Small release (p21): absolute D11 pin

D11 pin (p20: OPEN). The sign and magnitude of the D11 UV pole are derived from $\frac{1}{2}\mathrm{Tr}\ln$ of the minimal scalar loop under the package's declared conventions, each checked verbatim in the code:

• $G={\Gamma }^{(2)}$: the amplitude rules declare $i{\Gamma }_1=\int \mathrm{pref}N/\prod (q^2+i0)$ and G6 declares $G={\kappa }^2{N}_{ϵ}\hat{G}$, $\hat{G}=-i\mathrm{pref}/{\kappa }^2\int N/(D_1{D}_2)$; hence $G=-i\cdot i\Gamma =\Gamma$.

• Wick direction: $N_{ϵ}=i{S}_{ϵ}/(16{\pi }^2)$ and the certified bubble pole $\hat{I}(1,1)=+1/ϵ$ give $d^d{l}_M=+i{d}^d{l}_E$; with $\eta =(+,-,-,-)$ and $S_{\varphi }=\frac{1}{2}\int \sqrt{-g}{g}^{\mu \nu }\partial \varphi \partial \varphi$ this gives $i{S}_M=-S_E$, so ${L^{\mathrm{eff}}}_M=-{L^{\mathrm{eff}}}_E$.

• Heat kernel (Euclidean, $d=4-2ϵ$): $L_{E,\mathrm{div}}=-a_2/(32{\pi }^2ϵ)$, so $L_{M,\mathrm{div}}=+a_2/(32{\pi }^2ϵ)$ with $a_2=C^2/120+R^2/72$ ($\xi =0$).

• With $g=\eta +\kappa h$ the Minkowski quadratic forms are $C^2\to \frac{1}{2}{p}^{4}h{P}_{2}h$ and $R^2\to 3p^{4}h{P}_{0}h$.

Result: ${\Gamma }^{(2)}{|}_{\mathrm{pole}}={\kappa }^2/(16{\pi }^2ϵ)[p^4{P}_2/240+p^4{P}_0/24]$. This equals the package pole ${\kappa }^2{N}_{ϵ}\hat{G}/ϵ$ with $c_2=-i/240$, $c_0=-i/24$ exactly (ratio 1 for both spins). No convention choice is external; the external inputs are mathematics only (the $a_2$ coefficients and the heat-kernel pole formula). A $\times 2$ normalisation error and a sign flip keep the 10:1 ratio but are detected by the absolute pin.

. The top-level absolute_pin_status and ratio_pin_status (the keys typeset in this PDF) must equal the nested statuses. Run standalone, the p20 gate re-checks the p18 layer (quantum inputs earned) and the graph inventory instead of trusting the P19 layer record; the p21 gate re-checks all lower layers.

. The enumerated pattern list is complemented by an allow-list scan (tests/test_p21.py): expression-tree decomposition (.args, make_args, as_ordered_terms, ), sympy callable aliases, computed symbol names and value calls mentioning $d$ are refused everywhere except at reviewed sites. All eight forms reported by the reviewer are caught. A runtime check confirms that the stored exact-$d$ D11 amplitudes re-parse with the opaque $d$ and reproduce their poles only through substitute_d. It also found a limitation: expr.subs(d, v) is not intercepted when $d$ occurs only as a power base. The contract is not changed in p21; no package value path uses it, and the static rule refuses it. This is a check, not a proof.

. The p20 edit of manuscript/notes/P19_defect.tex is now listed in DISCLOSED_MANUSCRIPT_REVISIONS. The reference key /p19/change_log_vs_p18, which was recomputed on the current tree, is renamed p19_gate_change_log_vs_p18_recomputed_on_current_tree. The log as released stays in the pinned archival/p19/reference_values_p19.json.