Counterterms Are Not the Answer: Coherence, Descent, and Global Consistency in Higher-Dimensional Holography

Counterterms Are Not the Answer: Coherence, Descent, and Global Consistency in Higher-Dimensional Holography Andrei T. Patrascu 1 1 FAST Foundation, Destin FL, 32541, USA email: andrei.patr[email protected] Recent progress in higher–dimensional holography has demonstrated that physically relevant configurations—including bubbling geometries and defect backgrounds—often lie outside the scope of any finite–field AdS d+1 truncation, forcing renormalization to be performed directly in ten or eleven dimensions. In such settings, finite observables are obtained only after introducing specific cutoff prescriptions and higher–dimensional boundary counterterms, whose justification is typically validated a posteriori by agreement with known field–theoretic data. While successful as computations, these constructions remain intrinsically prescription–based: they provide neither a criterion for the uniqueness of the counterterms employed nor a principle governing transport between different regulator choices, and they leave the physical meaning of the higher–dimensional cutoff surface itself unresolved. In this work we argue that this situation reflects a missing global consistency structure rather than a technical limitation of holographic renormalization. We show that choices of cutoff, asymptotic presentation, and subtraction scheme naturally assemble into a context groupoid, whose morphisms encode admissible changes of regulator. Within this framework, counterterms acquire a precise interpretation as coherence (descent) data required to glue patchwise–defined renormalized functionals into a global object. Global consistency requires coherence under composition of context changes, but does not require strictification: a coherent but non–strictifiable assignment already defines a well–posed global object. We prove that strictification is possible if and only if the coherence data is trivial; when strictification fails, the remaining obstruction class is invariant and is identified with the anomaly. Applying this perspective to recent ten–dimensional computations of defect anomalies, we demonstrate that the appearance of ad hoc higher–dimensional boundary counterterms is neither accidental nor arbitrary, but reflects the presence of a genuine obstruction to strictification rather than a failure of patching. Our results supply the organizing principle missing from purely computational approaches and provide a sharp criterion distinguishing removable scheme dependence from intrinsic, physically meaningful anomalies in higher–dimensional holography. I. EXECUTIVE SUMMARY FOR PHYSICISTS The purpose of this paper is not to introduce new computational techniques, but to isolate a missing consistency principle that has become unavoidable in higher–dimensional holography. The recent proliferation of prescription–dependent renormalization schemes is not a temporary technical inconvenience; it is a structural signal. This section summarizes the diagnosis and the remedy in the most economical terms. 2 •Regulator choices are not gauge unless functorial transport is specified. Changing a cutoff surface, coordinate prescription, or subtraction scheme is not a redundancy by default. Such changes must be accompanied by a well–defined transport rule relating renormalized quantities across schemes. Absent such a rule, “scheme independence” is an empirical coincidence rather than a structural statement. •Counterterms supply gluing data, not merely divergence cancellation. In higher– dimensional holography, counterterms are not fixed solely by local power counting. Their essential role is to relate patchwise–defined renormalized functionals across different regulator choices. Interpreted correctly, counterterms provide the coherence data required to glue local definitions into a global object. •Anomalies are obstructions, not leftovers. Universal finite terms that survive all admissible subtractions do not signal an incomplete renormalization. They signal the failure of global strictification: a genuine obstruction to defining a regulator–independent observable. Anomalies are therefore invariant obstruction classes, not artifacts of poor scheme choices. •Higher–dimensional holography forces this issue into the open. When holographic observables are sensitive to intrinsically ten– or eleven–dimensional structure, no finite lower– dimensional truncation can supply a universal renormalization dictionary. The dependence on cutoff geometry and boundary prescriptions is unavoidable and must be organized, not eliminated. Taken together, these points lead to a simple conclusion: higher–dimensional holography requires a global notion of consistency that goes beyond local counterterm subtraction. The absence of such a notion explains why recent computations succeed only after introducing apparently ad hoc boundary terms, and why the meaning and uniqueness of these prescriptions remain obscure. This paper supplies the missing consistency principle. We show that regulator choices naturally assemble into a context groupoid, that counterterms function as coherence (descent) data on this groupoid, and that anomalies arise precisely as obstruction classes to global strictification. This framework converts prescription–dependent renormalization into a well–posed structural problem and sharply distinguishes removable scheme dependence from intrinsic physical invariants. II. THE SYMPTOM: WHEN HOLOGRAPHY STOPS BEING UNIVERSAL The standard holographic dictionary is often presented as if it were a universal renormalization machine: choose Fefferman–Graham gauge, introduce a radial cutoff, add local counterterms determined by covariance and power counting, remove the regulator, and obtain a finite generating functional whose finite part is unambiguous up to the usual scheme choices [ 2 – 4 ]. This picture is accurate in the regime for which it was engineered: asymptotically AdS d+1 solutions of a fixed lower–dimensional effective theory (or a consistent truncation) with a controlled asymptotic expansion. The symptom addressed in this paper is that a rapidly growing class of physically relevant holographic observables lives outside that regime, and the traditional notion of “scheme dependence” ceases to be the right organizing concept. A. Failure of effective AdSd+1 truncations The breakdown is not primarily computational; it is conceptual. Many of the most informative probes of AdS/CFT—defects, heavy operators, and bubbling geometries—are not naturally described as solutions of a finite–field AdS d+1 bulk theory. Rather, they are intrinsically ten– or eleven–dimensional supergravity solutions whose asymptotics approach AdS d+1 ×Mq but whose defining data and global structure do not reduce to a lower–dimensional truncation [ 1 , 5 ]. In such backgrounds, the part of the on–shell action that is required for certain observables is not controlled by the universal near–boundary expansion of a small set of fields. It is sensitive to genuinely higher–dimensional information: global topology, internal flux data, and the detailed way the geometry “fills in” away from the asymptotic region. This matters because many observables of interest are finite observables. They are not merely ratios in which divergences cancel, nor are they protected by a reduction to local boundary data. They probe 3 the renormalized on–shell action (or its logarithmic terms) itself, and therefore they probe the global consistency of the renormalization prescription, not just the cancellation of power divergences. In other words: the regime that is traditionally called “holographic renormalization” implicitly assumes that the relevant observable is already expressible in the language of a universal AdS d+1 effective theory. Once that assumption fails, universality is no longer guaranteed by the formalism. B. The rise of prescription–based renormalization A clear manifestation of the new regime is the appearance of renormalization procedures that succeed only as prescriptions. A representative example is the recent ten–dimensional computation of the Euler conformal anomaly coefficient for half–BPS surface operators in N = 4 SYM, obtained by evaluating the on–shell Type IIB supergravity action on the relevant bubbling geometry while introducing a specific higher–dimensional boundary counterterm and cutoff prescription [ 1 ]. This work is technically successful and conceptually revealing: it reproduces an observable known exactly in field theory and highlights that the corresponding information is not accessible within any finite–field AdS5truncation [1]. At the same time, the logic of such constructions is irreducibly procedural. The renormalized answer depends on a chain of choices: •a choice of cutoff surface (in the higher–dimensional geometry), • a choice of asymptotic presentation/coordinates (which determines what counts as “radial” and what data are held fixed), • a choice of higher–dimensional boundary counterterms (not fixed by the usual AdS d+1 power– counting logic), • and a normalization/subtraction convention (often calibrated by vacuum subtraction or matching to a protected CFT quantity). In this regime, “scheme independence” cannot be asserted by citing covariance of the near–boundary expansion alone. It can only be tested by comparison with independent field–theory input or by repeating the computation under a different set of choices and checking that the answers agree. That is not a renormalization principle; it is an empirical verification loop. This is the precise sense in which recent higher–dimensional holographic renormalizations are computations without a controlling mechanism: they show the existence of a prescription that works, but they do not provide (and often explicitly set aside) a criterion that would decide which prescriptions are admissible, when they are equivalent, and why a given counterterm is compelled rather than merely convenient. In particular, the physical meaning of the higher–dimensional cutoff surface and the classification/uniqueness of allowed boundary terms remain opaque in purely computational approaches [1]. C. The key observation The core observation is simple and sharp: The problem is no longer ultraviolet divergence; the problem is comparison across schemes. In the traditional AdS d+1 setting, one may treat the regulator as a technical scaffold because the renormalized functional is controlled by a universal asymptotic expansion and the admissible counterterms are classified by locality and covariance [ 2 – 4 ]. In higher–dimensional holography beyond truncations, by contrast, the regulator choice is part of the definition of the computation unless one also specifies a rule for transporting renormalized data between different choices. Without such transport, “scheme dependence” is not a small ambiguity; it is a failure to state what is being compared. This paper is built around the following diagnosis. The correct object to organize higher–dimensional holographic renormalization is not a single preferred cutoff prescription, but the space of prescriptions together with their admissible changes. We will formalize this as a context groupoid: objects are regulator contexts (cutoff surface, asymptotic presentation, subtraction/normalization, allowed boundary terms), and morphisms are admissible changes of context. In this language, counterterms are not merely subtraction devices; they supply the coherence (gluing) data needed to relate patchwise definitions across contexts. Anomalies appear precisely when this coherence cannot be strictified globally, leaving 4 an invariant obstruction class. The rest of the paper makes this statement precise and proves the corresponding global consistency criterion. III. WHAT A COMPUTATION ALONE CANNOT TELL YOU The higher–dimensional results motivating this paper are not “wrong,” nor are they to be dismissed as technical exercises. On the contrary, they deliver an important fact: certain observables can be computed holographically only by leaving the comfort zone of finite-field AdS d+1 truncations and working directly in ten or eleven dimensions. A representative example is the ten–dimensional evaluation of the Type IIB on–shell action on a bubbling geometry dual to a half–BPS surface operator, which reproduces the Euler anomaly coefficient by supplementing the action with a specific higher–dimensional boundary term and a carefully chosen cutoff prescription [1]. This is a genuine success. The point of the present section is surgical: to separate what such a computation establishes from what it cannot, by itself, establish. The distinction is not rhetorical. It is the difference between a result that is a one–off calibration and a result that is controlled by an internal principle. A. Existence is not mechanism A computation can establish existence: There exists a prescription (cutoff choice, coordinate presentation, boundary term) that yields a finite answer matching the expected field–theory quantity. This existence statement is already nontrivial in higher–dimensional holography beyond truncations, and it is exactly what is achieved in [1]. However, existence is not a mechanism. It does not tell you whether the prescription is compelled or accidental, canonical or engineered, stable or fragile. In the traditional AdS d+1 setting, this gap is largely closed because the admissible counterterms are classified by locality and covariance and the asymptotic expansion provides a universal control parameter [ 2 – 4 ]. Outside that regime, the gap reappears, and it is precisely the gap that produces the sensation of “ad hoc counterterms.” B. Four questions a computation cannot answer Once one leaves a universal truncation, a renormalized answer comes with a hidden burden: one must control the space of permissible regulator choices and their relations. A computation, by itself, cannot provide that control. Concretely, it cannot answer the following four questions. (i) Why is this prescription required? A working prescription may be one point in a large space of possible prescriptions. Without a principle, there is no distinction between a counterterm that is forced by consistency and one that is merely sufficient to match a known datum. In [ 1 ] the higher–dimensional boundary term is introduced because it makes the computation work; the deeper reason it should be present (and in what precise sense it is the right object) is not supplied by the computation itself [1]. (ii) Is the prescription unique? Even if one accepts the need for additional higher–dimensional boundary terms, one must still decide whether the choice is unique under a clear equivalence relation. In the absence of a classification of admissible boundary terms in the higher–dimensional setting, “uniqueness” reduces to a practical question: can one modify the prescription without changing the result? Pure computation answers this only by repeated trial. The conceptual task is to define the equivalence relation and show whether the answer is invariant under it. That task is not part of the computational pipeline. (iii) How does it generalize? A prescription calibrated on one class of backgrounds does not automatically transport to another. In particular, once the relevant observable depends on global bulk structure, the question of generalization becomes a question of transport: how do choices made in one context map to another context, and what data must be carried along? The ten–dimensional computation of [ 1 ] exhibits the need for such transport precisely by confronting the mismatch between the four–dimensional locus supporting the defect CFT data and the higher–dimensional cutoff surface used in the renormalization [1]. But again: exhibiting the need is not the same as providing the transport law. 5 (iv) What invariant is being measured? Matching a protected field–theory quantity identifies a number, but it does not identify the structural status of that number. Is it an intrinsic invariant of the holographic setup, or an artifact of the chosen regulator context? Is it stable under admissible changes, or only under the particular changes that were tried? In the traditional framework, universal terms (notably anomalies) are understood as invariant data because the space of admissible local counterterms is classified and the remaining ambiguity is controlled [ 2 , 4 ]. In the higher–dimensional regime highlighted by [ 1 ], the computation reproduces the anomaly coefficient, but the computation alone cannot supply the structural argument that this coefficient is the invariant remainder of a well–posed comparison problem rather than the residue of a convenient subtraction scheme [1]. C. The clinical conclusion The preceding four questions are not optional philosophical upgrades. They are the minimal conditions under which “scheme dependence” becomes a controlled statement rather than a slogan. The recent higher–dimensional computations are therefore best understood as high–quality symptoms: they show clearly where the traditional universality story fails, and they pinpoint the exact location where additional structure must enter. What is missing is a global notion of consistency. IV. CONTEXTS ARE NOT GAUGE: THE MINIMAL STRUCTURE The prescription–dependence encountered in higher–dimensional holography is often discussed in the loose language of “scheme choices.” That language is too weak for the regime of interest. A scheme is usually treated as an auxiliary choice that can be changed without conceptual consequence. In the present setting, by contrast, the regulator choice becomes part of the definition of the computation unless one also specifies how results are transported between choices. The minimal structure required to make this precise is small: one needs only a notion of context and a notion of admissible change of context. The rest is bookkeeping—but it is the bookkeeping that turns a computation into a principle. A. Definition: context Acontext is the full set of data required to state a renormalized holographic observable as a well–posed problem. Concretely, a context consists of the following components: 1. Cutoff surface. A choice of regulating hypersurface Σ  in the bulk geometry (or family thereof) on which boundary conditions are imposed and on which boundary terms are evaluated. In higher– dimensional setups this surface may not coincide with a canonical “radial” slice and may depend nontrivially on internal coordinates. 2. Asymptotic presentation. A choice of asymptotic coordinate system and field parametrization used to identify the regulated region and to define what it means to hold boundary data fixed. In AdS d+1 this is often encoded in Fefferman–Graham gauge; beyond truncations, the relevant presentation data can be more intricate. 3. Subtraction/normalization scheme. A choice of reference background and normalization convention (e.g. vacuum subtraction, choice of counterterm renormalization scale, or other calibration) used to define the finite part. 4. Allowed boundary terms. A specification of the admissible local (or quasi–local) boundary functionals that may be added on Σ  as counterterms, including the symmetry and covariance requirements they must satisfy. 5. Boundary conditions. A specification of the boundary conditions imposed on Σ  (Dirichlet/Neumann/mixed, ensemble choice, and any additional constraints required for a well–posed variational principle). 6 We denote a context by a symbol such as c , with the understanding that c packages the above five ingredients. Two remarks are essential. First, the point is not to proliferate data, but to acknowledge data that is already present implicitly in any actual computation. Second, in the higher–dimensional regime exemplified by [ 1 ], several of these ingredients are not fixed by the lower–dimensional AdS d+1 dictionary; they must be specified independently. This is precisely why one encounters apparently ad hoc boundary counterterms and cutoff prescriptions [1]. B. Allowed changes of context Given two contexts c and c0 , a change of context is an admissible map that relates the corresponding regulated setups. The word “admissible” is doing the work: we only allow changes that preserve the class of problems under consideration (symmetries, variational well–posedness, and the notion of fixed boundary data). The basic moves are: 1. Coordinate and field redefinitions. Changes of asymptotic coordinates and field variables that preserve the asymptotic class of the solution and the identification of sources/VEVs. In the AdS d+1 setting these include the familiar PBH transformations; in higher dimensions they include reparametrizations mixing “radial” and internal directions, provided they preserve the admissibility criteria. 2. Cutoff deformations. Deformations of the regulating surface Σ 7→ Σ 0  within the admissible family. Physically, these are changes of how one slices the bulk near the boundary (or near a degenerate boundary), and they generally induce changes in the induced metric and other boundary fields on the cutoff surface. 3. Induced counterterm shifts. Since the cutoff surface and presentation data change, the allowed boundary terms must transform accordingly. Even when the space of admissible counterterms is fixed abstractly, a change of context generically induces a shift in the representative counterterm functional used to define the finite part. A change of context therefore does not mean “pick a different regulator and hope nothing changes.” It means: specify a map between regulators and carry along the induced transformation of all contextual data. In particular, if a computation yields the same finite answer under two different contexts, this is meaningful only if the contexts are connected by an admissible change and if the corresponding transport is properly accounted for. C. Contexts form a groupoid The collection of contexts together with admissible changes of context naturally forms a groupoid. We denote this groupoid by Cand summarize its structure as follows: •Objects: contexts c (cutoff, presentation, subtraction, admissible boundary terms, boundary conditions). •Morphisms: admissible changes of context φ:c→c0. •Composition: concatenation of admissible changes, ψ◦φ:c→c00. •Inversion: each admissible change is reversible at the level of contextual data (at least locally in the space of contexts), yielding φ−1:c0→c. The groupoid viewpoint is the minimal mathematical upgrade required to make “scheme dependence” a controlled statement. It records not only the existence of multiple contexts, but also the admissible relations between them. In particular, loops in C encode the possibility of returning to the “same” context by different routes, a mechanism by which nontrivial obstruction data (anomalies) can arise. Key message. If you do not track morphisms, you cannot talk about invariance. 7 A renormalized observable is not merely a number assigned to one preferred regulator choice. It is, at minimum, an assignment that is stable under the admissible changes of context encoded by C . The remainder of the paper makes this statement precise: counterterms will be identified as the coherence data governing transport along morphisms, and anomalies will emerge as the obstruction to global strictification over the context groupoid. V. PATCHWISE DEFINITIONS VERSUS GLOBAL OBJECTS With the context groupoid C in place, one can state the central issue in one line: higher–dimensional holographic renormalization currently produces patchwise finite answers, while physics requires a global object. The gap between “patchwise” and “global” is not semantics. It is exactly where counterterms acquire their real meaning and where anomalies arise. Throughout this section we deliberately avoid categorical jargon. The only concepts used are the ones holographers already employ informally: local definitions, comparisons on overlaps, and consistency under repeated comparison. A. Patchwise renormalized functionals Fix a context c∈Ob ( C ), i.e. a choice of cutoff surface, asymptotic presentation, subtraction convention, allowed boundary terms, and boundary conditions. A higher–dimensional holographic computation then produces a regulated functional Sreg c (  )(for example, the on–shell action with boundary terms evaluated on the cutoff surface Σ  ), and after adding counterterms and subtracting reference contributions one extracts a finite quantity, which we denote schematically by Sren c:= lim →0Sreg c() + Sct c()−Sref c().(1) The defining feature is not the formula but the logic: the finite answer Sren c is produced relative to c . Change c and one typically changes intermediate steps and, in general, may change the finite remainder unless additional structure is imposed. This is precisely the situation in higher–dimensional computations beyond truncations: one obtains a clean finite answer in a chosen context, but one has not yet earned the right to call it context–independent. The output of the computation is therefore best viewed as a family of finite answers indexed by contexts, c7−→ Sren c.(2) Physics, however, does not ask for a family indexed by regulator choices. It asks for an observable. B. Overlaps and transition data Suppose now that two contexts c and c0 are related by an admissible change of context φ : c→c0 (a morphism in C ). Then φ specifies how cutoff surfaces, asymptotic data, and boundary conditions are transported. Crucially, it must also specify how the renormalized functionals are compared. In general, Sren c and Sren c0 are not equal as raw functionals. They may differ by a finite, local boundary term (or, more generally, by a controlled functional determined by the admissibility criteria). We encode this by introducing transition data T ( φ )assigned to each admissible change φ , such that the comparison takes the form Sren c0=Sren c+T(φ).(3) Equation (3) should be read as a definition of the problem, not as an identity that automatically holds. It states: to compare two contexts, one must specify extra data T(φ). This is the first incision: Counterterms do not merely cancel divergences; they live in the transition data. The usual view of counterterms as subtraction devices corresponds to focusing on a single context and enforcing finiteness of (1) . But once one asks for a context–independent object, counterterms acquire a 8 second role: they control how finite parts shift under admissible changes of regulator. In other words, counterterms are not an afterthought—they are the glue that relates patchwise finite answers across the space of contexts. In practice, higher–dimensional computations often fix T ( φ )implicitly by choosing one preferred context and calibrating it against field theory. That produces one Sren c with a desired value, but it does not produce a systematic rule for T ( φ )for all admissible φ . The rule is what turns the family (2) into an observable. C. Triple comparisons and consistency The decisive consistency check appears as soon as one compares three contexts. Let c0,c1,c2 be contexts connected by admissible changes φ01 :c0→c1, φ12 :c1→c2, φ02 :c0→c2, with φ02 comparable to the composed change φ12 ◦φ01 . If the transition rule is coherent, then comparing c0to c2directly or in two steps must agree: T(φ02) = T(φ12 ◦φ01)and hence T(φ12 ◦φ01) = T(φ12) + T(φ01).(4) Equation (4) is the minimal consistency condition: transport in two steps must equal transport in one step. Nothing more sophisticated is required to see its force. Once (4) fails, the comparison of finite answers becomes path–dependent. The sharpest way to expose the pathology is to consider a loop in the context groupoid. Let γ be a sequence of admissible changes that starts and ends at the same context c: c φ1 −→ c1 φ2 −→ · · · φn −−→ c, γ := φn◦ · · · ◦ φ1. If the renormalized object were genuinely context–independent, transporting around the loop would produce no net shift: T(γ)! = 0.(5) When (5) fails, the failure is not a computational error. It is an invariant residue of the comparison problem: Failure to close around loops is an obstruction. This obstruction is precisely what is measured by anomalies in the situations we care about. The point is structural: anomalies appear when one cannot choose counterterms (i.e. transition data) such that all admissible comparisons are simultaneously consistent. This section has done only three things. First, it has rephrased the output of higher–dimensional holographic renormalization as a family (2) of patchwise finite answers. Second, it has identified counterterms as the transition data (3) needed to compare those answers across contexts. Third, it has isolated the minimal consistency condition (4) and its loop test (5) , which is where obstructions live. The next section turns these statements into a precise criterion: when the transition data can be chosen coherently, a global renormalized object exists; when it cannot, the residual obstruction class is the anomaly. VI. MAIN RESULT: COHERENCE AND OBSTRUCTION Sections IV–V isolate the problem with surgical clarity: higher–dimensional holographic renormalization beyond universal truncations produces finite answers only relative to a context, and without a transport law between contexts there is no meaning to “scheme independence.” This section supplies the missing principle. It does not add another prescription. It makes prescriptions unnecessary by turning the comparison problem into a structural criterion. The core upgrade is this: 9 Counterterms are coherence data on the context groupoid, and anomalies are the obstruction to strictifying that data globally. This is precisely what purely computational approaches cannot provide: they can exhibit a working counterterm in one context, but they cannot decide whether the counterterm is forced, whether two prescriptions are equivalent, or what invariant is being measured. Here we give the mechanism. A. Counterterms as coherence data Fix the context groupoid C from Section IV C. For each context c∈Ob ( C )we have a patchwise renormalized functional Sren c defined by some regularization, counterterm subtraction, and normalization convention. To compare Sren c across contexts, we introduced transition data T ( φ )assigned to each morphism φ:c→c0: Sren c0=Sren c+T(φ).(6) The function φ7→ T ( φ )is not decoration; it is the minimal data needed to interpret the collection {Sren c} as describing a single object. In practice, T ( φ )is implemented by the finite pieces of boundary counterterms and normalization choices that must accompany a change of cutoff surface and asymptotic presentation. The essential point is that T(φ)is constrained: 1. It must be admissible: compatible with the definition of the context class (symmetries, locality/covariance requirements, and well–posed boundary conditions). 2. It must be coherent: consistent under composition of admissible changes of context. The first constraint is what physicists usually mean by “allowed counterterms.” The second constraint is what is missing from purely local discussions: it forces counterterms to behave as gluing data rather than as arbitrary subtractions. Concretely, coherence requires that when two morphisms compose, c φ −−→ c0ψ −−→ c00,(7) the induced comparison from c to c00 be independent of whether one compares in one step or two steps. This is the additive law T(ψ◦φ) = T(ψ) + T(φ),(8) already anticipated in (4) . Equation (8) is the strict form of coherence: it says that T is compatible with composition. First incision (principle versus prescription). A purely computational construction can pick one counterterm functional and verify that it cancels divergences and yields the desired finite part in one context. It cannot decide whether the induced transition data satisfies (8) on the full class of admissible morphisms. That decision requires posing renormalization as a global comparison problem. That is the meat supplied here. B. The strictification problem Even if one has coherent transition data T ( φ ), there remains the question of whether the context dependence can be eliminated by a redefinition of the patchwise functionals Sren c . This is the strictification problem. Consider shifting each patchwise functional by a finite counterterm (a “choice of representative”) depending only on the object c: Sren c7−→ e Sren c:= Sren c+B(c),(9) where B ( c )is an admissible finite boundary functional in the class allowed by the context definition. Under such a shift, the transition data changes as T(φ)7−→ e T(φ) := T(φ) + B(c0)−B(c), φ :c→c0.(10) 16 your transition data is coherent. If it fails, you have detected path dependence: the “observable” depends on how you move through context space. That path dependence is not a nuisance. It is the anomaly mechanism. Output of Step 4. A yes/no answer: •YES: Tis coherent; proceed to strictification. •NO: coherence fails; extract the obstruction from loops (Step 5). F. Step 5: If closure fails, compute the anomaly as a loop invariant If (16) fails, do not “fix” it by adding more ad hoc counterterms. Instead, compute the obstruction by evaluating T on loops. Pick a set of generating loops γ in C (typically arising from nontrivial relations between cutoff deformations and coordinate changes) and compute A(γ) := T(γ).(17) By Theorem VI.1, A ( γ )is invariant under admissible object–wise redefinitions and therefore defines a physical obstruction class. This is the operational definition of an anomaly in our framework: it is the strictification–invariant residue of a comparison problem, not a subtraction failure. Output of Step 5. A set of loop invariants A ( γ ). In defect holography, the Euler coefficient extracted in [ 1 ] is precisely of this type: it is stable not because one found a lucky subtraction, but because it is the invariant remainder of the context comparison problem. G. Step 6: If closure holds, strictify and obtain a global observable If (16) holds, compute an object–wise shift B(c)such that e T(φ) = T(φ) + B(c0)−B(c)=0 for all φ:c→c0.(18) This is strictification. It produces a globally defined renormalized functional e Sren independent of context within the admissible class. This is the point at which “scheme independence” becomes a theorem rather than a claim. No amount of single–context counterterm engineering can substitute for this step, because strictification is a global statement. Output of Step 6. A global renormalized observable (a strictified representative) together with a proof of its invariance under all admissible context changes. H. The scalpel conclusion The recipe can be summarized in a single line: Build the groupoid, compute the transitions, check coherence, and read off either a strictified observable or an obstruction class. This is how one upgrades higher–dimensional holographic renormalization from “find a counterterm that works” to “classify what is removable and what is invariant.” It is precisely what computation–only approaches cannot do, and it is exactly how past results—including [ 1 ]—can be systematized, sharpened, and pushed beyond the regime where calibration by known field–theory data is available. IX. SCOPE, LIMITS, AND PREDICTIONS The framework developed in this paper is deliberately narrow in one sense and deliberately strong in another. It is narrow because it does not attempt to replace explicit holographic calculations. It is strong because it decides, before and independently of detailed computation, what kind of answer one should expect. This section makes that separation precise and spells out the predictive content. 17 A. What this framework does not do First, a clear boundary. • This framework does not compute on–shell actions, correlation functions, or anomaly coefficients by itself. • It does not eliminate the need for detailed bulk analysis, supersymmetry constraints, or careful treatment of boundary conditions. • It does not replace the traditional machinery of holographic renormalization where that machinery is sufficient. Any claim to the contrary would be misguided. Explicit calculations remain indispensable, and nothing here diminishes their technical or conceptual value. What this framework replaces is something else: the idea that renormalization prescriptions can be justified post hoc by matching a known answer. It replaces calibration by principle. B. What this framework does decide The decisive gain is that the framework determines what can be made canonical. Given a class of admissible contexts and admissible changes between them, Theorem VI.1 answers a question that computation alone cannot even formulate: Is there a globally well–defined renormalized observable associated to this problem, or is any finite answer necessarily the residue of an obstruction? This distinction has concrete consequences. • If strictification is possible, then any two admissible prescriptions are equivalent, and differences can be removed by object–wise redefinitions. In this case “scheme dependence” is spurious and can be eliminated once and for all. • If strictification fails, then no amount of counterterm engineering will produce a globally invariant object. The remaining finite quantity is physical precisely because it is unavoidable. This is not a matter of taste or convention. It is a structural statement about comparison across regulator contexts. C. Predictions Because the framework is structural, its predictions are qualitative but sharp. They concern when certain phenomena must appear, not their numerical values. When counterterms must appear. Additional counterterms beyond the standard AdS d+1 list are unavoidable whenever the admissible context space is larger than the space controlled by a universal near–boundary expansion. In practice, this occurs precisely when the observable probes global bulk data that cannot be encoded in a finite–field truncation. In such cases, higher–dimensional boundary terms are not optional fixes; they are the coherence data required to define comparisons between contexts. When anomalies are inevitable. Anomalies are inevitable whenever the transition data fails the closure condition under composition. Equivalently, if there exists a nontrivial loop in the context groupoid along which the net transition shift does not vanish, then no global observable exists. The anomaly is the strictification–invariant residue associated to that loop. This predicts, without computation, that certain finite terms must survive all admissible subtractions. When scheme dependence is physical. Scheme dependence is physical precisely when it is detected by loop transport. If changing schemes along two different admissible paths leads to different finite answers, and if that difference cannot be removed by object–wise redefinitions, then the dependence is not a flaw. It is the observable content. Conversely, if all such differences can be strictified away, then apparent scheme dependence is an artifact of incomplete bookkeeping. These predictions convert vague expectations into testable criteria. They explain why certain coefficients are robust across wildly different computational setups, while others fluctuate with regulator choice. 18 D. Connections and extensions We briefly indicate how the framework interfaces with several active themes, without developing them here. Defects. Defect observables are a natural arena for this analysis because they force a mismatch between the locus supporting the field–theory data and the hypersurface on which holographic renormalization is performed. The context groupoid makes this mismatch explicit and explains why defect anomalies often emerge as obstruction classes rather than removable finite terms. Higher–form symmetries. Higher–form symmetries enlarge the space of admissible boundary conditions and therefore enlarge the context space. From the present viewpoint, this increases the number of admissible morphisms and potential loops, making obstruction phenomena more, not less, likely. The framework predicts that anomaly matching for higher–form symmetries is naturally phrased as a strictification problem over an enlarged context groupoid. Non–invertible structures. Non–invertible symmetries provide a particularly sharp illustration of why group–based intuition fails. They naturally generate context spaces with nontrivial composition laws and obstructions. From our perspective, this is not exotic behavior but the generic situation once invertibility is dropped: nontrivial loop transport is the rule rather than the exception. A detailed analysis of non–invertible structures in holography fits naturally into the present framework and will be pursued elsewhere. E. Outlook The unifying message of this section is simple. This framework does not compete with computation; it governs it. It tells you, before the first integral is evaluated, whether you should expect a canonical observable, an unavoidable anomaly, or genuine physical scheme dependence. In regimes where holography is no longer universal by construction, that distinction is not optional. It is the difference between engineering prescriptions and understanding what the theory is actually telling you. X. CONCLUSION Higher–dimensional holography has entered a regime in which successful computations increasingly rely on carefully chosen cutoff prescriptions and higher–dimensional boundary counterterms that cannot be justified by the traditional AdS d+1 renormalization dictionary alone. This development is not a temporary technical complication; it is a structural signal that the notion of universality inherited from lower–dimensional truncations has reached its limit. The central result of this paper is to identify what is missing and to supply it. We have shown that regulator choices in higher–dimensional holography assemble into a context groupoid, that counterterms function as coherence data governing transport between contexts, and that anomalies arise precisely as obstruction classes to global strictification. This reframes renormalization from a prescription–dependent procedure into a well–posed comparison problem with a sharp consistency criterion. From this perspective, recent ten–dimensional computations that reproduce protected field–theory data are neither mysterious nor ad hoc. They are successful precisely because they implicitly choose coherence data in a regime where strictification fails, leaving an invariant residue. What was previously justified only by calibration against known results is now explained as the unavoidable outcome of a global consistency obstruction. Higher–dimensional holography does not need more prescriptions; it needs a coherence principle. Appendix A: Optional categorical remarks (for completeness) This appendix is intentionally optional. Nothing in the main text requires categorical language beyond the elementary groupoid bookkeeping introduced in Section IV. The purpose here is only to record, in compact form, the standard structural interpretation of that bookkeeping, and to indicate how one would generalize it if needed. 19 1. The context groupoid as a category of regulators A groupoid Cis a small category in which every morphism is invertible. In the present paper: • objects c∈Ob ( C )are regulator contexts (cutoff surface, asymptotic presentation, subtraction/normalization, allowed boundary terms, boundary conditions); • morphisms φ : c→c0 are admissible changes of context (coordinate/presentation changes, cutoff deformations, and the induced adjustments of boundary data and boundary terms); •composition is concatenation of admissible changes. The content of the main text can be phrased as follows: a “renormalized observable” should not be understood as a number attached to a preferred object c, but as a globally consistent assignment on C. 2. Transition data as a 1–cocycle and strictification as trivialization The transition terms T ( φ )of Sections V–VI may be viewed as a 1–cocycle on C with values in an abelian group Aof admissible finite boundary functionals (or, after projecting to the relevant sector, in the corresponding coefficient space). Concretely, coherence under composition, T(ψ◦φ) = T(ψ) + T(φ),(A1) is precisely the cocycle condition. An object–wise redefinition by B(c)(Section VI B) acts as a coboundary shift, T(φ)7−→ T(φ) + B(c0)−B(c),(A2) which is the standard notion of equivalence of cocycles. Strictification is the statement that the cocycle class is trivial: one can choose B so that the cocycle vanishes identically. Nontriviality is detected by loop transport, and the resulting loop invariants are exactly the obstruction data identified with anomalies in Theorem VI.1. 3. Beyond groupoids: why one might want higher coherence (not used here) The main text uses a groupoid because it is the minimal structure needed to make scheme comparisons well–posed and to extract obstructions. In more elaborate situations one may encounter genuinely higher coherence conditions, for example when: •the admissible changes of context are naturally defined only up to controlled equivalence, • boundary condition data involves nontrivial extensions (e.g. mixing of ensembles or defects with junctions), • or one wishes to organize families of contexts parametrized by moduli with nontrivial higher identifications. In such cases it may be natural to replace the groupoid by a higher groupoid (or by a stacky refinement) and replace the cocycle/trivialization problem by its higher analogue. None of this is required for the results of the present paper: the groupoid–level criterion already isolates the obstruction mechanism and provides the practical recipe of Section VIII. [1] R. Izquierdo García, “Higher dimensional holography,” arXiv:2512.12696 [hep-th] (2025), DOI: 10.48550/arXiv.2512.12696. [2] M. Henningson and K. Skenderis, “The Holographic Weyl anomaly,” JHEP 07 (1998) 023, arXiv:hepth/9806087, DOI: 10.1088/1126-6708/1998/07/023. [3] S. de Haro, K. Skenderis and S. N. Solodukhin, “Holographic Reconstruction of Spacetime and Renormalization in the AdS/CFT Correspondence,” Commun. Math. Phys. 217 (2001) 595–622, arXiv:hep-th/0002230, DOI: 10.1007/s002200100381. 20 [4] K. Skenderis, “Lecture notes on holographic renormalization,” Class. Quant. Grav. 19 (2002) 5849–5876, arXiv:hep-th/0209067, DOI: 10.1088/0264-9381/19/22/306. [5] H. Lin, O. Lunin and J. Maldacena, “Bubbling AdS space and 1 / 2BPS geometries,” JHEP 10 (2004) 025, arXiv:hep-th/0409174, DOI: 10.1088/1126-6708/2004/10/025.