13. Stress tests, objections and the cost of the architecture

Chapter 13 of 15 · ≈ 35 min read · 55 paragraphs · section 19 of 40 · search the book

An architecture is credible not when it claims to have no internal tension, but when it can show where those tensions lie and how they are resolved.

An architecture is credible not when it claims to have no internal tension, but when it can show where those tensions lie and how they are resolved. The design process behind this document maintained a public register of inter-institutional inconsistencies — detected by interface audits, an adversarial review and a full-scale simulation in which distinct internal teams played each institution against realistic cases — and now tested by the integrated demonstrator of Section 9. That demonstrator does not validate the architecture in the real world; it makes observable the possible breaks between measurement, qualification, consequence, financing, certification and redress. Six tensions deserve to be recorded here, together with their resolutions.¶ 1

13.1 Design tensions and resolutions

13.1.1 Can fragmentation protect?

Constitutional antithesis. The fragmentation of the global system is not always a gap to be repaired. It may constitute a deliberate protection against the concentration of authority, premature standardisation, the propagation of a common error and the disappearance of competing forums. Separate mandates permit experimentation, subsidiarity, exit and resistance to central capture. A regime complex may be more adaptable than an integrated organisation, particularly under high uncertainty (Ostrom, 2010; Keohane and Victor, 2011; Alter and Meunier, 2009).¶ 2

The framework’s response. The Global Accountability Chain must not therefore seek the unification of decisions. It seeks a coordinated fragmentation: centres that remain legally and operationally autonomous, but that make explicit the evidence formats, the reasons, the incompatibilities and the avenues of redress. The interface does not impose agreement; it prevents disagreement from disappearing into an opaque transmission.¶ 3

Firewalls. Interoperability must be accompanied by circuit breakers: the possibility for a function to refuse an insufficiently documented input; no cross-cutting machine learning by default; the maintenance of multiple models and suppliers; publication of divergences; a right of re-examination; and the capacity to revert to a local or manual procedure. Useful redundancy is not eliminated where it provides an independent counter-assessment or an alternative avenue.¶ 4

Rejection criterion. The architecture must be simplified or rejected if the common standard propagates errors faster than it detects them, if the portability of the case file reduces the diversity of forums, if interoperability becomes a de facto centralisation, or if coordination costs remove the adaptability that justified pluralism. Completeness is not uniformity.¶ 5

VALIDATION STATUS The interface audits, the red-team exercises and the full-scale simulation reported here constitute validation internal to the programme, unless the protocol, the scenarios, the participants, the independence guarantees, the evidence, the scoring and the complete results are published. Internal adversarial tests improve the design; they are not a substitute for external validation.¶ 6

The orphan price. In the early versions, the border fairness mechanism expected a “reference price” from the rating authority — which, by virtue of its own negative mandate, prices nothing. The resolution created the most important doctrinal line of the architecture: rating is not pricing. The rating authority issues a dimensionless coefficient; the pricing authority publishes the price; the border fairness mechanism multiplies. One institution per transformation (P5).¶ 7

Double penalty. If the pricing authority prices an externality and the border fairness mechanism makes an adjustment for it, the same harm could be billed twice. Resolution: a scope contract — flows already corrected at the border are excluded from the tax base, and conversely; the two apply alternatively, never cumulatively, and the Court settles boundary disputes.¶ 8

Budgetary circularity. A fund financed by border revenues, feeding institutions that generate those same revenues, is a perpetual motion machine that fails in the first year in which revenues disappoint. Resolution: P6 — an independent seed endowment with a designated guarantor; revenues grow the regime, they never establish it.¶ 9

Consent versus justiciability for private actors. States may accede to a treaty; but how could a global platform or a commodity trader ever be made subject to the Court? This is the most structural weakness that the adversarial simulation brought to light. Resolution: a layered consent — contractual arbitration clauses in any instrument that incorporates the outputs of the architecture; a derived adherence through national law where parliaments incorporate the Charter as a model law; and, residually, the reputational force of a public and contestable rating, whose publication requires no one’s consent.¶ 10

Who certifies the certifiers’ algorithms? The rating pipeline, the allocation engine and the adjustment computations are themselves algorithmic systems of exactly the kind that the architecture proposes to govern. Resolution: the reflexive certification mandate of the digital authority (Section 5.7), under a three-tier risk regime with judicial review at the highest tier.¶ 11

Feedback without capture. Downstream institutions learn things that upstream institutions need — but binding feedback would allow the taxed entity to shape the rating. Resolution: all feedback loops running upstream are expressly non-binding improvement signals; methodological authority remains with the issuing institution; the loop is published.¶ 12

When to compress a signal — and when never to do so. The rating authority compresses the measurements into a coefficient and a class; the watchtower refuses to compress its six dimensions into a single score. The asymmetry is deliberate, and the resolution comes down to a general rule: a signal may be compressed where the downstream decision requires a single number and where an adversarial procedure exists in which to contest it (a rating feeds a price, and is open to redress); it must never be compressed where the compression would itself constitute the judgement (a single “danger score” for a digital system would pre-empt the qualification that falls to the deliberative and judicial process). Compression follows contestability — never the reverse. This asymmetry has a theoretical foundation: Arrow (1951b) showed that no rule can aggregate plural values into a single ordering without sacrificing one of the minimal conditions of coherence; the answer is not to force aggregation, but to enrich the informational base and to keep the dimensions separate where compression would make the judgement (Sen, 1970). The watchtower therefore does not aggregate preferences into a social welfare function — it measures impacts along declared dimensions; Arrow’s problem does not apply to it.¶ 13

The general lesson generalises beyond this project: it is interface failures, and not failures of institutional design, that make multi-body architectures fail — and interface auditing, including adversarial simulation before any deployment, should be treated as a first-order design discipline, with AI-assisted coherence verification as a supporting instrument (a dedicated methodological document is in preparation).¶ 14

These tensions belong, in the last analysis, to contract theory. No interface between institutions can foresee everything: contracts are necessarily incomplete, and what then matters is the allocation of residual control rights — who decides in the unwritten contingencies (Grossman and Hart, 1986). The resolutions set out above are precisely such allocations: methodological authority remains with the issuing institution, feedback remains a non-binding signal, a single institution prices. Likewise, the refusal to compress six dimensions into a single score is not an aesthetic scruple: placing a strong incentive on a single measurable metric distorts effort to the detriment of what is less measurable — the multitasking problem (Holmström and Milgrom, 1991) —, whereas the informativeness principle justifies, conversely, retaining all the relevant signals in the nineteen-field data contract (Holmström, 1979). The negative mandates, finally, are the institutional version of deliberately incomplete contracts: by strictly bounding what each body may do, they allocate decision rights cleanly and prevent the conflicts that the unforeseen would give rise to.¶ 15

13.2 Objections and responses

“This is world government. ” It is the reverse, by construction. Each mandate is single-purpose and negatively enclosed (P2); nothing is binding without incorporation or consent (P3); primacy is purely internal (section 5); the Court reviews legality, never policy; and the whole replaces no existing institution (P4). The precedents are instructive: States already accept binding dispute settlement (WTO), binding arbitration (thousands of investment treaties) and border price adjustment (CBAM) where the arrangement is reciprocal and adherence voluntary. The architecture generalises forms of binding force that the world has already accepted — it invents no new coercion. To centralise is not to govern: following Koremenos, Lipson and Snidal (2001), the centralisation of a function — normative, informational or analytical — determines neither coercive power nor political control, which remain distinct variables. A single global stratum of measurement and standards (HuIB) is therefore not a world bureaucracy: definitions, identifiers and methods are common, while collection, verification and enforcement remain federated.¶ 16

Two classic frameworks explain the importance of this design choice: Rodrik’s trilemma (2011) warns that deep integration, national sovereignty and democratic politics cannot be maximised simultaneously — this architecture favours consent over depth precisely in order to respect that constraint rather than to deny it; and Ostrom’s polycentric governance (2010a) names the theoretical family to which this arrangement belongs: multiple centres of partial authority, checking one another, none of which is sovereign over the whole. Far from eroding sovereignty, the architecture relies on voluntary cooperation between national institutions — a networked governance in which States retain control and consent to each commitment (Slaughter, 2004). The interparliamentary digital authority illustrates this: it links parliaments in order to strengthen democratic control over digital systems, not to bypass it. The network serves sovereignty by coordinating it; it does not substitute itself for it.¶ 17

“This will never be adopted. ” The response of section 12 is empirical: the general mechanism (incorporation) is the one by which every successful transnational norm of the past century actually spread; most of the adaptations of the incumbents require no statutory change; and the pilot route requires only a single volunteer country. The right question is not whether the regime arrives fully formed from the outset — it is whether each phase is useful individually. A measurement board is useful on its own. A proto-rating is useful on its own. The sequencing principle (P7) constitutes the theory of adoption. This objection has its scholarly form: the record of the great global architectures — universal, binding and comprehensive treaties — is a record of deadlock, and the route that succeeds is rather that of small clubs of the willing, of experimentation and of building on national interests (Victor, 2011). The architecture does not contest that diagnosis; it espouses its method. It rests on no founding treaty, begins with volunteer jurisdictions (P3), pilots components (§12.5) and builds coalitions that make abstention costly (§12) — precisely the bottom-up strategy that the failure of the great treaties recommends. What it adds to the bottom-up is the chain: enough to turn scattered clubs into a connected accountability.¶ 18

“The measurements will be contested. ” They will be, and the arrangement anticipates it: ranges rather than points, declared gaps, published methods, adversarial contestation and judicial review of the measuring institution itself. The disaggregation rule of Box 2 — never to add together market costs, welfare valuations and transfers — constitutes in itself a response to the principal honest criticism addressed to externality aggregates. What is not credible is the implicit alternative: that the best treatment of an unmeasured harm consists in leaving it unmeasured.¶ 19

“It will be captured. ” Capture is the most serious risk to the architecture, and four features respond to it structurally: parity of governance between impact-source and impact-recipient jurisdictions; the prohibition of payment by the issuer; glass-box methods with versioned changes; and non-solidarity between institutions, so that the capture of one does not deliver up the others. The reflexive certification of the architecture’s own algorithms (section 5.7) extends this response to the technical layer. This is the public choice objection in its hardest form — every new institution risks becoming a captured bureaucracy (Buchanan and Tullock, 1962); the architecture does not dismiss it, it is designed against it, each of the features above answering a failure mode that public choice predicts.¶ 20

“This duplicates what already exists. ” Table 11 and the Sevilla mapping respond quantitatively: the existing landscape saturates financing and measurement, and leaves rating, binding pricing and adjudication vacant. The architecture is deliberately complementary — it consumes IPCC-type science (IPCC, 2023), CBAM-type instruments and Sevilla-type financing, rather than replacing them. The existing landscape is what political science calls a regime complex: a set of loosely coupled institutions, without clear hierarchy (Keohane and Victor, 2011). Such a complex may be more adaptable than a single regime, but it pays for that flexibility in inconsistencies, in gaps and in forum shopping. Adding an institution may therefore aggravate fragmentation as much as reduce it — and that is why the architecture does not add one more loosely coupled piece: it joins the existing links into a coherent chain (non-substitution P4, severability, one function per body), filling the voids without duplicating what already holds.¶ 21

“These are merely known components, reassembled under new names.” The observation is partly right at the level of the building blocks — a scientific council, a rating agency, Pigouvian pricing, border adjustment, a fund, parliamentary oversight and a tribunal exist separately. But in institutional design, originality does not reside in the components: it resides in the architecture, the separation of functions, the conflict rules, the interfaces and the sequence of decisions. An aircraft is not original by virtue of its metals or its engines, but by virtue of the organisation that draws from those parts a new capacity. The claim is therefore not to have invented each of these elements, but an institutional capacity — a closed accountability chain — that does not exist so long as they remain separate.¶ 22

“The new institutions will be accountable to the States and the funders that create them, not to the people they claim to protect.” This objection is eliminated neither by transparency alone nor by the assertion of a universal interest. It calls for a differentiated architecture that combines control by the delegators, the procedural rights of affected parties, independent supervision, peer review, budgetary consequences, judicial review and public reputation. The risk remains that the best-resourced actors will dominate the participatory avenues themselves; this is why the pilot programme measures not only the number of participants, but representativeness, accessibility and the rate at which objections from low-capacity groups are taken into account. The chain removes none of these mechanisms: it synchronises them so that a measured harm reaches a consequence and a remedy.¶ 23

“Why would these seven institutions not develop their own interests, routines and pathologies, like the existing international organisations?” The objection is a serious one and does not dissolve in negative mandates alone. An organisation endowed with expertise and rational-legal authority can acquire an autonomy of its own and, without ever formally violating its rules, progressively redefine its mission, turn its procedures into ends and favour the problems it already knows how to handle (Barnett and Finnemore, 1999). The Charter and the negative mandates are initial guarantees, not guarantees of duration. That is precisely why the architecture does not stop at creation: each institution is bounded by revision and sunset clauses, exposed to external feedback loops — affected parties, peers, judicial review — rather than to its own internal evaluation alone, held to a professionally heterogeneous composition against the homogeneity that feeds the blind spot, and bound to a precise mission any redefinition of which must be publicly justified. The prevention of pathologies is not an added corrective: it is part of the mandate.¶ 24

“By multiplying institutions, forums, audits, parliamentary scrutiny and judicial avenues, the architecture risks a multiple accountabilities disorder: incompatible control requirements that no institution could satisfy at once.” The risk is real — but it is handled by discipline, not by reduction. To the separation of functions is added a separation of the modes of accountability: each institution receives a distinct accountability profile, which ranks the forms of control compatible with its function and sets aside those that would corrupt it (Koppell, 2005). A requirement that is legitimate for one institution may be pathological for another: the measurement board must answer to scientific standards and to the data of affected parties, not to the instructions of a State that would like a more favourable measurement; the Court must remain faithful to the law, never commanded by opinion or by its delegators. Failing such an ordering, an organisation oscillates between contradictory expectations and leaves its leaders to choose after the fact the conception of accountability that justifies the decision already taken. The density of the architecture is therefore an ordered strength, not a confusion.¶ 25

“You are adding fragmentation to an already crowded system: new actors, overlapping competences, forums that may be chosen at one’s convenience.” The architecture responds to this fear by its very construction. It does not duplicate existing functions; it assigns each transformation to a principal responsible body; it employs the institutions in place as sources and operators rather than replacing them; it imposes an interface contract on each link; it confines normative conflicts and provides a rule of redress where two competences meet. Turning an unmanaged fragmentation into an organised fragmentation does not add to the disorder: it disciplines it (Biermann et al., 2009).¶ 26

“Why create seven separate institutions ex ante if a division of labour can emerge of itself among the existing institutions?” Order can indeed emerge within an institutional complex: common members, playing nested games, lead overlapping bodies to specialise and to recognise one another (Gehring and Faude, 2014). But this emergent order has three decisive limits. It emerges from the convenience of the most powerful members, not from the affected parties. It coordinates the functions that already exist, not those that are missing — independent rating, binding consequence, adjudication. And it equips itself with no constitutional core that would make it coherent and contestable. The 7+3 provides precisely what emergence does not produce: the absent functions, the constitutional foundation and contestability by those who bear the effects. Where institutions overlap, the architecture employs them as sources and operators; it does not wait for a convenient division of labour to take the place of a chain that must be complete, separated and contestable.¶ 27

“Is this not simply ESG reporting?” ESG cleared the ground — it imposed the idea that a firm has impacts beyond its balance sheet, built the disclosure and data infrastructure on which this architecture relies, and its regulatory descendants (CSRD, ISSB) carry forward what it achieved. The arrangement does not overturn it; it completes what a voluntary, self-declared regime of financial materiality could not accomplish alone, on three counts. Materiality: dominant ESG rating measures the risk that the world places on the firm, whereas internalising a harm requires measuring the firm’s footprint on third parties — MSCI acknowledges as much (Bloomberg Businessweek, 2021), and the architecture rates that footprint (GIRA, §5.3). Source: ESG is self-declared and rated by agencies that the entity pays, and its ratings diverge (correlation 0.38–0.71 against close to 0.99 for credit, the gap being mainly a matter of measurement; Berg, Kölbel and Rigobon, 2022) — the architecture separates independent measurement (HuIB) from a rating never paid for by the rated (P6). Consequence: ESG discloses, where the chain prices, adjusts, finances and adjudicates. These three steps are not internal adjustments to ESG but functions that it has, by construction, never carried: to provide them requires institutions, not one more score. The architecture therefore consumes the data and the rails that ESG built — a framework still maturing (the first global assurance standard in force at the end of 2026, IAASB, 2025) and being recomposed (the lightening of the CSRD by the “Omnibus”, European Union, 2026) — rather than starting again from scratch. The human impacts ledger (conclusion) is not one more ESG score: it is the register, measured and rated by third parties, that ESG made thinkable without being able to keep it.¶ 28

“Where does the money come from? ” For the institutions: modest and non-circular budgets (P6) — the measurement and rating layers cost millions, not billions, and it is they that create the option value for everything else. For the fund: a sequenced growth, from an endowed pilot through to the revenues generated by the pricing and border layers — the regime finances protection out of the very flows that today externalise the harm: the fund is not paid for by humanity as a whole; it is paid for by the mechanisms that destabilise humanity as a whole.¶ 29

“Pricing these harms confuses ordinary economics with market failure. ” A recurrent objection from welfare economics holds that a pecuniary externality — a cost transmitted through prices rather than outside them — is not a market failure but the normal working of competition; that where a genuine technological externality exists, prevailing prices and Coasian bargaining already internalise the essential part of it; and that the failures of a corrective authority usually exceed the failure it corrects. The document treats this as its sharpest challenge. Its response is threefold. First, the harms at issue are priced nowhere for the party that bears them: a monetary spillover effect, an unpriced data levy or a cross-border emission reaches a community that never entered the market on which the price was formed, so that the Coasian precondition — a forum in which the parties can bargain — is precisely what is missing and what the chain is built to supply. Second, where a cost is authentically pecuniary and where no welfare loss remains under zero-friction assumptions, the architecture does not price it: entry into the system requires that an externality satisfy declared criteria (§2.2, §9.6), and a mere transfer is never counted as a global loss. Third, the risk of government failure is conceded, not brushed aside, and that is the reason why the institutions are negatively enclosed, contestable, bounded by consent and sequenced by pilots, rather than empowered to fix outcomes — the arrangement is calibrated on the possibility that the corrector is wrong. The claim is not that public action is costless, but that a status quo which leaves a measurable, uncompensated and cross-border harm entirely outside any price or any forum is not the neutral baseline that the objection assumes.¶ 30

“Putting a price on harm amounts to authorising it. ” A distinct objection, coming from the ecological and degrowth left, holds that pricing a wrong authorises it against payment and launders extraction into compliance. The response is already inscribed in the principles of the document rather than added here: gravely unlawful harms are placed entirely outside the pricing logic under a non-compensable exclusion (P8), which ring-fences what is not for sale and distinguishes the pricing of a measurable externality from the authorisation of what must on the contrary be prohibited. On this point, the architecture is deliberately reformist rather than revolutionary, and it owns that choice: to render a harm visible, costly and accountable is not to endorse it.¶ 31

“This is a regime of taxes and barriers that stifles trade and financial flows, and pushes towards sovereigntist closure.” The objection inverts the mechanism. Trade and investment are capped by trust: weaker bilateral trust measurably reduces trade, portfolio investment and FDI — all the more so for trust-intensive goods (Guiso, Sapienza and Zingales, 2009); and where contract enforcement is weak, exchange is under-produced for want of credible institutions (Dixit, 2004). By making measurable, rated and contestable wrongs that are today invisible, the architecture completes missing markets and levels the playing field, removing the grievance of unfair competition through externalities that is precisely what feeds retaliatory protectionism. Three qualifications against the opposite excess. First, the effect depends on the design: only rules perceived as legitimate, reciprocal and uncaptured produce trust; asymmetric rules would fragment — hence the parity of governance and the anti-capture safeguards already set out, and the differentiated treatment that protects low capacities. Next, the architecture does not claim to increase all exchange: it reallocates externality-intensive exchange towards clean and fair exchange, and externality-laden trade shrinks — an intended effect, not collateral damage. Finally, this is not a matter of more tax but of better-placed tax: where the conditions of §9.7 are met, a Pigouvian pricing of the harmful corrects a distortion instead of creating one. Where they are not, the chain measures, rates and publishes — and does not price. The same logic holds for financial flows, which the GFEA disciplines: trust affects portfolio investment and FDI as much as trade (Guiso, Sapienza and Zingales, 2009). One nuance specific to finance, however: the objective is not more flows but better flows — the GFEA reallocates the speculative and the destabilising (regulatory arbitrage, hot money, high-frequency rent) towards productive investment, and disciplining instability itself (capital surges, sudden stops, §9) reduces what provokes capital flight and financial fragmentation; capital flees the arbitrary, not the fair and predictable rule. There remains a risk more acute here than in trade — migration towards non-participating jurisdictions (arbitrage). The response is the coordinated minimum: the global minimum tax of 2021 showed that one hundred and thirty-six jurisdictions can agree on a floor (OECD, 2021); to which are added reputational rating (GIRA) and conditions of access. There remains the counterfactual, and it is decisive: the alternative is not perpetual openness but the fragmentation already under way — unilateral tariffs, interdependence turned into a weapon, sovereignty brandished as protectionism (Aiyar et al., 2023). This enclosure occurs without the architecture, by the law of the strongest. The architecture does not close the system: it offers the only ruled, reciprocal and fair way of reopening it on a basis of trust (see also “the objection to the objections”, infra: the status quo is not neutral).¶ 32

And the objection to the objections. Each of the arguments above measures the architecture against the yardstick of the status quo as if the latter were neutral. It is not. The status quo is a choice — one that currently costs, on the basis of conservative third-party figures alone, trillions a year in unpriced harms, borne disproportionately by those least able to bear them and least represented where the decisions are taken. The externalities we fail to price today become the crises that the next generation will have to manage tomorrow. Complexity and resistance are signals inviting the design of better systems — not reasons to abandon the field to inertia.¶ 33

“The architecture costs more legitimacy and coordination than it creates.” This objection cannot be refuted by the scale of the harms alone. It requires comparative indicators of administrative cost, decision time, burden of redress and representation. The design response lies in modularity, common data contracts, severability and the sunset clause; the empirical response must come from the pilots.¶ 34

“Powerful actors will stay outside.” Some will. The architecture must not presuppose universal participation. Its test consists in determining whether voluntary measurement, reputational ratings, public procurement, financing clauses and a coalition of incorporating jurisdictions create useful coverage without coercive universality — and whether the costs of non-participation remain lawful, proportionate and reversible. This is the prognosis of game theory — non-participation is an equilibrium so long as free-riding remains costless; the issue is therefore not to force adherence but to make abstention costly (rating, access, border adjustment), so that useful coverage is built without universality (Nash, 1951; Barrett, 2003).¶ 35

“The system will reproduce the North-South hierarchy through data and expertise.” This is a central risk. Open methods do not suffice if data ownership, computing power, language, the selection of experts and the capacity for redress remain concentrated. Capacity financing, distributed research nodes, multilingual procedure, rights over local data and the representation of affected communities are constitutive safeguards, not development-aid accessories. This risk has a name in the literature: data colonialism — the appropriation of human life through data extraction, which reproduces, in informational form, the asymmetries of colonial extraction (Couldry and Mejias, 2019). The safeguards above are aimed precisely at preventing the architecture, in measuring the world, from itself becoming an instrument of that appropriation. This objection extends the critique long conducted by Third World Approaches to International Law (TWAIL) and by postcolonial studies: international institutions, even well intentioned, have historically reproduced the asymmetries of power that they claimed to correct. The risk is real, and the architecture cannot avert it by decree; it responds to it through its design — differentiated treatment according to capacity to contribute, representation of affected parties within the bodies, and sequencing that does not place the burden from the outset on low-capacity jurisdictions (§12.6) — while acknowledging that these safeguards are judged in use, not in intention.¶ 36

“What governs local commons does not scale to the world.” The objection is a serious one: the enduring institutions documented by Ostrom (1990) rested on proximity, shared norms and low-cost monitoring — three conditions absent between States and multinationals on a planetary scale. The architecture’s response is not to deny the gap, but to fill it by design: where community and trust are lacking, it is independent measurement (HuIB), contestable publication (GIRA), justiciability (Court) and incorporation by consent (P3) that stand in the place of shared norms and proximate monitoring — not a supposed global civic spirit. This is also the response to the question of costs raised by transaction cost theory (Williamson, 1985): a chain of institutions does indeed have an interface cost, but that cost is measured against the counterfactual (§4.6, §13), not against zero, and interface auditing (§13) makes it a design discipline rather than a blind spot.¶ 37

“Institutions are shaped by power, and power will not constrain itself.” This is the deepest objection. The institutional literature teaches that institutions are endogenous to the distribution of power: the powerful resist those that would hold them to account (Acemoglu and Robinson, 2012). The architecture does not claim to dissolve that fact. It responds to it by three routes already set out: it begins without the consent of the most powerful (incorporation by the willing, P3); it alters the payoffs so that abstention becomes costly rather than costless (§12); and it distributes power within its own design — parity, contestability, non-capture — so as not to be extractive itself. As regards the non-consenting major players, enforcement remains limited, and the document owns that; but the history of inclusive institutions shows that they are rarely built through the initial assent of the powerful, and more often through coalitions that make their resistance progressively more costly.¶ 38

“The world has already refused this, where the science was strongest.” The objection is the weightiest of all, and it must be stated without attenuation. Climate is the only domain in which the scientific attribution of a cross-border externality has reached maturity. Yet in 2015, decision 1/CP.21, which adopted the Paris Agreement, “agrees that Article 8 of the Agreement” — the one on loss and damage — “does not involve or provide a basis for any liability or compensation” (United Nations, 2015, §51; official English text). The clause was obtained by the United States, and it was the price of their adherence. If payment was explicitly ruled out where the science is most robust and the historic emitter most clearly identified, how would it be any less so where attribution is more contested and where the presumed debtor is an independent central bank?¶ 39

The answer lies in examining what the climate regime has actually built. It built an authoritative measurement stratum: the Intergovernmental Panel on Climate Change. It built a rating: global warming potential, the social cost of carbon. It built a pricing: allowance markets, carbon taxes. It built a border adjustment: the Carbon Border Adjustment Mechanism. It built a financing: the Green Fund, then the loss and damage fund. Measure, rate, price, adjust, finance — five links out of seven. And it built neither the sixth nor the seventh: the enforceable consequence and the judge.¶ 40

Paragraph 51 is therefore not a counter-example: it is the amputation, expressly dated and expressly attributed, of the last link of an almost complete chain. It confirms, with a precision that this document would not have dared hope for, the law of attrition stated in Box 9: what survives is what obliges no one; what disappears is what obliges someone who has not consented to it. Measurement survived, because it does not oblige. Pricing survived, because it proceeds from the consent of whoever incorporates it. The loss and damage fund survived, because it is voluntary. Liability alone was excised — and it was excised by the text, at the request of the party that would have borne it.¶ 41

The present document does not claim to escape that law. It claims to know it. That is the raison d’être of principle P3 and of the feasibility scale of §12.1: one first builds the links that oblige no one — measurement, rating, publication —, and the link that obliges comes about only through incorporation, jurisdiction by jurisdiction, contract by contract. The chain is not decreed: it closes by consent, or it does not close. This is not a concession wrung from prudence: it is, to this day, a route that no negotiation has closed off.¶ 42

“Incoherence protects the weak; your coherence would expose them.” The objection is advanced by the most serious critique in the field, and it deserves to be quoted in its own terms. Grabel (2017) argues that the incoherence of global financial governance is productive and not debilitating: a denser, fragmented and multipolar governance widens the space for institutional experimentation and the autonomy of developing economies, where a coherent design would be captured by the powerful. Ocampo formulates the same preference in terms of architecture: “in a heterogeneous international community, the creation of networks of global, regional, and national institutions can provide a better system of governance than arrangements based on single global organizations”; the International Monetary Fund should be conceived as “the apex of a network of regional reserve funds”, within a “dense” and multi-level architecture (Ocampo, 2017, pp. xiii, 200 and 231). The argument is not one of convenience: it is one of justice. Single global organisations give, he says, a weaker voice to small countries.¶ 43

The response does not consist in removing functions, but in distinguishing the function from the body. The chain requires that each function have a duty-holder, and that its negative mandate forbid it the neighbouring functions. It does not require that this duty-holder be single, global and centralised. Nothing in §4.7 requires the measurement stratum to be a monopoly: it is designed as an accreditation body — it approves the methods, aggregates knowledge, preserves provenance and uncertainty, and re-measures nothing (§5.2). Nothing prohibits a plurality of accredited raters, under the prohibition of issuer-pays (P6). Nothing prohibits regional chapters of the measurement stratum, regional windows under the solidarity fund, or regional chambers of the Court. Ostrom’s polycentricity, invoked in §4.6, is not opposed to the chain: it describes the way in which each of its links must be organised.¶ 44

There remains the substance of the objection — the risk of capture —, which is also principle P6 of this document. Two responses are owed to it. The first is structural. A single institution that held measurement, rating, pricing and adjudication all at once would be capturable once only, and neutralised in its entirety. That is precisely what became of the Palais-Royal chain: it was entrusted to the International Monetary Fund, an institution governed by the very countries that it ought to have penalised (Box 9). The separation of functions is therefore not a design elegance: it is the principal anti-capture device available. To capture seven institutions, endowed with seven governance structures, seven negative mandates and a court that reviews them all, is not to capture one organisation: it is to capture seven.¶ 45

The second response is empirical, and it is owed to an adversary. Incoherence has lasted for eighty years. It has not protected the weak: it has left them the bill. Self-insurance through reserves costs developing countries on the order of one point of output a year — and it is Rodrik (2006) who puts the figure on it, the very author whose preface opens Grabel’s book. Sevilla produced seventy-one financing initiatives and no binding instrument (§3.3). Incoherence is productive for those who can arbitrage it; and those who can arbitrage it are those who have the lawyers, the negotiators and the balance sheets. The weakest do not take advantage of the disorder: they are the ones who cannot afford to find their way through it.¶ 46

13.3 The cost of inaction

The last objection of section 13 deserves a section of its own — for it does not target this architecture, but every alternative, including the tacit alternative of carrying on as before. Four observations make it possible to assess that alternative.¶ 47

Externalities are stocks, not passing flows. The annual figures of section 2 read like the weather; they behave like a debt. An unpriced harm does not expire at the end of the year — it accumulates: in the atmosphere and the water tables, in morbidity and disability, in sovereign balance sheets weighed down by the spillover effects of other actors, and — least visibly but most corrosively — in institutional trust. And the stocks compound with one another: each year without credible measurement makes the next measurement more contested; each uncompensated transfer makes the next cycle of cooperation politically more costly in the country that has borne it; each unpunished harm lowers the expected cost of the next. The status quo is not a stationary state. It is a trajectory.¶ 48

Let us project the five modes through time. Extending each transmission mode of Table 4 over a decade, with rules unchanged: market mediation — rents concentrate further, extraction compounding while being accounted for as creation (Mazzucato, 2018) — and systemic fragility reproduces itself, each rescue being socialised; the pattern is documented, not speculative (the World Bank, 2020, recorded in real time the scale of the pandemic contraction). Policy spillover effects: a world of more frequent shocks and of more assertive domestic mandates reproduces the 2022–2025 episode with growing amplitude, and the fiscal buffers of the recipient countries, already thinned by the previous cycle, absorb the next from a lower position. Technological mediation: informational externalities compound within the very infrastructures that democracies use to correct everything else — the only stock that degrades the mechanism of repair itself. Embedded in flows: on current trends, a further decade will again yield ten trillion dollars of illicit outflows from the countries least able to afford them. Commons: the non-linearities are the best documented in science, and that is why this mode alone already has its scientific institution — what it lacks is everything that comes after measurement.¶ 49

What past crises actually cost. The 2008 crisis and the pandemic each cost the world economy cumulative output losses running into trillions of dollars — recorded figures, not projections. Neither of the two was a black swan. Each was an accumulation — of unpriced leverage in one case, of unpriced biological risk in the other — measured too partially, priced nowhere, and justiciable before no one until it detonated. The honest counterfactual must be stated with care: this architecture would not have prevented 2008, and it claims no omniscience. What it would have done is to make the accumulation visible, rated, and increasingly costly to ignore — early warning is not prophecy, but insurance has never needed prophecy to justify its premium. Set against annual institutional budgets counted in millions, the arithmetic of averted cost requires no optimism; it requires only that visibility bend a fraction of a single one of these trajectories.¶ 50

The argument is not new in climate economics: the most cited analysis of the cost of inaction concluded that the benefits of strong and early action far outweigh its costs, and described climate change as the greatest market failure ever observed (Stern, 2007). The scope of this document is broader — it deals with the whole set of invisible externalities, not climate alone —, but the logic is the same: a harm left outside any price is not free; it is simply billed later, and more.¶ 51

Who pays, and when the window closes. The costs of inaction are not distributed uniformly. They fall, first and most heavily, on those who are least represented where the choices are made — the uncompensated incidence quantified in section 9, the populations situated behind the last column of Table 4, and the generations yet to be born, who have neither a seat, nor a vote, nor a hedge; the harms left unpriced this decade arrive as budget lines of the next. And the window for building the corrective stratum is not indefinite. The multilateral openings on which this document relies — the Pact for the Future, the Sevilla agenda, a universal tax convention under negotiation — exist today, and the political capital that carries them is dated. The realistic alternative to a reformed multilateralism is not the status quo; it is a world increasingly governed by the balance of force rather than by law, in which the adjustments occur in any event — unilaterally, without equity clauses, and without anyone’s consent being sought. To do nothing, in other words, does not avoid the question of architecture. It answers it — with the answer of the strongest actor. And the institutional record is unambiguous as to where this road leads: prosperity follows inclusive institutions, and their absence is never a neutral state (Acemoglu and Robinson, 2012).¶ 52

It is here that the criterion of governance choice set out in §4.6 finds its measure. An architecture is not justified because its institutions are virtuous, nor condemned because they have a cost: it is judged net, against the right counterfactual. The transaction costs and the human costs that a chain of institutions saves are not to be compared with a costless world, but with this one — a world in which unpriced harm accumulates and in which the crises that it prepares arrive in any event. Measured against that baseline, the cost of acting is an investment in an option; the cost of doing nothing, a premium paid without being insured.¶ 53

Counterfactual discipline.¶ 54

The cost of inaction is not the benefit of adopting the 7+3. The harm averted depends on adherence, on compliance, on behavioural response, on administrative cost, on substitution and on unintended effects. Future work should compare at least three counterfactuals: unchanged fragmentation; incremental reform by the incumbents; and the progressive adoption of the 7+3. Option value claims should state their assumptions and their ranges rather than treating the entire stock of externalities as avoidable.¶ 55

@incollection{yaiche_hc_chapitre_13,
  author    = {Yaïche, Nizar},
  title     = {13. Stress tests, objections and the cost of the architecture},
  booktitle = {The Humanity Compact},
  year      = {2026},
  version   = {2.3},
  doi       = {10.5281/zenodo.22746533},
  publisher = {Zenodo},
  url       = {https://humanitycompact.org/ouvrage/chapitre-13/}
}

A book by Nizar Yaïche, former Minister of Finance of the Republic of Tunisia. Who is the author?