15. From corrective accountability to anticipatory accountability: the prevention loop

Chapter 15 of 15 · ≈ 10 min read · 32 paragraphs · section 21 of 40 · search the book

An accountability chain that begins only after the harm has materialised remains incomplete where the mechanisms of propagation are reasonably foreseeable.

Status of this chapter. This chapter develops an exploratory extension of the framework: a cross-cutting prevention and alert loop (Preventive Accountability Loop, PAL). It is a refutable design proposal, of the same evidentiary status as the application to peace — neither a result, nor an eighth institution. Prevention is treated here as a cross-cutting function, distributed across the seven existing functions, whose definitive hosting remains open and must be tested against a federated solution.¶ 1

The prior link: to prevent. An accountability chain that begins only after the harm has materialised remains incomplete where the mechanisms of propagation are reasonably foreseeable. In interdependent systems, the first decisions alter the conditions to which the other actors react; those reactions in turn become shocks. The risk is therefore not only that of an isolated harm, but that of a cascade produced by several decisions that are locally rational and globally uncoordinated. Responsibility cannot then be confined to measuring after the event: it must also organise the manner in which credible forward-looking information enters the decision before the buffers are exhausted.¶ 2

A two-loop architecture. The present work accordingly distinguishes two complementary loops. The Global Accountability Chain remains the principal loop: it organises the treatment of an observed or sufficiently measurable effect, from evidence to redress. The Preventive Accountability Loop is cross-cutting: it organises anticipation, the detection of cascade risks, the qualification and notification of the alert, the reasoned response of the decision-maker, any anticipatory action and ex post learning. This second loop does not create a general competence to prevent risky decisions. It creates a procedural duty of diligence: reasonably foreseeable risks must be capable of being known, transmitted, examined and reasoned, while forecasting errors must be measured and made public.¶ 3

15.1 Two temporalities, one and the same accountability

Two temporalities. The Global Accountability Chain and the Preventive Accountability Loop answer two moments of the same problem. The first intervenes where an effect is observed, measurable or sufficiently materialised to be investigated: it preserves the evidence through to qualification, to any consequence and to the remedy. The second intervenes where the harm has not yet occurred, but where a combination of decisions, exposures and vulnerabilities makes a cascade reasonably foreseeable. It transforms forward-looking knowledge into an alert, and then the alert into an obligation of examination and response.¶ 4

One and the same discipline. The two loops apply the same principles: data provenance, declaration of uncertainty, separation of roles, participation, reasoned justification, the possibility of independent counter-assessment, control of conflicts of interest and stopping rules. Prevention therefore constitutes no exception to accountability; it brings forward its point of entry in time.¶ 5

No absolute separation. The distinction is not strictly chronological. An effect already realised may reveal a future risk; an alert may be revised as the data become observable; an anticipatory action may produce its own externalities and enter the principal chain. The two loops share a versioned case file and may reactivate one another.¶ 6

15.2 Preventive accountability

Definition. Preventive accountability exists where an actor or a system is required to account for the manner in which it identified, transmitted and examined a reasonably foreseeable risk before its materialisation. It does not create an obligation of result consisting in preventing all harm. It creates a procedural obligation of diligence: to seek out the relevant signals; to test several scenarios; to flag the uncertainties; to notify the competent recipients; to examine the available measures; to give reasons for the decision; and to assess after the event the quality of the alert and of the response.¶ 7

Activation threshold. The loop is not activated for every abstract possibility. Four minimal conditions are required: a plausible cross-border or intergenerational link; identifiable transmission mechanisms; a material potential harm in the light of pre-registered thresholds; and a time window within which reasonable action could still reduce the exposure, the vulnerability or the magnitude of the cascade. Where these conditions are not met, the correct outcome is monitoring or closure.¶ 8

Nature of the alert. An alert is an act of knowledge and of procedure, not a policy decision. It describes a scenario, a horizon, mechanisms, exposures, thresholds, uncertainties and a set of options. It determines neither the policy rate, nor fiscal policy, nor the legal adjustment to be adopted. The decision-maker retains its mandate and its competence; it must only render visible the manner in which the alert was examined.¶ 9

15.3 The seven operations of the preventive loop

1. Anticipate. To construct several plausible trajectories, including a central scenario, adverse scenarios and regime breaks, without converting radical uncertainty into artificially precise probability.¶ 10

2. Detect. To identify the exposures, vulnerabilities, thresholds, feedbacks and second-round reactions liable to transform an initial impulse into a cascade.¶ 11

3. Qualify the alert. To assess the plausibility, the magnitude, the urgency, the reversibility and the distributive reach of the risk, distinguishing signal, extreme scenario and established result.¶ 12

4. Notify. To transmit an actionable alert to the authorities and parties concerned under a graduated disclosure regime, with the data, assumptions, uncertainties, recipients and time limits.¶ 13

5. Respond. To obtain from the competent decision-maker an acknowledgement of receipt and a reasoned response indicating the elements accepted, those contested, the options examined and the reasons for the decision.¶ 14

6. Act. To prepare or activate, where this is proportionate and legally authorised, measures of mitigation, of capacity, of liquidity, of social protection, of phasing or of anticipatory financing.¶ 15

7. Evaluate. To compare the alert with the outcome, to measure false positives and false negatives, to identify the specific effects of the alert and of the response, and then to correct the models, thresholds and procedures.¶ 16

15.4 Articulation with the principal chain

Entries into the chain. The preventive loop may feed the principal chain in three ways. An alert may lead to mere monitoring; it may produce a measure of prevention or of solidarity without attribution of a harm; or it may open a case file in the chain where effects become observable. Conversely, a case file processed by the chain may reveal a recurrent mechanism and feed the prevention scenarios.¶ 17

Shared case file. The preventive case file preserves the same architecture of evidence as the accountability case file: source data, freeze date, scenarios, models, assumptions, uncertainties, disagreements, recipients, responses and revisions. A common identifier makes it possible to follow the passage from the alert to the observed harm without rewriting the initial assumptions after the event.¶ 18

Non-automaticity. No alert automatically creates a pricing, a sanction, a financing obligation or a presumption of liability. It may, by contrast, trigger procedural obligations: examination, notification, consultation, preparation and reassessment. Any substantive consequence remains subject to the classes, to the evidence, to the applicable law and to the forums of the principal chain.¶ 19

Download this table (CSV)

15.5 Preventing without governing by forecast

P10 — Preventing without governing by forecast. Where a cross-border or intergenerational risk of cascade is reasonably foreseeable and a window for action remains, the architecture must be able to anticipate it, to qualify it, to notify it and to obtain a reasoned response. The alert never transfers the competence of the decision-maker to the producer of the forecast; it creates neither a presumption of fault, nor a pricing, nor a sanction. The models, thresholds, recipients, uncertainties, dissents, false positives and false negatives are published or audited under a proportionate regime. The preventive loop is subject to the same rights, negative mandates, stopping rules and avenues of redress as the principal chain.¶ 20

15.6 Illustration: the monetary cascade alert

Purpose. The monetary alert protocol seeks to inform decision-makers before a combination of tightenings, of defensive reactions and of exhaustion of the buffers transforms several rational decisions into a systemic contraction. It does not recommend a global policy rate and does not subordinate domestic mandates to an external authority. It produces structured information on the plausible cross-border consequences, the thresholds and the mitigation options, and then requires a reasoned response from the participants that have consented to the protocol.¶ 21

Unit of analysis. The alert bears on an episode and a time window, not on an isolated decision. It links the impulse of the source jurisdiction, the synchronisation among source jurisdictions, the exposures of the recipient countries, their policy reactions, the state of the buffers and the distributive effects. The protocol preserves the distinction between anticipated causality, stress scenario and observed effect.¶ 22

The complete protocol — inputs, cascade risk memorandum, action options and stopping rules — is specified in Annex L.¶ 23

15.7 Objection, limit and refutability

Objection — Prevention becomes government by forecast. An authority may multiply alarming scenarios, impose costs on the basis of unrealised risks, disrupt markets or acquire a power greater than that of the responsible decision-makers. False positives may lead to alert fatigue; false negatives may create a false sense of security; publication may become self-fulfilling; confidentiality may become a pretext for opacity.¶ 24

Response. The PAL confers no original power of decision. The thresholds, levels, models, recipients and time limits are pre-registered; the alert is contestable; decisions remain with the competent authorities; no financial or legal consequence follows automatically from the scenario; alerts expire; performance is published; and a pilot compares the value of the arrangement with lighter alternatives.¶ 25

Rejection criterion. The loop must be suspended or simplified if its alerts are unstable, if the recipients cannot interpret them, if the false positives exceed the benefits, if disclosure causes significant harm, if the models converge artificially towards a single doctrine or if the obligation to respond persistently diverts decision-makers from their mandate without demonstrated gain in preparedness.¶ 26

17. Prevention may produce the risk that it announces. The loop depends on forward-looking models, on thresholds and on a disclosure regime whose effects are themselves endogenous. An alert may stabilise the system by permitting rapid action; it may also provoke a run, displace the risk, stigmatise a jurisdiction or create a dependence on the central scenario. The research programme must therefore jointly estimate the predictive performance, the decisional usefulness and the reflexive effects of the alert. It must compare models, levels of disclosure, horizons and institutional architecture, publish the errors and provide for operation without artificial intelligence.¶ 27

Refutation criterion for the preventive loop. The proposal is weakened if a light information-sharing protocol produces the same usefulness without institutional qualification; if the alerts improve neither the timing nor the preparedness; if decision-makers respond formally without modifying their understanding; or if the costs and reflexive effects exceed the plausibly avoidable harms. In those cases, the prevention function must be reduced to a monitoring capacity, merged with existing arrangements or abandoned.¶ 28

15.8 Foundations and specific contribution

Existing literatures. The loop draws on several traditions that often remain separate. Anticipatory governance links foresight, participation and learning in order to act while trajectories remain modifiable (Guston, 2014; OECD, 2024). Early warning systems organise risk knowledge, detection and forecasting, the dissemination of the alert, and then preparedness for action (UNDRR, 2024). Macroprudential policy assesses warning indicators in the light of the decision problem, the timing and the costs of error (Drehmann and Juselius, 2014). The Early Warning Exercise of the IMF and the Financial Stability Board examines low-probability, high-impact global risks and their interactions (IMF-FSB, 2010). The law on the prevention of transboundary harm offers, finally, analogies of diligence, prior assessment, notification and consultation, in a legal field that cannot be transposed automatically to monetary policy or to any economic decision (International Law Commission, 2001).¶ 29

Proposed contribution. The originality claimed therefore lies neither in early warning nor in foresight taken in isolation. It lies in their insertion into an architecture of accountability: separation between the producer of the evidence, the qualifier, the decision-maker and the funder; obligation of reasoned response; contestability of the alert itself; preservation of versions and of dissents; measurement of false positives and false negatives; and an explicit interface with a chain capable of treating the effects that materialise.¶ 30

Status. This proposal belongs to institutional design and to lex ferenda where it aims at new obligations. It does not claim that a general legal duty to prevent monetary spillover effects already exists. It proposes a procedural standard to be tested by consent, protocol, contract, internal policy or a suitable legal instrument.¶ 31

“An institution is not protected by the nobility of its mission; it is protected by the rules that prevent that mission from becoming a power that answers to no one.”¶ 32

@incollection{yaiche_hc_chapitre_15,
  author    = {Yaïche, Nizar},
  title     = {15. From corrective accountability to anticipatory accountability: the prevention loop},
  booktitle = {The Humanity Compact},
  year      = {2026},
  version   = {2.3},
  doi       = {10.5281/zenodo.22746533},
  publisher = {Zenodo},
  url       = {https://humanitycompact.org/ouvrage/chapitre-15/}
}

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