Controlled Penumbral Inflation from Monodromic Valleys Pirzada, Tianjun Li Abstract Long monodromic valleys arise in the penumbra of complex-structure moduli space. We show that their local branch data already determine whether they support controlled inflation, and thereby isolate the first controlled penumbral inflationary window. In the axion–saxion effective theory given in Eq. ( 4 ), a branch-displacing odd term generates a plateau when Δ ≡ p + 2 ​ ν − q > 0 \Delta\equiv p+2\nu-q>0 , while covariant single-clock control further requires p < 2 p<2 , or p = 2 p=2 with 12 ​ A p ​ m 2 / V 0 ≫ 1 12A_{p}m^{2}/V_{0}\gg 1 over the observational window. This splits penumbral valleys into no plateau, uncontrolled plateau, and controlled plateau before global completion is attempted. We identify a minimal analytic family with a closed-form valley and an invariant attractor equation for the full two-field dynamics, providing the first exactly solvable penumbral realization that remains predictive under the next penumbral order. The penumbra is thus promoted from a geometric suggestion to a predictive search principle. Introduction. –Axion monodromy remains one of the clearest routes to parametrically long inflationary trajectories in string theory 42 ; 35 ; 28 . The decisive obstruction is no longer field range alone. A long valley may still fail because heavy-sector backreaction, canonical distance, or loss of adiabaticity destroys the single-clock description 15 ; 13 . F-term constructions, hierarchy analyses, universal axion backreaction, four-form descriptions, and large-distance EFT studies established how monodromy reshapes potentials and exposes UV sensitivity 34 ; 6 ; 25 ; 24 ; 5 ; 45 ; 29 ; 21 ; 22 ; 30 ; 38 . Hyperbolic attractors explain why logarithmic canonical variables generate plateau-like observables 26 ; 7 ; 16 ; 40 ; 27 . The decisive need for top-down work is a local discriminator: which long monodromic valleys are controlled inflationary targets near, though not at, an infinite-distance boundary? Recent work isolated the penumbra of complex-structure moduli space as the crossover region where uplift-like positive terms coexist with surviving polynomial monodromic corrections and long axion valleys 31 ; 12 . Throughout this Letter, umbra denotes the strict asymptotic full-shadow regime near an infinite-distance boundary, whereas penumbra denotes the near-boundary partial-shadow regime where those terms still coexist; Fig. 1 fixes the terminology. That analysis made the regime concrete and showed why geometry alone does not secure inflation: the late de Sitter slope remains too large, other sectors run away, or the orthogonal mode fails to stay adiabatically heavy. The decisive local input is a criterion that either upgrades a monodromic penumbral valley to a controlled inflationary EFT or rejects it locally. Here we derive that criterion. Near moduli-space boundaries, curved-field EFT is rich enough to reveal flattening yet deliberately local: it cannot by itself settle global viability 5 ; 45 ; 24 ; 22 ; 30 ; 21 ; 38 ; 23 ; 14 . If local data already decide which valleys deserve the harder step of global completion, top-down model building becomes a filtered search problem. An odd monodromy-preserving branch term displaces the heavy minimum, rotates the light direction saxionically, and exponentiates inverse powers into a plateau. The same data determine whether the plateau is a trustworthy single-clock EFT. We obtain a covariant local control theorem, an analytic family whose attractor structure is visible directly in the full two-field equations, and a narrow observable corridor that survives the complete next penumbral order. The result is a local decision principle for top-down multifield inflation: existence, control, and predictive stability. Rather than replacing global completion, it sharpens the local burden of proof before global completion is attempted. Figure 1: Umbra and penumbra terminology. The schematic fixes the language used throughout: the umbra is the strict asymptotic full-shadow regime, while the penumbra is the near-boundary partial-shadow crossover where uplift-like terms and polynomial monodromy still coexist. Local EFT and Universality Class. — Write the local complex modulus as z=a+is , where a is the periodic axion and s>0 its saxionic partner. The one-modulus penumbral sector is locally described by the hyperbolic metric \frac{{\cal L}_{\rm kin}}{\sqrt{-g}}=\frac{1}{4s^{2}}\big[(\partial a)^{2}+(\partial s)^{2}\big],\qquad X\equiv e-ma, (1) with m the local monodromy charge and e a constant branch-location parameter. The coordinate X is the invariant heavy branch variable: motion in a shifts the branch, while s controls how strongly that branch sector feeds back into the low-energy potential. Through quadratic order in X , the local potential is written as V=V_{0}-c_{q}V_{0}s^{-q}+A_{p}s^{-p}X^{2}+B_{p+\nu}s^{-(p+\nu)}X+\cdots. (2) The four leading structures respectively control the uplift, saxionic falloff, heavy restoring force, and branch displacement. V_{0}>0 is the uplift-like positive piece, c_{q}V_{0}s^{-q} sets the leading saxionic falloff of the plateau, A_{p}s^{-p}X^{2} is the quadratic restoring operator in the heavy direction, and the odd term B_{p+\nu}s^{-(p+\nu)}X displaces the heavy minimum away from X=0 . The integers q , p , and \nu control, respectively, the plateau exponent, the asymptotic softness of heavy stabilization, and the decay rate of the branch displacement. Equation ( 2 ) is the generic local branch expansion of a Type-IIB complex-structure flux potential near a Calabi–Yau boundary: the flux potential is bilinear in monodromy-invariant axion polynomials with coefficients scaling as inverse powers of s , so a chosen branch contains an X -independent sector together with linear and quadratic heavy terms unless symmetry or flux tuning removes them 21 ; 23 ; 20 ; 32 SM . We assume V_{0}>0 , A_{p}>0 , q\geq 1 , and \nu\geq 1 . For weak mixing with additional moduli or the axio-dilaton, the same local projection defines effective (p,\Delta) up to perturbative renormalization; strong mixing places the system outside the present one-modulus universality class. Minimizing with respect to X gives \displaystyle X_{v}(s) \displaystyle=-\frac{B_{p+\nu}}{2A_{p}}s^{-\nu}+\mathcal{O}(s^{-\nu-1}), (3) \displaystyle\varphi \displaystyle=\frac{\log s}{\sqrt{2}}+\mathcal{O}(s^{-2\nu-2}), so the valley is asymptotically tangent to the saxion rather than to the axion. The branch displacement therefore rotates the light direction. Substituting the valley back into Eq. ( 2 ) yields U(\varphi)=V_{0}-c_{q}V_{0}\mathrm{e}^{-q\sqrt{2}\varphi}-\frac{B_{p+\nu}^{2}}{4A_{p}}\mathrm{e}^{-(p+2\nu)\sqrt{2}\varphi}+\cdots. (4) Inverse powers of s become exponentials because the canonically normalized light field is logarithmic. The plateau is the backreacted image of the monodromic valley itself. The saxionic branch converts the late-slope obstruction into a parametrically small plateau slope: earlier effectively axionic valleys can flatten while retaining \gamma_{\rm late}\equiv\frac{|dV_{\rm val}/d\phi_{\rm val}|}{V_{\rm val}}=\mathcal{O}(1), (5) whereas the saxionic branch gives \gamma_{\rm saxionic}\equiv\frac{|U_{,\varphi}|}{U}\simeq\frac{1}{\sqrt{2}\,q\,N_{*}}\simeq 1.3\times 10^{-2} (6) for q=1 and N_{*}=55 . The leading plateau is universal whenever \Delta\equiv p+2\nu-q>0, (7) in which case, with N_{*} the number of e-folds before the end of inflation, n_{s}=1-\frac{2}{N_{*}}+\mathcal{O}(N_{*}^{-2}),\qquad r=\frac{4}{q^{2}N_{*}^{2}}+\mathcal{O}(N_{*}^{-3}). (8) Equation ( 8 ) defines the local penumbral universality class. The same data that generate the plateau also test whether the orthogonal sector remains heavy enough for the reduced one-field description SM . Figure 2: Mechanism. (a) Benchmark potential and leading q=1 plateau. (b) Representative two-field trajectories in X=e-ma with the analytic valley. The restoring force acts on X-X_{v}(s) , driving the system to the valley before the cosmic-microwave-background (CMB) window ends. Control Theorem. — Adiabatic single-clock control is imposed covariantly: the orthogonal mode stays heavy, the turn rate stays small, and the reduced light trajectory captures the inflationary dynamics throughout the observational window 10 . In curved field space the relevant quantity is the entropy mass 19 ; 41 ; 3 ; 2 ; 39 , the basis-independent mass of fluctuations orthogonal to the trajectory, m_{s}^{2}\equiv m_{\perp}^{2}\equiv N^{I}N^{J}\nabla_{I}\nabla_{J}V+\epsilon H^{2}R_{\rm fs}-\Omega^{2},\qquad R_{\rm fs}=-4, (9) with N^{I} the unit normal, \Omega the turn rate, and R_{\rm fs} the field-space Ricci scalar. The three terms encode, respectively, Hessian stabilization normal to the path, the correction from field-space curvature, and the loss of decoupling caused by bending. In the (X,s) coordinates one has G^{XX}=2m^{2}s^{2} , so along the valley m_{s}^{2}=4A_{p}m^{2}s^{2-p}\big[1+\mathcal{O}(s^{-2\nu-2})\big]. (10) Since H^{2}\to V_{0}/3 on the plateau, one obtains \frac{m_{s}^{2}}{H^{2}}=\frac{12A_{p}m^{2}}{V_{0}}s^{2-p}\big[1+\mathcal{O}(s^{-2\nu-2})\big]. (11) Thus control fails asymptotically for p>2 , becomes asymptotically strong for p<2 , and is boundary-sensitive at p=2 . Local Control Theorem. A penumbral valley supports a controlled plateau if and only if \Delta>0 and either p<2 , or p=2 with 12A_{p}m^{2}/V_{0}\gg 1 in the inflationary window. Equivalently, local valleys split into three EFT regimes: no plateau ( \Delta\leq 0 ), plateau without control ( \Delta>0 and p>2 ), and controlled plateau. The theorem converts valley geometry into a local EFT selection rule. A long valley is not enough. The same backreaction that creates the plateau also invalidates it when the quadratic heavy operator softens too quickly: the orthogonal sector then ceases to decouple and the apparent plateau is not a controlled single-clock regime. Existence and control must be imposed together. At the boundary p=2 , control is supplied by the large prefactor 12A_{p}m^{2}/V_{0} across the CMB window. At the same point the canonical slope is small, |U_{,\varphi}|/U=\sqrt{r/8}\simeq 1.25\times 10^{-2} , so the model avoids both the late-slope and heavy-decoupling obstructions. Earlier penumbral analyses identify candidate finite-distance windows; the displaced-minimum condition together with the heavy-sector inequality decides whether such a window is already a controlled inflationary EFT. Figure 3: Theorem and covariant control. (a) Local theorem map in the (p,\Delta) plane. The benchmark (p,\Delta)=(2,3) sits on the asymptotic control boundary but inside the controlled region. (b) The covariant entropy mass tracks the heavy eigenvalue, confirming that control is basis-independent. Analytic Attractor Family. – A minimal closed-form penumbral realization is V_{q}(a,s)=V_{0}\Bigl(1-\frac{\beta}{s^{q}}\Bigr)^{2}+\frac{A}{s^{2}}(e-ma)^{2}+\frac{B}{s^{3}}(e-ma), (12) where \beta sets the leading plateau amplitude, A fixes the heavy restoring force, and B measures the odd branch displacement that tilts the valley. For this family, \displaystyle X_{v}(s) \displaystyle=e-ma_{v}(s)=-\frac{B}{2A}s^{-1}, (13) \displaystyle U_{q}(s) \displaystyle=V_{0}\Bigl(1-\frac{\beta}{s^{q}}\Bigr)^{2}-\frac{B^{2}}{4A}s^{-4}. (14) The full two-field background equations admit the closed form \partial_{\mathcal{N}}^{2}X+\Bigl(3-\epsilon-2\partial_{\mathcal{N}}\ln s\Bigr)\partial_{\mathcal{N}}X+\frac{4Am^{2}}{H^{2}}\bigl(X-X_{v}(s)\bigr)=0, (15) where \mathcal{N}\equiv\ln a_{\rm FRW} denotes e-fold time. This is the invariant attractor equation : written directly for the branch variable X , it exposes the damped heavy oscillator that locks X to the displaced valley X_{v}(s) independently of the light-field basis. The attractor is visible inside the full two-field system, not only after asymptotic reduction. The valley is dynamically attractive and adiabatically controlled SM . For the cosmic-microwave-background (CMB)-normalized benchmark (\beta,A,B,m,e)=(1,10^{-10},2\times 10^{-11},1,0) and V_{0}=3.9187\times 10^{-11} , one finds at N_{*}=55 (n_{s},r)=\bigl(0.96434,1.2493\times 10^{-3}\bigr) , m_{s}^{2}/H^{2}\simeq 30.9 , and \Omega/H\lesssim 1.4\times 10^{-7} , with negligible heavy-turn corrections. The benchmark requires only a modest odd-to-even branch hierarchy within the local flux expansion: B/A=0.2 fixes the odd-to-even branch weight, while V_{0} sets the observed scalar amplitude and flux quantization discretizes admissible coefficient ratios in candidate embeddings 9 SM . Varying the reheating history shifts N_{*} by the usual \mathcal{O}(5\text{--}10) and moves (n_{s},r) along the same narrow corridor. The boundary case p=2 remains safely controlled because the heavy prefactor is numerically large in the observational window. Predictive Stability and Observable Discriminator. — Predictive stability requires the local control filter to lock higher penumbral corrections into a compact observable bundle rather than reopening a wide phenomenological landscape 11 . To test the theorem across the local EFT rather than at a single benchmark, we keep the complete monodromy-preserving local EFT through the next penumbral order, V_{\rm NLO}=V_{0}\Bigl(1-\frac{\beta}{s}+\frac{\lambda}{s^{2}}\Bigr)^{2}+\Bigl(\frac{A}{s^{2}}+\frac{D}{s^{3}}\Bigr)X^{2}+\Bigl(\frac{B}{s^{3}}+\frac{C}{s^{4}}\Bigr)X, (16) together with the leading saxionic metric deformation (2s^{2})^{-1}[(1+\gamma_{a}/s)da^{2}+(1+\gamma_{s}/s)ds^{2}] . The coefficients \lambda , C , D , and \gamma_{a,s} parameterize the most general local next-order monodromy-preserving corrections retained here. Across the accepted deformation set we find at N_{*}=55 0.96386\leq n_{s}\leq 0.96503 and 1.20\times 10^{-3}\leq r\leq 1.29\times 10^{-3} , while m_{s}^{2}/H^{2} remains safely above unity throughout the CMB window. In addition, the same branch yields the leading-order running relation \alpha_{s}\equiv\frac{dn_{s}}{d\ln k}=-\frac{2}{N_{*}^{2}}+\mathcal{O}(N_{*}^{-3})=-\frac{q^{2}}{2}r+\mathcal{O}(N_{*}^{-3}), (17) which for the q=1 benchmark reduces to \alpha_{s}\simeq-r/2 . Control does more than establish existence: once imposed, the complete next penumbral order does not open a broad phenomenological swath; it collapses into a compact corridor in both the (n_{s},r) and (\alpha_{s},r) planes SM . Oscillatory monodromy remnants give an additional discriminator. Local monodromic potentials frequently carry corrections of the schematic form \delta V_{\rm osc}\sim s^{-\gamma}\cos(\omega a+\delta) 15 ; 28 . Along the penumbral valley one has a_{v}^{\prime}(s)=\mathcal{O}(s^{-\nu-1}) , hence a_{v}(s)=a_{\infty}+\mathcal{O}(s^{-\nu}) at large s . Pulling the oscillatory term back to the valley gives \delta U_{\rm osc}(s)=s^{-\gamma}\Bigl[\cos(\omega a_{\infty}+\delta)+\mathcal{O}(s^{-\nu})\Bigr]. (18) Such terms renormalize leading exponential coefficients when they enter at comparable order, but they do not destroy the plateau unless they already compete at order s^{-q} . In the controlled region they decorate the corridor rather than define it. Tameness, Scope, and Remaining Global Input. — The penumbra and tameness literature impose a sharper requirement: local couplings must belong to a controlled definable class rather than exhibit arbitrarily wild behavior near the boundary 31 ; 23 ; 20 ; 14 ; 32 . Our construction isolates a finite-distance penumbral patch on which inverse powers of s , \log s , exponentials of \log s , and periodic functions of the compact axion remain definable 23 ; 20 . At the quantum-gravity level, because the inflationary window lies at finite distance, the strict infinite-distance obstruction is avoided at the level of the local EFT. Any admissible completion must still satisfy m_{\rm tower}(\varphi_{*})>H_{*},\qquad\Lambda_{\rm EFT}(\varphi_{*})>H_{*}, (19) throughout the observational window 37 ; 36 ; 17 . Compatibility with refined de Sitter bounds and explicit Kähler stabilization remain global completion requirements rather than local assumptions. This perspective sharpens the classification of penumbral valleys. Long uplift-like valleys do not automatically yield inflation 31 ; 12 ; 8 ; 33 : the slow-roll slope remains too large, other sectors such as Kähler moduli or the axio-dilaton run away, or the orthogonal mode fails to remain adiabatically heavy. The theorem isolates the third obstruction locally and splits valleys into no plateau, uncontrolled plateau, and controlled plateau. Within the one-modulus local EFT, it is a necessary-and-sufficient control criterion; for compactification, it becomes the filter deciding which valleys merit a global embedding and which can be excluded at the local level. It is a local reducer of the compactification burden of proof. The implications are broader than one benchmark. The distinction from standard attractor logic is structural 7 ; 16 ; 27 : hyperbolic geometry supplies the logarithmic field map, while the monodromic branch displacement supplies the saxionic rotation of the tangent mode, adiabatic control of the orthogonal sector, and predictive stability of the observable corridor under the next penumbral order. Those features follow from the local monodromic EFT together with the heavy-sector criterion. The integer q labels discrete penumbral families with common leading tilt and parallel running–tensor lines \alpha_{s}\simeq-(q^{2}/2)r . In that sense the penumbra is promoted from an intriguing region of moduli space to a filtered target space for top-down searches. Observationally, the benchmark lies in the concave small- r region favored by Planck plus BICEP/Keck 2018 (BK18) constraints 4 ; 44 , while the predicted tensor scale is targeted by LiteBIRD and future CMB-S4 analyses 18 ; 1 . The observational discriminator is whether future data select the narrow penumbral bundle rather than the broader swath admitted by unconstrained attractor models. A correlated pattern near r\sim 10^{-3} and \alpha_{s}\simeq-r/2 would make penumbral inflation sharply testable and turn definable moduli-space geometry into an observational target. Conclusion. —This work establishes a local, covariant, and falsifiable selection rule for top-down inflationary model building. A penumbral construction that realizes this mechanism must exhibit three linked signatures: branch-induced saxionic rotation, the heavy-control inequality, and the associated (n_{s},r,\alpha_{s}) family. The local analysis is the decision rule for which penumbral valleys deserve elevation to full compactifications. The immediate target is a survey of known Calabi–Yau orientifolds for explicit flux choices realizing the (p,\Delta)=(2,3) benchmark, together with an extension of the local control theorem that incorporates Kähler-moduli stabilization effects at next-to-leading order in the large-volume expansion. Acknowledgement. —TL is supported in part by the National Key Research and Development Program of China Grant No. 2020YFC2201504, by the Projects No. 11875062, No. 11947302, No. 12047503, and No. 12275333 supported by the National Natural Science Foundation of China, by the Key Research Program of the Chinese Academy of Sciences, Grant No. XDPB15, by the Scientific Instrument Developing Project of the Chinese Academy of Sciences, Grant No. YJKYYQ20190049, by the International Partnership Program of Chinese Academy of Sciences for Grand Challenges, Grant No. 112311KYSB20210012, and by the Henan Province Outstanding Foreign Scientist Studio Project, No.GZS2025008. Supplementary Material This Supplemental Material gives the derivations and diagnostics underlying the Letter. Section I derives the local universality class and heavy-control filter. Section II gives the covariant entropy-mass derivation. Section III develops the closed-form benchmark family and invariant attractor equation. Section IV treats the complete next penumbral order and the documented deformation scan. Section V derives the local Type-IIB origin of the ansatz, fixes the scope of the control theorem, and collects the remaining diagnostics. Appendix A Local reduction and the universal plateau The local monodromy-preserving penumbral EFT used in the Letter is V(X,s)=V_{0}-c_{q}V_{0}s^{-q}+\sum_{n>q}\frac{v_{n}}{s^{n}}+\sum_{n\geq p}\frac{A_{n}}{s^{n}}X^{2}+\sum_{n\geq p+\nu}\frac{B_{n}}{s^{n}}X+\mathcal{O}(X^{3}), (A1) with hyperbolic metric G_{IJ}=\delta_{IJ}/(2s^{2}) and X\equiv e-ma . Minimization with respect to X gives \partial_{X}V=2A_{p}Xs^{-p}+B_{p+\nu}s^{-(p+\nu)}+\cdots=0, (A2) and therefore X_{v}(s)=-\frac{B_{p+\nu}}{2A_{p}}s^{-\nu}+\mathcal{O}(s^{-\nu-1}). (A3) Since a=(e-X)/m , the valley slope is a_{v}^{\prime}(s)=-\frac{X_{v}^{\prime}(s)}{m}=\mathcal{O}(s^{-\nu-1}), (A4) so the induced line element along the valley is d\sigma^{2}=\frac{1+a_{v}^{\prime}(s)^{2}}{2s^{2}}ds^{2}=\frac{1}{2s^{2}}\left[1+\mathcal{O}(s^{-2\nu-2})\right]ds^{2}, (A5) which integrates to \varphi=\frac{\log s}{\sqrt{2}}+\mathcal{O}(s^{-2\nu-2}). (A6) Substituting the valley back into the local potential yields U(s)=V_{0}-c_{q}V_{0}s^{-q}-\frac{B_{p+\nu}^{2}}{4A_{p}}s^{-(p+2\nu)}+\cdots, (A7) so the canonical potential becomes U(\varphi)=V_{0}-c_{q}V_{0}\mathrm{e}^{-q\sqrt{2}\varphi}-\frac{B_{p+\nu}^{2}}{4A_{p}}\mathrm{e}^{-(p+2\nu)\sqrt{2}\varphi}+\cdots. (A8) Whenever \Delta\equiv p+2\nu-q>0 , the heavy-branch correction is asymptotically subleading and the leading potential is a plateau determined entirely by q . The e-fold integral then gives, with N_{*} the number of e-folds before the end of inflation, N_{*}\simeq\int\frac{U}{U_{,\varphi}}\,d\varphi\simeq\frac{\mathrm{e}^{q\sqrt{2}\varphi_{*}}}{2q^{2}c_{q}}, (A9) so that \epsilon_{*}\simeq\frac{1}{4q^{2}N_{*}^{2}},\qquad\eta_{*}\simeq-\frac{1}{N_{*}}, (A10) and therefore n_{s}=1-\frac{2}{N_{*}}+\mathcal{O}(N_{*}^{-2}),\qquad r=\frac{4}{q^{2}N_{*}^{2}}+\mathcal{O}(N_{*}^{-3}). (A11) Appendix B Covariant entropy mass and the heavy-control filter In the (X,s) variables the field-space metric is ds_{\rm field}^{2}=\frac{1}{2s^{2}}\left(\frac{dX^{2}}{m^{2}}+ds^{2}\right). (A12) Let y(s)\equiv X_{v}^{\prime}(s)/m . The orthonormal tangent and normal vectors are T^{I}=\frac{\sqrt{2}s}{\sqrt{1+y^{2}}}(my,1),\qquad N^{I}=\frac{\sqrt{2}s}{\sqrt{1+y^{2}}}(m,-y), (A13) with G_{IJ}T^{I}T^{J}=G_{IJ}N^{I}N^{J}=1 and G_{IJ}T^{I}N^{J}=0 . The covariant entropy mass is m_{s}^{2}=N^{I}N^{J}\nabla_{I}\nabla_{J}V+\epsilon H^{2}R_{\rm fs}-\Omega^{2}, (A14) with R_{\rm fs}=-4 . Because X_{v}^{\prime}(s)=\mathcal{O}(s^{-\nu-1}) , the normal vector aligns asymptotically with the X direction. Using \nabla_{X}\nabla_{X}V=2A_{p}s^{-p}+\cdots and G^{XX}=2m^{2}s^{2} , one obtains m_{s}^{2}=4A_{p}m^{2}s^{2-p}\left[1+\mathcal{O}(s^{-2\nu-2})\right]+\mathcal{O}(s^{-p-2\nu})+\mathcal{O}(\epsilon H^{2}). (A15) Since H^{2}\to V_{0}/3 on the plateau, one has \frac{m_{s}^{2}}{H^{2}}=\frac{12A_{p}m^{2}}{V_{0}}s^{2-p}\left[1+\mathcal{O}(s^{-2\nu-2})\right]. (A16) Thus asymptotic control is automatic for p<2 , fails for p>2 , and is boundary-sensitive at p=2 , where one requires the prefactor 12A_{p}m^{2}/V_{0} to be large in the inflationary window. The turn rate also decays. For small valley slope, the geodesic curvature obeys k_{g}\sim s^{-\nu-1} , while on a q -plateau one has \epsilon\sim s^{-2q} . Therefore \frac{\Omega}{H}=k_{g}\sqrt{2\epsilon}=\mathcal{O}\!\left(s^{-q-\nu-1}\right), (A17) so heavy-turn corrections are strongly suppressed in the controlled region. Appendix C Closed-form benchmark family and invariant attractor equation The minimal closed-form family is V_{q}(a,s)=V_{0}\Bigl(1-\frac{\beta}{s^{q}}\Bigr)^{2}+\frac{A}{s^{2}}(e-ma)^{2}+\frac{B}{s^{3}}(e-ma). (A18) Its heavy valley is analytic, a_{v}(s)=\frac{e}{m}+\frac{B}{2Am}\frac{1}{s},\qquad X_{v}(s)=-\frac{B}{2A}\frac{1}{s}, (A19) and the reduced potential is U_{q}(s)=V_{0}\Bigl(1-\frac{\beta}{s^{q}}\Bigr)^{2}-\frac{B^{2}}{4A}\frac{1}{s^{4}}. (A20) For q=1 this becomes U(\varphi)=V_{0}\Bigl(1-\beta\mathrm{e}^{-\sqrt{2}\varphi}\Bigr)^{2}-\frac{B^{2}}{4A}\mathrm{e}^{-4\sqrt{2}\varphi}. (A21) The full background equations admit a closed-form rewriting in the invariant coordinate, \partial_{\mathcal{N}}^{2}X+\Bigl(3-\epsilon-2\partial_{\mathcal{N}}\ln s\Bigr)\partial_{\mathcal{N}}X+\frac{4Am^{2}}{H^{2}}\bigl(X-X_{v}(s)\bigr)=0, (A22) where \mathcal{N}\equiv\ln a_{\rm FRW} denotes e-fold time and \epsilon=\frac{(\partial_{\mathcal{N}}a)^{2}+(\partial_{\mathcal{N}}s)^{2}}{4s^{2}}. (A23) In terms of the canonical heavy fluctuation X=X_{v}+\sqrt{2}ms\chi , one finds m_{s}^{2}=4Am^{2}+\mathcal{O}(s^{-4}), (A24) which realizes the saturation case p=2 . For the benchmark values (\beta,A,B,m,e)=\left(1,10^{-10},2\times 10^{-11},1,0\right),\qquad V_{0}=3.9187164325\times 10^{-11}, (A25) the reduced single-field valley EFT gives \displaystyle N_{*}=50: \displaystyle\quad(n_{s},r)=(0.96083,1.5046\times 10^{-3}), (A26) \displaystyle N_{*}=55: \displaystyle\quad(n_{s},r)=(0.96434,1.2493\times 10^{-3}), (A27) \displaystyle N_{*}=60: \displaystyle\quad(n_{s},r)=(0.96727,1.0540\times 10^{-3}) (A28) with m_{s}^{2}/H^{2}\simeq 30.9 across the CMB window and \Omega/H\lesssim 1.4\times 10^{-7} . The benchmark requires only a modest hierarchy within the local flux expansion. The ratio B/A=0.2 fixes the odd-to-even branch weight, V_{0} fixes the observed scalar amplitude, and in a compact embedding the allowed coefficient ratios are discretized by flux quantization rather than scanned as arbitrary continuous numbers. Full two-field integrations rapidly approach the analytic valley and produce only small numerical shifts, consistent with adiabatic decoupling and the negligible turn rate. Appendix D Complete next penumbral order and documented scan The complete local monodromy-preserving EFT through the next penumbral order, still truncated at quadratic order in X , is V_{\rm NLO}=V_{0}\Bigl(1-\frac{\beta}{s}+\frac{\lambda}{s^{2}}\Bigr)^{2}+\Bigl(\frac{A}{s^{2}}+\frac{D}{s^{3}}\Bigr)X^{2}+\Bigl(\frac{B}{s^{3}}+\frac{C}{s^{4}}\Bigr)X, (A29) with the leading saxionic metric deformation ds_{\rm field}^{2}=\frac{1}{2s^{2}}\Bigl[(1+\gamma_{a}/s)\,da^{2}+(1+\gamma_{s}/s)\,ds^{2}\Bigr]. (A30) The heavy branch remains analytic, X_{v}(s)=-\frac{Bs+C}{2s(As+D)}=-\frac{B}{2A}s^{-1}+\Bigl(-\frac{C}{2A}+\frac{BD}{2A^{2}}\Bigr)s^{-2}+\mathcal{O}(s^{-3}), (A31) and the reduced potential is U_{\rm NLO}(s)=V_{0}\Bigl(1-\frac{\beta}{s}+\frac{\lambda}{s^{2}}\Bigr)^{2}-\frac{(Bs+C)^{2}}{4s^{5}(As+D)}. (A32) Thus the leading plateau coefficient is unchanged. The effective kinetic map also stays logarithmic at leading order, \frac{d\varphi}{ds}=\frac{1}{\sqrt{2}s}\left(1+\frac{\gamma_{s}}{2s}+\mathcal{O}(s^{-2})\right), (A33) so metric deformations move only subleading coefficients in the exponentiated potential. We scan the dimensionless deformations \hat{\lambda}\in[-0.35,0.35],\qquad\hat{C}\in[-1,1],\qquad\hat{D}\in[-0.55,0.55],\qquad\hat{\gamma}_{s}\in[-0.8,0.8], (A34) where \hat{\lambda}\equiv\frac{\lambda}{\beta^{2}},\qquad\hat{C}\equiv\frac{C}{Bs_{p}},\qquad\hat{D}\equiv\frac{D}{As_{p}},\qquad s_{p}=225. (A35) In the scan shown here we set \gamma_{a}=0 , since it only renormalizes the heavy-sector kinetic normalization and does not modify the leading plateau data. Models are retained only if the reduced potential is positive, the effective kinetic coefficient is positive, 50 – 60 e-fold inflation exists, and m_{s}^{2}/H^{2}>5 throughout the CMB window. Among the accepted models we obtain at N_{*}=55 0.96386\leq n_{s}\leq 0.96503,\qquad 1.20\times 10^{-3}\leq r\leq 1.29\times 10^{-3}, (A36) with representative control values m_{s}^{2}/H^{2}\gtrsim 34 and \Omega/H\lesssim 10^{-11} . The same accepted set gives a narrow running corridor centered on the leading q=1 relation \alpha_{s}\simeq-r/2 , which is why the main-text observational bundle remains compact once the local EFT constraints are imposed. Figure 4: Predictive stability and observational corridor. (a) Slow-roll tracks for the complete next-order monodromy-preserving EFT and mild metric deformations; the benchmark is the thick black curve. (b) The same models in the (r,\alpha_{s}) plane compared with the leading q=1 relation \alpha_{s}=-r/2 . After control is imposed, the local EFT predicts a compact observational corridor rather than a broad attractor swath. Figure 5: Control envelopes for the accepted scan. The upper envelopes of \Omega/H remain tiny throughout the CMB window. Appendix E String-theoretic origin, scope, and additional diagnostics Type-IIB origin of the local ansatz. Near a one-parameter Calabi–Yau complex-structure boundary the period vector is controlled by a nilpotent-orbit expansion, so the prepotential and periods are polynomial in the local modulus z=a+is up to exponentially small corrections. After fixing the axio-dilaton and using the no-scale complex-structure potential, one obtains a scalar potential built from inverse powers of s multiplying monodromy-invariant axion polynomials 21 ; 23 ; 20 ; 32 . A convenient way to state the general local structure is V_{\rm flux}(a,s)=\mathcal{G}_{IJ}(s)\,\rho^{I}(a)\rho^{J}(a),\qquad\mathcal{G}_{IJ}(s)\sim s^{-n_{IJ}}, (A37) where the \rho^{I} are invariant under the axionic monodromy action on a chosen branch. Expanding those polynomials around a local branch coordinate X\equiv e-ma gives \rho^{I}(X)=\rho_{0}^{I}+\rho_{1}^{I}X+\rho_{2}^{I}X^{2}+\cdots, (A38) and therefore V_{\rm flux}(X,s)=V_{\rm const}(s)+B(s)X+A(s)X^{2}+\mathcal{O}(X^{3}), (A39) with V_{\rm const}(s)=\mathcal{G}_{IJ}\rho_{0}^{I}\rho_{0}^{J},\qquad B(s)=2\mathcal{G}_{IJ}\rho_{0}^{I}\rho_{1}^{J},\qquad A(s)=\mathcal{G}_{IJ}\rho_{1}^{I}\rho_{1}^{J}+\cdots. (A40) Because the matrix \mathcal{G}_{IJ}(s) carries the asymptotic saxionic weights, this local branch expansion generically yields an X -independent sector together with linear and quadratic heavy terms whose coefficients scale as inverse powers of s . Equation (A1) of the Letter is therefore the natural local branch expansion of a Type-IIB flux potential in the penumbral regime. The odd term appears whenever the chosen branch has a nonzero invariant offset \rho_{0}^{I} and a nonzero linear monodromy component \rho_{1}^{I} ; monodromy permits it, while symmetry or flux tuning may remove it. What remains for an explicit compactification. The derivation above establishes the structure of the local EFT. A full top-down embedding still requires a concrete Calabi–Yau orientifold, a flux choice, stabilization of the axio-dilaton and Kähler sector, and an explicit extraction of the coefficients (V_{0},c_{q},A_{p},B_{p+\nu}) . The local theorem controls adiabatic decoupling of the orthogonal mode inside the complex-structure EFT; the full Distance-Conjecture tower remains an additional compactification-level constraint. Any explicit embedding must therefore satisfy m_{\rm tower}(\varphi)\gg H(\varphi),\qquad\Lambda_{\rm EFT}(\varphi)\gg H(\varphi), (A41) throughout the CMB window. The scope of the Letter is deliberately local: before global completion is attempted, the penumbral branch data already decide whether a given valley is a controlled inflationary EFT worth pursuing. Tameness, and the scope of the inflationary claim. The Letter isolates a finite-distance inflationary window inside the penumbra, not an inflationary solution in the strict asymptotic regime. This finite-distance scope is compatible with the tame-geometry logic emphasized in Refs. 23 ; 20 ; 14 . On a penumbral patch U=\{|a|<1,\ s>s_{0}\}, (A42) the local ingredients used in the Letter are definable in the same o-minimal structures relevant for complex-structure sectors: inverse powers of s , logarithms, exponentials of \log s , restricted analytic functions, and periodic functions of the compact axion. In particular, s^{-n}=\exp[-n\log s] (A43) is tame on s>s_{0}>1 , while \cos(\omega a+\delta) is tame because a remains compact. Ref. 31 identified long uplift-like penumbral valleys and clarified why such valleys need not yet yield full slow-roll inflationary solutions. The present Letter sharpens that analysis by identifying the displaced-minimum criterion and heavy-sector inequality that decide when a finite-distance penumbral window is already a controlled local inflationary EFT. The Letter uses the covariant entropy mass rather than a shortcut mass because the field space is curved. Figure 6 shows that the projected entropy mass tracks the heavy eigenvalue of the full covariant mass matrix. The same figure also shows representative oscillatory envelopes along the valley. For corrections of the form \delta U_{\rm osc}\propto s^{-\gamma}\cos(\omega a+\delta) , the cases \gamma>1 are parametrically subleading to the leading q=1 plateau term, while \gamma=1 only renormalizes the coefficient of the leading exponential. Figure 6: Covariant entropy-mass diagnostics and oscillatory envelopes. (a) m_{s}^{2}/H^{2} , the heavy eigenvalue \lambda_{+}/H^{2} , the light eigenvalue, and the shortcut approximation 4Am^{2}/H^{2} . (b) Representative oscillatory envelopes normalized to the leading plateau term. 43