SG-1 — Geometry specification: full dossier

The 30–50 page treatment of the gate that commits the framework's starting object. This is the deep version of the live 30-second popup: the rigorous math behind the freeze, the insights that produced it, the certificates that witness it, and an honest, specialist-grade work plan for every open hole. It expands the brief closure-attack packet (DOSSIER_SG1_GEOMETRY_SPEC_CLOSURE_ATTACK) without contradicting it. STATUS-UPGRADES:0 — this dossier reflects SG-1's honest current grade (DECLARED-FROZEN); it upgrades nothing.

Binding discipline, carried verbatim. No status was ever upgraded. The frozen branch (dcc66f1b2685 / a5b1e6f9d951) is READ-ONLY. SG-1 is a freeze-and-reproduce certificate, NOT a derivation and NOT a uniqueness theorem. Minimality is selector-minimal inside the pre-declared search category — SHAPE is selected, not forced absolutely. The honest charged input cost is ≈4 anchors + 9–10 injected reals (~13–14 measured reals), NOT the old "4-input" / "4→22" headline. Carrier forcedness is mixed and explicitly graded. given-E ≠ derivation of E; selection ≠ derivation; dissolved ≠ solved; AXIOM-CLOSED ≠ proven; category-relative ≠ absolute.

Table of contents

1. Executive summary + honest status
2. The community gap
3. The construction — rigorous math
4. The insights we used
5. Evidence & reproducibility
6. Open gaps + closure path (the specialist work plan)
7. Honest ceiling & scope

1. Executive summary + honest status

Headline. This framework writes down its entire starting universe in public, to the last byte, and a machine rebuilds it from scratch with zero mismatches — so the one object every downstream prediction is tested against can never be quietly retuned after the fact.

The honest grade (matches the live popup chip): DECLARED-FROZEN. Direction: held. SG-1 is a serious candidate geometry, frozen and reproducible, not yet proven unique.

SG-1 is the gate that commits the object. Everything downstream — SG-2 (gauge-group recovery), SG-3 (the family index $-3$), SG-6/7/8 (Higgs, thresholds, flavor), and on through the GUT and TOE papers — is evaluated against this one frozen branch. SG-1 is therefore the load-bearing precondition of the entire program: if the starting object can be silently edited after the data is seen, no downstream claim means anything. The whole point of SG-1 is to make that editing impossible, in public, mechanically.

The committed object is a three-layer active branch in which only the first ("Stage") layer carries metric dimension, with total dimension $D = 4 + 6 + 2 + 1 = 13$:

 𝔅_active = [ M₄ × K₆ × S² × S¹_Y/Z₂ ]_×                       ← STAGE   (metric geometry, 13 dims)
          ⊕ [ F⁺_finite ⊕ C_admiss ]_⊕                         ← RULEBOOK (finite admissibility, 0 dims)
          ⊗ [ E_matter ⊕ E_gauge ⊕ E_Higgs ⊕ E_proton ]_⊗     ← ACTORS   (bundle/operator, 0 dims)

with $K_6 = SU(3)/T^2$ the flag manifold (color source), $S^2$ the weak source, and $S^1_Y/\mathbb{Z}_2$ the hypercharge source with an orbifold chirality filter.

What this dossier establishes, and what it does not. It establishes three — and only three — certified things about $\mathfrak{B}_{\rm active}$: (1) specificity — the branch is a fully written, layer-complete object, pinned to the last index by a 33-row manifest, committed before any downstream gate is evaluated; (2) no-layer-smuggling — no gate is closed by content not declared in its proper layer (the $\times/\oplus/\otimes$ discipline, certificate B2); and (3) reproducibility — every primitive and derived object is content-addressed by SHA-256, and a regenerator rebuilds every hash byte-for-byte, independently re-run target-blind during the SG-1 completion (exit 0, all 33 hashes recomputed without the script's own comparator, both frozen hashes byte-equal, deterministic across reruns). It does not establish that the geometry is derived, that 13 dimensions are forced, or that the branch is unique. The technical witness is a freeze-and-reproduce certificate; it pins the object and closes no physics.

On top of the freeze, the dossier carries the genuinely strong structural results the program has earned, each labelled at its true strength: the weak carrier $S^2$ is forced within the grammar by a hand-checkable theorem (F1); the hypercharge carrier $S^1_Y/\mathbb{Z}_2$ is forced by another (F2); and the color carrier $K_6=SU(3)/T^2$ is clean by an architecture-neutral representation-theory argument (abelian-isotropy uniqueness), with the one concrete cheaper rival — the $K_6 \to CP^2$ swap — built end-to-end and shown to break at the gauge gate.

The edge, stated as a confident bet, not a hedge. The freeze is a reproducibility certificate, not a uniqueness proof — and we say so flat out. There is exactly one way to prove us wrong, and we name it publicly: build a genuinely simpler theory — a clean 4D effective theory, or a cheaper $SU(3)$ carrier — that reproduces the same particle content while injecting fewer independent measured reals, once the geometry→observables generator map is fully charged on both sides. A bare dimension-count win ($4 < 13$) is already inadmissible as the record-cost metric (it calls "simpler" the theory that needs more finite records, and it is foreclosed by the granularity⇒MDL result up to one bridge axiom — do not waste effort there); the live escape is a smaller charged-information ledger with the generator charged — the REFUTED-ECONOMY outcome, which we would report as an equally valuable discovery. The one thing we can never claim — that no geometry under any possible mathematics is shorter — is uncomputable for everyone (it is a Kolmogorov universal negative); it is a hard limit on all of physics, not a gap in ours. The honest input cost is roughly 13–14 measured real numbers, not the old 4-input headline, and we charge every one of them in the open.

One-line status table.

(Sources: handoff SG1.md; existing article DOSSIER_SG1_GEOMETRY_SPEC_CLOSURE_ATTACK.md §0–§2; SG1_COMPLETION_RESULT.md §1–§3; SHAPE_FINAL_STATUS_SPLIT_RESULT.md.)

2. The community gap

2.1 The precise open problem

Every candidate theory of everything must, before any prediction means anything, commit to one definite starting object — a geometry, an algebra, a vacuum. The field's standing failure mode is not a shortage of candidate objects; it is the moving target: the starting structure gets nudged after the data is seen, so apparent successes are unfalsifiable. The community gap SG-1 addresses is therefore twofold:

1. The commitment gap. No widely accepted framework commits its starting object, to the last index, in advance of the fit — in a form a third party can re-derive byte-for-byte and confirm was not edited later.
2. The selection gap. No framework anywhere proves its starting geometry is the one nature had to use. This is the deep, possibly-permanent half of the problem.

These are different gaps with different difficulties. The first is a discipline and reproducibility gap — hard to do honestly, but mechanically closable. The second is a uniqueness/forcedness gap — and, in its absolute form, provably unreachable for everyone. SG-1 closes the first and is openly honest that it does not close the second.

2.2 History and state of the art

String / M-theory and the landscape. The most developed TOE program hosts a vast landscape of vacua — by widely-quoted estimates of order $10^{500}$ flux compactifications — with no selected member. The structure (Calabi–Yau, flux integers, brane content, quotient) is chosen from enormous catalogs, and the absence of a selection principle is the defining open problem of the program. No string vacuum is committed, to the last index, in advance of matching the Standard Model; the moving target is structural.

Grand unified theories (4D GUTs). $SU(5)$, $SO(10)$, $E_6$ and their kin do commit a gauge group, and they buy real relations (e.g. $b$–$\tau$ unification). But the flavor sector remains injected: Yukawa textures, family replication, and mixing are put in by hand. A 4D GUT is a partial commitment with the hardest part (flavor) left as free input.

Noncommutative geometry (NCG, the spectral Standard Model). Connes' program commits an algebra and a finite spectral triple, and recovers the SM gauge group and much of its content with striking economy. But the finite Dirac operator carries the Yukawa couplings as input data — flavor is again injected, not generated. NCG is a serious, genuinely economical competitor on the qualitative gauge content and an honest one on its limits.

Lattice / causal-set / asymptotic-safety programs. Each commits a different starting object (a discretization, a partial order, a fixed-point action). None claims, or proves, that its starting object is forced over all alternatives; each leaves substantial SM content as input.

The honest summary of the state of the art: every framework recovers the SM gauge group (so "we get $SU(3)\times SU(2)\times U(1)$" is a filter every framework passes — a TIE, not a discriminator), and every framework injects flavor data in some layer. What no framework has done is (a) commit its starting object reproducibly to the last byte in public, and (b) prove that object is selected over a declared competitor class. SG-1's contribution is (a) outright, plus a bounded, category-relative down-payment on (b), with the absolute version openly left as the shared ceiling.

2.3 Prior attempts and why each falls short — including ours

• Landscape scans fall short on selection: they enumerate vacua but supply no principle picking one; the moving target is the catalog index.
• Anthropic / measure arguments attempt selection but require a probability measure on the landscape that is itself undefined; they convert a structural gap into a measure-theory gap.
• 4D GUT unification falls short on flavor: the economy is real in the gauge sector and absent in the flavor sector.
• Our own first headline ("4 inputs → 22 outputs") fell short on honesty and we retired it. The true charged cost is ~13–14 measured reals (§2.4 below). Overstating the economy is exactly the moving-target failure in reverse — quietly under-counting inputs — and the program's own honesty pass caught and corrected it.
• Our absolute-minimality ambition ("13D is the geometry nature had to use") falls short on computability: it is a universal negative over all conceivable architectures, equal to the Kolmogorov complexity $K(T)$ of the constant set, which is uncomputable. We do not claim it; we name the achievable, bounded substitute.

2.4 The honest cost, stated up front (the retired headline)

The public-facing economy claim was once "4 anchors → ~22 outputs." The corrected, authoritative figure is:

≈4 named anchors $\{M_{\rm Pl},\ \alpha_i(M_Z),\ y_t,\ |V_{us}|\}$ + ~9–10 independently fitted reals (the species normalizations $N_d, N_e, N_\nu$; the $\delta$ threshold triple; the Higgs-protection angle $\theta_H^\star$) ≈ 13–14 measured reals total.

Both wrong figures are retired: the overstated ~18-economy headline (which counted fitted injections as outputs) and a ~1.6× over-correction (which double-counted frozen within-sector ratios that the family-level-normalization ban forbids treating as independent). The middle figure — ~9–10 injected reals beyond the four anchors — is the binding one.

(Sources: SHAPE_FINAL_STATUS_SPLIT_RESULT.md §"Honest caveats" items 1–2; DIMENSION_LADDER_MDL_AUDIT_CERTIFICATE.md Cert 1 target ledger ≈25 reals and Cert 4 note on $S_{13}$; SG1_COMPLETION_RESULT.md §2.)

3. The construction — rigorous math

This is where the pages live. We give the object end-to-end, the three-layer algebra and its necessity proof, the carrier-forcing theorems (F1, F2, abelian-isotropy uniqueness), the freeze/reproducer machinery, and the MDL scoring scaffold that turns "shortest" into a decidable statement. Common material lives on the published manuscript (GUT.html §2/§2B/§4/§6.1 and Appendices A0–A3, B1, B2, C1–C10, R0, GS, N.4); here we recap at attack/build depth.

3.1 The object, layer by layer

$$ \mathfrak{B}_{\rm active} = \big[\,\mathcal{M}_4 \times K_6 \times S^2 \times S^1_Y/\mathbb{Z}_2\,\big]_\times \;\oplus\; \big[\,F^+_{\rm finite} \oplus C_{\rm admiss}\,\big]_\oplus \;\otimes\; \big[\,E_{\rm matter}\oplus E_{\rm gauge}\oplus E_{\rm Higgs}\oplus E_{\rm proton}\,\big]_\otimes . $$

The $\times$-layer (STAGE) — the metric geometry, 13 dimensions. - $\mathcal{M}_4$: the observed $(3{+}1)$ Lorentzian spacetime — an observational primitive, the external comparison surface KK reduction lands on. It is declared, not selected (this is residual R9; honest by design, not a defect). - $K_6 = SU(3)/T^2$: the flag manifold of $SU(3)$, a 6-dimensional compact homogeneous space. Its isometry group sources color $SU(3)$: in the coset-space dimensional-reduction (CSDR) picture, gauge forces are isometries of the internal factors. The isotropy subgroup is the maximal torus $T^2$. - $S^2$: the 2-sphere, isometry $SU(2)$ — the weak source. - $S^1_Y/\mathbb{Z}_2$: a circle for hypercharge $U(1)_Y$, with a $\mathbb{Z}_2$ orbifold identification that acts as the chirality filter (it projects out the mirror partners that an unorbifolded closed odd-dimensional factor would keep).

The dimension ledger is $D = 4 + 6 + 2 + 1 = 13$. This is the rung whose forcedness is the entire SG-1 attack surface: $D=13$ is selected inside the declared search category, not derived (the headline residual, R1/R2/R5).

The $\oplus$-layer (RULEBOOK) — finite admissibility, 0 metric dimensions. - $F^+_{\rm finite}$: the flavor chamber — modulus pinned at $\tau=\omega$ (a cube root of unity), with projectors, action ladders, and operators. This is the SG-8 object; flagged by the program as its weakest link ($F^+$ is the highest-risk minimality lemma; see R10 and SG-8 R1). - $C_{\rm admiss}$: the admissibility set — the anti-fitting firewall, the selector, the freeze discipline, and the no-smuggling rule. It constrains which configurations and deformations are legal.

The rulebook carries zero metric dimension; it is a constraint layer, not a geometry. A live hazard (R10): some of $C_{\rm admiss}$ is method/governance (freeze-before-compare, gate-status discipline), not architecture-neutral physics. Counting governance in the minimality burden risks F6 circularity (the constraint set presupposing the submitted architecture). Only the physics the rulebook enforces — anomaly admissibility, flavor closure, chamber constraints — may be charged.

The $\otimes$-layer (ACTORS) — bundle/operator content, 0 metric dimensions. - $E_{\rm matter}$: the matter bundle (chiral fermion content). - $E_{\rm gauge} = (P, \mathrm{ad}(P), A, F, \rho_{\rm rep}, \text{KK spectrum})$: the gauge bundle and its data. - $E_{\rm Higgs}$: the Wilson-line / Hosotani Higgs (symmetry breaking from boundary conditions on the internal space). - $E_{\rm proton}$: the proton-stability structure.

A key actor-layer result that is computed, not posited: the $\mathbb{Z}_6$ center-kernel that identifies the centers of the $SU(3)$ and $SU(2)$ factors is read off the matter content as a discrete index, $$ q \equiv 3 z_2 - 2 z_3 \pmod 6 , $$ i.e. the hypercharge assignment is fixed (mod 6) by the $SU(2)$ and $SU(3)$ center charges of each multiplet. This is a discrete integer relation given E, not a tuned real — which is exactly why the actor layer has real (if incomplete) support (R8).

3.2 The three-layer algebra and its necessity (certificate B2)

The $\times/\oplus/\otimes$ structure is not decorative; it is a load-bearing discipline. Each operator has a distinct role: - $\times$ (Stage): the metric product of factors — only this layer carries dimension. - $\oplus$ (Rulebook): the direct sum of admissibility constraints — finite, dimensionless. - $\otimes$ (Actors): the tensor of bundle/operator content over the stage.

The no-smuggling claim is that no downstream gate is closed using content that was not declared in its proper layer. The certificate is B2: the proper-subset null-space theorem. Define, for any proper subset $L \subsetneq \{\times, \oplus, \otimes\}$ of the layers, the set $\mathcal{N}_L$ of branch configurations that close all scoped-GUT gates using only the layers in $L$. B2 proves $$ \mathcal{N}_L = \varnothing \quad \text{for every proper } L \subsetneq \{\times,\oplus,\otimes\}, $$ inside the declared category. In words: you cannot drop any one layer and still close the gates — every layer is necessary. Crucial honest scoping (B2.0.3.4): this is an in-category null-space result, explicitly not a universal no-go. It asserts load-bearingness inside the declared category, not for any conceivable architecture. Upgrading it to architecture-neutral necessity is residual R7 (the functional-role necessity theorem).

Complementing B2 at the term level, the C1–C10 failure-if-removed ledger records, for each named term, at least one Gate 1–10 that breaks if the term is deleted. These are auditable conditional claims — conditional on the declared term-construction and the failure ledgers — not theorems. They are hand-checkable and strong, but category-relative (R7).

3.3 Carrier forcing — the three internal rungs, graded honestly

Not all three internal dimensional choices are equally forced. This is the single most important honest refinement in the gate, and it is stronger than the blanket "all category-relative" claim because it says exactly where the program is strong and where it is exposed.

(a) Weak carrier $S^2$ — FORCED within the grammar by F1.

Fact F1. No abelian / torus carrier of any dimension has non-abelian $SU(2)$ among its isometries.

The isometry group of a flat torus $T^k$ is (up to discrete factors) $T^k \rtimes \text{(finite)}$ — abelian connected component. A non-abelian $SU(2)$ gauge factor cannot arise as the isometry of any abelian carrier, of any dimension. So any attempt to source the weak interaction from a "cheaper" abelian/torus factor fails for all such carriers at once — the cheaper direction is closed over a whole shelf, not against a named candidate. $S^2$ (the lowest-dimensional carrier with $SU(2)$ isometry) is the forced choice within the grammar. This is a hand-checkable general theorem. (Source: SHAPE_FINAL_STATUS_SPLIT_RESULT.md §"forcedness gradient", F1.)

(b) Hypercharge carrier $S^1_Y/\mathbb{Z}_2$ — FORCED within the grammar by F2.

Fact F2. A closed odd-dimensional factor keeps both handednesses, producing mirror fermions excluded by the LEP $Z$-width.

A closed odd-dimensional internal factor (e.g. an unorbifolded $S^1$) admits a Dirac operator whose spectrum is non-chiral: it keeps left- and right-handed partners symmetrically, producing mirror fermions. Mirror fermions are excluded experimentally by the measured invisible $Z$ width at LEP (which counts exactly three light neutrino species and tightly constrains extra chiral matter). The $\mathbb{Z}_2$ orbifold projection on $S^1_Y$ is what removes the mirrors. So the cheaper direction (a bare circle, no orbifold) is closed for all closed odd-dimensional carriers. Hand-checkable. (Source: same, F2.)

(c) Color carrier $K_6 = SU(3)/T^2$ — CLEAN by an architecture-neutral theorem (abelian-isotropy uniqueness); the cheaper rival was built and broke.

This was historically the weak link — the color rung is where a genuinely cheaper competitor lives. The minimal-dimension $SU(3)$ carrier is not $K_6$ (6D) but $CP^2 = SU(3)/U(2)$ (4D). A naive dimension-first reading says $CP^2$ wins ($4 < 6$). The corpus's original exclusion of $CP^2$ — "its 3-family count is a tunable bundle modulus, hence an injected real, hence a longer recipe" — turned out to be asymmetric and unsound as stated: $CP^2$'s 3 families are a discrete $\mathrm{Spin}_c$ index ($r(r+1)/2 = 3$ at $r=2$), not a continuous dial, and $K_6$ also needs its own bundle choice (Borel–Weil–Bott weight $(1,0)$) to land on 3 families. Treating one as tunable and the other as forced was a target-fitting error, and the program caught it.

The honest resolution came from building the cheaper competitor end-to-end (run wsmnjvt55) and watching where it actually breaks. The result is a new, architecture-neutral exclusion principle:

Abelian-isotropy uniqueness. Among $SU(3)$ cosets $SU(3)/R$, the maximal torus $T^2$ is the unique isotropy that is purely abelian. Its centralizer is $C_{SU(3)}(T^2) = T^2$ (the Cartan only) — so by the CSDR centralizer rule it injects no spurious non-abelian gauge factor, keeping color / weak / hyper separable. $CP^2$'s isotropy is $U(2) = (SU(2)\times U(1))/\mathbb{Z}_2$, which is non-abelian and a subgroup of color $SU(3)$. By the CSDR centralizer rule, $U(2)$ is gauge-active, forcing a lose-lose fork: - keep $S^2, S^1$ alongside $CP^2$ → over-produce an extra $SU(2)+U(1)$ → the Gate-2 gauge-group equality fails; or - drop $S^2, S^1$ → $SU(2)_L / U(1)_Y$ become isotropy-locked inside $SU(3)$ → binding A1.4 violated ($C_{SU(3)}(U(2)) = U(1)$ only, by Schur).

So $K_6 = SU(3)/T^2$ is the unique clean $SU(3)$ color carrier — derived from $\mathrm{Rep}(T^2)$ vs $\mathrm{Rep}(U(2))$ and the centralizer rule alone, with no flavor or coupling number in sight. This is stronger than the original "tunable family count" reason, and it passes the target-blindness (κ³/π) falsification test by construction. Four independently-built sectors of the $CP^2$ swap converged on this one break. (An M-theory mesh check came back superficial: $CP^2 \times S^2 \times S^1$ is a legitimate 11D Freund–Rubin KK ansatz but not a $G_2$-holonomy compactification — a dimension coincidence, not a structural mesh.)

The honest residual on (c): abelian-isotropy uniqueness closes the named shelf $\{K_6, CP^2\}$. Full-shelf completeness — that no other admissible compact $SU(3)$-homogeneous carrier below 6D has a clean centralizer-surviving abelian isotropy — is corpus-uncertified (this is residual R4 / N.4). A partial classification already shows $\dim H \le 4 \Rightarrow \dim M \ge 4$, with $S^5$/Wu spaces killed by odd-dimensionality, so the shelf is plausibly the whole sub-6D $SU(3)$ story — but the certification is the work.

(Sources: SHAPE_FINAL_STATUS_SPLIT_RESULT.md UPDATE block 2026-06-24; SHAPE_COLOR_RUNG_CP2_REFUTATION.md; ELEVEN_D_CP2_VARIANT_CONSTRUCTION.md; existing article §2.1 W9 and §4.4.)

3.4 The freeze and the reproducer (the certified core)

The branch and every derived object are content-addressed by SHA-256. The frozen artifacts are: - branch content hash dcc66f1b2685 (the canonical string of the active branch); - manifest meta-hash a5b1e6f9d951 (Appendix A0, 33 rows, the roll-up over all per-item hashes in file order); - the orbifold-freeze sub-hash ac4d2df3e708 (R1.3).

The R0 reproducer (reproduce_all.py + manifest_hashes.json) regenerates each hash with no manual steps. Content-addressing plus a reproducer that regenerates every hash is the strongest, most defensible leg of the gate: it makes the object a reviewer attacks byte-identical to the object the gates ran on. The discipline (freeze-before-compare, no-smuggling) is the entire purpose of SG-1, and it is genuinely sound by construction.

The frozen UV/boundary package the branch carries (load-bearing for downstream gates, recorded here for traceability, READ-ONLY): - unification scale $M_U \sim 10^{16}$ GeV; - compactification radius $R_0 = 1.592 \times 10^{-17}\ \mathrm{GeV}^{-1}$; - threshold triple $\delta = (+4.8424,\ -3.1112,\ -1.7313) \pm 1.6\times 10^{-3}$; - family index $\chi(K_6, E) = -3$.

(Sources: existing article §0 header and §5.3 hash index; SG1_COMPLETION_RESULT.md §3; GUT.md cited lines for $M_U$/$R_0$/$\delta$/$\chi$.)

3.5 The MDL scoring scaffold — turning "shortest" into a decidable statement

The minimality claim needs a metric. The program uses minimum description length (MDL): in a finite-record universe, admissible descriptions are finite bit-strings, and "simplest" = "shortest such string." The scaffold (Certificates 1–4 of the dimension-ladder audit) is:

Certificate 1 — the target ledger $T$ (architecture-neutral, $E$ cancels). Both the 13D branch and every competitor must reproduce the same $T$, stated with no geometry names. $E$ (chiral spectrum + gauge group + quantum numbers) is presupposed on both sides and cancels — it is not counted. What MDL counts is the number of independent measured reals each side must inject:

Certificate 2 — the codebook (frozen before scoring). Fixed neutral bit-costs, chosen so no architecture is syntactically cheap: an independent measured real anchor costs $b = \log_2(1/\Delta_0)$ bits (large — set by the operational resolution $\Delta_0$); a manifold factor from a named catalog, a coset $G/H$, a bundle/representation, a finite quotient, or a named generator rule (RG running, heat-kernel, index theorem) each cost $O(1)$; a topological integer costs $O(\log)$; and a fitted table / tuned normalization is charged like anchors ($\approx$ entries $\times\, b$). The decisive consequence is the anti-fitting ⇄ MDL bridge: a tunable match is not free — the tuned value must be injected as a measured real, costing $\sim b$. So "adjustable = fail" is an MDL penalty, which is what makes a considered-class win non-circular.

Certificate 3 — the normal-form / exhaustion theorem (the hard, OPEN target). The claim to prove: every admissible architecture $B \models T$ reduces, without increasing $I(B)$, to a normal form determined by its choice of mechanism for each functional role, over a finite taxonomy of five axes: - gauge-origin ∈ {posited-4D · internal-isometry (KK/coset) · holonomy/Wilson-line · algebraic (NCG)} - chirality-origin ∈ {posited · index-theorem · domain-wall/overlap} - family-count ∈ {posited integer · topological index} - flavor-origin ∈ {posited Yukawas · geometric overlap · fitted normalization} - scale-origin ∈ {posited anchors · dynamical}

The two proof obligations are (1) reduction-without-cost-increase (the no-smuggling metric: a role hidden in notation is charged after unfolding, so a rewrite cannot lower cost by hiding) and (2) taxonomy completeness, declared (a genuinely novel mechanism is unconsidered and extends the grammar; the theorem is exhaustion relative to the declared taxonomy). This is a classification theorem in the spirit of the classification of simple Lie groups: tractable iff the role-mechanism invariants cut the space into finitely many classes.

Certificate 4 — the lower-bound matrix (per class, the OPEN job). For each class $N_{D,j}$ prove a lower bound $I(B) \ge L_{D,j}$ from one of: (A) failure to generate $T$; (B) anchor floor $I \ge n_{D,j} b + O(1)$; (C) hidden-rule floor (projectors/tuned conditions/lookup tables add $I_{\rm hidden}$); (D) generator floor (the map {anchors+structure}$\to T$ must be supplied and fully charged). Writing $I(B_{13}) = S_{13} + 4b$ and $I(B_{D,j}) \ge S_{D,j} + n_{D,j} b$, the whole theorem is the boxed inequality $$ \boxed{\ \forall\, D = 4..12,\ \forall j:\quad (n_{D,j} - 4)\, b \;+\; (S_{D,j} - S_{13}) \;>\; 0\ } $$ where $S_{13}$ is the 13D branch's fully-charged structure+generator cost. Honest note from the audit: $B_{13}$ is not "$4b$" — it injects $\approx 9$–$10$ reals beyond the 4 anchors, so the honest $S_{13}$ carries those. The economy is real but modest; the matrix must use the honest $S_{13}$, not the "4-in" headline.

Where this stands today. A first-pass population of the matrix (ladder audit wst99pv9j) scored 11 considered competitors and gave 0 REFUTED, 1 FAILS_TO_GENERATE_T (the 6D rung), 10 LOSE to 13D under MDL — the 13D branch beats the plain 4D EFT (~13–14$b$ vs ~25$b$), beats heterotic/M/F-theory on structural bits, and beats NCG and lattice. But this is a survey, not a certified classification (Certificate 3 is OPEN), so the correct verdict label is CATEGORY_RELATIVE: 13D wins every competitor considered, under MDL, inside the grammar. Not "no competitor below 13D is shorter."

(Sources: DIMENSION_LADDER_MDL_AUDIT_CERTIFICATE.md Cert 1–4 verbatim; SHAPE_LADDER_MDL_VERDICT.md (10 LOSES / 1 FAILS / 0 REFUTED); SHAPE_FINAL_STATUS_SPLIT_RESULT.md.)

4. The insights we used

The progress in SG-1 came from a small number of reusable moves. Sharing them is what makes the gate believable and the closure paths actionable.

Insight 1 — Freeze first, compare second; content-address everything. The deepest fix for the moving-target failure is not an argument, it is mechanism: hash the object before any gate is run, publish the hashes, and ship a regenerator. This converts "trust us, we didn't retune" into a byte-identity check anyone can re-run. It is the cheapest and strongest leg of the gate precisely because a self-witness asserts nothing about nature — it only pins the object — so it cannot be wrong about physics.

Insight 2 — Declare a finite grammar to tame an uncomputable question. "Is there any shorter geometry, under any possible mathematics?" is a universal negative equal to $K(T)$, the Kolmogorov complexity of the constant set — uncomputable for everyone. The move that rescues a real claim is to declare a finite role-mechanism grammar $\mathcal{G}$ (the five mechanism axes). This converts the uncomputable universal negative into a finite, decidable problem: a lower-bound matrix over finitely many normal-form classes. The conversion is the achievement; it does not by itself close the matrix. The grammar is living and extensible: a genuinely novel mechanism extends $\mathcal{G}$, it does not refute the claim.

Insight 3 — Anti-fitting is an MDL penalty (the bridge that makes a win non-circular). The selector's anti-fitting rule ("adjustable = fail") looks like a separate discipline. Under the MDL codebook it is the same thing: a tunable match must inject the tuned value as a measured real, costing $\sim b$ bits. So a competitor that "matches the data by adjusting a knob" is automatically charged for that knob. This is what stops a considered-class win from being circular — you cannot beat the ledger by hiding fitted reals as "free" structure.

Insight 4 — Build the cheaper rival end-to-end, then watch where it breaks. The color rung was rescued not by a clever a-priori argument but by constructing the $K_6 \to CP^2$ swap fully and adversarially. It broke at the gauge gate, and the break revealed the right exclusion principle — abelian-isotropy uniqueness — which is architecture-neutral and far stronger than the asymmetric "tunable family count" reason it replaced. The lesson: a built-and-broken competitor yields a sounder exclusion than an unbuilt dismissal, and it self-corrects target-fitting.

Insight 5 — Grade forcedness rung by rung, not in bulk. "All category-relative" was true but weak. Separating the three internal rungs — weak forced by F1, hyper forced by F2, color clean by abelian-isotropy uniqueness with N.4 open — is both stronger (two rungs are forced by general theorems) and more honest (it names the single exposed completeness obligation). Fine-grained grading beats blanket caveats.

Insight 6 — Charge the generator map (the guard against the easiest self-deception). The 13D win evaporates if the geometry→observables generator map is itself a large injected object. The discipline that protects the claim is guard G2: the generator map must be fully charged in the ledger. Symmetrically, the EFT competitor must be charged only for reals it must inject as independent facts, and anything that is an automatic consequence of $E$ + 4D renormalizable gauge invariance (anomaly cancellation, written-spectrum chirality, the $\mathbb{Z}_6$ center-kernel, accidental proton stability) is free-for-both-and-cancels — charging the EFT for those would be reverse tuning to the known answer.

Insight 7 — The cautionary precedent (a posit that looks derivable can be circular). The program's SCALE-firewall verdict found that the operational-cell scale $\mu_{\rm cell}$ has no $v$-independent readout, so anchoring it at $dV/d\sigma = 0$ is circular. That is the standing warning for SG-1's metric question: the granularity ⇒ MDL bridge must be derived without knowing 13D should win (guard G1), or it will be the same kind of circular self-confirmation. A root posit that looks derivable has already been caught being circular once.

5. Evidence & reproducibility

5.1 The witness ledger

5.2 The reproducibility leg, verified — and its load-bearing caveat

The one genuinely derived leg of SG-1 is reproducibility (R6), and it was machine-verified by an independent re-run during the SG-1 completion, not merely asserted: - the regenerator (reproduce_all.py + manifest_hashes.json) was run target-blind: exit 0; it reports all 33 per-item hashes matching and the manifest meta-hash matching a5b1e6f9d951; - adversarial recheck (not trusting the script's own comparator): the SHA-256 of every canonical description was recomputed from scratch → 33 rows, zero mismatches; the active-branch canonical string hashes to dcc66f1b2685; the meta-hash over all 33 rows in file order is a5b1e6f9d951; - determinism: two fresh reruns produce byte-equal artifacts.

Caveat (load-bearing, carried verbatim from the completion result). This is a content-addressed self-witness: a SHA-256 of canonical English descriptions of the geometry. It commits the object a reviewer attacks to the object the gates ran on. It lowers no assumption floor and closes no physics. It is "derived" precisely because a self-witness asserts nothing about nature. Separately, the reproducer's physics CSV outputs (lepton ratios, $m_b$ target, NuFIT values, the threshold vector) are reverse-pinned to known targets and are outside the hash-verified self-witness — "the reproducer runs green" is not output validation.

This is the cleanest honesty point in the gate: the strongest mechanical leg is verified, and its scope (object-commitment, not physics-validation) is stated plainly.

5.3 How a reader re-derives / re-runs

1. Re-run the reproducer (R6). Mount reproduce_all.py + manifest_hashes.json + the A0 manifest (33 rows); run target-blind; confirm both hashes regenerate byte-equal and A1's ≥16-sig-fig reconstruction recomputes from the frozen primitives. Fail-closed: if any hash cannot be regenerated, the gate downgrades from claimed certificate pass to open / not claimed.
2. Hand-check F1 and F2. F1: write the isometry algebra of any torus and confirm it has no non-abelian $SU(2)$ subalgebra. F2: write the Dirac spectrum on a closed odd-dim factor and confirm both handednesses survive; cross-check against the LEP invisible-$Z$-width bound.
3. Hand-check abelian-isotropy uniqueness (W9). Compute $C_{SU(3)}(T^2) = T^2$ (Cartan only) and $C_{SU(3)}(U(2)) = U(1)$ (Schur); apply the CSDR centralizer rule; confirm the $CP^2$ lose-lose fork.
4. Re-score the MDL survey. Use the Cert 2 codebook and Cert 1 target ledger; reproduce the 10-LOSE / 1-FAILS / 0-REFUTED first pass with the honest $S_{13}$ ($n_{13} \approx 13$–$14$).

5.4 Numbers in this dossier and their sources

Every number above is traceable to a file read for this dossier. Where a quantity is uncomputed (the certified lower-bound matrix; N.4 full-shelf completeness; the granularity ⇒ MDL theorem), it is marked OPEN in §6 — no fabricated value substitutes.

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

This is the most load-bearing section. Each open hole is a work-package: precise statement, why it's hard + traps the verifier already caught, exactly what closes it (with success and refuting criteria — a negative is a valid close), the machinery and corpus files to start from, and the cross-gate leverage. Physics only; the firewall holds throughout.

The two decisive seams are R5 (which metric is correct) and R2/R4 (the classification and shelf completeness). R5 is a single theorem that can flip the entire ladder; close it the wrong way and SHAPE-minimality folds for everyone — which we would report honestly as REFUTED-ECONOMY.

6.1 R5 — the MDL-vs-dimension-first metric-selection theorem (the decisive seam)

(a) Precise statement. Every "13D wins" verdict holds under the MDL / description-length metric. Under a dimension-first lexicographic order (with $k_{\rm dim}$ the leading key) a clean 4D chiral-gauge EFT beats 13D ($4 < 13$) irrespective of injected reals, and SHAPE-minimality folds for everyone. The open object is an architecture-neutral principle that selects MDL over dimension-first. Candidate: granularity ⇒ MDL, with three sub-claims: - (A) Granularity ⇒ MDL. Derive the description-length functional from the cost-floor / recordability — do not assume it. (A finite-record universe cannot specify infinite-precision continuum data; its admissible descriptions are finite bit-strings; "simplest" = shortest such string = MDL.) - (B) Anchors dominate dimension under $\mathcal{I}$. A measured real to cell resolution costs $\Theta(\log(1/\Delta_0))$ bits; a discrete structural choice (a coset, a $\mathbb{Z}_6$, an integer index) costs $O(1)$ bits. - (C) Economy-win, quantitative + symmetric. Build both ledgers $I(B_{13})$ and $I(B_{\rm EFT})$ against the same pre-declared target set $T$, with the geometry→observables generator map fully charged.

(b) Why it's hard / trap lessons. The first-pass plug [REDUCED] this and found the precise residual: granularity fixes the domain (finite bit-strings) and the per-anchor cost $b = \log_2(1/\Delta_0)$ (verified: DIMENSION_LADDER_MDL_AUDIT_CERTIFICATE.md Cert 2), but is silent on the aggregation rule. Both MDL (additive sum) and dimension-first lex ($k_{\rm dim}$ leading key) are well-defined on the same finite-bit-string space; the granularity premise $P$ does not discriminate between them. A lexicographic leading key is mathematically equivalent to an unbounded / infinite weight ratio on that field relative to an additive sum — and granularity (a finite common resolution $\Delta_0$ per record) provides no weight asymmetry between a dimension-bit and an anchor-bit. So the honest residual is one named axiom, AXIOM-COMMON-CURRENCY (equivalently AXIOM-GRANULARITY-MDL-BRIDGE): "all brute-fact bits at the common operational resolution are charged additively; no theory-category, including dimension, carries infinite lexicographic weight over another." Under it, MDL holds and 13D wins given $E$; reject it and the metric fork reopens and SHAPE folds.

Traps the verifier flagged, name them so they are not repeated: - Do not let granularity enter as "the shape must geometrize / chirality must come from an internal index." That is F6 circularity — it renames the answer by narrowing the class to KK architectures. The metric must be architecture-neutral. - Do not reverse-engineer the metric to favor low anchor count. That is the κ³/π failure (guard G1: the metric counts only if it would be written without knowing 13D should win). The SCALE-firewall precedent ($\mu_{\rm cell}$ with no $v$-independent readout, circular at $dV/d\sigma=0$) is the cautionary case. - Do not leave the generator map uncharged (guard G2). If the geometry→observables map is a large injected object, the 13D win evaporates and the EFT wins. - Do not charge the EFT for $E$-automatic facts (anomaly cancellation, written-spectrum chirality, the $\mathbb{Z}_6$ kernel, accidental proton stability) — that is reverse tuning to the known answer (guard G5, symmetric ledgers).

(c) Exactly what closes it. Prove sub-claims A/B/C under guards G1/G2/G5. Success (DERIVED-CLOSED): A + B + C hold with the generator charged → the 4D EFT is excluded on a principled basis → SHAPE upgrades from "selected" to "realization-minimal under the granularity-induced information metric, given $E$." Refuting result (equally valuable, REFUTED-ECONOMY): under a fair $\mathcal{I}$ with the generator charged, the 4D EFT injects no more than 13D → the simpler 4D realization is correct → the 13D geometry is selected, not minimal. AXIOM-CLOSED (likely interim): A holds but the aggregation rule cannot be derived from granularity alone → name AXIOM-COMMON-CURRENCY as the single target-blind posit and stop. The first-pass plug already established this is the honest current ceiling.

(d) Machinery & inputs. Derive $\mathcal{I}$ from recordability (A); establish the bit-cost ordering (B); compute the two symmetric ledgers (C) with the honest $n_{13} \approx 13$–$14$. Start from: SHAPE_REALIZATION_GRANULARITY_METRIC_TARGET.md (the theorem target, guards G1/G2/G5); SHAPE_METRIC_RESULT.md (the five operational-record-cost axioms claiming MDL unique up to $O(1)$); DIMENSION_LADDER_MDL_AUDIT_CERTIFICATE.md Cert 2 (codebook); SHAPE_LADDER_CONTEXT_WITH_SELECTOR.md line ~238 (dimension-first defined as lex with $k_{\rm dim}$ leading); SHAPE_LADDER_MDL_VERDICT.md §2.3.

(e) Leverage. Closing R5 gates R2 (the matrix is only meaningful under a justified metric), resolves R4's category-fairness sub-residual (the fairness question routes to the metric), and decides the headline forced-given-E vs merely-selected for the whole gate. This is the single highest-leverage object in SG-1.

6.2 R2 — the dimension-ladder lower-bound matrix / grammar exhaustion (the whole remaining job)

(a) Precise statement. Under MDL the 13D branch wins every considered rung (10 LOSES / 1 FAILS / 0 REFUTED), but this is a first-pass survey, not a certified classification. The open object is Certificate 3 (normal-form / exhaustion) + Certificate 4 (per-class lower bound): prove every $B \models T$ reduces cost-non-increasingly to a role-mechanism tuple over the five axes, then prove the boxed inequality $\forall D=4..12, \forall j: (n_{D,j}-4)b + (S_{D,j}-S_{13}) > 0$ for each class.

(b) Why it's hard / trap lessons. It is a genuine classification theorem (Lie-group-classification-grade), bottlenecked on the two proof obligations: (O1) reduction-without-cost-increase (the no-smuggling metric — a role hidden in notation is charged after unfolding) and (O2) declared taxonomy completeness. Trap: do not give the 13D branch a privileged row — it is itself a tuple (gauge = isometry/coset; chirality = index; family = index; flavor = geometric overlap + fitted normalization; scale = posited anchors) and must be scored by the same Cert-2 codebook. Trap: do not use the "4-in" headline for $S_{13}$ — use the honest $n_{13} \approx 13$–$14$ (the audit explicitly flags this).

(c) Exactly what closes it. Prove O1 + O2. Success (DERIVED-CLOSED): grammar-relative LADDER_FORCED (still grammar-relative, still bottoms on $E$ — not absolute). AXIOM-CLOSED-as-survey (likely interim): name AXIOM-ROLE-MECHANISM-GRAMMAR and carry the first-pass matrix as a survey. Refuting result: a class is found with a strictly shorter MDL recipe → 13D is not minimal even in-grammar (valuable negative). sharper-OPEN: some classes lower-bounded, others surveyed → partial matrix.

(d) Machinery & inputs. (1) Compress the 13D generator into pseudocode normal form and measure $S_{13}$ honestly. (2) For each $D=4..12$ + non-dim, prove (A) failure / (B) anchor floor / (C) hidden-rule floor / (D) generator floor. (3) Publish the completed matrix. Start from: DIMENSION_LADDER_MDL_AUDIT_CERTIFICATE.md Cert 3/4 (the scaffold + matrix template); ROLE_MECHANISM_TAXONOMY_POSITS.md (the grammar $\mathcal{G}$, five axes, the 13D branch as a tuple); SHAPE_LADDER_MDL_VERDICT.md (first-pass rows); SHAPE_REALIZATION_DERIVATION_TARGETS.md (Lemmas 1–5, Gate 0 / defeat-F6).

(e) Leverage. Closing R2 upgrades the ladder verdict from CATEGORY_RELATIVE to grammar-relative LADDER_FORCED. R2's normal-form subsumes R4's SU(3)-carrier seam (the sub-shelf must be exhausted, not enumerated). Conditional on R5.

6.3 R4 — search-category completeness (R2.5 fairness + the SU(3)-carrier shelf / N.4)

(a) Precise statement. Two sub-residuals, both the corpus's own declared first review targets (GUT.md:382): (a) is the declared category R2.5 (forces = isometries, compact, classified structures, admissible bundles) a fair category, or does it exclude natural competitors (string bundle-sourced gauge, NCG spectral triples, 4D chiral-gauge EFT)? (b) Is the named $SU(3)$-carrier shelf $\{K_6, CP^2\}$ complete (N.4) — i.e. no other admissible compact $SU(3)$-homogeneous carrier below 6D has a clean centralizer-surviving abelian isotropy?

(b) Why it's hard / trap lessons. Category fairness is a wall (no canonical "fair-category" principle exists); the color sub-shelf is a bounded gap. Trap already caught: the original "CP² is tunable" exclusion was asymmetric and unsound — $CP^2$'s 3 families are a discrete $\mathrm{Spin}_c$ index $r(r+1)/2 = 3$, not a continuous dial, and $K_6$ needs its own BWB weight $(1,0)$ too. Do not reinstate an asymmetric exclusion. The sound replacement (abelian-isotropy uniqueness, W9) is architecture-neutral and passes the κ³/π falsification test — use it, and certify completeness over the shelf, not against a named rival.

(c) Exactly what closes it. Certify N.4: enumerate all admissible compact $SU(3)$-homogeneous carriers and prove none below 6D has a clean abelian isotropy other than the $K_6/CP^2$ pair, and that $CP^2$ breaks at Gate 2. Success (DERIVED-CLOSED): N.4 complete + abelian-isotropy uniqueness over the full shelf → the color rung is forced within the grammar. AXIOM-CLOSED (already standing): AXIOM-ABELIAN-ISOTROPY-COLOR as a theorem on the named shelf + CP²-built-and-broken — stronger than the original corpus position. Refuting result: a cheaper admissible $SU(3)$ carrier with a clean abelian isotropy is exhibited → the color rung folds. sharper-OPEN: N.4 stays uncertified → $K_6$ is the cheapest clean carrier on a named shelf only. For category-fairness (R2.5): state the claim as category-relative and route fairness to R5 — do not defend R2.5 as the unique fair category (that would be circular).

(d) Machinery & inputs. A bounded coset-classification + CSDR centralizer check. The partial classification already shows $\dim H \le 4 \Rightarrow \dim M \ge 4$, with $S^5$/Wu killed by odd-dimensionality. Start from: SHAPE_COLOR_RUNG_CP2_REFUTATION.md (CP² classification + the $r(r+1)/2$ index + the bundle-admissibility one-move); ELEVEN_D_CP2_VARIANT_CONSTRUCTION.md (the end-to-end build that broke); SHAPE_FINAL_STATUS_SPLIT_RESULT.md UPDATE block (abelian-isotropy uniqueness, the centralizer computations).

(e) Leverage. A natural in-category competitor strictly preferred under the metric is the standing reopen → documented downgrade to category-relative diagnostic. Closing N.4 hardens the one historically-exposed rung.

6.4 R8 — actor-layer minimality (SHAPE Lemma 3) + E un-forced (demoted-OPEN)

(a) Precise statement. Is $E_{\rm matter}\oplus E_{\rm gauge}\oplus E_{\rm Higgs}\oplus E_{\rm proton}$ the minimal actor realization given $E$? Two undischarged obligations: the no-cheaper-competitor obligation (competitor matrix all Tier-1 "Unknown"; NCG's finite Dirac operator un-scored as a role-equivalent), and $E$ itself un-forced (the SM chiral content presupposed on both sides).

A sharper established finding to surface, not gloss: the generation count is $E$-forced, not geometry-forced. The corpus carries a concrete, bounded result (SHAPE_FINAL_STATUS_SPLIT_RESULT.md lines 57–59; SHAPE_COLOR_RUNG_CP2_REFUTATION.md §4): the index-3 family count requires a chosen bundle on both carriers — Borel–Weil–Bott weight $(1,0)$ on $K_6$, and the $\mathrm{Spin}_c$ choice $r=2$ on $CP^2$ — and the two carriers are symmetric under $E$. So the family number $3$ traces to $E$ (it is the bundle picked to reproduce the observed three generations), not to the geometry being forced. This is the bundle-admissibility symmetry that retired the asymmetric "$CP^2$ family count is tunable" exclusion (§3.3c); it deepens the bottoms-on-$E$ finding by localizing which observable (generation count) is $E$-derived. It is a falsifiable banked fact, carried here at its true strength — it does not make $\chi(K_6,E)=-3$ a forced rung.

(b) Why it's hard / trap lessons. The closure campaign DEMOTED an earlier AXIOM_CLOSED to OPEN under the overclaim test: Lemma 3's success criterion is a universal negative ("no admissible competitor supplies lower $k_{\rm actor}$"), which is UNMET. The trap (already caught): folding two open obligations (undischarged universal negative + un-forced $E$) into one axiom is overclaiming. Do not re-bank AXIOM_CLOSED here. The partial support in hand is genuine: the $\mathbb{Z}_6$ center-kernel is computed ($q \equiv 3z_2 - 2z_3 \bmod 6$, a discrete index given $E$), not posited.

(c) Exactly what closes it. Score the actor-layer competitor matrix target-blind — especially NCG's finite Dirac operator as a role-equivalent — and prove no competitor supplies lower $k_{\rm actor}$ given $E$. Success (DERIVED-CLOSED, unlikely — universal negative): matrix scored + no-alternative proven → actor-minimality given $E$. Refuting result: a lower-$k_{\rm actor}$ competitor (e.g. NCG) is scored cheaper → actor-minimality folds. sharper-OPEN (honest standing): reduced to AXIOM-ACTOR-MDL-MINIMALITY-GIVEN-E + un-forced $E$ + the undischarged audit. Note $E$ stays un-forced regardless — the corpus T3 (an attempt to force $E$) was REFUTED; do not present $E$ as derived.

(d) Machinery & inputs. The narrowest layer — attack first (if actor-minimality fails, realization-minimality fails fast). Start from: SHAPE_REALIZATION_DERIVATION_TARGETS.md (Lemma 3); CLOSURE_CAMPAIGN_RESULT_2026-06-24.md (the demotion to OPEN; the κ³/π falsification test; AXIOM-CLOSED ≠ proven); the $\mathbb{Z}_6$ kernel computation (existing article §1.1/§4.8).

(e) Leverage. Cross-links to the bottoms-on-E finding shared with GEO-02 (is $E$ forced?) and to SG-3 (the family index, also $E$-forced). A cheaper-actor refutation would localize realization-minimality's failure to the narrowest layer.

6.5 R7 — B2 / C1–C10 are category-relative + conditional (the functional-role bridge)

(a) Precise statement. B2 (three-layer necessity) is an in-category null-space result, explicitly not a universal no-go (B2.0.3.4); C1–C10 (term necessity) are conditional on the declared term-construction. They assert load-bearingness inside the declared category, not for any architecture. The open object: upgrade to a functional-role necessity theorem — any admissible architecture must carry functional equivalents of Stage + Rulebook + Actors ($k_{\rm role} \ge 3$), proven by contradiction per role with no hidden reference to the submitted factor set.

(b) Why it's hard / trap lessons. The bridge from category-relative to architecture-neutral must defeat F6 circularity per role (Gate 0): each role-requirement must be stated architecture-neutrally, with no hidden reference to the $\times/\oplus/\otimes$ notation. Trap: stating a role-requirement using the submitted factor set smuggles the answer.

(c) Exactly what closes it. Prove $\mathcal{C}_{\rm phys} \Rightarrow \mathrm{Stage} + \mathrm{Rulebook} + \mathrm{Actors}$ by contradiction on each role (a role-free admissible competitor is impossible). Success (AXIOM-CLOSED at the floor, essentially standing): name AXIOM-FUNCTIONAL-ROLE-FLOOR; B2 upgraded to architecture-neutral role-necessity — necessary, not sufficient. Refuting result: a role-free admissible competitor is exhibited → the floor fails. sharper-OPEN (likely): role-necessity holds but realization-minimality (the exact factor set is the unique minimum) stays open — that is R2/R8 territory.

(d) Machinery & inputs. Start from: T_SHAPE_FUNCTIONAL_ROLE_NECESSITY_THEOREM.md (the bridge, already PARTIAL); SHAPE_REALIZATION_DERIVATION_TARGETS.md (Gate 0, defeat F6); certificate B2 (the in-category result to upgrade).

(e) Leverage. Yields the architecture-neutral floor the corpus genuinely earns ($k_{\rm role} \ge 3$). Distinguishing the floor (necessary) from realization-minimality (sufficient) is what keeps the gate honest about what is proven.

6.6 R10 — rulebook physics-vs-governance separation (the F6 hazard)

(a) Precise statement. Several $C_{\rm admiss}$ requirements (freeze-before-compare, gate-status discipline) are governance/method rules, not architecture-neutral physics. Counting them in the minimality burden risks F6 circularity (the constraint set presupposing the submitted architecture). The open object: restate $\mathcal{C}_{\rm phys}$ architecture-neutrally, separating the physics the rulebook enforces (anomaly admissibility, flavor closure, chamber constraints) from method discipline, and charge only the physics burden.

(b) Why it's hard / trap lessons. $F^+ \subset C_{\rm admiss}$ is flagged the program's weakest link (highest-risk minimality lemma; tied to SG-8 R1). Trap: charging method discipline to a candidate is double-dishonest — it inflates the burden and it presupposes the submitted architecture.

(c) Exactly what closes it. Restate $\mathcal{C}_{\rm phys}$ architecture-neutrally (Gate 0), formally splitting physics from governance. Success (AXIOM-CLOSED, realistic): name AXIOM-RULEBOOK-PHYSICS-ONLY; the physics/governance split made formal → the rulebook burden is countable non-circularly. Refuting result: the rulebook's physics burden is shown reducible by a cheaper competitor → Lemma 2 folds. sharper-OPEN: the split is partial; some requirements resist clean classification.

(d) Machinery & inputs. Start from: CONSTRAINT_SET_CLASSIFICATION.md (the $C_{\rm admiss}$ inventory); SHAPE_REALIZATION_DERIVATION_TARGETS.md (Gate 0); the SG-8 dossier (the $F^+$ weakest-link analysis, R1).

(e) Leverage. A prerequisite for any architecture-neutral minimality upgrade (R2/R7). Not a physics gap — a methodology guard — but it gates the rest.

6.7 R1 — absolute irreducibility (the Kolmogorov wall + bottoms-on-E)

(a) Precise statement. "No competitor anywhere, under any conceivable architecture or math, is shorter" is a universal negative equal to $K(T)$, the Kolmogorov complexity of the constant set — uncomputable in principle, for any object in any field. And $T$ presupposes $E$ on both sides, so the strongest reachable claim is "shortest-recipe generator given E," never "geometry from nothing."

(b) Why it's hard. It is not a theorem to be proven — it is a wrong target. The catastrophe is not a math problem; it is a misframed demand (the Einstein-audit move). Trap: presenting any bounded result as the absolute claim is the over-promotion failure the program guards against most carefully.

(c) Exactly what closes it. Nothing closes it as stated — and that is the point. The closure is a reframe: declare the finite role-mechanism grammar $\mathcal{G}$ explicitly, which converts the uncomputable universal negative into the bounded R2, and state minimality as $\mathcal{G}$-relative + living/extensible. Honest endpoint (sharper-OPEN, essentially standing): name AXIOM-GRAMMAR-RELATIVE-FORCING — "minimality is asserted only relative to the declared finite grammar $\mathcal{G}$; a competitor using a mechanism not in $\mathcal{G}$ forces $\mathcal{G}$ to extend, not the claim to refute." This retires the wrong target and names the right one (R2). Refuting "result": a competitor outside $\mathcal{G}$ that is shorter → $\mathcal{G}$ extends (the certificate works as designed), not a refutation.

(d) Machinery & inputs. Framing only; no new math. Start from: T_SHAPE_ABSOLUTE_IRREDUCIBILITY_FORK_THEOREM.md (the fork; CERTIFIED-SCOPED ≠ CERTIFIED-ABSOLUTE); SHAPE_FINAL_STATUS_SPLIT_RESULT.md (the uncomputability statement).

(e) Leverage. R1 is defined down into R2 — strictly easier — and bounds the honesty of every minimality statement SG-1 makes. given-E ≠ derivation of E binds here permanently.

6.8 R3 / R6 / R9 — the mechanical / honesty residuals (strictly easier)

• R3 (the ~9–10 injected reals). Not physics — an honesty/claim-boundary residual. Endpoint DISCLOSED-CORRECTED: state "~22 outputs from ~13–14 effective inputs (4 anchors + 9–10 injected reals)" wherever "4→22" appears; retire both wrong figures (the ~18-economy overclaim and the ~1.6× over-correction; the latter is forbidden by the family-level-normalization ban, SG-8 I.4). Owner countersign only. Named principle: AXIOM-HONEST-CHARGED-COST.
• R6 (the reproducer). VERIFIED (machine, target-blind; §5.2) — the cheapest, highest-value-per-effort leg, now done. A re-runner confirms both hashes regenerate byte-equal and A1 recomputes. Fail-closed if any hash fails. Self-witness only — closes no physics.
• R9 ($\mathcal{M}_4$ as declared axiom). Endpoint DISCLOSED-CONSISTENT: keep $\mathcal{M}_4$ labelled an observational primitive (AXIOM-M4-OBSERVATIONAL-PRIMITIVE); never count it as a forced rung or cite it as a derivation. Labeling sweep only — honest by design.

6.9 Attack-order summary

1. R5 (metric-selection) — single theorem, flips the whole ladder; gates R2 and resolves R4's fairness.
2. R2 (lower-bound matrix) — the bounded "whole remaining job"; conditional on R5.
3. R4 (N.4 shelf completeness) — bounded coset-classification; hardens the historically-exposed color rung.
4. R8 (actor-minimality) — narrowest layer; if it fails, realization-minimality fails fast.
5. R7 / R10 (functional-role floor / rulebook F6 guard) — category-relative bridges, prerequisites for any architecture-neutral upgrade.
6. R1 (absolute irreducibility) — reframe, not closure (sharper-OPEN).
7. R3 / R6 / R9 — honesty + reproducibility + labeling (mechanical; R6 done).

REDUCE-vs-RELOCATE check. The plan does not turn one hard problem into three harder ones. R1 is defined down into R2. R4 (color) is a genuine REDUCE — abelian-isotropy uniqueness replaced an asymmetric hand-wave with a stronger, smaller theorem. R2 and R5 are each a single bounded theorem (a classification; a metric derivation), both carrying explicit κ³/π + G1/G2/G5 guards so a target-fitted or reverse-engineered win cannot be banked. R3/R6/R9 are mechanical. No DERIVED-CLOSED is promised on the forcedness question. A full campaign moves the gate from DECLARED-FROZEN with asserted reproducibility to DECLARED-FROZEN with machine-verified reproducibility + a named axiom floor + an architecture-neutral color-carrier theorem — a real honesty/reproducibility/forcedness gain, not a promotion. A REFUTED-ECONOMY (the simpler 4D realization wins) remains a live, honest, equally valuable outcome.

7. Honest ceiling & scope

What SG-1 is. A freeze-and-reproduce certificate for a fully specified, layer-complete, reproducible three-layer object — $\mathfrak{B}_{\rm active}$ frozen as content hash dcc66f1b2685 / meta-hash a5b1e6f9d951, with no-layer-smuggling (B2), term-level load-bearing (C1–C10), and an R0 reproducer that regenerates every hash (now machine-verified target-blind). The weak ($S^2$, F1) and hypercharge ($S^1_Y/\mathbb{Z}_2$, F2) carriers are forced within the grammar; the color carrier is clean by an architecture-neutral theorem (abelian-isotropy uniqueness), and the one cheaper competitor ($CP^2$) was built end-to-end and breaks at Gate 2.

What SG-1 is not — stated flat out. - It is not a derivation. The gate declares and freezes an object; it derives no physics. selection ≠ derivation; declared + frozen ≠ derived. - It is not a uniqueness theorem. Minimality is selector-minimal, category-relative only (inside R2.5: forces = isometries, compact, classified structures, admissible bundles). category-relative ≠ absolute. - Absolute irreducibility is OPEN and uncomputable — the universal negative "no competitor anywhere is shorter" equals $K(T)$ and is unprovable for everyone. This is a hard limit on all of physics, not a gap in ours. The bounded claim (grammar-relative forcing) is the ceiling, not a hedge. - It bottoms on E. The target ledger $T$ has the SM chiral content $E$ on both sides; $E$ cancels. The strongest reachable claim is "shortest-recipe generator of the flavor/coupling data, given E." given-E ≠ derivation of E. $E$ itself stays un-forced (the corpus attempt to force it was REFUTED). - The dimension-ladder lower-bound matrix is a survey, not a certified classification (Certificate 3 OPEN). - The search category and the SU(3)-shelf completeness (N.4) are the corpus's own intended first review targets. - The MDL metric under which 13D wins is itself unproven against the dimension-first alternative — the decisive seam (R5). The first-pass plug pinned the residual to one named axiom (AXIOM-COMMON-CURRENCY / the granularity-MDL bridge); reject it and the metric fork reopens and SHAPE folds for everyone.

The anchors paid (charged in the open). SG-1 charges the four headline anchors $\{M_{\rm Pl},\ \alpha_i(M_Z),\ y_t,\ |V_{us}|\}$ plus ~9–10 injected reals ($N_d, N_e, N_\nu$; the $\delta$ triple; $\theta_H^\star$) — ~13–14 measured reals total, not the retired "4-in" headline. It witnesses itself on the frozen-object hashes (the DERIVED reproducibility leg, a self-witness that closes no physics). The forcedness question additionally depends on the un-anchored GRANULARITY root (R5) and bottoms permanently on spectrum-E. It anchors on no new measured number — which is precisely why the forcedness claim is selected, not forced.

The single sentence. SG-1 freezes a fully specified, layer-complete, reproducible three-layer object and certifies it is the selector-minimal survivor inside the declared category — it does not prove the object is unique, forced, or derived, and the honest input cost is ~13–14 measured reals, not 4.

Dissolved unicorns — shared ceilings, never claimed as proven, never listed as weakness: - Absolute irreducibility (no competitor anywhere under any math is shorter) = $K(T)$, uncomputable for everyone — a limit on all knowledge. - "The unique geometry nature could have used" — an absolute-uniqueness claim over all possible geometries; no framework in physics can prove it, so the bounded claim (selector-minimal survivor inside the declared category) is the ceiling. - "No future theory could do better" — an open-ended universal negative over all future theories; unprovable in principle and correctly never claimed.

Binding close. No status was ever upgraded. Frozen branch dcc66f1b2685 / a5b1e6f9d951 READ-ONLY. given-E ≠ derivation of E; selection ≠ derivation; dissolved ≠ solved; AXIOM-CLOSED ≠ proven; category-relative ≠ absolute. Nothing applied, nothing deployed. SG-1 remains DECLARED-FROZEN — a serious candidate, frozen and reproducible, not yet proven unique.

Full dossier built from the SG-1 handoff, the existing closure-attack article, the SHAPE realization packet (final-status split, dimension-ladder MDL audit certificate, color-rung CP² refutation, granularity-metric target), the SG-1 completion result, and the first-pass plug. Common material referenced to the published manuscript (GUT.html / TOE.html), not duplicated. Our geometry (the frozen 13D K₆ branch) only.