Gap-02 — Yang–Mills mass gap: full dossier

What this is. The full closure-attack dossier for Gap-02 — the continuum Yang–Mills mass gap, one of the Clay Mathematics Institute's seven Millennium Prize Problems. It is the deep version of the 30-second status popup: it recaps what is rigorously established, localizes the exact missing object, records every executed numerical and structural result honestly (wins and losses), and lays out a specialist-grade work plan for each open hole. It synthesizes and expands the program's frozen Gap-02 corpus; it invents nothing.

Binding discipline (carried verbatim from the corpus). STATUS-UPGRADES:0. This is the GENUINE OPEN Clay Millennium Problem — no proof exists, not here, not anywhere. The program's established results REDUCE Gap-02 to a precisely-localized open object; they do NOT solve it. Frozen branch dcc66f1b2685 / manifest a5b1e6f9d951 is READ-ONLY. Every symbolic quantity ($\delta, K, \kappa, c, c', \Delta, \rho_\star, z_\star$) is symbolic — never assigned a value; any $\Delta>0$ would be inherited from partner lattice work, not derived. The geometry supplies NO Clay lever (Lemma 1, operator-level); the frozen package $F^+$ is INERT for the constructive pure-glue domain. Given-$E \neq$ derivation of $E$; dissolved $\neq$ solved; anchored $\neq$ derived; AXIOM-CLOSED $\neq$ proven.

1. Executive summary and honest status

Headline. We did not solve the Yang–Mills mass gap — nobody has — but we squeezed the entire unsolved Clay wall down to a single finite inequality, and proved that our own geometry gives no shortcut to it.

Honest grade (matches the live popup chip). Two honest framings of one fact:

• Clay side: Precisely-OPEN [X] — EXTERNAL / Clay. The continuum problem is unsolved here and everywhere.
• Scoreboard side: Reduced-to-axiom (granularity) — axiom-conditional, NOT a Clay solution. The program's granularity posit (the Finite Operational Cell Law — a smallest physical action step, so the $a\to0$ continuum limit is never physically taken) dissolves the continuum half by choice, leaving a finite, checkable mass-gap residual plus the granularity axiom itself.

These are the same situation described from two directions. Either way: dissolved $\neq$ solved; axiom-conditional; NOT a Clay solution. No status was ever upgraded.

What this dossier establishes — and what it does not. It establishes (rigorously, citably): (i) the implication "uniform clustering $\Rightarrow$ the mass gap" via an all-operator lemma; (ii) the localization of the entire remaining continuum obstruction to one inequality — the uniform-gap bridge $\Delta(a,L)/\Lambda_{\rm YM}\ge c>0$ uniformly as $a\to0$; (iii) an audit-grade "Lemma 1" descent proving the frozen 13D geometry adds no IR-relevant operator, so it supplies no proof lever; (iv) a rigorous pure-reflection-positivity no-go; (v) a rigorous combinatorial bound $E_{\rm conn}\le e\cdot7\approx19.03$ on one of the four $O(1)$ constants; and (vi) two executed, honestly-reported negative numerical results. It does not establish — and explicitly does not claim — a proof of the continuum mass gap, a continuum-measure construction, a satisfied certificate inequality, or that the geometry contributes a constructive lever. The cost-floor reframe changes the Clay question rather than answering it, and we say so out loud.

The testable bet, stated flat. The whole Clay wall is now localized to one finite, local inequality — the uniform-gap bridge — and, under the granularity axiom, to a floored certificate inequality $z_\star < 1/(E_{\rm conn}\cdot A_{\rm fluc})$. The honest sharpening: before that floored inequality can be proved or refuted, its activity weight $w(\gamma)$ must first be pinned by a derivation, not a convention choice, because the natural circularity-clean readings currently disagree on the verdict. The inequality is finite and could genuinely come out $\ge1$ — there is no guaranteed sign. And even a favorable sign closes only one horn of one sufficient condition, still conditional on the separately-open OS-reconstruction and SL-3 positivity gates.

2. The community gap

2.1 The precise open problem

Pure $SU(3)$ Yang–Mills theory in four spacetime dimensions — the gauge sector of the strong interaction with the quarks removed — is observed to have a mass gap: a strictly positive lowest excitation above the vacuum. This is what makes the strong force short-ranged, the lightest glueball massive rather than massless, and the strong coupling run the way it does. It is an observation-locked fact (short-range nuclear force, massive hadrons, the measured running of $\alpha_s$), not a modeling artifact.

Yet no one has ever proven it. The Clay Millennium Prize problem (Jaffe & Witten, Quantum Yang–Mills Theory, Clay Mathematics Institute, 2000) asks for a mathematically rigorous construction of continuum 4D $SU(N)$ Yang–Mills as a quantum field theory satisfying the Wightman (or Osterwalder–Schrader) axioms, together with a proof that the theory has a positive mass gap $\Delta>0$. The conjunction — existence of the continuum theory and a uniform gap for that same theory — is the prize. The Clay statement itself anticipates that closing it likely requires "important new ideas both in physics and in mathematics."

Stated in the standard constructive-Euclidean arena, the target theorem is:

Target theorem (verbatim, genuinely open). There exists $\Delta>0$ such that $\operatorname{Spec}(H|_{\mathcal H_{\rm phys}})\cap(0,\Delta)=\varnothing$ — there is no sequence of normalized, vacuum-orthogonal, physical states whose energy $\to0$ — for the Hamiltonian $H$ obtained by Osterwalder–Schrader reconstruction of a genuine 4D continuum $SU(3)$ pure-gauge measure, with (a) OS reconstruction of that measure, (b) a unique vacuum $\Omega$ with $H\Omega=0$, (c) self-adjoint $H\ge0$, and (d) the gap $\Delta\ge c'\Lambda_{\rm YM}>0$ in physical units.

2.2 State of the art and best bounds

The honest landscape — every entry a real, standardly-cited result quoted at the scope it actually establishes:

(All scope columns and citations are reproduced from GAP02_STEP2_STANDARD_YM_PROOF_LANE/05_uniform_gap_bridge.md §1–6 and .../06_first_attack_choice.md §6.2, which carry the citation-honesty discipline verbatim.)

2.3 Prior attempts and why each falls short

• Strong-coupling expansion (the "easy route"). At large bare coupling (small $\beta$) the convergent character/cluster expansion gives a rigorous area law and a positive gap in lattice units. But the expansion has a finite radius of convergence in $\beta$ and diverges before $\beta\to\infty$ — its breakdown is associated with the roughening transition. The continuum-limit window $\beta_R\le\beta<\infty$ is precisely the window the strong-coupling argument cannot reach. So strong-coupling confinement, real as it is, does not imply the continuum gap. The hard part is everything between $\beta_R$ and $\infty$ — the fine-lattice / weak-bare-coupling regime where no convergent series controls the gap. (05_uniform_gap_bridge.md §2–3.)
• Bałaban's RG program. Rigorous block-spin renormalization-group control of the weak-coupling, small-field regime, uniformly in the cutoff — a landmark, genuinely 4D, genuinely non-abelian. But it is a UV-stability program; it controls the small-field end and does not deliver the completed infinite-volume continuum measure with a uniform gap. The large-field / IR end (where confinement and the gap are generated) is not closed by it.
• Magnen–Rivasseau–Sénéor. A 4D YM phase-space / multiscale expansion constructing the theory with an explicit IR cutoff. The cutoff's removal with a gap bound is exactly the open part.
• Area-law-to-gap (Route B). Even a continuum area law for the fundamental Wilson loop bounds the static quark–antiquark sector; converting it to a bound on the glueball / physical gauge-invariant spectrum requires the constructed continuum measure and OS reconstruction — an area law is a statement about one observable; the gap is a statement about the whole spectrum of $H$. So this route presupposes most of the bridge rather than bypassing it.
• Anomaly-matching arguments. A 't Hooft anomaly (e.g. a mixed one-form-center / time-reversal anomaly) constrains the IR but does not by itself force a gap — it can be satisfied by a gapless theory (a conformal or TQFT phase can saturate the anomaly). So even a nonzero anomaly is necessary-not-sufficient.

The structural lesson, verified across machineries: the obstruction is intrinsic to $d=4$ marginality at order-one coupling, not an artifact of any one method.

3. The construction — the rigorous math

This section is the core. It recaps the modular conditional chain that reduces Gap-02 to a single localized inequality, shows the actual equations, and records the intermediate results.

3.1 The constructive arena and dimensional transmutation

The standard arena (Glimm–Jaffe machinery; Osterwalder–Schrader reconstruction) is the Euclidean lattice regularization. A finite hypercubic lattice $\Lambda\subset(a\mathbb Z)^4$ with spacing $a$ and linear size $L$, link variables $U_\ell\in SU(3)$, the Wilson plaquette action $$S_W=\beta\sum_p\Bigl(1-\tfrac13\,\mathrm{Re}\,\mathrm{tr}\,U_p\Bigr),\qquad \beta=6/g_0^2,$$ and the product Haar measure $d\mu_{a,L}\propto e^{-S_W}\prod_\ell dU_\ell$. The mass gap on the regularized theory is the inverse correlation length of connected gauge-invariant correlators, equivalently the lowest nonzero eigenvalue of the transfer-matrix Hamiltonian $H=-a^{-1}\log T$.

Pure $SU(3)$ has no classical mass scale; by asymptotic freedom the only scale is the dynamically generated $\Lambda_{\rm YM}$, set via the two-loop $\beta$-function, $$a\,\Lambda_{\rm YM}=(b_0g_0^2)^{-b_1/2b_0^2}\,e^{-1/2b_0g_0^2}\,(1+O(g_0^2)),\qquad b_0=\tfrac{11}{16\pi^2}.$$ The continuum limit is $a\to0\Leftrightarrow g_0\to0\Leftrightarrow\beta\to\infty$, holding $\Lambda_{\rm YM}$ fixed in physical units. Any honest gap statement must be expressed relative to $\Lambda_{\rm YM}$ — a bare lattice number $\Delta(a)$ in lattice units is meaningless without scale-setting. (05_uniform_gap_bridge.md §0.)

3.2 The reduction chain (SL-0 through SL-2): what is established

SL-0 (variational identity — ESTABLISHED, pure functional analysis). The statement "no soft physical sequence," $$\inf\{\langle\psi,H\psi\rangle:\psi\perp\Omega,\ \|\psi\|=1\}=\Delta>0\;\equiv\;\operatorname{Spec}(H)\cap(0,\Delta)=\varnothing,$$ is an exact equivalence — the spectral gap and the absence of low-energy normalized vacuum-orthogonal states are literally the same statement. Hand-checkable; no open content.

SL-1 / M4D (the all-operator Lemma — CLEAN conditional). The implication $$[\,H1\wedge H2\wedge H3\wedge H4\,]\Rightarrow\text{no soft physical sequence}\Rightarrow\text{gap}$$ holds, where H1/H2/H3 are the OS-reconstruction conditions (continuum measure existence, RP survival, nontriviality + unique vacuum) and H4 is uniform exponential clustering of the renormalized glueball correlator. This conditional reuses OS reconstruction + the spectral theorem + the Källén–Lehmann representation (all cited, not re-derived). It derives no numerical value. "Uniform clustering $\Rightarrow$ the gap" is genuinely rigorous. (01_DOSSIER.md §0.1; witness W3.)

SL-2 (finite-$(a,L)$ foundations — ESTABLISHED). At fixed spacing $a$ and finite volume, four results hold rigorously:

At finite $a$ both the gap and reflection positivity are trivial — the entire wall IS the $a\to0$ continuum limit. The finite-lattice theorems are the floor the continuum limit must not collapse below; they are not the bridge.

3.3 The brutal one-line bridge — the exact missing theorem

The reduction localizes the entire remaining continuum obstruction to a single inequality:

$$\boxed{\;\frac{\Delta(a,L)}{\Lambda_{\rm YM}}\;\ge\;c\;>\;0\quad\text{uniformly as }a\to0,\;L\to\infty,\;\beta\to\infty\;}$$

with the same $c$ for the same measure for which existence, reflection positivity, and nontriviality also hold in the limit. No such uniform lower bound is known. This lone inequality is the entire wall. (05_uniform_gap_bridge.md §0 box, §6.)

The quantifier order is decisive and must be stated exactly: $\exists(\delta,K,\kappa,a_0,L_0,n_0)\,\forall(a,L,\beta,n)$ — a single $\delta$ uniform over all regulators and all scales. The hard core is the conjunction: existence and a uniform gap for the same theory in the weak-bare-coupling regime, where YM is renormalizable-but-not-super-renormalizable and the limit is genuinely non-perturbative.

3.4 Equivalent reformulations (the same open theorem, three faces)

The bridge has three equivalent faces — restating it does not advance it:

1. Spectral form (the box above): the transfer-matrix gap, scale-set, bounded below uniformly.
2. Correlator form (H4). For the glueball operator $O(x)=\mathrm{tr}\,F_{\mu\nu}F^{\mu\nu}(x)$ (lattice: plaquette / smeared-clover), the connected correlator obeys $$\langle O(x)O(0)\rangle_c\le C\,e^{-\Delta_0|x|},\qquad \Delta_0=-\lim_{|x|\to\infty}\tfrac1{|x|}\log\langle O(x)O(0)\rangle_c>0,$$ with $\Delta_0/\Lambda_{\rm YM}\ge c>0$ for the continuum-limit measure (decay rate not collapsing to zero, not a finite-$a$ artifact).
3. Marginal Kotecký–Preiss (KP) form. A convergent polymer/activity expansion requires $$\rho_\star=E_{\rm conn}\cdot A_{\rm fluc}\cdot e^{-s_{\rm marg}}<1\quad(\text{equivalently }\delta>0),$$ uniformly through the marginal band $g(2^na)=O(1)$.

3.5 The decisive obstruction: rarity $\not\Rightarrow$ domination in the marginal band

This is the precise reason the wall is hard. In the deep UV ($g\to0$), the large-field weight $\mu_n\le e^{-c/g^2}\to0$ banks the KP inequality automatically — large fields are exponentially rare. But rarity is not domination: the bare activity of those configurations can compensate. Inside the $d=4$ marginal band $g(2^na)=O(1)$ the inequality becomes a strict finite comparison between two $O(1)$ constants, with no spare coercive margin because $d=4$ marginal renormalizability leaves no tree-level smallness — the only smallness is loop-induced running. $\delta=0$ is the $d=4$ stall; this is the wall. (01_DOSSIER.md §2.3; GAP02_AFRP_MONOTONE_DISPOSITION.md §3.)

Both available routes hit the same $\delta=0$ object — method-invariance, verified: - Polymer / cluster route (expansion side): in the band, $s_{\rm block}=O(1)$ versus $\log(E_{\rm conn}A_{\rm fluc}N_{\rm cert})=O(1)$; the contribution $C_n=\mu_n\|\Phi_n\|$ decays only if the activity exponent $b<c$, and $b=c$ is the stall. The band is plausibly outside both convergence radii — the Bałaban Gaussian-UV disk and the strong-coupling product-measure disk — a no-man's-land where no single uniform $\rho_\star<1$ need exist. - AF-RP monotone route (flow side): reaches the identical wall. The live target is a one-step contraction of a large-field disorder functional $M_n$, $$M_{n+1}\le(1-c\,b_0\,g_n^2)\,M_n+O(g_n^4)\,M_n,\qquad c>0,\ b_0=\tfrac{11N}{3}>0\ (\text{SU(3): }b_0=11),$$ anchored at the $b_0>0$ UV-free fixed point. The driver $-c\,b_0\,g_n^2$ is non-abelian antiscreening (asymptotic freedom). The entire mathematical content is the uniform-in-$n$, uniform-in-$a$ bound on the $O(g_n^4)M_n$ remainder plus a fixed $c>0$ through $g=O(1)$ — which is exactly the $\delta=0$ marginal stall renamed. At $g=O(1)$ the $O(g^2)$ and $O(g^4)$ terms are the same order, the perturbative truncation is uncontrolled, and "$c>0$ uniform through the band" is the open problem, not a derived result. (Full derivation: GAP02_AFRP_MONOTONE_DISPOSITION.md §2.)

A crucial honest note on what this monotone gives even if it worked: $\prod_k(1-c\,b_0\,g_k^2)\sim\exp(-c\,b_0\sum_k g_k^2)$; with one-loop running $g_k^2\sim1/(2b_0k)$ the harmonic sum gives $\sim n^{-c/2}$ — a power-law decay in RG-step count in the UV regime, not $e^{-mn}$. The physical exponential gap $m\sim\Lambda_{\rm YM}$ is generated by dimensional transmutation $\Lambda=\mu\exp(-1/(2b_0g^2(\mu)))$ — a scale set at the band where the running coupling becomes $O(1)$, exactly where the contraction control is lost.

3.6 The pure-reflection-positivity NO-GO (a rigorous negative)

A clean, rigorous reductio rules out the tempting shortcut of trying to get the gap from positivity alone:

Claim. No reflection-positivity-ONLY (gauge-group-agnostic) uniform-gap argument can exist, because it would prove a false gap for 4D compact $U(1)$, which has an established gapless Coulomb phase.

Proof. (1) RP is gauge-group-agnostic: the Osterwalder–Seiler transfer-matrix construction proves RP for the Wilson action for any compact $G$, using only Haar measure on $G$ and positivity of the reflected single-plaquette action — it never invokes $b_0$, non-abelianity, or structure constants, and holds verbatim for $G=U(1)$ and $G=SU(N)$. (2) 4D compact $U(1)$ is gapless: Guth (1980, PRD 21 2291) rigorously proved a non-confining weak-coupling massless-photon phase (strengthened by Fröhlich–Spencer 1982, CMP 83 411) — a theorem, not a conjecture. (3) Any uniform-gap argument whose only group-sensitive input is RP would apply identically to $U(1)$ and prove a gap there, contradicting (2). Hence no RP-only method can establish the $SU(N)$ uniform mass gap. $\blacksquare$

Sharpening (tightens, does not weaken): the no-go is against RP-only mechanisms. RP itself remains true and is needed downstream for the H4 $\Rightarrow$ gap / OS reconstruction step. The honest statement is "RP is insufficient / non-discriminating," not "RP is false." The gap must come from a genuinely group-sensitive nonperturbative input — asymptotic freedom $b_0=11N/3>0$ — not from positivity that compact $U(1)$ shares. (GAP02_AFRP_MONOTONE_DISPOSITION.md §1.)

3.7 Lemma 1 — the geometry supplies no Clay shortcut (audit-grade descent)

The frozen 13D object is the three-layer $$M_4\times K_6\times S^2\times S^1_Y/\mathbb Z_2\ \oplus\ F^+_{\rm finite}\oplus\mathcal C_{\rm admiss}\ \otimes\ E_{\rm matter}\oplus E_{\rm gauge}\oplus E_{\rm Higgs}\oplus E_{\rm proton},$$ with $K_6=SU(3)/T^2$ the full flag manifold $F_3$ — the geometric origin of color. The descent map $\Pi$ KK-expands on $K_6\times S^2\times S^1_Y/\mathbb Z_2$, keeps 4D zero modes, integrates out modes $\ge M_{\rm KK}\sim1/R\sim M_{\rm GUT}\sim10^{16}$ GeV, and projects $A_M=(A_\mu,A_A)$ keeping $A_\mu^a\in su(3)_c$ (color glue) while $A_A\to E_{\rm Higgs}$ (Wilson-line scalars, not color glue).

Lemma 1B (pure-glue Clay projection — AUDIT-GRADE / TARGET). Dropping the quark matter $E_{\rm matter}$, the descended pure-gauge sector is ordinary 4D pure $SU(3)$ Yang–Mills plus a topological $\theta$-term and an $M_{\rm KK}$-suppressed irrelevant tail: $$S_{\rm eff}^{\rm 1B}=\int d^4x\Bigl[\tfrac1{4g_3^2}F^a_{\mu\nu}F^{a\,\mu\nu}+\tfrac{i\theta_{\rm QCD}}{32\pi^2}F^a_{\mu\nu}\tilde F^{a\,\mu\nu}+\sum_{d>4}\tfrac{c_d}{M_{\rm KK}^{d-4}}\mathcal O_d\Bigr].$$ No extra IR-relevant constraint survives in the pure-glue $SU(3)$ projection — with the single explicit exception of the R4 row (§3.8).

Every settled 13D-specific residue lands in exactly one bin: (i) a renormalization of $g_3$ (from $\mathrm{Vol}(K_6)$ + the KK threshold); (ii) topological $\theta$-data; (iii) physical quark matter (1A only, absent in 1B); (iv) an $M_{\rm KK}$-suppressed irrelevant operator $\mathcal O_d$ vanishing in the IR as $(\mu/M_{\rm KK})^{d-4}\to0$; or (gov) admissibility governance ($\mathcal C_{\rm admiss}$ restricts model-building moves, not the path-integral measure). None of (i)–(iv) or (gov) is an IR-relevant constraint. By Wilsonian universality the IR confining dynamics is indifferent to the UV completion. The geometry settles the gap's origin ($SU(3)$ from $K_6=SU(3)/T^2$) and pins $\Lambda_{\rm YM}$ as boundary data, but supplies no IR constructive lever; $F^+$ is inert. (Full gate-by-gate matrix: gap02_lemma1_descent_audit/10_final_lemma1_statement.md §1–5; three-layer reconciliation: 00_GAP02_CURRENT_STATUS_AND_RECONCILIATION.md, net verdict MIXED_WITH_OPEN, 5-way tally INERT:10 / ORDINARY-YM:3 / WILSONIAN-IRRELEVANT:1 / TOPOLOGICAL→R4:1 / LEVER:0.)

3.8 The one open descent row (R4) and its well-posing

The single row the descent audit leaves genuinely open: does the frozen $K_6$ spin-c bundle carry a nontrivial mixed 't Hooft anomaly between the $\mathbb Z_3$ one-form center symmetry of $SU(3)_c$ and the spin-c data? Its family-count / chiral-anomaly channels are projected out or inert in pure glue (the FP ghosts are adjoint scalars with no spinor twist; $A_{333}=0$ leaves $Q_{\rm BRST}^2=0$ untouched); its holonomy phase is $\theta$-data (Gap-03's life, not the mass gap). But the $\mathbb Z_3$-center one-form anomaly survives the pure-glue projection in principle (center symmetry is exact only in pure glue) and is uncomputed. Held OPEN_CANDIDATE — neither closed to Lemma 1 (inertness unproven) nor promoted to a lever.

The R4 object was first checked and found not well-posed as stated (verdict R4_NOT_WELL_POSED_AS_STATED), then well-posed (Theorem 11):

Theorem 11 (R4 Well-Posing). (I) The anomaly background is not a detached flat two-form in $B^2\mathbb Z_3=K(\mathbb Z_3,2)$; it is a $PSU(3)=SU(3)/\mathbb Z_3$ bundle whose lift obstruction is $u_2=w_2^{PSU(3)}\in H^2(BPSU(3),\mathbb Z_3)$ (the 't Hooft flux / fractional-instanton datum). (II) The $K_6$ datum enters as a local-coefficient twist $\tau_{K_6}:=\tau(\bar c_1(L_{K_6}))$ of the Spin-c bordism spectrum, not a cup product to be integrated. (III) The invariant lives, with correct degree, in $$\xi_{R4}\in\Omega_5^{\rm Spin\text{-}c}\bigl(B(SU(3)\to PSU(3));\,\tau_{K_6}\bigr),\quad\text{a finite abelian group.}$$

Two elementary checks force the well-posing. The originally-written cup product $\bar c_1(L_{K_6})\cup B^{(2)}$ is malformed: $\bar c_1\in H^2$ (degree 2) and $B^{(2)}\in H^2$ (degree 2) give degree 4, but a 4D anomaly-inflow term needs a degree-5 class on a bounding 5-manifold $W_5$ — and the product lives on the 10-real-dimensional $K_6\times M_4$, not on $W_5$. Malformed on both degree and locus. The lone-$\bar c_1$ fiber pushforward over the 6-dimensional $K_6$ lowers degree by 6, giving $2-6=-4<0$, ill-defined. The structurally correct cure (Freed–Hopkins) is to twist the theory, not multiply classes up to degree 5.

Supporting frozen / standard facts (verified by direct reduction): $K_6=F_3$ has $b_2=\mathrm{rank}\,SU(3)=2$, so $H^2(K_6;\mathbb Z_3)=(\mathbb Z/3)^2$; $c_1(TK_6)=2\rho=\sum(\text{positive roots})=(2,2)$ in the fundamental-weight basis, which mod 3 is $(2,2)\ne(0,0)$, so $\bar c_1\ne0$ and the twist $\tau_{K_6}$ is nonzero. (Correction carried verbatim: the frozen "spin-c index $-3$" is $\chi(K_6,E)$, the holomorphic index of the specific bundle $E$, a different object from the manifold Euler characteristic $\chi(F_3)=+6=|\mathrm{Weyl}(SU(3))|$ — do not cite "$-3$" as the manifold Euler characteristic.)

The operative differential was subsequently recomputed (GAP02_R4_STEENROD_COMPUTATION.md, which supersedes the earlier Theorem-11-note guess of an AHSS $d_3$). Two load-bearing structural corrections hold: (i) the anomaly carrier is the degree-3 class $\bar x_1\in H^3(BPU(3);\mathbb Z/3)$ — there is no degree-2 mod-3 class ($H^1(BPU(3);\mathbb Z/3)=H^2(BPU(3);\mathbb Z/3)=0$), so the earlier "carrier $u_2\in H^2$" does not exist; (ii) the 2-primary $d_3=Sq^3_{\mathbb Z}=\beta\circ Sq^2\circ\rho_2$ is identically zero on 3-torsion (it factors through mod-2 reduction $\rho_2$, and $\rho_2(\text{3-torsion})=0$). The genuinely operative differential is therefore the 3-primary $d_5=Q_1=\beta P^1$ (the integral Milnor primitive, $|Q_1|=2p-1=5$ at $p=3$), not $d_3$. But $Q_1(\bar x_1)$ lands in degree $3+5=8$ (off the $p+q=5$ bordism lines), so $d_5$ kills the degree-5 survivor neither as source nor target — and the untwisted value is $\Omega_5^{\rm Spin}(PSU(3)\times B^2\mathbb Z_3)=\mathbb Z_3\ne0$ (cited literature anchor), i.e. the survivor persists. (Do not cite the older "untwisted $\Omega_5^{\rm Spin\text{-}c}(B^2\mathbb Z_3)=0$" reading — that was the wrong/superseded object.) The sole remaining lever is whether the frozen $\tau_{K_6}=(2,2)$ twist-correction to $d_5$ — a degree-2 class cupping the degree-3 carrier to land on the degree-5 line, carrying Massey-product / $v_1$-filtration structure at $p=3$, not a bare cup — supplies a nonzero, on-line, degree-lowering image; this is a finite $\mathbb Z_3$-linear-algebra question the cited (untwisted-only) literature does not settle. Value $\xi_{R4}$: OPEN — STILL_SUBTLE, not certified $0$, not forced $\ne0$; untwisted default $=\mathbb Z_3$. (GAP02_R4_WELLPOSING_THEOREM.md Theorem 11 + §2–4 for the well-posing home; GAP02_R4_STEENROD_COMPUTATION.md §1–5 for the corrected operative differential, which voids the earlier $d_3$ guess.)

3.9 The cost-floor reframe and the four $O(1)$ constants

Under the program's granularity / cost-floor root axiom, the continuum $a\to0$ idealization is out of scope by choice; the floored target is the finite certificate-alphabet inequality $$z_\star:=\sum_{\gamma\ne0}w(\gamma)<\frac1{E_{\rm conn}\cdot A_{\rm fluc}}\qquad\text{at fixed }\ell_\star\sim\Lambda_{\rm YM}^{-1},$$ equivalently the master inequality $s_{\rm block}>\log(E_{\rm conn}A_{\rm fluc}N_{\rm cert})$. The four $O(1)$ constants are built only from local gauge-invariant configuration data (binned $\mathrm{Re/Im}\,\mathrm{tr}\,U_p$, $\mathrm{tr}\,U_p^2$, $\mathbb Z_3$ center-flux class, Wilson-loop class, block-curvature class, Wilson block action), modulo gauge — no energy eigenstate, no transfer-matrix spectrum, no correlation length enters anywhere (the circularity guard, confirmed by code inspection). The constants:

The rigorous $E_{\rm conn}\le e\cdot(2d-1)=e\cdot7\approx19.03$ is the one constant with a clean proof — a Penrose/Klarner-type bound on the connective constant of lattice animals on $\mathbb Z^4$, pure combinatorics, no fields and no spectrum, $\beta$-independent. (GAP02_MASTER_INEQUALITY_RESULT.md §1.)

4. The insights we used

These are the load-bearing ideas — now shareable — that made the progress believable and reproducible.

• Localize, don't gesture. Rather than claim partial progress on "the gap," the program decomposed the Clay statement into a modular conditional chain (SL-0 $\to$ SL-2 $\to$ M4D/H1–H4) and asked: which single link carries all the open content? The answer is the uniform-gap bridge / H4. Everything else is either established literature (the finite-$a$ floor) or a clean conditional. This is what converts a famous wall into a named, finite object a specialist can attack.

• The quantifier order is the whole problem. The bridge is hard not because a gap is hard to find at fixed $a$ (it is trivial there), but because the single constant $c$ must be uniform over all regulators and all scales — $\exists\,c\,\forall(a,L,\beta)$, not $\forall(a,L,\beta)\,\exists\,c$. Making that order explicit is what reveals that the difficulty lives entirely in the $d=4$ marginal band.

• Rarity $\not\Rightarrow$ domination — the category error that is the wall. In the UV the dangerous configurations are exponentially rare ($\mu_n\le e^{-c/g^2}$), which tempts one to conclude they cannot dominate. But the bare activity can compensate; rarity bounds a weight, not a contribution. Naming this distinction precisely (a weight bound vs a contribution bound) is what shows why Bałaban's small-field control does not extend, and what defines the exact sharpening a proof would need.

• Method-invariance as evidence of intrinsic-ness. Two independent machineries — the polymer/cluster expansion and the AF-RP block-spin monotone — were pushed to their limits and both stall on the identical $\delta=0$ marginal object at $g=O(1)$. That coincidence is the signal that the obstruction is intrinsic to $d=4$ marginality, not an artifact of one tool — which is exactly the honest content the Clay statement's "needs a new idea" anticipates.

• Group-sensitivity is mandatory; positivity is not enough. The compact-$U(1)$ reductio is a sharp filter: any argument that would gap $U(1)$ is wrong on its face, because $U(1)$ is provably gapless. This kills the entire class of RP-only / group-blind arguments and forces the proof to route through asymptotic freedom $b_0=11N/3>0$ — the one genuinely group-sensitive, non-abelian input.

• Wilsonian universality as a discipline, not a hope. The Lemma 1 descent does not assume the geometry is irrelevant; it audits every 13D structure operator-by-operator and shows each is UV-origin, $\theta$-data, $M_{\rm KK}$-suppressed, or governance. The honest payoff is a negative: the geometry fixes the gap's origin and pins the scale, but cannot shortcut the proof. Reporting that negative — rather than manufacturing a lever — is what makes the rest of the dossier trustworthy.

• The $\kappa^3/\pi$ falsification test (no target-fitting). Every constant ($E_{\rm conn}$, $A_{\rm fluc}$, $s_{\rm marg}$, $N_{\rm cert}$) is admitted only if it would be written without knowing the target verdict. A constant chosen to land the inequality subcritical is target-fitted and relocates the problem rather than solving it. This blade is why the program reports $z_\star$ as OPEN rather than picking the convention that gives PASS (§5.3).

5. Evidence and reproducibility

5.1 What reproduces, plainly

• The reduction chain reproduces. SL-0 is hand-checkable; SL-2 is established literature; SL-1/M4D is a clean cited conditional. "Uniform clustering $\Rightarrow$ the gap" is genuinely rigorous.
• The negatives reproduce. The pure-RP no-go is a rigorous reductio against the established gapless 4D compact $U(1)$ Coulomb phase. The Lemma 1 no-lever result is operator-level. The R4-lever-dead result is a cited fact (anomaly-matching does not force a gap).
• The executed lattice runs reproduce. A validated $SU(3)$ Wilson Monte-Carlo engine, plaquette-gated, returns the honestly-recorded INCONCLUSIVE / convention-dependent verdicts.

5.2 The executed master-inequality Monte-Carlo run (an honest negative)

A finite-lattice evaluation of the sufficient-condition master inequality was run on a fixed $4^4$, $\beta=6.0$ ensemble. Hard gate (precondition for any verdict): the engine must reproduce the known $SU(3)$ Wilson plaquette average before any decidability verdict is emitted. It did:

Engine: Cabibbo–Marinari + Kennedy–Pendleton heatbath + $SU(2)$-subgroup overrelaxation + reunitarization. (Disclosed caveat, low severity: $\beta=5.7$ fails by $\sim8\sigma$ on $L=4$ — finite-volume bias; the verdict rests on the $\beta=6.0$ anchor; a clean second anchor needs $L\ge8$.)

The decidability margin $D\equiv s_{\rm block}-\log(E_{\rm conn}A_{\rm fluc}N_{\rm cert})$ (natural log; $D>0$ closes the horn, $D\le0$ INCONCLUSIVE):

Every $D$ is negative — on both paths (effective-count and raw-UB), under both $E_{\rm conn}$ choices (rigorous $e\cdot7=19.028$ and empirical MC $\lambda_4=13.5$), at all three binnings; smallest $|D|/\sigma\approx67$. The critical count for $D=0$ is $N_{\rm cert}<\exp(1.0413)/(19.028\cdot0.05264)=2.83$ letters, while the smallest realized $N_{\rm eff}=155\gg2.83$. The negative sign was measured (small activation cost $\approx1.04$ vs certificate log-volume $\approx5$–18), not assumed subcritical.

EMPIRICAL VERDICT: INCONCLUSIVE ($D\le0$, sign-stable / binning-stable). The master inequality is not satisfied on this ensemble, so the large-field horn does not close at $\ell_\star$ here. This is NOT a refutation of the gap: master $\Rightarrow$ target is sufficient, not necessary; failing the sufficient condition does not deny the gap. A genuine refutation would need a direct lower bound $z_\star\ge1/(E_{\rm conn}\cdot A_{\rm fluc})$, which is not the object computed here.

An honest granularity correction is recorded: an initial run mis-defined $s_{\rm block}$ as the summed whole-block action over $\sim81$–96 plaquettes ($s_{\rm block}\approx189$, $N_{\rm cert}\approx10^{461}$, $D\approx-872$), conflating "block" with a $2^4$ sub-lattice. Corrected to the per-certificate single-cell footing the master inequality requires ($s_{\rm block}\approx1.04$, per-cell $D\approx-4$). The correction moved $D$ toward $D>0$ but still landed negative — it did not cherry-pick a favorable sign. (GAP02_MASTER_INEQUALITY_RESULT.md §0–7.)

The companion block-spin scale-scan ($b=2$ chain, 15 $(\beta,k)$ points, plaquette gate PASS at every point) returned $D\le0$ at every sampled coupling $g_{\rm eff}\in[0.942,2.449]$ spanning weak(UV) $\to$ marginal-$O(1)$ $\to$ strong(IR/blocked), with no positive-$D$ island and no sign crossing: NO-MANS-LAND. The overlap inequality $g_{\rm up}<g_\star$ is undefined — there is nothing to match because both expansion domains are empty over the band. (GAP02_BLOCKSPIN_SCAN_RESULT.md §1–4. Sub-floor disclosure: only the five $k=0$, $L=4$ points are torus-clean; $k\ge1$ are sub-floor artifacts; the verdict holds on the physical $k=0$ points alone.)

5.3 The $z_\star$ convention-underdetermination (a deeper, more honest finding)

A separate direct computation of the boxed target $z_\star:=\sum_{\gamma\ne0}w(\gamma)$ (rather than the sufficient KP proxy) found that $z_\star$ has three natural circularity-clean readings that disagree on the verdict (threshold $1/(E_{\rm conn}A_{\rm fluc})=0.998$):

Reading (A) diverges because as $\delta_{\rm tr}\to0$ each realized cell becomes its own letter, so $\sum_{\rm letters}e^{-\min}\to\sum_{\rm cells}e^{-\text{cost}}=2678.2=n_{\rm ref}\cdot z_\star(B)$ — an extensive, divergent count, not a per-site activity. Its finiteness is purchased only by the arbitrary binning $\delta_{\rm tr}$, and the inequality's truth value is whatever $\delta_{\rm tr}$ one picks. This exposes an internal inconsistency: the corpus equated $E_{\rm conn}A_{\rm fluc}e^{-s_{\rm marg}}<1\Leftrightarrow E_{\rm conn}A_{\rm fluc}\sum_\gamma w(\gamma)<1$, which holds only if $\sum_\gamma w(\gamma)=e^{-s_{\rm marg}}$ (one effective activity), but the $N_{\rm cert}$ route makes $\sum_\gamma w(\gamma)\sim N_{\rm cert}e^{-s_{\rm block}}\gg e^{-s_{\rm block}}$. Since $E_{\rm conn}=e\cdot7$ is pure $\mathbb Z^4$ graph combinatorics containing no per-block letter count, the multiplicity $N_{\rm letters}$ must either live inside $z_\star$ (reading A, divergent) or be dropped (B / single-activity) — and the boxed inequality does not say which.

Disposition: OPEN — named obstruction: $w(\gamma)$ is convention-underdetermined at the floor; the natural circularity-clean conventions DISAGREE on the verdict. Picking reading B (or single-activity) to report PASS is naked target-fitting; picking reading A to report FAIL is equally illegitimate (and ill-posed). No posit fixes $w(\gamma)$ that would be written without knowing which verdict you want. Until $w(\gamma)$ is pinned by a derivation, $z_\star$ is not a decidable number. (GAP02_B2_COSTFLOOR_ZSTAR_DISPOSITION_2026-06-24.md §1–6.)

5.4 How to re-run / re-derive

• Hand-checkable, no numerics: SL-0 (variational identity), the pure-RP no-go reductio, the Lemma 1 degree-counting for R4 (degree 4 $\ne$ 5; lone-$\bar c_1$ pushforward $2-6<0$), and the $E_{\rm conn}\le e\cdot7$ combinatorial bound.
• Numerical lane (all scripts under C:/tmp/massgap_computation/): su3_lattice.py (validated $SU(3)$ Wilson MC engine; heatbath + overrelaxation; plaquette gate driver); massgap_pipeline.py (5-stage pipeline computing the four constants + $N_{\rm cert}$ + $D$); CERTIFICATE_SPEC.md (certificate alphabet, local gauge-invariant data only); verify_D.py (independent $D$ recomputation, reproduces to 3 decimals); plus scratchpad zstar_direct.py / zstar_wellposed.py for the three $z_\star$ estimators and the $\delta_{\rm tr}$ divergence. Pipeline self-reported [done in 287.5s]; verify CLEAN. The plaquette hard gate is the binding precondition: no $D$ verdict is trustworthy unless $\langle P\rangle$ reproduces the standard value at that point.

5.5 Frozen hashes (boundary data only — none is a Clay lever)

Branch dcc66f1b2685 / manifest meta a5b1e6f9d951 / spin-c bundle on $K_6$ 0fd19c9ae0c1. Radii at $\vec u=(1,1,1)$: $R_0=R_{K_6}=R_{S^2}$ 634438ce0776, $R_{S^1_Y}=R_0/2$ 0e8b8dba2cf0. $\Lambda_{\rm YM}$ pinned by $\alpha_3(M_Z)$ + $M_U\sim1.0\times10^{16}$ GeV + threshold $\delta_3=-1.7313$ via two-loop $\overline{\rm MS}$. $R_0=1.592\times10^{-17}$ GeV$^{-1}$. (01_DOSSIER.md §0; 00_GAP02_CURRENT_STATUS_AND_RECONCILIATION.md Refinement 1.)

6. Open gaps and closure path — the specialist work plan

This is the most load-bearing section: a concrete, target-blind work-package for each open hole. A specialist should be able to pick up any one and start.

Hole R1 — The uniform-gap bridge (THE CLAY WALL)

(a) Precise statement. Prove $\Delta(a,L)/\Lambda_{\rm YM}\ge c>0$ uniformly as $a\to0$, $L\to\infty$, $\beta\to\infty$ (equivalently H4 = the Marginal-KP inequality $\rho_\star=E_{\rm conn}A_{\rm fluc}e^{-s_{\rm marg}}<1$, equivalently $\delta>0$) — with the single $c$ uniform over all regulators and all scales, holding through the marginal band $g(2^na)=O(1)$, for the same measure whose existence, RP, and nontriviality survive the limit.

(b) Why it's hard / prior-attempt lessons. Inside $g=O(1)$ the inequality is a strict $O(1)$-vs-$O(1)$ comparison with no spare coercive margin (d=4 marginal renormalizability). Bałaban-RG and constructive-SPDE both succeed only at $d\le3$; the marginal band is plausibly outside both convergence radii (Gaussian-UV disk and strong-coupling product-measure disk). Rarity ($\mu_n\le e^{-c/g^2}$) does NOT imply domination — the bare activity compensates; this category error is the wall. Method-invariance is verified: the polymer route and the AF-RP monotone stall on the same $\delta=0$ object. Traps the program's own passes caught (do not repeat): (1) the $\kappa^3/\pi$ pattern — reverse-engineering any of $E_{\rm conn},A_{\rm fluc},s_{\rm marg},\rho_\star$ to land subcritical (true-by-construction; it relocates, never closes); (2) reading the executed $D<0$ as a refutation (it is a sufficient proxy failing, INCONCLUSIVE); (3) using RP alone (the rigorous $U(1)$ no-go forbids it); (4) smuggling in the frozen geometry as a lever (Lemma 1 proves it supplies none; $F^+$ inert).

(c) Exactly what closes it. A target-blind proof of $\rho_\star<1$ uniform through the band, via one of three live lanes: Lane A — sharpen the Bałaban large-field step from a weight bound to a contribution bound, contracting by a fixed $e^{-\delta}$ through the band; Lane B — a convergent activity/cluster expansion at $g=O(1)$ with strictly positive $\delta$; Lane C — matching/overlap of the UV (Gaussian) and IR (product-measure) expansion domains across the band, OR a genuinely group-sensitive non-expansion method using $b_0=11N/3>0$ (the AF-RP monotone with the $O(g^4)$ remainder controlled uniformly through $g=O(1)$). Success criterion: $c>0$ established with no target-fitted constant; then SL-2/SL-1/M4D/SL-0 deliver the spectral consequence (with R2/R3 separately closed). A REFUTING result is also valid: a direct lower bound $\rho_\star\ge1$ (or $z_\star\ge1/(E_{\rm conn}A_{\rm fluc})$) uniformly would kill this sufficient route — a legitimate negative close.

(d) Machinery & inputs. Constructive RG (Bałaban CMP 1984–89); 4D YM multiscale (MRS 1993); KP polymer convergence; OS reconstruction (OS 1973/75); the AF-RP monotone schema and the four-constant pipeline. Start from GAP02_STEP2_STANDARD_YM_PROOF_LANE/05_uniform_gap_bridge.md (§4 names the missing RG estimate exactly) and GAP02_AFRP_MONOTONE_DISPOSITION.md (§2 the live monotone, §3 method-invariance). The Clay statement (Jaffe–Witten 2000) anticipates a genuinely new idea.

(e) Leverage. This is the gate. Closing it (with R2/R3) is the Clay prize. Even alone it would convert the entire reduction chain into a true Hamiltonian gap statement.

Hole R2 — SL-3: interacting nonperturbative BRST/Gribov kernel positivity

(a) Precise statement. Establish the interacting nonperturbative positivity certificate on $\mathcal H_{\rm phys}=\ker(s)/\mathrm{im}(s)$ surviving $a\to0$ — the physical-kernel positivity of the BRST cohomology in the continuum limit.

(b) Why it's hard. The Gribov–Singer–Neuberger triad: no global continuous gauge section on a nontrivial bundle (Gribov 1978, Singer 1978); lattice BRST gives an indeterminate $0/0$ (the FP determinant sign-flips across Gribov copies; Neuberger 1987). Established only perturbatively (Kugo–Ojima quartet).

(c) Exactly what closes it. Construct a positivity certificate for the interacting nonperturbative kernel that survives the continuum limit, resolving the FP/Gribov ambiguity (e.g. via a Gribov-region-restricted measure with controlled positivity, or a manifestly positive gauge-invariant formulation). Refuting outcome: a proof that physical-kernel positivity fails in the limit would itself be a major (negative) result.

(d) Machinery & inputs. BRST cohomology; Gribov–Zwanziger; Kugo–Ojima; constructive positivity methods. This is an independent co-gate — do not fold into R1. Export to a constructive-positivity specialist. (01_DOSSIER.md §4.2.)

(e) Leverage. Independent and load-bearing: even a proved R1 yields no gap without R2.

Hole R3 — OS-reconstruction / continuum-RP survival (H1·H2·H3)

(a) Precise statement. Construct the continuum measure $\mu=\lim_{a\to0,L\to\infty,\beta\to\infty}\mu_{a,L}$ on gauge-invariant observables satisfying OS-0/1/2/3, with RP surviving $a\to0$ (the open part — F1 gives RP only at fixed $a$), nontriviality (interacting, not Gaussian), and a unique clustering vacuum; then run OS$\to$Wightman reconstruction.

(b) Why it's hard. No 4D $SU(3)$ continuum measure has ever been constructed. Bałaban controls only the small-field UV; MRS retains an IR cutoff. The non-negotiable quantifier order — existence FIRST, then the gap — means a "continuum mass gap" asserted before steps 1–4 is fabrication (there is no Hamiltonian to gap).

(c) Exactly what closes it. A controlled construction of $\mu$ with the OS axioms verified in the limit and nontriviality proved. Refuting outcome: a proof of triviality (Gaussian fixed point) in the limit would refute the interacting theory's existence.

(d) Machinery & inputs. Constructive-QFT machinery (Glimm–Jaffe); OS reconstruction; tightness/compactness of the regulated sequence; convergence of Schwinger functions. The §4 RG bridge of 05_uniform_gap_bridge.md is what would supply MO-1. Export to a constructive-measure specialist. (01_DOSSIER.md §4.3; 06_first_attack_choice.md §6.3 MO-1..MO-7.)

(e) Leverage. Independent co-gate; converts clustering (H4) into a true Hamiltonian gap. Recommended division of labor (from the proof-lane analysis): Route A builds the measure (MO-1..4 + clustering); Route C converts that clustering into the spectral gap (MO-5/6/7) via glueball-correlator decay + OS reconstruction — the cleanest framing of the open problem, not a weakening of it.

Hole R4 — The R4 bordism anomaly value $\xi_{R4}$

(a) Precise statement. Evaluate $\xi_{R4}\in\Omega_5^{\rm Spin\text{-}c}(B(SU(3)\to PSU(3));\tau_{K_6})$. The untwisted default is the nonzero survivor $\Omega_5^{\rm Spin}(PSU(3)\times B^2\mathbb Z_3)=\mathbb Z_3$; the open question is whether the frozen $\tau_{K_6}=(2,2)$ twist-correction to the operative differential $d_5=Q_1=\beta P^1$ deforms / kills that degree-5 $\mathbb Z_3$ class (finite $\mathbb Z_3$-linear-algebra, carrying Massey-product structure at $p=3$).

(b) Why it's hard / prior-attempt lessons. The object was first written malformed (degree 4 cup product on the wrong locus); the lone-$\bar c_1$ pushforward gives negative degree. Theorem 11 well-poses it but does not evaluate it; the subsequent Steenrod recomputation (GAP02_R4_STEENROD_COMPUTATION.md) corrects the operative-differential guess but lands on STILL_SUBTLE (value OPEN). Traps caught (do not repeat): (i) the carrier is the degree-3 class $\bar x_1\in H^3(BPU(3);\mathbb Z/3)$ — there is NO degree-2 mod-3 carrier ($H^2(BPU(3);\mathbb Z/3)=0$), so the earlier "$u_2\in H^2$" is void; (ii) the operative differential is NOT a $d_3$ — the 2-primary $d_3=\beta\mathrm{Sq}^2\rho_2$ is identically zero on 3-torsion; the genuinely operative one is the 3-primary $d_5=Q_1=\beta P^1$ ($|Q_1|=2p-1=5$), whose target $Q_1(\bar x_1)$ sits in degree 8, off the $p+q=5$ lines; (iii) do NOT read the older "untwisted $\Omega_5^{\rm Spin\text{-}c}(B^2\mathbb Z_3)=0$" as $\xi_{R4}=0$ — the correct untwisted value is $\mathbb Z_3\ne0$ (survivor persists), and the twist is the only lever; (iv) do NOT transcribe $\mathbb Z_2/\mathbb Z_4$ Pontryagin-square arithmetic onto $\mathbb Z_3$ (on $\mathbb Z_3$, $P$ lands in $\mathbb Z_3$); (v) do NOT read "R4 lever dead" as "R4 solved" — only the lever is closed-negative; the value is open. And the binding caveat (Hole R5 below): even a nonzero, universal $\xi_{R4}$ is necessary-not-sufficient for a gap.

(c) Exactly what closes it. Supply the three specialist inputs — the correct degree-5 bordism object ($B^2\mathbb Z_3$ vs $B(PSU(3))$ / 2-group), the $\tau_{K_6}$ twist mechanism, and the $c_1(L_{K_6})$ representative bit — then evaluate $\xi_{R4}$ via the twisted $d_5=Q_1$ correction on the degree-5 line (the bare $Q_1$ being off-line, only the $[\tau_{K_6}]$-correction can act there). Success criterion: a definite element of $\mathbb Z/3$. Refuting outcome: $\xi_{R4}=0$ is a perfectly valid close (the row is then fully inert); equally, confirming the untwisted $\mathbb Z_3$ persists is a valid value.

(d) Machinery & inputs. Twisted spin-c bordism; Atiyah–Hirzebruch spectral sequence (operative differential $d_5=Q_1=\beta P^1$ at $p=3$, NOT $d_3$); Freed–Hopkins (anomalies = twisted bordism invariants); Kapustin–Seiberg ($SU(N)/\mathbb Z_N$ flux); GKSW (one-form symmetries); GKKS (mixed anomaly at $\theta=\pi$); twisted-Morava / odd-prime twisted-AHSS structure (Westerland; Grady–Sati) for the $\tau_{K_6}$-correction. Start from GAP02_R4_WELLPOSING_THEOREM.md (Theorem 11 + §3) and GAP02_R4_STEENROD_COMPUTATION.md (§1–5, the corrected operative differential). Note the shared thread: the same $\xi_{R4}$ object is also the R4 thread under SG-4 (the geometry's own candidate anomaly) and the Born T-1(a) leg — the value computation lives on the SG-4 / Born ledgers, because it is inert on the gap here.

(e) Leverage. LOW for the gap (the lever is dead regardless of value — see R5). HIGH cross-gate: settling $\xi_{R4}$ closes a shared residual for SG-4 and Born T-1(a).

Hole R5 — Even a nonzero R4 anomaly is necessary-not-sufficient for a gap

(a) Precise statement. A 't Hooft anomaly can be matched by a gapless theory; so even $\xi_{R4}\ne0$, descending and universality-preserving, does not force a gap.

(b) Why it's hard. Anomaly-matching constrains the IR phase but admits conformal or TQFT saturations that are gapless. So R4 alone cannot deliver the gap even if nonzero.

(c) Exactly what closes it. Conditional on R4 turning out nonzero and universality-preserving: separately prove the matched IR phase is gapped/confining rather than gapless — e.g. exclude a conformal or TQFT saturation of the anomaly — before treating R4 as a Clay handle. Refuting outcome: exhibiting a gapless anomaly-matching phase would confirm R4 is not a gap handle.

(d) Machinery & inputs. Anomaly-matching ('t Hooft); conformal/TQFT phase classification; the full conjunctive chain $\alpha$(descends to 4D) $\wedge\ \beta$(survives continuum) $\wedge\ \gamma$(constrains FP/Gribov) $\wedge$ universality-preserving. (01_DOSSIER.md §4.4; 10_final_lemma1_statement.md §3 adjudicator.)

(e) Leverage. Gates whether any future nonzero $\xi_{R4}$ could ever become a gap lever.

Hole R6 — The granularity / cost-floor axiom (the deep root)

(a) Precise statement. The granularity axiom (the Uniform Operational Cell Law: a single system-independent $\Delta_0>0$ such that no stable record occupies an operational cell smaller than $\Delta_0$) licenses the continuum dissolution but is itself unproven — the program tried and failed to derive a uniform floor from mere finiteness, and a countermodel exists.

(b) Why it's hard / prior-attempt lessons. First-pass plug result (folded honestly): this hole is REDUCED-to-named-axiom — a status that matches the already-frozen grade, not an advance on it (the verifier caught an overclaim attempt here; keep to what is proven). The corpus already holds, as banked losses: (i) finite resources buy only $\eta$-dependent total boundedness, NOT a uniform $\Delta_0$ — the basin-shallowing countermodel (disjoint stable basins of depth $d_n=B\cdot2^{-n-1}$, $\sum d_n=B/2<\infty$, $\inf_n d_n=0$: finitely many basins above any tolerance $\eta$, but no smallest cell) is a rigorous counterexample to the uniform claim; (ii) the no-target-fitting blade kills every candidate deeper premise — basin-packing (RELABEL: takes $\Delta_0$ as hypothesis), pre-quantum Margolus–Levitin (CIRCULAR + keystone $\tau\ge\pi\hbar/2E$ pinned only by $\hbar$ = the target), finite-information (RELABEL: "smallest bit" is the cell in costume), finitism ($\mathbb Q\cap[0,1]$ is dense-but-finite-in-window with $\inf$ gap 0), GPT finite dimension (assumed by fiat); (iii) the uniform-positive cell $\Delta_0>0$ is a named UNPROVEN posit with $\hbar$ as its residue value. Trap: do not relabel "REDUCED-to-axiom" as a derivation or a count reduction — the axiom count does NOT drop.

(c) Exactly what closes it. A derivation of a uniform positive minimum-action floor from a strictly weaker principle than positing it — one that (a) is not the cell law in costume [RELABEL], (b) is not QM smuggled in, (c) is writable without knowing the answer is $\hbar$. Two precisely-named residuals remain, both posited not derived: (R-uniformity-across-arenas) inter-arena uniformity — self-reference gives one cell per arena but does not bound the cell uniformly across all bounded causal arenas (the basin-shallowing countermodel CM2, $\inf_{\rm arenas}\Delta=0$, is unrefuted; composition-closure was tested and does NOT force the budget/cell ratio bounded); (R-positivity) strict positivity $\Delta_0>0$ vs the continuum $\Delta_0=0$. Refuting/honest-standing outcome: an explicit standing disclosure that it is a named posit (the current state) is a legitimate terminal — this is the deep-root axiom, not a unicorn, so a concrete weaker-premise derivation would genuinely close it.

(d) Machinery & inputs. All steps hand-checkable, no numerics. Start from RESULT_A1_FTC_UNIFORM_CELL_PREQUANTUM_DERIVATION_ATTACK_2026-06-24.md (the AXIOM_CLOSED disposition + the basin-shallowing countermodel + the no-target-fitting falsification test table), RESULT_FTC_FORK_B_CELL_LAW_2026-06-23.md, and the DeepRoot-granularity row of GATE_REGRADE_BESTCASE_AND_HOLES_2026-06-29.md.

(e) Leverage. Closing this would convert the scoreboard's "Reduced-to-axiom (granularity)" from axiom-conditional to derivation-grounded — it is the deepest root of the whole stack, shared across many gates. It does not close the continuum Clay problem (that remains R1–R3); it would only ground the dissolution side.

The cardinal honest point for §6. R1 is the Clay wall; R2/R3 are independent co-gates; R4/R5 are not a gap lever (only a shared cross-gate value + its necessary-not-sufficient caveat); R6 is the axiom the dissolution rests on. The wall — $d=4$ marginal coercivity at order-one coupling — is invariant under every legal, observation-preserving move; every such modification relocates into a strictly harder subproblem. Physics only; the device-engineering applications are firewalled out of this dossier entirely.

7. Honest ceiling and scope

What is genuinely won (non-promoting). (1) The rigorous all-operator "uniform clustering $\Rightarrow$ gap" lemma (SL-1/M4D). (2) The precise localization of the entire wall to ONE named inequality — the Marginal-KP / uniform-gap bridge through the $d=4$ marginal band. (3) The rigorous pure-RP NO-GO (any RP-only argument would falsely gap compact $U(1)$). (4) The executed INCONCLUSIVE master-inequality run + the $z_\star$ convention-underdetermination finding (honest negatives, not closures). (5) The proven no-lever theorem (the geometry fixes the gap's origin and pins $\Lambda_{\rm YM}$ as boundary data, but supplies no IR constructive lever; $F^+$ inert). (6) The rigorous combinatorial bound $E_{\rm conn}\le e\cdot7\approx19.03$, and the well-posing of R4 as a finite twisted spin-c bordism target.

Dissolved $\neq$ solved. Declining the continuum limit by the granularity axiom changes the Clay question rather than answering it. The continuum (Clay) mass gap remains an open question; it is simply not the program's objective under the cost-floor scope. We state this out loud.

Selection $\neq$ derivation; given-$E\neq$ derivation-of-$E$. The geometry selects $SU(3)$ as the color group and pins $\Lambda_{\rm YM}$ from the frozen anchors ($\alpha_3(M_Z)$, $M_U$, $\delta_3$) as boundary data. Deriving the scale would be deriving $E$; it is not done inside Gap-02. The gap itself is a measured fact (short-range strong force, massive glueballs, running $\alpha_s$) — observation-locked, not a Gap-02 output.

Dissolved unicorns (shared ceilings, never claimed as proven, never listed as open weakness). (i) "No future theory could ever solve the continuum Clay mass gap by a different route" — an unprovable universal negative over all possible mathematics; the bounded ceiling is "no accepted constructive proof exists today, community-wide." (ii) "The granularity / cost-floor axiom is THE unique provably-irreducible bottom of the stack" — an undischargeable universal negative; the ceiling is "earned-irreducible under known reductions, a confessed posit." (iii) "No geometric / observation-preserving modification of the frozen shape could EVER supply an IR proof lever" — proof over an open-ended space of modifications; the bounded, banked claim is Lemma 1: on every settleable row, the geometry supplies no lever (R4 the one unsettled row, held honestly open).

What is explicitly NOT claimed. The Yang–Mills mass gap is not solved, closed, or proven here (no Clay solution). The cost-floor reframe does not answer the Clay problem. The frozen 13D geometry does not provide a proof lever or shortcut. The certificate / master inequality is not satisfied ($z_\star$ is convention-underdetermined and OPEN; could be $\ge1$). The INCONCLUSIVE MC result is not a partial win and not a refutation. A nonzero $\xi_{R4}$ would not by itself prove the gap.

The anchors paid. $\Lambda_{\rm YM}$ and the UV package are given-E boundary data (SM/declared inputs); the granularity cost-floor is a named posit ($\hbar$ as its residue). The reduction is rigorous; the wall and the two co-gates stay OPEN.

Ceiling, stated flat. Serious candidate / the gap is observed and its origin is geometry-fixed, but the continuum gap is NOT proved. STATUS-UPGRADES:0; frozen branch READ-ONLY; $F^+$ inert; given-$E\neq$ derivation of $E$; rarity $\not\Rightarrow$ domination; a captured log $\neq$ an independent reproduction; anchored $\neq$ derived; AXIOM-CLOSED $\neq$ proven.

Sources synthesized (all read in full): TOE/PER_GATE_DOSSIERS/GAP02_COMPLETION_HANDOFF/01_DOSSIER.md; TOE/gap_02_yang_mills_mass_gap/GAP02_STEP2_STANDARD_YM_PROOF_LANE/05_uniform_gap_bridge.md and 06_first_attack_choice.md; TOE/gap_02_yang_mills_mass_gap/gap02_lemma1_descent_audit/10_final_lemma1_statement.md; TOE/gap_02_yang_mills_mass_gap/00_GAP02_CURRENT_STATUS_AND_RECONCILIATION.md; TOE/GAP02_AFRP_MONOTONE_DISPOSITION.md; TOE/GAP02_R4_WELLPOSING_THEOREM.md and the superseding TOE/GAP02_R4_STEENROD_COMPUTATION.md (corrected operative differential $d_5=Q_1$; the earlier GAP02_R4_DELTA3_TRIVIAL.md $d_3$ transcript is voided); TOE/GAP02_B2_COSTFLOOR_ZSTAR_DISPOSITION_2026-06-24.md; TOE/computational_runs_2026-06-23/massgap_computation/GAP02_MASTER_INEQUALITY_RESULT.md and GAP02_BLOCKSPIN_SCAN_RESULT.md; TOE/RESULT_A1_FTC_UNIFORM_CELL_PREQUANTUM_DERIVATION_ATTACK_2026-06-24.md; and the existing brief articles/GATE_BRIEF_GAP02.html. External: Jaffe & Witten, "Quantum Yang–Mills Theory," Clay Mathematics Institute Millennium Problem (2000). No number, hash, sign, or citation in this dossier was fabricated; every uncomputed quantity is marked OPEN.