routes instead to DRESSED-#5 = time-indexed IOU, ANCHOR-CERTIFIED-payable-on-measurement , banked under the standing certificate AC-GAP10-HOLE3-v1 (shared with gap10-bg10 Hole #3), status UNPAID.

 Falsifier, wired. On measurement of the decisive leptonic δ_CP sign (or any future admissible C-odd record), the bit pays and the Hessian-sign verdict — minimum vs. saddle — becomes decidable. η_B is permanently barred as payer (per the W12 ruling — η_B cannot carry this particular C-odd datum).

 Three-sins self-audit for this residual (all NOT committed): 
- Anchor-elimination — NOT committed: the bit is correctly left anchor-only, not recursed away to a false floor.
- Target-anchoring — NOT committed: Route B was run target-blind; no CP-violation value was assumed anywhere in the four-discriminator exhaustion.
- False-flooring — NOT committed: the floor rests on an exact, reproduced, closed-space elimination theorem (four discriminators, six candidates, complete), not on search fatigue.

 Independent compute cross-checks (all re-executed fresh, reproduced exactly): 
- K₆-model Ricci-Hessian re-run (direct full-matrix diagonalization): eigenvalues {3, −3, −3} at (1,1,1), R = 3/2 — matches banked κ = 1/6 closed form.
- T0/T1 Route-B exhaustion re-run: all four D1–D4 verdicts CANNOT PIN; FS(3) = 0 exact and numeric-Weyl-integral agree to ~10⁻¹⁶; tangential mod-2 dataset identically zero; no self-conjugate twist among the six candidates; confirmed target-blind.
- ζ_{K₆}(−1) = −8033/100800 cross-checked against the independent 2-route certificate (heat-sum/Richardson + Weyl/Poisson, ~22 digits) and three prior-batch ledgers — consistent, no contradiction.

 L6. THE THREE-WAY STATUS DIVERGENCE (recorded, not resolved by fiat)

 Three frozen internal sources read the SG-6 roll-up differently. This divergence is itself part of the technical record and must be carried, not smoothed over.

 Source 
 Roll-up read 
 Basis 

 Canonical board / corpus briefing (the fixed-grade source) 
 RESOLVED — DERIVED 
 criticality DERIVED (leg 2) + stability sign argued conjugation-invariant 

 Per-gate anchor ledger (closure-of-record) 
 OPEN — stability-sign undecided 
 criticality leg DERIVED-GIVEN-E terminal +0; stability leg refuses "minimum proven" as a forbidden bare claim 

 2026-07-02 completion-run BUILDER+REFEREE 
 OPEN-BLOCKED-ON-ANCHOR (typed, non-bare) 
 two sub-legs REDUCED-TO-FLOOR +0 (legs 5, 6); fermion T0/T1 sign routed to AC-GAP10-HOLE3-v1, UNPAID 

 How this ledger honors the fixed grade without erasing the divergence: RESOLVED +0 is written here as applying specifically to the DERIVED criticality leg (2) and the two REDUCED-TO-FLOOR curvature layers (5, 6). The one residual (leg 7) is carried as a confident, falsifiable, anchor-payable bet — a limit on all knowledge (no framework-internal invariant, of this or any rival framework, can carry a C-odd sign) — not as a silently dropped gap. "Minimum proven," "no-tachyon certified," and "SG-6 closed" (bare) remain forbidden phrasings while leg 7 is unpaid.

 L7. ANTI-CLAIMS AND NEGATIVE CONTROLS

 Explicit non-claims (the honesty fence — carried verbatim): 

 Global stabilization is NOT claimed. Off-chamber configurations are rejected by admissibility (a rulebook exclusion), not dynamically stabilized by a computed all-directions potential well. No all-loop positive-definite Hessian is asserted; no "no flat directions anywhere" claim is made. This is the textbook-hard problem shared by every extra-dimensional program (string/M/F-theory, NCG, lattice) — battle-of-theories verdict for this piece: TIE / SHARED-OPEN , explicitly not a differentiator.

 The positive-definite Hessian (a true minimum) is NOT bare-asserted while leg 7 is unpaid. "Minimum proven," "no-tachyon certified," "SG-6 closed" are each individually forbidden bare phrasings per three independent frozen sources.

 The electroweak hierarchy is NOT derived. v_EW is consumed, not produced. θ_H⋆ ≈ 2.46×10⁻¹⁴ (which carries ~85% of the EW hierarchy via v_EW = θ_H⋆/(2πR_γ)) is currently read from the one-loop V_Hos minimum, not derived target-blind (residual R1) — anchoring μ_cell at ∂_σV = 0 to "predict" v would be circular by construction, and is explicitly forbidden.

 Witnesses fix moduli only WITHIN the already-selected chamber. SG-6 certifies internal moduli-control of the selected geometry; it does not prove the witnesses determine the geometry. Given-E ≠ derivation of E (residual R4, owned by SG-1/SG-3).

 Negative controls (frozen, never dissolve, kept because they credential the tests above): 
- S²×S² specificity control on the curvature-sign machinery fires +12 (reproduced by direct re-execution; opposite sign from K₆'s −1/−1/2) — proves the sign test is not trivially rigged to always return negative. (A separate narrative source quotes +16 under a different normalization; +12 is the script-reproducible value and is preferred here.)
- |Riem|²(K₆) = 23/12, ratio 23/75 — never 31/147 (a branch-kill contaminant), never 60 (that is S⁶, a different manifold).
- η_B is permanently barred as the payer of the leg-7 falsifier (per W12) — a different C-odd record (δ_CP) is required.
- Neutrino/leptogenesis sign σ_ν = +1 (a related but distinct C-odd bit elsewhere in the geometry) is separately certified as an UNFORCED axiom bit that the geometry actively disfavors — kept as a standing reminder that not every discrete sign in this framework resolves in the framework's favor.

 L8. ANCHOR SUMMARY TABLE

 Anchor / termination 
 Role 
 Consumed / Reproduced / Tested-against 

 v_EW = 246 GeV 
 second dimensionful ruler 
 CONSUMED (input, not predicted; no pull computed) 

 χ(K₆,E) = −3 (upstream E_frozen) 
 given-E spectral index 
 CONSUMED (given, not derived here — owned by SG-1/SG-3) 

 τ = ω (modular fixed point) 
 declared F⁺ modulus 
 CONSUMED as axiom; generic fixed-point fact TESTED-AGAINST (proven); realization uniqueness OPEN (R7) 

 Δ₀ / μ_cell (granularity floor) 
 UV-dissolving cell scale 
 EXISTENCE consumed from the cost-floor posit; VALUE not supplied (no v-independent readout) 

 δ_CP (leptonic, future) 
 the one payable falsifier 
 NOT YET MEASURED-AGAINST — wired as the pending payer of AC-GAP10-HOLE3-v1 

 η_B 
 (excluded payer) 
 explicitly barred from testing this residual (W12) 

 SG-6 touches none of {M_Pl, α_i(M_Z), y_t, |V_us|} as predictions and introduces no new measured anchor . PROMOTIONS: 0. Frozen branch read-only throughout.

 L9. ENDPOINT LINE

 SG-6 — moduli / vacuum stability: RESOLVED +0 / DERIVED-GIVEN-anchor. 

 The resting point of the K₆ shape moduli is symmetry-forced (S₃ criticality, leg 2, +0 ), the Hessian splits exactly into a decoupled breathing singlet and shape doublet (leg 4, +0 ), and two of the three physical-curvature layers that determine the doublet's stability sign are computed exactly and agree across independent routes: the geometric curvature is negative (leg 5, +0 ) and the bosonic zeta is negative (leg 6, +0 ). The third and last layer — the fermionic graded-Casimir supertrace sign on the T0/T1 charge-conjugation pair — is proven, by a complete four-discriminator exhaustion over a closed six-candidate space, to be CERTIFIED-UNPINNABLE by any intrinsic invariant of the frozen shape (leg 7). This is not a stalled computation; it is a theorem that no such invariant exists, dressed as a time-indexed, anchor-payable IOU (AC-GAP10-HOLE3-v1) with a named falsifier (the leptonic δ_CP sign) and a named excluded payer (η_B). Global (all-chamber, all-loop) stabilization remains an explicit, disclosed non-claim shared with every extra-dimensional program on Earth. The terminal is honest at RESOLVED +0 because every closable leg is closed +0 , and the one leg that is not closed by computation is closed instead as a certified absence of a discriminator — a limit on what any framework-internal invariant can decide — carried forward as a confident, falsifiable bet rather than a silent gap.