Le texte de l’ouvrage est en anglais, tel qu’il a été écrit ; seule l’interface du site est en français.

14. Limits, refutability and research programme

Chapitre 14 sur 15 · ≈ 10 min de lecture · 35 paragraphes · 20ᵉ des 40 sections de lecture · rechercher dans l’ouvrage

Sixteen limits bound the claims of this document, and each defines a research front on which collaboration is invited.

Sixteen limits bound the claims of this document, and each defines a research front on which collaboration is invited.¶ 1

14.1 The sixteen limits and their research fronts

Empirical validation and the hierarchy of results. The formal model and the estimation protocol are specified; their general empirical execution is not. The conformity of the equations, the reconciliation of the reference case and the availability of an identification plan constitute necessary research conditions, not a validation of the amounts. A result can change status only after publication of the data and scripts, estimation of uncertainties, sensitivity tests, comparison of counterfactuals, independent replication and examination of null or contradictory results.¶ 2

1. From measurement to price. The conversion chain (metric → coefficient → price) is specified institutionally, but not yet econometrically: the formation of the reference price under uncertainty, and its incidence, call for formal treatment. The attribution model, by contrast, is now written formally (Annex F) and its estimation protocol — identification conditions, fallback scale — is specified (Annexes G and J); what remains open is the econometric estimation itself, on published data.¶ 3

2. Resistance to capture. The anti-capture devices are design claims; they call for testing by game theory and by historical comparison with the record of existing rating agencies, standard-setters and courts.¶ 4

3. Distributive effects of border adjustment. The Mechanism’s special and differential treatment must be calibrated so that the equity correction does not turn into a penalty on poverty; the emerging CBAM evidence base constitutes the natural laboratory.¶ 5

4. The citizen instrument. The design of the permanent body of the citizen instrument (selection, mandate, agenda-setting power) remains the least specified component — deliberately, since it is the one component that should not be designed by experts alone.¶ 6

5. Pilot design. The national platforms pathway requires a complete protocol: which country characteristics, which sectors, which success indicators.¶ 7

6. An intermediate continental scale. The anatomy of the spillover effects in section 9 is quantified at the national scale; the intermediate (continental) estimate and the global extrapolation alike remain to be constructed. The magnitudes of the Tunisian case in Section 9 are, since revision 2.1 of Annex F, requalified pre-Configuration M scenarios (§F.16); their evidentiary status, the reconciliation obligation and the recalculation trajectory are consolidated in Annex I. The reconciliation of the reference case with the programme’s national net is, for its part, now discharged with a zero residual (Annex I, §I.5).¶ 8

7. The political economy of adoption. The subsection devoted to the production and the discipline of political will specifies a first theory of activation — visibility, coalitions, graduated commitment, incorporation, peer review, transition financing and safeguards against capture. It remains a design proposal: the causal effect of each of these mechanisms on accession, on coalition stability, on the behaviour of dominant actors and on perceived legitimacy must be tested on the pilots.¶ 9

8. Reproducibility.¶ 10

The review of sources, the Sevilla mapping, the aggregate synthesis of externalities, the spillover model and the simulation results require public data, codebooks, equations and changelogs before independent replication is possible. The model, its multidimensional identification protocol and its extension to the forty-four least developed countries are now specified (Annexes F, G, J); publication of the data, codebooks and scripts remains the outstanding condition.¶ 11

9. Legal feasibility.¶ 12

The legal pathways are reasoned hypotheses, not opinions covering the whole set of relevant jurisdictions; questions of trade, taxation, treaties, arbitration, immunity, data protection and constitutional law require validation by specialists.¶ 13

10. Governance and strategic behaviour.¶ 14

Appointment, voting, funding, withdrawal, the strategic choice of jurisdiction (forum shopping), coalition breakdown and coordinated capture are not yet modelled.¶ 15

11. Dependence on data and on AI.¶ 16

Continuous global measurement depends on lawful access to data, on concentrated infrastructure and on model performance across languages and contexts; operation in degraded mode remains to be tested.¶ 17

12. Cost and institutional burden.¶ 18

The operating cost of the architecture — bottom-up budget, staffing model, decision times and interface costs — has not yet been costed. It is plausible that the cost of a pilot arrangement, or of a federated architecture relying on existing institutions, remains modest in relation to the macroeconomic costs of the recurrent systemic crises it seeks to prevent, running into the trillions (§13). This proportionality hypothesis must nonetheless be established by a bottom-up budget and by an estimate of the share of those crises that is genuinely avoidable; it does not, at this stage, permit any determinate quantitative ratio to be asserted. The feasibility question therefore concerns the sizing of the instantiation — and the demonstration of that sizing — not a trade-off settled in advance between the benefit of separation and its cost.¶ 19

13. Inter-coder reliability. The two classification exercises — the Sevilla mapping and the register of United Nations sources — were coded by a single coder within the programme; the register was built pillar by pillar and required deduplication (each merger being documented in the supporting evidence pack, the original being archived in full), and neither exercise yet includes an independent second coding with published agreement statistics. Both remain open precisely to that test.¶ 20

14. Asymmetry of quantitative easing and quantitative tightening. The specific effect of quantitative tightening on spillovers, distinct from that of rate increases, remains poorly identified: its only large-scale episode is recent (2022-2025), it is not the exact mirror image of the easing deployed in crisis on illiquid markets, and the empirical separation of its flow, stock and announcement components has not been established. Isolating the balance-sheet component of the shock, propagating it through the fifteen mechanisms by way of predetermined exposures, and testing the asymmetry on the recipient economies constitute a dedicated front; the end of the Federal Reserve’s tightening, in December 2025, opens the window of observation for it.¶ 21

15. Validation of the integrated demonstrator. The artificial intelligence data centre case of Section 8 is a parameterised simulation, not a field evaluation. It verifies the coherence of the interfaces and reveals the data that are missing; it demonstrates neither that the rating thresholds are empirically valid, nor that the proposed consequences would be legally incorporable in a determinate jurisdiction. Its transformation into evidence requires a real project, a pre-registered protocol, audited operator data, participation by the affected communities, an independent comparison of competing architectures and the publication of errors and challenges.¶ 22

16. The appointment of the independent panels. The architecture rests, in several of its links, on independent panels: measurement, qualification, the assessment of alerts. Their independence is addressed through negative mandates, the publication of methods, adversarial procedure and the right of challenge. None of these devices answers the prior question: who appoints.¶ 23

The problem is structural, not practical. A panel appointed by the States it is charged with rating reproduces at the entrance the conflict that the separation of functions eliminates at the exit — this is the identified cause of failure of the 2016 rating proposal (§9.9). A self-appointed panel, or one co-opted by its peers, escapes that conflict but has no opposable legitimacy and finds itself exposed to the first serious objection on representativeness. The known intermediate solutions — random selection from qualified lists, appointment by academies, non-renewable terms, mixing of regional panels — mitigate the risk without resolving it: each displaces the question towards the authority that draws up the list or recognises the academies.¶ 24

No satisfactory solution is proposed at this stage. This limit is of a different nature from the fifteen preceding ones: those call for work — replication, costing, legal opinions —, this one calls for an institutional solution that does not yet exist, and that no quantity of data will produce. It must therefore figure among the open questions of the architecture, not among its points to be consolidated.¶ 25

It also constitutes a stopping criterion: so long as a mode of appointment resistant to the double objection — capture by the rated, illegitimacy of the self-appointed — is not specified and tested, the qualification function should not pass beyond the pilot stage, whatever the quality of the measurement that feeds it.¶ 26

Each of these limits corresponds to a planned companion document; the Sevilla mapping note and the methodological document are the most advanced.¶ 27

These additional limits broaden this document into a staged research programme covering reproducibility, legal feasibility, governance, dependence on data and on AI, as well as institutional cost.¶ 28

The architecture and “bad GDP”. What national accounting records the wrong way round — defensive expenditures counted as positive — and what it leaves off its books — the costs displaced onto others and onto the future — forms a gap between what is counted and what counts, which the genuine progress tradition has named and measured without ever institutionalising it (§5.2). The 7+3 architecture acts on that gap along its chain: the HuIB makes it visible (measurement); GIRA designates who produces what share of it; GFEA and the Border Fairness Mechanism price that share — contracting externality-intensive and purely speculative activity; the Solidarity Fund makes the causer pay and makes reparation to the injured party — internalising the displaced cost; the Digital Authority and the Court govern and adjudicate. The result is not a smaller GDP but a GDP of better composition: less illth, as much real wealth. Two reservations: the architecture reaches only the cross-border and intergenerational share of that gap — purely domestic defensive expenditures escape it —, and the objective remains the quality of growth, not degrowth. Above all, narrowing this wedge is not the raison d’être of the architecture: it is what a chain designed first for accountability does to the books of the world. What counts ends up being counted — because the body that measures it and the chain that draws the consequences from it have been built.¶ 29

14.1.1 Minimal external validation package

Change of status. The passage from a work of design to an empirically and legally consolidated architecture requires a minimal validation package. No isolated element — publication of the data, legal opinion, demonstrator or institutional support — suffices on its own.¶ 30

Eight deliverables. The package comprises: (1) the open corpora and codebooks, with a version log; (2) an independent double coding and the agreement statistics; (3) the equations, data, scripts and execution environments; (4) an independent replication of the main case and the publication of null results; (5) at least four distinct legal opinions covering public international law, trade and taxation, immunities and enforcement, data and fundamental rights; (6) a bottom-up budget comparing configurations M, F and I; (7) a pre-registered multi-arm pilot; (8) a documented consultation of the populations and institutions concerned.¶ 31

Transparency rule. Each deliverable is published with its commissioning party, its conflicts of interest, its assumptions, its inaccessible data and its divergences. External assessors are not invited to approve the model, but to identify the conditions under which it must be simplified, merged or rejected.¶ 32

Threshold of claim. Before the completion of this package, the document may claim an explicit theory, a candidate architecture and specified protocols. It must not claim the empirical superiority of the 7+3, the general causal validity of the estimates, universal legal compatibility or field effectiveness.¶ 33

This package encompasses and extends the minimal reproducibility package of Annex F (§ F.14, Table F3): it adds to it legal validation, the pre-registered multi-arm pilot and the documented consultation of the populations concerned.¶ 34

14.2 Headwinds accepted

In preparing and publishing this document, I fully appreciate the headwinds, and I accept the risks taken. The first is disciplinary: a paper that calls at once upon public and international economics, finance, international law, international relations theory, institutional design, the statistics of measurement, the governance of artificial intelligence and ethics fits into almost no classical journal category, nor within the remit of a customary panel of reviewers. The second is methodological: an ambition as cross-cutting as this, carried by a single article, mechanically implies less depth per discipline — which is precisely what the empirical programme of the present report (Annexes G, I and J) is charged with correcting. The third is political: the interests that the architecture proposes to discipline, and several currents of thought that it unsettles, may judge the undertaking problematic — each for different reasons, which is in itself an indication of its independence. The fourth is geopolitical: the trend of the moment runs towards a multipolar world, less organised rather than more. Despite all this, I forbid myself to keep to myself what practice has taught me, when unpriced harms run into trillions of dollars a year, when the associated avoidable mortality is counted in millions of lives (§2.4), and when humanity, better organised, already possesses, for the most part, the remedies.¶ 35

@incollection{yaiche_hc_chapitre_14,
  author    = {Yaïche, Nizar},
  title     = {14. Limits, refutability and research programme},
  booktitle = {The Humanity Compact},
  year      = {2026},
  version   = {2.3},
  doi       = {10.5281/zenodo.22746533},
  publisher = {Zenodo},
  url       = {https://humanitycompact.org/ouvrage/chapitre-14/}
}

Ouvrage de Nizar Yaïche, ancien ministre des Finances de la République tunisienne. Qui est l’auteur ?