# SG-7 — Threshold Unification (GUT Gate 7): Per-Gate Closure-Attack Dossier

> **🔻 POST-PUBLICATION UPDATE (2026-06-24) — Gate-7 DOWNGRADED to *Diagnostic only*.** A blind
> reviewer-reproduction (independently verified) found the G.3.2 / G.3.2a threshold ledger rows are
> **fitted-to-target, not reproduced by the cited heat-kernel formula** — the unambiguous $S^2$ gauge row
> is ~18× off, and a convention-free ratio test fails on both compact factors — so the strong claim "no row
> is a free parameter" is **withdrawn**. The binding label below is updated **PARTIAL → SCHEME-ANCHORED /
> Diagnostic-only**: the *signs* of the threshold corrections are geometric, the *magnitudes* are not
> derived; unification is **reopened-as-diagnostic, not refuted**. Live-corrected in GUT.md (Gate-7, G.3.2a)
> and consolidated in the Gaps & Walls Register. The body below is the original PARTIAL-era analysis,
> preserved verbatim — read it through this downgrade.

> **What this is.** The closure-attack packet for **SG-7 (Threshold unification / the heat-kernel $\delta$-triple
> to $M_U$)** on the **frozen 13D K₆ branch ONLY**. It recaps how our geometry closes coupling unification
> (concise; the full derivation lives on the published site), verifies where the status actually stands, names
> every open residual, and lays out the attack plan to push the gate further. It is **not** a rival comparison
> (that is `BATTLE_GATES/`), **not** the one-line status ledger, **not** a reprint of the manuscript.
>
> **Binding discipline (carry verbatim).** No status was ever upgraded. Frozen branch `dcc66f1b2685` / `a5b1e6f9d951`
> READ-ONLY. Honest throughout: the three $\delta$ are **injected reals** quoted at five sig figs, NOT
> regenerated from the named invariants; the **G.3.2a row formulas do NOT reproduce the $\delta$ decimals**
> (off by **~2× to ~18×**, recomputed in §2.3 of this dossier); the load-bearing executable proof —
> `reproduce_all.py` + `appendix_F_heat_kernel_ledger.csv` / `appendix_F_threshold_outputs.csv` with their
> VERIFY row — is **referenced throughout but ABSENT** from this audit (and the captured terminal output in
> R0.4.1 verifies only the **trivial re-sum of the printed rows**, not the deep regeneration of the rows from
> invariants); so **"no row is a free parameter" is a declared-not-proven assertion** (the manuscript's own
> G.9.6 referee-facing caveat says exactly this); the **SM beta coefficients $b_i$ and the comparison/
> unification scales $M_Z, M_U$ are inputs / declared-target conventions**, not predictions. This is a
> consistency / pruning pass evaluated GIVEN the selected survivor, NOT a determination that the spectrum is
> unique. Closure paths must be **non-target-loaded** (the κ³/π kill-test: a proposed axiom counts only if it
> would be written WITHOUT knowing the target $\delta$). **given-E ≠ derivation of E.** AXIOM-CLOSED ≠ proven;
> selection ≠ derivation; dissolved ≠ solved; *a captured terminal log ≠ an independent reproduction.*

---

## 0. Header & Verdict

| Field | Value |
|---|---|
| **Gate id** | SG-7 — Threshold unification (the heat-kernel $\delta$-triple to $M_U$) |
| **GUT manuscript gate** | Gate 7 (§6.7; narrative module §5.6; certificate `certificates/G07_thresholds/`) |
| **Status label (binding)** | **SCHEME-ANCHORED / Diagnostic-only** (downgraded 2026-06-24 from PARTIAL — see the update note at top; rows fitted-to-target, not formula-derived) |
| **Target anchor(s)** | **No measured invariant** — Diagnostic-only, rows fitted-to-target. Signs terminate on **E** + measured **$\alpha_i(M_Z)$** (geometric); magnitudes terminate on **one heat-kernel SCHEME-ANCHOR** — the single geometry-unfixed Seeley–DeWitt finite object (KK zeta-normalization $Z_{\mathcal X}$ + overlap $c_{a,b,i}$, shared with $c_{\rm loop}$/$a_6$). Status: **SCHEME-ANCHORED / OPEN**. See §A "🎯 Target anchor(s) for this gate." |
| **Manuscript card status** | *Claimed certificate pass under frozen scheme, regulator, spectrum, and comparison scale* (the manuscript's own conditional phrasing in §6.7 / G.4 / G.11; PARTIAL is the honest scoped-GUT roll-up of it, downgrading on the AUDIT residual) |
| **Frozen hashes it rides** | Branch `dcc66f1b2685` / manifest meta `a5b1e6f9d951`. Threshold-sector primitives: radii $R_{K_6}$ `634438ce0776`, $R_{S^2}$ `2381d472c62e`, $R_Y$ `0e8b8dba2cf0`; RG transport (two-loop $\overline{\rm MS}$) `f531205a9159`; comparison scale $M_Z$ `a6852c7a6b00`; uncertainty rule `61b0d93507e7`; declared inputs $M_{\rm Pl}$ `df5976a365c3`, $\alpha_i^{-1}(M_Z)$ `6a3b6ef06697`; orbifold action $\mathbb{Z}_2$ on $S^1_Y$ `ac4d2df3e708`; $\mathbb{Z}_6$ identification `a68ee92a75be`; Wilson-line cycle $\gamma$ `640e1d7f7773`, winding $n_H$ `f65094fd8fd1`; $\eta_{BK}$ `84e94518d3f5`. Bundle/CSV file-hashes (REFERENCED, not re-run here): `reproduce_all.py` `30d4d3049051`; `appendix_F_threshold_outputs.csv` `a9b61c5f8049`; `appendix_F_heat_kernel_ledger.csv` `9a6c7c08dc3c`. **$\delta$-triple: $(+4.8424,-3.1112,-1.7313)\pm1.6\times10^{-3}$ — no per-row first-principles hash; injected reals.** |

### 0.1 Abstract — established / open / what would close it

**Established (given-E, given the selected & frozen survivor).** Gate 7 is the program's "not quite" gate: the
three measured couplings $\alpha_1,\alpha_2,\alpha_3$, run up from $M_Z$ under two-loop $\overline{\rm MS}$ SM
running, *almost* meet — and the survivor's own heavy KK spectrum on $K_{\rm gauge}=K_6\times S^2\times
S^1_Y/\mathbb{Z}_2$ must supply *exactly* the finite corrections that close the gap, with nothing adjustable.
The witness is the **8-row G.3.2 heat-kernel ledger** (matter packets, gauge+ghost nets, hypercharge zero-mode +
packet, Higgs Wilson-line at $n_H=1$, orbifold-boundary term at $\theta\in\{0,\pi\}$) whose three **column sums**
are the frozen **threshold vector** $(\delta_1,\delta_2,\delta_3)=(+4.8424,-3.1112,-1.7313)\pm1.6\times10^{-3}$.
Inserted into the RG transport, these drive inverse-coupling equality at $M_U\sim10^{16}$ GeV with a residual at
the numerical-pipeline floor ($9.6\times10^{-11}$, far inside the propagated PDG band $\sim10^{-3}$). The gate
carries genuine **discipline machinery**: the G.10 No-Hidden-Knob audit (10-row knob table), the G.5 fail-closed
rules, the G.9 certificate checklist, and three structural cross-checks (gauge–ghost identity per factor;
family-count load-bearing-ness; Wilson-line load-bearing-ness). **The genuine, defensible content is the
qualitative-structural claim**: that finite KK thresholds of *this* frozen spectrum, computed under a frozen
regulator, are of the right sign and magnitude to close the three couplings, with the family count $-3$ and the
Higgs winding $n_H=1$ both demonstrably load-bearing.

**Open (the honest deductions).** (1) The three $\delta$ are **injected reals** — quoted to five sig figs (G.3 /
A1.11.3) but **not regenerated** from the named invariants anywhere in the manuscript. (2) The **G.3.2a row
formulas do NOT reproduce the $\delta$ decimals**: applied literally, the per-row gauge/ghost/matter
coefficient formulas miss the printed rows by **~2× to ~18×** (recomputed in §2.3) — so the chain
"invariant → row → column-sum" is **broken at the middle link**. (3) The **load-bearing executable proof is
ABSENT** from this audit: `reproduce_all.py` + the two CSVs are *referenced* (and a captured terminal log is
*pasted* in R0.4.1 / R0.7.2), but no independent re-run was performed here — and, decisively, the captured log
verifies only that the **printed rows re-sum to the printed $\delta$** (a tautology), NOT that the rows
regenerate from invariants. (4) **"No row is a free parameter" is declared-not-proven** — the manuscript's own
G.9.6 referee-facing note concedes this is "a freeze-and-derive claim, not an independent per-row proof of
forcedness." (5) The **SM beta coefficients $b_i$ and the scales $M_Z, M_U$ are inputs / declared-target
conventions**, not predictions; $M_U$ is explicitly "a declared closure-target convention (not a 16-sig-fig
prediction)" in A1.10. (6) The certificate is **given-the-selected-survivor**: a consistency/pruning pass, not a
proof that this spectrum is the unique one that unifies. (7) By the manuscript's own §6.12 falsification map, an
unreproduced ledger or a hidden knob downgrades the gate from *Claimed certificate pass* to *Diagnostic only* —
which is exactly why the honest roll-up is **PARTIAL**, with the reproduction-harness status **AUDIT/OPEN**.

**What would close it (the ladder).** The CHEAPEST genuine win is **machine-real reproduction**: mount
`reproduce_all.py` + the two CSVs and run the **deep** check — regenerate each of the 8 rows from the G.3.2a
invariants (Dynkin indices, Euler characteristics, $\sum Y^2$, $n_H$, the Seeley–DeWitt integrals) target-blind,
and confirm they sum to the $\delta$ — not merely re-sum the printed rows. The second win is the **G.3.2a
row-formula reconciliation**: re-derive the $\delta$ decimals from the named invariants so the ~2×–18× row
mismatch closes (today the formulas as written are under-specified — missing normalization / overlap /
projector-weight factors). DERIVED-CLOSED would require the rows to fall out of the invariants with no
per-row fudge; the realistic ceiling per residual is **AXIOM-CLOSED** (name the missing structural
normalization as one target-blind posit) or, more likely on the harness leg, **VERIFIED** (machine-real) or
**sharper-OPEN**.

### 0.2 Website source-of-truth (link, don't duplicate)

The full construction, the worked KK-threshold derivation, the No-Hidden-Knob audit, and the freeze records are
the **published manuscript**, Paper I:

- **GUT.html §6.7** (Gate-7 card) — <https://physics.magflowmeters.com/articles/GUT.html>
- **§5.6** narrative module; **Appendix G** (G.0–G.11: inputs, RG transport, the G.3.2 ledger + G.3.2a
  coefficient table, output table, fail-closed rules, certificate JSON, G.9 checklist, G.10 No-Hidden-Knob
  audit, G.11 authority summary); **Appendix A1.10–A1.11** (scale constants + threshold constants — the
  per-factor packet table A1.11.2 and the boxed $\delta$); **Appendix R0** (R0.4.1 captured `reproduce_all.py`
  output; R0.6 fail-closed rules; the bundle-file hash table). Same public article.
- Higgs-protection interface (does not own this gate but reads its comparison scale): **§6.8 / Appendix H**.
  Downstream Λ cross-reference (does not touch this gate): Paper IV, TOE.html —
  <https://physics.magflowmeters.com/articles/TOE.html>.

This dossier recaps only what is needed to attack; the site controls all common material.

---

## 1. How OUR geometry closes threshold unification — the closure claim (concise)

> Full derivation: **GUT.html §6.7 + §5.6 + Appendix G (G.1–G.4) + A1.11**. This section is the attack-grade
> recap, not the derivation.

### 1.1 The mechanism end-to-end

Gate 7 is a **consistency / pruning gate evaluated on the survivor** — no new geometry is searched. The pipeline
is one forward chain (§5.6, G.3):

```
 α_i⁻¹(M_Z) [3 measured]  --two-loop MS-bar RG-->  α_i⁻¹ near M_U  (almost meet)
        +  KK spectrum of K_gauge  --heat-kernel a_2 ledger (G.3.2)-->  δ = (δ_1,δ_2,δ_3)
        --insert into RG-->  α_1(M_U)=α_2(M_U)=α_3(M_U)  at  M_U ~ 1e16 GeV,  residual 9.6e-11
```

The load-bearing factors, named so a reader can attack each:

1. **The three running couplings + the SM beta functions.** $b_1^{\rm SM}=41/10$, $b_2^{\rm SM}=-19/6$,
   $b_3^{\rm SM}=-7$ (GUT-normalized $\alpha_1=(5/3)\alpha_Y$), *fixed by SM content* (3 generations + 1 Higgs +
   gauge sector). RG transport is frozen at two-loop $\overline{\rm MS}$ (`f531205a9159`). (Attack handle: the
   $b_i$ are *inputs* — SG-7 does not derive the SM content; it inherits E.)
2. **The KK spectrum on each compact factor** (G.3.1 Step 2). $K_6$ carries $SU(3)_c$ with Laplacian tower
   $m_n^2=n(n+2)/R_{K_6}^2$; $S^2$ carries $SU(2)_L$ with $m_n^2=n(n+1)/R_{S^2}^2$; $S^1_Y/\mathbb{Z}_2$ carries
   $U(1)_Y$ with orbifold-projected tower. The chiral families are zero modes (no KK chiral copies above $m_c$),
   so $n_q^{\rm KK}=0$ for matter but the gauge KK towers contribute fully. (Attack handle: the spectrum is
   *frozen at the chamber center* $\vec u=(1,1,1)$ — it inherits SG-6's "moduli read from a witness".)
3. **The heat-kernel finite remainder $\Delta_i^{\rm finite}$** (G.3.1 Step 3). With $m_c=M_U$ the logarithmic
   KK-running term **vanishes** and only the finite Seeley–DeWitt $a_2$ remainder survives:
   $$\delta_i=\frac{1}{2\pi}\,\Delta_i^{\rm finite}\!\left(K_6,S^2,S^1_Y/\mathbb{Z}_2,\text{Wilson-line}\right).$$
   This is the structural claim: thresholds are a *finite topological number*, not a tuned log. (Attack handle:
   the identification $m_c=M_U$ "up to an $\mathcal{O}(1)$ factor" is the soft seam — the $\mathcal{O}(1)$ is
   exactly where a hidden normalization could live.)
4. **The 8-row G.3.2 ledger.** Each row is a (compact factor × field class) contribution to one column
   ($\Delta_1,\Delta_2,\Delta_3$); the **three column sums** are the $\delta$-triple. The genuine prediction is
   *qualitative-structural*: (a) the gauge+ghost net is negative (asymptotic-freedom sign); (b) matter packets
   are positive; (c) the hypercharge zero-mode matter ($\sum Y^2=10/3$ per gen $\times$ 3) and the
   orbifold-fixed-point term drive $\delta_1>0$; (d) family count $-3$ and Higgs winding $n_H=1$ are
   *load-bearing* (cross-checks 2 and 3 in G.3.2 — flip either and the columns miss).
5. **The discipline layer.** G.10 No-Hidden-Knob audit (every knob frozen + listed); G.5 fail-closed rules;
   G.9 certificate checklist (10 rows, all "yes"); G.9.8 sensitivity analysis (degrades smoothly under input
   shifts, *catastrophically* under chamber-modulus shifts — the structural signature of a non-tuned packet).

### 1.2 What "closed" means here, and its conditionality

"Closed" for SG-7 is **a frozen-spectrum consistency pass under declared scheme/regulator/comparison scale** —
strictly **not** a prediction of $M_U$, **not** a derivation of the SM beta functions, **not** a proof that this
KK spectrum is the unique unifier. It is conditional on:

- **given-E** — the SM chiral content (hence the $b_i$) and the family index $-3$ supplied from SG-2/SG-3; SG-7
  does not derive E.
- **given-the-selected-geometry** — the frozen 13D K₆ branch and the chamber-center moduli (which fix the KK
  radii); the spectrum is read at the SG-6 witness, not independently derived.
- **given-the-declared-conventions** — $M_Z$ (comparison scale, frozen), $M_U$ (declared closure-target
  convention, *not* a 16-sig-fig prediction; A1.10), the two-loop $\overline{\rm MS}$ scheme, and the
  heat-kernel/proper-time regulator. Change any after comparison and the certificate voids (G.5).

**The genuine, defensible content** (the part that survives the honest accounting):

| Genuine output (frozen-spectrum structural claim) | Mechanism |
|---|---|
| Finite KK thresholds of the right **sign** per column | gauge+ghost (−) vs matter (+) heat-kernel signs |
| **Family count $-3$ is load-bearing** for the columns | G.3.2 cross-check 2 — scale the matter row by $n_{\rm gen}$, columns miss |
| **Higgs $n_H=1$ is load-bearing** for $\delta_1$ | G.3.2 cross-check 3 — $n_H=0$ misses by $(+1.047,-0.211,0)$ |
| Couplings cross at a single $M_U\sim10^{16}$ GeV | three-coupling equality under RG + the $\delta$ |
| Catastrophic sensitivity to chamber-modulus shift | G.9.8 — the structural (non-tuned) signature |

**The conditional / non-genuine content** (input or unreproduced, per the honest audit):

| Quantity | Honest reality |
|---|---|
| The exact $\delta$ decimals $(+4.8424,-3.1112,-1.7313)$ | **injected reals** — quoted, not regenerated from invariants (G.3.2a fails to reproduce them; §2.3) |
| $M_U=1.0\times10^{16}$ GeV | **declared closure-target convention** (A1.10), not a prediction |
| $b_1,b_2,b_3$ | **SM inputs** (fixed by E, which SG-7 imports) |
| $M_Z$, scheme, regulator | **declared conventions**, frozen |
| "no row is a free parameter" | **declared-not-proven** (G.9.6 own caveat) |

So the precise statement of closure: **the survivor's frozen KK spectrum supplies finite thresholds of the
correct structural sign/magnitude to close the three couplings at $M_U$, with family count and Higgs winding
demonstrably load-bearing — but the specific $\delta$ decimals are injected, the row-by-row derivation does not
reproduce them, and the executable proof that would make the whole chain machine-real is absent from this
audit.** That gap is the entire reason SG-7 is PARTIAL and not a claimed pass.

---

## 2. Verify the status — is PARTIAL real?

This is the verification a skeptic would run. Each witness gets a grade — **hand-checkable** / **symbolic** /
**machine-lane** — and an honest **reproduces?** flag.

### 2.1 The witness ledger

| # | Witness | What it asserts | Grade | Reproduces? |
|---|---|---|---|---|
| W1 | **SM beta functions** $b=(41/10,-19/6,-7)$ | fixed by SM content, not free | hand-checkable | **Yes** — standard, recompute by hand from the SM spectrum |
| W2 | **Two couplings almost meet under SM running** | the "not quite" the gate exists to close | machine-lane | **Yes in principle** (standard two-loop run); not re-run here |
| W3 | **8-row G.3.2 ledger column sums = $\delta$** | the printed rows sum to $(+4.8424,-3.1112,-1.7313)$ | hand-checkable | **Yes (trivially)** — summing the printed rows reproduces the printed $\delta$ to 4 dp (verified by hand, §2.3) |
| W4 | **G.3.2a row formulas → the row values** | each row is a deterministic function of named invariants | symbolic | **NO** — literal formulas miss the printed rows by **~2×–18×** (§2.3); the middle link is broken |
| W5 | **$\delta$ inserted → unification at $M_U$, residual $9.6\times10^{-11}$** | three couplings cross within the PDG band | machine-lane | **AUDIT** — asserted from `appendix_F_threshold_outputs.csv` row `residual_at_MU`; not independently re-run here |
| W6 | **Family-count load-bearing** (cross-check 2) | $-3$ is required; $-2/-4$ misses | hand-checkable | **Plausible but not re-run** — the row scaling is asserted, the recomputation was not performed here |
| W7 | **Higgs $n_H=1$ load-bearing** (cross-check 3) | $n_H=0$ misses $\delta$ by $(+1.047,-0.211,0)$ | hand-checkable | **Yes (consistent)** — row 7 equals exactly that miss; internally consistent |
| W8 | **G.10 No-Hidden-Knob audit** (10-row knob table) | every knob frozen + listed before comparison | symbolic/audit | **Conditional** — an auditable *claim*, not a theorem; a reviewer must run G.10.3 to falsify |
| W9 | **`reproduce_all.py` captured terminal output** (R0.4.1) | column sums + 33 hashes + meta-hash all match | machine-lane | **AUDIT** — a *pasted log*, not an independent run; and it verifies only the **trivial re-sum** (W3), not W4 |
| W10 | **Freeze hashes** + meta `a5b1e6f9d951` | every threshold object content-addressed before comparison | machine-lane | **Yes in principle** (re-hash the radii/RG/scale rows + recompute meta-hash); not re-run here |

### 2.2 What reproduces, plainly

- **The beta functions reproduce by hand.** $b=(41/10,-19/6,-7)$ is the standard SM result; not adjustable.
- **The printed-row column sums reproduce trivially.** Summing the 8 printed rows gives exactly
  $(+4.8424,-3.1112,-1.7313)$ to four decimals (hand-checked in §2.3). **But this is a tautology**: it checks
  that the *injected rows* add up, not that the rows come from anywhere.
- **Three internal-consistency checks hold.** Row 7 (Higgs) equals exactly the $n_H=0$ miss the cross-check
  asserts; the gauge–ghost identity $c^{\rm gauge}+c^{\rm ghost}=c^{\rm gauge}/2$ is algebraically structural;
  the catastrophic-under-modulus-shift sensitivity signature is the right shape for a non-tuned packet. These
  make the *structure* credible even where the *numbers* are injected.

### 2.3 What does NOT reproduce (honest gaps in the status) — the load-bearing arithmetic

This is the decisive verification. Two checks were run by hand for this dossier.

**(a) The trivial check (passes).** The printed G.3.2 / A1.11.2 rows:
$\delta_1$ column $\{-0.84,+3.2140,+1.0470,+1.4214\}\!\to\!+4.8424$; $\delta_2$ column
$\{-4.02,+0.92,-0.2110,+0.1998\}\!\to\!-3.1112$; $\delta_3$ column $\{-2.49,+0.79,-0.0313\}\!\to\!-1.7313$. **All
three reproduce to four decimals.** This is what the captured `reproduce_all.py` log (R0.4.1, R0.7.2) verifies —
and *all* it verifies.

**(b) The deep check (FAILS — off ~2×–18×).** Applying the G.3.2a coefficient formulas **literally** to
regenerate the rows from invariants:

- $c^{\rm gauge}+c^{\rm ghost}=-(1/3)\,T_{\rm adj}/(24\pi)\cdot\!\int R\sqrt g$, with
  $\int_{K_6}R\sqrt g=12\pi^3$ and $\int_{S^2}R\sqrt g=8\pi$ (both from G.3.2a).
- $c^{\rm matter}=+(1/3)\,T(R)/(24\pi)\cdot\!\int R\sqrt g\cdot n_{\rm gen}$.

| Row | Printed value | Literal-formula value | printed / formula |
|---|---:|---:|---:|
| 1 — $SU(3)$ gauge+ghost net ($\to\delta_3$) | $-2.4900$ | $-4.9348$ | **0.505** (~2× off) |
| 2 — quark matter on $K_6$ ($\to\delta_3$) | $+0.7900$ | $+2.4674$ | **0.320** (~3× off) |
| 3 — $SU(2)$ gauge+ghost net ($\to\delta_2$) | $-4.0200$ | $-0.2222$ | **18.090** (~18× off) |
| 4 — doublet matter on $S^2$ ($\to\delta_2$) | $+0.9200$ | $+0.1667$ | **5.520** (~5.5× off) |

This **confirms the SG-7 status-line claim ("off ~2×–18×") quantitatively.** The chain
*invariant → row → column-sum* is broken at the **middle link**: the column sums are right because the rows are
*injected to make them right*, but the rows themselves do not fall out of the stated formulas. The literal
formulas are evidently **under-specified** — they omit at least the per-factor normalization that converts the
Seeley–DeWitt $a_2$ on each factor into the running-coupling correction (the "$\mathcal{O}(1)$ factor" in the
$m_c=M_U$ identification of G.3.1 Step 3, and the projector/overlap weights $c_{a,b,i}$ named-but-not-valued in
G.3.2). **Until those weights are written down and shown to be forced, "each row is a deterministic function of
the invariants" is unverified, and the deep check fails.**

**(c) A minor internal inconsistency to flag.** G.3.2a states $\sum_f Y_f^2=10/3$ *per generation*, but the
G.3.2 row-6 "index" column reads "$\sum Y^2=10$" (i.e. $\times 3$ generations) — the two are reconcilable
(per-gen vs all-gen) but the labels are inconsistent as printed. A reviewer running G.10.3 step 1 will hit this.

**(d) The harness is ABSENT from this audit.** `reproduce_all.py`, `appendix_F_heat_kernel_ledger.csv`, and
`appendix_F_threshold_outputs.csv` are *referenced* (with file-hashes `30d4d3049051` / `9a6c7c08dc3c` /
`a9b61c5f8049`) and a terminal log is *pasted* — but no independent re-run was performed, and the pasted log
verifies only check (a). So the load-bearing executable proof of the gate is **declared, not independently
verified here.**

### 2.4 Why PARTIAL (not higher, not lower)

- **Not DERIVED-GIVEN-E** (the SG-2/SG-3 tier): SG-7 is not a rigid integer/representation recovery. The
  $\delta$ are injected, the row derivation fails the deep check, $M_U$ is a declared convention, and the harness
  is AUDIT — too many open residuals for the DERIVED tier.
- **Not OPEN/DECLARED-FROZEN:** the structural content is real and non-trivial — finite KK thresholds of the
  right sign, family count and Higgs winding load-bearing, a genuine discipline layer (G.5/G.9/G.10), and a
  near-perfect closure of the couplings. The gate is well clear of "nothing claimed."
- **PARTIAL is exactly right:** genuine structural unification claim, injected $\delta$ decimals, a broken
  row-derivation link, an AUDIT harness, and a downgrade rule with teeth (§6.12 row 7 → *Diagnostic only* on an
  unreproduced ledger / hidden knob). This matches the SCOPED_GUT ledger SG-7 line verbatim.

---

## 3. The attack surface — what to attack (ranked by leverage)

Every open residual as a named object with its precise obstruction. **Leverage** = how much closing it moves
the gate toward a machine-real, derivation-grade pass.

### 3.1 The residual register

| ID | Residual (named object) | Precise obstruction | Status | Leverage |
|---|---|---|---|---|
| **R1** | **Absent reproduction harness** (`reproduce_all.py` + the two CSVs) | The load-bearing executable proof is referenced + a terminal log pasted, but not independently re-run here; and the pasted log verifies only the trivial re-sum (W3/W9), not the deep regeneration (W4). By §6.12 row 7 an unreproduced ledger downgrades the gate to *Diagnostic only*. | AUDIT | **HIGHEST** — it is the cheapest win AND it gates the gate's headline status |
| **R2** | **G.3.2a row formulas do NOT reproduce the $\delta$ decimals** | Literal formulas miss the printed rows by ~2×–18× (§2.3); the chain invariant→row is broken at the middle link. "Each row is a deterministic function of the invariants" is therefore unverified. | OPEN (quantified) | **HIGH** — this is the difference between "injected" and "derived" |
| **R3** | **The three $\delta$ are injected reals** | Quoted to five sig figs (G.3 / A1.11.3) but never regenerated from first principles; no per-row first-principles hash. The whole numerical claim rests on three hand-entered numbers. | OPEN | **HIGH** — dependent on R2; closing R2 dissolves R3 |
| **R4** | **"No row is a free parameter" declared-not-proven** | The G.9.6 own caveat concedes this is "a freeze-and-derive claim, not an independent per-row proof of forcedness." The No-Hidden-Knob discipline (G.10) tests *enumeration*, not *forcedness*. | DISCLOSED (own caveat) | **MEDIUM** — honesty/claim-boundary; the forcedness is the real target (= R2) |
| **R5** | **$M_U$ is a declared closure-target convention, not a prediction** | A1.10 labels it "declared closure-target convention (not a 16-sig-fig prediction)." The "$m_c=M_U$ up to an $\mathcal{O}(1)$ factor" seam (G.3.1 Step 3) hides where a normalization could live. | DISCLOSED | **MEDIUM** — the $\mathcal{O}(1)$ seam is exactly where R2's missing weight may hide |
| **R6** | **SM $b_i$ + comparison scale are inputs (given-E)** | The beta functions and $M_Z$ are SM/PDG inputs; SG-7 inherits E from SG-2/SG-3 and does not derive it. | DISCLOSED (given-E) | **LOW** — structural, shared by every gate; not closeable within SG-7 |
| **R7** | **KK radii / spectrum frozen at the SG-6 chamber-center witness** | The KK tower masses are fixed by the moduli read at the SG-6 Weyl-rigid witness $\vec u=(1,1,1)$, not independently derived; SG-6's "read from a witness" soft spot propagates here. | OPEN (inherited from SG-6) | **LOW-MEDIUM** — shared with SG-6; not unique to thresholds |
| **R8** | **G.3.2a internal label inconsistency** ($\sum Y^2=10/3$ vs $=10$) | Per-gen vs all-gen labels printed inconsistently (G.3.2a vs G.3.2 row-6 index column); reconcilable but a reviewer trips on it. | DISCLOSED (cosmetic) | **LOW** — fix-on-sight; no physics |
| **R9** | **Spectrum uniqueness not claimed (consistency, not determination)** | The certificate is given-the-selected-survivor: a pruning/consistency pass, NOT a proof that this spectrum is the unique unifier. Rivals tie here (SHARED-OPEN on threshold). | OPEN (scoped) | **LOW** — a scope statement, correctly disclosed |

### 3.2 Leverage ranking (attack order)

1. **R1** (absent harness) — the cheapest win AND it converts the gate's headline from asserted to machine-real.
   Owner artifact, no new physics.
2. **R2** (row-formula non-reproduction) — the single physics target; closing it dissolves R3 and R4 together.
3. **R3** (injected $\delta$) — dependent on R2.
4. **R4 / R5 / R8** (forcedness caveat / $M_U$ convention / label fix) — honesty + cosmetic; mostly already
   disclosed.
5. **R6 / R7 / R9** — given-E / inherited-from-SG-6 / scope; structural, not closeable within SG-7.

> **The cardinal honest point.** R1 and R2 are where the gate's credibility lives. **The trap to avoid:** the
> captured terminal log (R0.4.1) is *not* a closure of R1 — it verifies the trivial re-sum (R3-as-tautology),
> not the deep regeneration (R2). A closure path that "mounts the harness" must run the **deep** check
> (invariant → row → sum), or it has closed nothing. And the κ³/π kill-test applies in full to R2: if the
> missing per-row normalization is **reverse-engineered** to make the rows sum to the already-known $\delta$,
> that is the κ³/π-manifest cautionary pattern (true by construction) and R2 RELOCATES rather than closes.

---

## 4. THE ATTACK PLAN — closure paths (the core)

For each residual: the **technique** (closure playbook T1 axiom-floor / T4 no-go / T5 firewall / T6
all-operator-conditional / T7 eliminative / T10 selector / machine-lane owner-artifact), the **named axiom it
could reduce to** (stated so it would be written WITHOUT the target $\delta$ — the κ³/π kill-test), the
**specialist target** (theorem to hand off) or the **owner artifact** (CSV / script / computation) needed, the
**math to attempt**, and the **success ladder** (DERIVED-CLOSED rare → AXIOM-CLOSED likely → VERIFIED /
sharper-OPEN → REFUTED).

---

### 4.1 R1 — The absent reproduction harness (the cheapest win)

**Why first.** This is the highest value-per-effort closeable item in the gate: it converts an *asserted* status
into a *machine-checked* one with no new physics — and it is what §6.12 row 7 and G.9.10 / G.10.3 demand. It is
also the residual most likely to be misclosed (by accepting the pasted log), so the discipline matters.

**Technique: owner artifact (machine-lane), not an axiom.** This is a **fail-closed reproducibility task**, not a
physics closure.

**Owner artifact needed.** (1) Mount `reproduce_all.py` (file-hash `30d4d3049051`),
`appendix_F_heat_kernel_ledger.csv` (`9a6c7c08dc3c`), and `appendix_F_threshold_outputs.csv` (`a9b61c5f8049`).
(2) Run **target-blind** against the frozen inputs (radii R1.2, RG transport R1.7, $M_Z$ R1.7, the three PDG
couplings R1.8). (3) Confirm **two distinct things**, not one:
- **(3a) Trivial check** — the CSV column sums equal $(+4.8424,-3.1112,-1.7313)$ to tolerance $5\times10^{-4}$
  and the VERIFY row reads `True,True,True,True`. *(This is what the pasted log already shows; it is necessary,
  not sufficient.)*
- **(3b) Deep check** — each of the 8 rows is **regenerated inside the script from the G.3.2a invariants**
  (Dynkin indices, Euler characteristics, $\sum Y^2$, $n_H$, the Seeley–DeWitt integrals + the per-row weights),
  not read from a hard-coded constant. If `reproduce_all.py` merely re-sums hard-coded rows, **R1 is NOT closed**
  — it has reproduced the tautology and left R2 fully open.
(4) Re-hash the threshold-sector R1 rows and confirm the meta-hash recomputes to `a5b1e6f9d951`.

**Math to attempt.** None new — execution + verification. **But the audit must distinguish (3a) from (3b).** The
single most important diagnostic: open the script and check whether the 8 row contributions are **computed** from
the G.3.2a primitives or **hard-coded**. The §2.3 deep-check failure predicts they are hard-coded; if so, the
honest disposition is *harness VERIFIED for the sum, R2 OPEN for the derivation.*

**Success ladder.**
- **BLOCKED_INPUTS** until the script + CSVs are mounted.
- **VERIFIED (sum-level)** if (3a) passes on an independent run → "the printed $\delta$ are the column sums of
  the printed ledger" becomes machine-real. This is the realistic, honest outcome and it **upgrades the
  harness status from AUDIT to machine-verified** without touching R2.
- **VERIFIED (derivation-level)** only if (3b) passes — the rows regenerate from invariants. **Unlikely given
  §2.3**; would simultaneously close R2.
- **REFUTED** if a value fails to regenerate at the declared tolerance → Gate 7 downgrades to *Diagnostic only*
  per G.9.10.

**Honest disposition: BLOCKED_INPUTS → VERIFIED (sum-level) is the realistic, cheap win; it does NOT close R2.**
The trap is reporting (3a) as if it closed the gate. The pasted log already shows (3a); the genuine new
information is whether (3b) holds, and §2.3 predicts it does not.

---

### 4.2 R2 — G.3.2a row formulas do not reproduce the $\delta$ decimals (the physics target)

**Why second.** This is the only genuine-physics residual in the gate. Closing it converts the $\delta$ from
**injected** to **derived** and dissolves R3 and R4 in one move. It is also where a tuned closure is most
tempting — so the κ³/π kill-test is load-bearing here.

**The target without loading.** The G.3.2a formulas, as written, must be completed with the **per-row weights
$c_{a,b,i}$** (named in G.3.2 but never valued) and the **factor-normalization** (the "$\mathcal{O}(1)$" in
$m_c=M_U$) so that the chain *invariant → row → column-sum* runs forward and lands on
$(+4.8424,-3.1112,-1.7313)$. The numbers it must reproduce **without a new knob**: rows 1–4 must move from the
literal-formula values $\{-4.9348,+2.4674,-0.2222,+0.1667\}$ to the printed
$\{-2.4900,+0.7900,-4.0200,+0.9200\}$ — i.e. the missing weights must supply exactly the ratios
$\{0.505,0.320,18.09,5.52\}$ from **structure**, not from fitting.

**Technique: T1 (axiom-floor) for the genuine path; T5 (firewall) to guard against relocation.**

**Route A (the missing structural normalization — the one real lead).** The most credible physical origin of the
~2×–18× discrepancies is that the literal G.3.2a formula uses the **bare** Seeley–DeWitt $a_2$ on each factor,
whereas the running-coupling correction needs (i) the **KK-tower zeta-regularized sum** (not the single-mode
$a_2$), (ii) the **per-factor volume / radius normalization** that converts a heat-kernel coefficient into a
$1/2\pi\ln$-type correction, and (iii) the **projector overlap** $c_{a,b,i}$ between the mode's representation
and the gauge factor. The ~18× on the $SU(2)/S^2$ row vs ~2× on the $SU(3)/K_6$ row is *suggestive*: $S^2$ and
$K_6$ have very different volumes and KK degeneracies, so a volume/degeneracy normalization is a natural origin
of a factor that is large for $S^2$ and small for $K_6$.

- **Specialist target:** hand the threshold/heat-kernel specialist: "Compute the **zeta-regularized KK-tower
  sum** of the one-loop gauge-coupling correction on each factor ($K_6$, $S^2$, $S^1_Y/\mathbb{Z}_2$) with the
  declared regulator, **target-blind**, and report the per-row contributions. Do they land on
  $\{-2.49,+0.79,-4.02,+0.92,\dots\}$?"
- **κ³/π kill-test:** PASS only if the normalization/overlap weights are computed from the geometry
  (volumes, degeneracies, overlaps) **before** and **independently of** the printed rows. **Current status:
  SUSPECT-until-shown.** If the weights are instead chosen to make the literal formula hit the printed rows,
  this is the κ³/π-manifest pattern (true by construction) and **Route A RELOCATES.**
- **Math to attempt:** the bounded computation above — recompute the four checkable rows (1–4) from the
  zeta-regularized tower sum + volume normalization, target-blind, and compare to the printed rows. If they
  match, R2 is on a path to DERIVED; if they miss in a *new* direction, the printed rows are themselves wrong.

**Route B (name the normalization as one axiom — the honest fallback).** If the full zeta-sum is intractable to
hand off cleanly, the honest reduction is to **name the missing structural factor as a single posit**:

> **AXIOM-THRESHOLD-NORMALIZATION** (target-blind form): *"the running-coupling correction from each compact
> factor is the bare Seeley–DeWitt $a_2$ coefficient times the factor's KK zeta-regularization normalization
> $Z_{\mathcal X}$ (a function of its volume, curvature integral, and KK degeneracy only) and the
> representation-overlap weight $c_{a,b,i}$ — both fixed by the geometry, independent of any coupling value."*

This is writable with **no** $\delta$ value in sight (it references only geometric data). It PASSES the
kill-test **as a statement**. The test is then whether the named $Z_{\mathcal X}, c_{a,b,i}$ *yield* the printed
rows — which is exactly Route A's computation. **If named but not verified to yield the values, R2 is
AXIOM-CLOSED-pending-verification, not closed.**

**Success ladder for R2.**
- DERIVED-CLOSED: Route A zeta-sum computed target-blind → printed rows fall out → $\delta$ becomes derived;
  R2/R3/R4 all close. **The prize, but gated on the computation landing without a fudge.**
- AXIOM-CLOSED: AXIOM-THRESHOLD-NORMALIZATION named and *verified to yield* the rows → the $\delta$ reduce to
  one named structural normalization. **Best realistic outcome.**
- sharper-OPEN: Route A misses in a new direction → either the printed rows are wrong (→ R1 REFUTES) or the
  normalization is more subtle; R2 stays OPEN with a sharpened obstruction.
- REFUTED: the target-blind zeta-sum gives definite rows that *miss* the printed $\delta$ beyond the band →
  the ledger is wrong; Gate 7 → *Diagnostic only*.

**Honest disposition: OPEN, with one concrete bounded computation (Route A zeta-regularized tower sum) and one
named-axiom fallback (Route B), both target-blind, both falsifiable. Do NOT bank any normalization that was
chosen to hit the printed rows — that is the κ³/π trap.**

---

### 4.3 R3 — The three $\delta$ are injected reals

**The obstruction.** $(+4.8424,-3.1112,-1.7313)$ are hand-entered to five sig figs with no per-row
first-principles hash; the entire numerical unification claim rests on three injected numbers.

**Technique: T6 (all-operator-conditional) — R3 is the dependent of R2.** There is no independent closure of R3:
the $\delta$ become non-injected exactly when R2's row-derivation runs forward. So R3's attack plan **is** R2's
(§4.2).

**Named principle (κ³/π-clean):** **AXIOM-DELTA-IS-COLUMN-SUM** — *"the threshold vector is, by definition, the
column sum of the heat-kernel ledger; it carries no independent degrees of freedom beyond the ledger rows."*
This is writable with no $\delta$ value (it is a definitional statement). It is **already essentially the
manuscript's position** — but it only pays a debt once the *rows* are derived (R2). Until then it relocates the
injection from three numbers to eight rows.

**The honest count.** Today: the $\delta$-triple is **3 injected reals** (or, equivalently, the 8 ledger rows
are injected and the 3 sums are derived-from-them). Closing R2 removes all of them as inputs. **This is the only
place a quantitative input lives in SG-7** (the $b_i$, $M_Z$, $M_U$ being SM/declared, not tunable fits).

**Success ladder.** Tied to R2: DERIVED-CLOSED if R2 derives the rows; otherwise R3 stays OPEN (the $\delta$ are
injected) — honestly labeled. No independent AXIOM-CLOSED that pays a debt.

**Honest disposition: OPEN, fully dependent on R2. The $\delta$ are injected today; the only honest way to
un-inject them is to derive the rows.**

---

### 4.4 R4 — "No row is a free parameter" declared-not-proven

**This is an honesty/claim-boundary residual that points at the physics.** The G.9.6 own caveat already concedes
it: *"'no row is a free parameter' is the manuscript's assertion that each row is forced … a freeze-and-derive
claim, not an independent per-row proof of forcedness."*

**Technique: T5 (no tuning to the known answer firewall) — separate enumeration from forcedness.** The G.10 No-Hidden-Knob
audit is a genuine and useful discipline, but it tests the **wrong predicate** for R4: G.10 verifies that every
knob is *enumerated and frozen before comparison* (no *undeclared* knob). It does **not** verify that the frozen
values are *forced* by the invariants rather than chosen. R2 (§4.2) is the forcedness test; R4 is the honest
acknowledgement that, until R2 closes, "frozen" ≠ "forced."

**Named axiom (κ³/π-clean):** **AXIOM-ROW-FORCEDNESS** — *"each heat-kernel ledger row equals its G.3.2a
invariant times the geometric normalization $Z_{\mathcal X}$ and overlap $c_{a,b,i}$, with no per-row residual
degree of freedom."* This is exactly AXIOM-THRESHOLD-NORMALIZATION (R2) re-stated as a forcedness claim; it
PASSES the kill-test (no $\delta$ value), and it is **the precise content R4 needs** — but it is **unverified**
until R2's computation lands.

**Action (countersign-gated; nothing applied here, No status was ever upgraded).** State plainly, wherever the No-Hidden-Knob
language appears, that the discipline proves **enumeration**, not **forcedness**, and that forcedness is the open
R2 target. The manuscript's G.9.6 already carries this; the residual is "propagate the distinction to the
headline," an owner edit, not a derivation.

**Success ladder.** **DISCLOSED-CORRECTED** (the enumeration-vs-forcedness distinction stated) is the cheapest
honesty win; **AXIOM-CLOSED** only when R2's AXIOM-ROW-FORCEDNESS is *verified to yield* the rows. The endpoint
of R4 is the endpoint of R2.

**Honest disposition: DISCLOSED (own caveat already present); the real closure is R2. Do not let "No-Hidden-Knob
passed" be read as "rows proven forced" — it is not.**

---

### 4.5 R5 — $M_U$ is a declared closure-target convention, not a prediction

**The obstruction.** A1.10 labels $M_U=1.0\times10^{16}$ GeV "declared closure-target convention (not a
16-sig-fig prediction)." The residual is that the gate's headline ("couplings unify at $M_U\sim10^{16}$ GeV")
can be misread as a prediction of $M_U$, and the "$m_c=M_U$ up to an $\mathcal{O}(1)$ factor" seam (G.3.1 Step 3)
is where an undeclared normalization could hide.

**Technique: T2-guarded restatement + T5 firewall on the $\mathcal{O}(1)$ seam.**

**Named principle (κ³/π-clean):** **AXIOM-MU-IS-CROSSING** — *"$M_U$ is defined as the scale at which the three
RG-transported inverse couplings (with the frozen thresholds) coincide; it is an output of the crossing
condition, not an independent input."* This is the honest framing and carries no target value. It is true **iff**
the $\mathcal{O}(1)$ factor in $m_c=M_U$ is itself fixed by geometry (not chosen). **Firewall task:** verify the
$m_c/M_U$ ratio is computed from the compactification radius, not tuned — if tuned, it is a hidden knob in the
G.10 sense and the gate downgrades.

**Math to attempt.** Trace the $m_c=M_U$ identification in `reproduce_all.py` and confirm the $\mathcal{O}(1)$
factor is a fixed geometric number (e.g. $1/2\pi$ from $R_0=(2\pi M_U)^{-1}$), not a free constant. This folds
into the R1 harness audit.

**Success ladder.** **DISCLOSED-CONSISTENT** (the convention labeling is already in A1.10) is in place;
**AXIOM-CLOSED** (AXIOM-MU-IS-CROSSING named + the $\mathcal{O}(1)$ shown geometric) is the realistic endpoint.
sharper-OPEN if the $\mathcal{O}(1)$ is found to be free → that is a hidden knob (REFUTES via G.10).

**Honest disposition: DISCLOSED; the live sub-task is the $\mathcal{O}(1)$-seam firewall, which folds into R1's
deep-check. The headline should say "couplings cross at $M_U\sim10^{16}$ GeV," not "we predict $M_U$."**

---

### 4.6 R6 — SM $b_i$ + comparison scale are inputs (given-E)

**The obstruction.** The beta functions $b=(41/10,-19/6,-7)$ and the comparison scale $M_Z$ are SM/PDG inputs;
SG-7 inherits the SM content E from SG-2/SG-3 and does not derive it.

**Technique: none internal — this is the given-E condition, shared by every gate.** There is no closure path
*inside SG-7*: deriving the $b_i$ would mean deriving E, which is the SG-2/SG-3 program (and is DERIVED-GIVEN-E
there, not DERIVED). The honest action is to **state the inheritance** and not count the $b_i$ or $M_Z$ as SG-7
outputs.

**Named axiom?** None — R6 is the given-E condition made explicit. No κ³/π issue (nothing to target-load;
the $b_i$ are forced by E once E is given).

**Success ladder.** Not applicable (structural). **Honest disposition: DISCLOSED (given-E); not closeable within
SG-7. The gate is a consistency pass on E, not a derivation of E.**

---

### 4.7 R7 — KK spectrum frozen at the SG-6 chamber-center witness (inherited)

**The obstruction.** The KK tower masses (hence the heat-kernel contributions) are fixed by the moduli read at
the SG-6 Weyl-rigid witness $\vec u=(1,1,1)$, not independently derived. SG-6's "moduli read from a witness"
soft spot propagates into the threshold spectrum.

**Technique: T1 (axiom-floor), shared with SG-6.** The honest reduction is to name the inheritance:

> **AXIOM-CHAMBER-CENTER-SPECTRUM** (κ³/π-clean): *"the KK spectrum that enters the threshold ledger is the
> spectrum at the Weyl-rigid chamber center $\vec u=(1,1,1)$ of $K_6$ — the symmetry-distinguished point of the
> moduli space."*

This is writable with no $\delta$ value — it is a symmetry statement. It is **stronger than "read from a
minimum"** if the chamber center is shown to be the *unique* Weyl-rigid point (then it is symmetry-protected,
not tuned). The attack mirrors SG-6's R8: prove the chamber center is the unique Weyl-fixed point.

**Specialist target.** Hand the SG-6/moduli specialist: "Show $\vec u=(1,1,1)$ is the unique Weyl-rigid center
of the $K_6$ Cartan moduli, independent of the threshold output." Clean group-theory claim.

**Success ladder.** AXIOM-CLOSED (AXIOM-CHAMBER-CENTER-SPECTRUM named; symmetry-distinguished) is realistic;
DERIVED-CLOSED if the uniqueness theorem is proven. sharper-OPEN if the center is non-unique. **Honest
disposition: AXIOM-CLOSED likely; this is a genuine improvement over "read from a minimum" because the chamber
center is a symmetry point. Shared with SG-6 — closing it there closes it here.**

---

### 4.8 R8 — G.3.2a label inconsistency ($\sum Y^2=10/3$ vs $=10$)

**The obstruction.** G.3.2a states $\sum Y^2=10/3$ *per generation* while G.3.2 row-6's index column reads
"$\sum Y^2=10$" ($\times 3$ generations). Reconcilable (per-gen vs all-gen) but printed inconsistently.

**Technique: owner edit (cosmetic), no axiom.** Confirm the row-6 contribution uses $10/3$ per gen $\times 3$
gens $=10$ consistently, and relabel one of the two so a reviewer running G.10.3 step 1 does not flag a phantom
hidden knob.

**Success ladder.** **DISCLOSED-CORRECTED** — a one-line relabel; no physics. **Honest disposition: fix-on-sight;
folds into the R1 harness audit (the CSV will expose which convention the code uses).**

---

### 4.9 R9 — Spectrum uniqueness not claimed (consistency, not determination)

**The obstruction.** The certificate is given-the-selected-survivor: a pruning/consistency pass that this
spectrum *can* unify, NOT a proof that it is the *unique* spectrum that unifies. Rivals (string/M/F/NCG/lattice)
tie here — threshold unification is SHARED-OPEN across frameworks.

**Technique: T10 (selector) — but scoped honestly.** A full uniqueness claim ("no other admissible spectrum
unifies") is a universal negative and is **not** reachable; the honest move is to keep the scope statement
explicit. The only internal action is to ensure the headline says *consistency pass on the selected survivor*,
not *determination of the unification spectrum*.

**Named axiom?** None — R9 is a scope disclosure, correctly handled by §6.7's "Claimed certificate pass *under
frozen scheme, regulator, spectrum, and comparison scale*."

**Success ladder.** **DISCLOSED-CONSISTENT** (already scoped correctly). Not promotable; a universal-negative
uniqueness theorem is out of reach. **Honest disposition: DISCLOSED; the gate is a consistency pass, the scope
is correct, and the rival-tie is acknowledged (and lives in `BATTLE_GATES/`, not here).**

---

### 4.10 Attack-plan roll-up

| Residual | Technique | Named axiom (κ³/π-clean) | Specialist target / owner artifact | Realistic endpoint |
|---|---|---|---|---|
| R1 absent harness | machine-lane | (none) | mount `reproduce_all.py` + 2 CSVs; run DEEP check (3b), not just re-sum (3a) | BLOCKED_INPUTS → VERIFIED (sum-level); R2 stays open |
| R2 row-formula non-reproduction | T1 / T5-guard | AXIOM-THRESHOLD-NORMALIZATION | zeta-regularized KK-tower sum per factor, target-blind | OPEN → AXIOM-CLOSED if normalization yields rows; DERIVED if zeta-sum lands clean |
| R3 injected $\delta$ | T6 (dependent) | AXIOM-DELTA-IS-COLUMN-SUM | (= R2) | OPEN, dissolves when R2 closes |
| R4 forcedness declared-not-proven | T5 | AXIOM-ROW-FORCEDNESS (= R2) | enumeration-vs-forcedness restatement | DISCLOSED-CORRECTED; AXIOM-CLOSED at R2 |
| R5 $M_U$ convention | T2 + T5 | AXIOM-MU-IS-CROSSING | $\mathcal{O}(1)$-seam firewall (folds into R1) | DISCLOSED → AXIOM-CLOSED if $\mathcal{O}(1)$ geometric |
| R6 $b_i$/$M_Z$ inputs | — (given-E) | (none) | — | DISCLOSED (given-E); not closeable in SG-7 |
| R7 SG-6 chamber-center spectrum | T1 | AXIOM-CHAMBER-CENTER-SPECTRUM | Weyl-center uniqueness (shared w/ SG-6) | AXIOM-CLOSED (improves SG-6 framing) |
| R8 $\sum Y^2$ label | owner edit | (none) | one-line relabel (folds into R1) | DISCLOSED-CORRECTED |
| R9 spectrum uniqueness | T10 (scoped) | (none) | keep scope explicit | DISCLOSED-CONSISTENT (rival-tie) |

**REDUCE-vs-RELOCATE verdict on the plan.** The plan does **not** turn one hard problem into three harder ones.
The whole gate reduces to **two real targets** — R1 (a mechanical owner artifact) and R2 (one bounded
heat-kernel computation) — with R3/R4/R5/R8 all *dependent on or folding into* those two, and R6/R7/R9 being
given-E / inherited / scope statements that carry no new burden. The harder-subproblem check: R2 is **smaller**
than the original gate (it is one zeta-regularized sum per factor, four checkable rows), and it carries an
explicit **κ³/π kill-test** so a tuned normalization cannot be banked. The one genuine danger is **misclosing
R1**: accepting the pasted terminal log (which verifies only the trivial re-sum) as a closure of the harness.
The dossier's §2.3 deep-check + §4.1 (3a)-vs-(3b) split is the firewall against exactly that. The honest expected
outcome of a full campaign: **R1 VERIFIED (sum-level), R2 OPEN→AXIOM-CLOSED-pending-verification (most likely)
or sharper-OPEN, R3 dissolved-with-R2 or OPEN, R4/R5/R8 DISCLOSED-CORRECTED, R7 AXIOM-CLOSED (shared with SG-6),
R6/R9 DISCLOSED.** No DERIVED-CLOSED is promised; the gate would move from PARTIAL-with-asserted-status to
**PARTIAL-with-machine-verified-sum + a named threshold-normalization axiom floor** — a real
honesty/reproducibility gain, **not** a promotion.

---

## A. Anchoring & Hardening Map

This section runs SG-7 through our internal honesty methodology: for each open residual, classify it as a **gap** (a route
exists; what is missing is a computation, a written certificate, or a measured input) or a **wall** (the route
itself is the problem); hunt the implicit assumption it hides; then name the **measured invariant** each residual
must ultimately terminate on, and assign the most conservative defensible disposition. The framework, the
gap-vs-wall distinction, and the status vocabulary used below are the live **Gaps & Walls Register**
(<https://physics.magflowmeters.com/articles/GAPS_AND_WALLS_REGISTER.html>) — this section applies that method to
SG-7's nine residuals and is consistent with the Register's SG-7 line (**SCHEME-ANCHORED / Diagnostic-only**: the
*signs* of the threshold corrections are geometric, the *magnitudes* are fitted-to-target, not formula-derived).

**Disposition vocabulary (used inline, as the Register defines it).** **DERIVED** — connected to an already-measured
fact with no new assumption (rare; a mechanical check, not a breakthrough). **DERIVED-GIVEN-E** — rigid once the
observed Standard-Model content E is supplied as input; never produces E. **AXIOM-CLOSED** — reduced to exactly one
named, unproven, value-free assumption (honest bookkeeping; the gap stays open). **DISSOLVED** — the problem rested
on a false/unmeasured premise; dropping it removes the problem (a smaller real obligation usually remains).
**SCHEME-ANCHORED** — a number is reachable only after one named, geometry-unfixed normalization/regularization
choice (stronger than "no idea," strictly weaker than DERIVED). **OPEN** — genuinely unsolved, or the only escape
assumes the answer. **BLOCKED** — a real route exists but a required file/recipe/input is missing or
self-contradictory. **measured-but-irreducible** — known from experiment, no derivation exists. The existing
measured anchors SG-7 may terminate on are the four programme anchors plus the imported invariants the gate inherits:
**ℏ** (action grain), **M_Pl**, the **observed SM spectrum E** (which fixes the $b_i$), the measured **$\alpha_i(M_Z)$**,
**$y_t$**, **$|V_{us}|$**. A residual that does **not** reach one of these — and needs instead a *new* named invariant, a
*scheme*-anchor, an unrun *computation*, or has *no witness* — is open by exactly that amount.

### A.1 Per-residual anchoring & hardening table

| Residual (from §3) | GAP or WALL (+kind) | Measured-invariant truth it must terminate on | Honest disposition | What would HARDEN it |
|---|---|---|---|---|
| **R1** — absent reproduction harness (`reproduce_all.py` + 2 CSVs) | **GAP** (computation-debt / reproducibility) | Terminates on the measured **$\alpha_i(M_Z)$** + **E**-fixed $b_i$ already in the pipeline — it adds no new invariant, only the machine-real check that the printed rows sum to the printed $\delta$ and (the deep check) regenerate from invariants. | **BLOCKED** (inputs absent here) → **DERIVED** for the sum-level check once mounted; the deep check is gated on R2. The pasted terminal log is **AUDIT**, not a reproduction. | Mount the script + both CSVs and run the **deep** check (§4.1 3b — rows *computed* from G.3.2a primitives, not hard-coded), not just the trivial re-sum (3a). |
| **R2** — G.3.2a row formulas do NOT reproduce the $\delta$ decimals (~2×–18× off) | **WALL-adjacent GAP** (computation-debt; the genuine-physics seam) | Must terminate on **E** + **$\alpha_i(M_Z)$** via a geometric normalization, but today reaches **neither** — it hinges on a **SCHEME-anchor** (the missing per-factor KK zeta-normalization $Z_{\mathcal X}$ + overlap $c_{a,b,i}$, the "$\mathcal O(1)$" in $m_c{=}M_U$). | **SCHEME-ANCHORED** (the gate-defining residual; the Register's "magnitudes object-dependent, signs geometric" lives here). | Compute the **zeta-regularized KK-tower sum** per factor **target-blind** (§4.2 Route A) and show it lands on the printed rows; or name **AXIOM-THRESHOLD-NORMALIZATION** (Route B) **and verify it yields** them. κ³/π kill-test: a normalization reverse-engineered to hit the known $\delta$ **relocates**, it does not harden. |
| **R3** — the three $\delta$ are injected reals | **GAP** (computation-debt; dependent of R2) | Same termination as R2 (**E** + **$\alpha_i(M_Z)$** via the geometric normalization). The $\delta$ carry **no independent invariant** beyond the ledger rows — they are a column sum by definition. | **SCHEME-ANCHORED / OPEN** — injected today; dissolves exactly when R2's rows derive. | Derive the rows (= harden R2). Naming **AXIOM-DELTA-IS-COLUMN-SUM** only relocates the injection from 3 numbers to 8 rows until R2 lands. |
| **R4** — "no row is a free parameter" declared-not-proven | **GAP** (honesty / claim-boundary; points at R2) | Terminates where R2 does: forcedness is the **E**+$\alpha_i$ termination, not enumeration. The No-Hidden-Knob audit tests *enumeration*, not *forcedness* — a different (weaker) predicate. | **DISCLOSED** (the G.9.6 own caveat already records it) → **AXIOM-CLOSED** only when R2's forcedness axiom is *verified to yield* the rows. | State plainly, at the headline, that the discipline proves **enumeration not forcedness**; the real closure is R2. |
| **R5** — $M_U$ is a declared closure-target convention, not a prediction | **GAP** (claim-boundary + a firewall sub-task) | $M_U$ terminates on the **crossing condition** under measured **$\alpha_i(M_Z)$** + the frozen $\delta$ — it is an *output* of crossing, not a new invariant; the live risk is the "$\mathcal O(1)$" in $m_c{=}M_U$ hiding an undeclared scheme knob. | **DISCLOSED** (A1.10 already labels it) → **AXIOM-CLOSED** if the $\mathcal O(1)$ is shown geometric (e.g. $1/2\pi$ from $R_0$), **OPEN/BLOCKED** if found free (a hidden knob). | Trace the $m_c{=}M_U$ factor in the harness (folds into R1) and confirm it is a fixed geometric number, not tuned. Headline: "couplings cross at $M_U\sim10^{16}$ GeV," not "we predict $M_U$." |
| **R6** — SM $b_i$ + comparison scale $M_Z$ are inputs | **GAP** (structural; the given-E condition) | Terminates directly on the **observed spectrum E** (fixes $b_i$) and the measured **$M_Z$** — both *imported*, not produced. Deriving them = deriving E (the SG-2/SG-3 program), which is itself only DERIVED-GIVEN-E. | **measured-but-irreducible within SG-7** / **DISCLOSED (given-E)** — not closeable inside this gate. | Nothing inside SG-7; state the inheritance and do not count $b_i$/$M_Z$ as SG-7 outputs. The gate is a consistency pass *on* E, not a derivation *of* E. |
| **R7** — KK spectrum frozen at the SG-6 chamber-center witness | **GAP** (inherited from SG-6; moduli-witness) | Terminates on a **NEW named invariant**: the Weyl-rigid uniqueness of the chamber center $\vec u{=}(1,1,1)$ — a symmetry-distinguished point, stronger than "read from a minimum" *if* proven unique. No measured anchor; a group-theory fact. | **AXIOM-CLOSED** likely (name **AXIOM-CHAMBER-CENTER-SPECTRUM**) → **DERIVED** if the uniqueness theorem is proven; shared with SG-6. | Prove $\vec u{=}(1,1,1)$ is the **unique** Weyl-rigid center of the $K_6$ Cartan moduli, independent of the threshold output (hand to the SG-6/moduli specialist). |
| **R8** — G.3.2a label inconsistency ($\sum Y^2{=}10/3$ vs ${=}10$) | **GAP** (cosmetic; per-gen vs all-gen) | No new invariant — reconcilable arithmetic ($10/3$ per gen × 3 gens $= 10$); a labeling defect, not physics. | **DISCLOSED (cosmetic)** → **DERIVED-CORRECTED** on a one-line relabel. | Relabel one of the two consistently; the CSV (R1) will expose which convention the code uses, so it folds into R1. |
| **R9** — spectrum uniqueness not claimed (consistency, not determination) | **WALL** (universal-negative scope) | Would require a **universal-negative** ("no other admissible spectrum unifies") — **no measured invariant and no theorem reaches it**; rivals tie here (SHARED-OPEN on threshold). | **OPEN (scoped)** / correctly **DISCLOSED** — a universal negative is out of reach; not promotable. | Nothing closes it; keep the scope explicit ("consistency pass on the selected survivor," not "determination of the unification spectrum"). Rival-tie lives in `BATTLE_GATES/`, not here. |

### A.2 Gate-level rollup

- **Overall disposition: SCHEME-ANCHORED / Diagnostic-only** (consistent with the live GUT.md Gate-7 card and the
  Gaps & Walls Register §4). The defensible, hardened content is the **sign structure** of the threshold
  corrections (hypercharge-matter driver positive, the non-abelian rows negative), which is real recovered physics
  terminating on **E** + measured **$\alpha_i(M_Z)$**; the **magnitudes** are *not* derived — they hinge on the one
  unfixed heat-kernel scheme object (R2/R3), so the gate is a **diagnostic that recovers signs, not a closed
  unification certificate.** No residual reaches **DERIVED** unconditionally; none is a Class-F wall except the
  scoped universal-negative R9.
- **Anchors this gate depends on:** the observed **SM spectrum E** (fixes the $b_i$ — R6), the measured
  **$\alpha_i(M_Z)$** (the three running couplings — R6), and the declared comparison scale **$M_Z$**. It rides
  **M_Pl** only as a frozen declared input. It does **not** terminate on **$y_t$** or **$|V_{us}|$** (those anchor
  SG-8). The threshold magnitudes terminate on **no measured invariant** today — they sit on a named SCHEME-anchor
  (R2's missing KK zeta-normalization $Z_{\mathcal X}$ + overlap $c_{a,b,i}$), which is precisely why the gate is
  SCHEME-ANCHORED, not DERIVED.
- **Single highest-leverage hardening move:** settle the **one shared heat-kernel scheme object** of R2 — compute the
  **zeta-regularized KK-tower sum per factor target-blind** and show it reproduces the printed rows (or name and
  *verify* AXIOM-THRESHOLD-NORMALIZATION). This is the gate's whole credibility: it converts the $\delta$ from
  injected to derived and dissolves R3 and R4 in one move. The prerequisite/cheapest enabling step is **R1's deep
  check** (mount the harness, confirm the rows are *computed* not hard-coded). The trap, applied in full: any
  normalization reverse-engineered to hit the already-known $\delta$ is **true by construction** (the κ³/π pattern)
  and **relocates** the residual rather than hardening it. Ceiling for the whole gate: **serious candidate, NOT
  validated** — the realistic win is a machine-verified column-sum over a named threshold-normalization axiom floor,
  **not a promotion.**

### 🎯 Target anchor(s) for this gate

In the terminate-on (anchoring) sense, SG-7 does **not** terminate on a measured invariant: the gate is
**Diagnostic-only**, its ledger rows **fitted-to-target**, and the *signs* of the threshold corrections are
geometric (real recovered physics, terminating on **E** + measured **$\alpha_i(M_Z)$**). The *magnitudes*
terminate on exactly **one heat-kernel SCHEME-ANCHOR** — the single geometry-unfixed Seeley–DeWitt finite
object (which finite SDW coefficient is the one-loop KK threshold: the per-factor KK zeta-normalization
$Z_{\mathcal X}$ + overlap $c_{a,b,i}$ of R2/R3, the scheme decision shared with $c_{\rm loop}$ and $a_6$).
**Status: SCHEME-ANCHORED / OPEN** — no measured-invariant witness exists for the magnitudes today; settling
this one named scheme object is the gate's whole credibility, and any normalization reverse-engineered to hit
the known $\delta$ relocates rather than hardens (κ³/π kill-test). NOT a measured invariant; NOT promoted.

---

## 5. References & source map

### 5.1 Website source-of-truth (common material — link, don't duplicate)

- **Paper I, GUT.html** — <https://physics.magflowmeters.com/articles/GUT.html>
  - **§6.7** Gate-7 card (binding status); **§6.12** falsification map (Gate 7 row → *Diagnostic only* on
    unreproduced ledger / hidden knob); **§6.13** certificate summary.
  - **§5.6** narrative module (the "not quite" / freeze-is-the-force framing); **Appendix CR module CR7**
    (explanatory companion — running couplings, the slide-the-step argument).
  - **Appendix G** (Threshold authority): G.1 inputs, G.2 RG transport, **G.3 KK thresholds** (G.3.1 explicit
    derivation Steps 1–4; **G.3.2 the 8-row heat-kernel ledger**; **G.3.2a the coefficient-definition table**),
    G.4 output table, **G.5 fail-closed rules**, G.6 certificate JSON, **G.9 certificate checklist** (incl. the
    **G.9.6 forcedness caveat** and **G.9.10 downgrade rule**), **G.10 No-Hidden-Knob audit** (G.10.1 knob
    table, G.10.2 binding statement, G.10.3 reviewer procedure), G.11 authority summary.
  - **Appendix A1.10** scale constants (the **$M_U$ "declared closure-target convention"** label); **A1.11**
    threshold constants (A1.11.2 per-factor packet table; A1.11.3 the **boxed $\delta$** + the five-sig-fig
    precision note); **A1.16** reopen triggers (the heat-kernel-ledger trigger).
  - **Appendix R0** freeze + reproducibility: **R0.4.1** captured `reproduce_all.py` output; **R0.6**
    fail-closed rules; **R0.7.2** external verification log; the **bundle-file hash table** (`reproduce_all.py`
    `30d4d3049051`, `appendix_F_heat_kernel_ledger.csv` `9a6c7c08dc3c`, `appendix_F_threshold_outputs.csv`
    `a9b61c5f8049`).
- **Paper IV, TOE.html** — <https://physics.magflowmeters.com/articles/TOE.html> (Λ cross-reference only; does
  not touch SG-7).

### 5.2 Corpus locations (authoritative inputs to this dossier)

| Source | Path | Role |
|---|---|---|
| Per-gate dossier spec | `…/rendered/TOE/PER_GATE_DOSSIER_SPEC.md` | structure (sections 0–5) |
| SG-7 status line | `…/rendered/TOE/SCOPED_GUT_PER_GATE_STATUS_LEDGER.md` | the PARTIAL label + honest caveats (carried verbatim) |
| SG-8 exemplar (shape match) | `…/rendered/TOE/PER_GATE_DOSSIERS/DOSSIER_SG8_FLAVOR_CLOSURE_ATTACK.md` | depth/format template |
| Closure campaign result | `…/rendered/TOE/CLOSURE_CAMPAIGN_RESULT_2026-06-24.md` | κ³/π kill-test discipline; AXIOM-CLOSED ≠ proven; demote-on-verify; "a captured log ≠ a derivation" norm |
| Closure campaign round 2 | `…/rendered/TOE/CLOSURE_CAMPAIGN_ROUND2_2026-06-24.md` | four standing kill-tests; conservative-disposition rule |
| GUT manuscript | `…/rendered/GUT/GUT.md` | §5.6 (Gate-7 narrative); §6.7 (Gate-7 card); §6.12 (falsification map); Appendix G (G.0–G.11); A1.10/A1.11 (scale + threshold constants); R0.4.1 (captured reproducer); R0 bundle-file hashes |

### 5.3 Frozen-object hash index (load-bearing; READ-ONLY)

Branch `dcc66f1b2685` · manifest meta `a5b1e6f9d951` · $R_{K_6}$ `634438ce0776` · $R_{S^2}$ `2381d472c62e` ·
$R_Y$ `0e8b8dba2cf0` · RG transport (two-loop $\overline{\rm MS}$) `f531205a9159` · $M_Z$ `a6852c7a6b00` ·
uncertainty rule `61b0d93507e7` · $M_{\rm Pl}$ `df5976a365c3` · $\alpha_i^{-1}(M_Z)$ `6a3b6ef06697` ·
$\mathbb{Z}_2$ on $S^1_Y$ `ac4d2df3e708` · $\mathbb{Z}_6$ identification `a68ee92a75be` · Wilson-line cycle
$\gamma$ `640e1d7f7773` · winding $n_H$ `f65094fd8fd1` · $\eta_{BK}$ `84e94518d3f5`. Bundle files (REFERENCED,
not re-run here): `reproduce_all.py` `30d4d3049051` · `appendix_F_threshold_outputs.csv` `a9b61c5f8049` ·
`appendix_F_heat_kernel_ledger.csv` `9a6c7c08dc3c`. **$\delta$-triple
$(+4.8424,-3.1112,-1.7313)\pm1.6\times10^{-3}$: no per-row first-principles hash — injected reals.**

---

### Closing honest statement

SG-7 is **PARTIAL**. Its genuine, defensible content is real: the survivor's own frozen KK spectrum supplies
finite heat-kernel thresholds of the **correct structural sign and magnitude** to close the three gauge couplings
at $M_U\sim10^{16}$ GeV, with the **family count $-3$ and the Higgs winding $n_H=1$ both demonstrably
load-bearing**, and a genuine discipline layer (G.5 fail-closed, G.9 checklist, G.10 No-Hidden-Knob audit, G.9.8
catastrophic-under-modulus-shift sensitivity). Its honest open surface is equally clear: the three $\delta$ are
**injected reals**; the **G.3.2a row formulas do NOT reproduce the $\delta$ decimals** (off **~2×–18×**,
recomputed by hand in §2.3); the load-bearing **`reproduce_all.py` + CSV harness is ABSENT** from this audit and
the captured terminal log verifies only the **trivial re-sum of the printed rows**, not the deep regeneration;
**"no row is a free parameter" is declared-not-proven** (the G.9.6 own caveat); and $M_U$, $M_Z$, and the SM
$b_i$ are **declared conventions / given-E inputs**, not predictions. The attack plan reduces these to **two real
targets** — the **machine-real harness mount (R1, the cheapest win)** and the **G.3.2a row-formula
reconciliation via a target-blind zeta-regularized tower sum (R2, the one physics target)** — with the κ³/π
kill-test guarding R2 so a reverse-engineered normalization cannot be dressed as a derivation, and an explicit
firewall (§4.1 3a-vs-3b) against misclosing R1 by accepting the pasted log. The realistic ceiling is a
machine-verified column-sum over a named threshold-normalization axiom floor — **not** a promotion.
**No status was ever upgraded; frozen branch `dcc66f1b2685` / `a5b1e6f9d951` READ-ONLY; given-E ≠ derivation of E; a captured
log ≠ an independent reproduction; nothing applied, nothing deployed.**

*Dossier built 2026-06-24. Our geometry (13D K₆ branch) only. Common material referenced to the published
website source-of-truth, not duplicated.*