Crisis Science Institutions

Established Crisis science institutions are the machinery that turns an emergency into evidence, and the single most important thing known about them is that the speed-versus-rigour trade-off is largely an artefact of trial bureaucracy rather than of statistics. In 2020 one British platform trial went from protocol conception to first randomised patient in nine days, enrolled more than forty thousand people, and changed clinical practice worldwide within three months on a fully randomised result. No methodological corner was cut. What it took was a two-page data form, outcomes pulled from routine health records, and permission to randomise inside ordinary care.

Established The corresponding failure was not too little research but far too much of the wrong kind. Thousands of trials were registered worldwide within months, a large majority too small to answer anything; one drug alone attracted over two hundred registered studies. The scarce resource in a crisis is not funding or enthusiasm. It is patients, and uncoordinated enrolment consumes them.

Frontier The question this brief owns is what part of the 2020 performance was a system and what part was an emergency. Scientific Advisory Institutions holds how states take technical advice and whether any of it improves policy; Future Civil Services, whether the permanent bureaucracy reforms itself; Civilization Resilience Planning, the extreme tail. None holds the research machinery itself: the trial platforms, shared protocols, data standards, living syntheses and after-action inquiries that decide whether a crisis produces knowledge or noise.

Established The answer, on the evidence, is that the fast parts were the parts written down before the crisis. The clinical characterisation protocol that gave Britain patient data in January 2020 had been drafted years earlier and sat waiting. The adaptive platform trial that delivered the corticosteroid, immunomodulator and anticoagulation results had been designed in 2016 with a pandemic appendix that simply activated. Everything improvised from scratch was slow, and most of it produced nothing.

Frontier The strongest negative finding in the field is about preparedness measurement. The most prominent pre-pandemic index of national health security ranked the United States and the United Kingdom first and second in the world in 2019; published analysis found the index was not predictive of pandemic outcomes among comparable countries. The instrument that was supposed to tell governments whether they were ready did not.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established Start with the demonstration, because it is the only large one this field has. The British RECOVERY platform trial randomised its first patient nine days after the protocol was conceived in March 2020, ran across most acute hospitals in the country, and reported by mid-June that dexamethasone cut twenty-eight-day mortality substantially in ventilated patients and modestly in those on oxygen alone, with no benefit where no respiratory support was needed. The drug was off-patent and cost pennies. The same trial showed hydroxychloroquine and lopinavir-ritonavir did not work — negatives that stopped other trials consuming patients.

Established The design features that made it possible are all boring and all transferable. Eligibility was a clinician’s judgement that a patient might benefit with no clear indication or contraindication. Randomisation data fitted on a short form. Outcomes came from national record linkage rather than site staff. Review was expedited, not waived. None of this is methodological innovation; it is the removal of accumulated procedure, removable in a crisis because somebody had authority to remove it.

Established The pre-written protocols were the other half. The ISARIC clinical characterisation protocol had been drafted after earlier outbreaks and activated in the United Kingdom in January 2020, eventually enrolling hundreds of thousands of patients across dozens of countries; it is the reason clinical descriptions of the disease existed within weeks. The REMAP-CAP adaptive platform trial had been designed years earlier for community-acquired pneumonia with a pandemic stratum written in, and that stratum activated; it delivered the immunomodulator and anticoagulation results. Both were investments made in the absence of a crisis, by people with no way to prove the investment would pay.

Established Against that, the aggregate research response was wasteful on a scale that is itself a finding. Thousands of registered studies, most underpowered, many observational where randomisation was feasible, and a long tail of single-arm reports. Contemporaneous analysis of the registered trial portfolio found the great majority incapable of changing practice. The failure mode is a commons problem: every investigator has a private incentive to run their own study, and the depleted shared resource is eligible patients.

Established The international coordinated trial worked but showed the cost of coordination. The World Health Organization’s Solidarity trial enrolled over fourteen thousand patients across thirty countries and reported that remdesivir, hydroxychloroquine, lopinavir and interferon had little or no effect on mortality. It was genuinely global, which is what a multilateral trial is for; it was also slower to stand up and thinner in its data than the national platform, which is what multilateral coordination costs.

Established Living evidence synthesis worked and then stopped, and the stopping is the evidence. Australia’s national clinical evidence taskforce produced living guidelines updated on a weekly cycle across hundreds of recommendations, and was widely regarded as the best national example of the model. It wound down when the emergency funding ended. The World Health Organization’s living therapeutics guideline went through many versions. The method is proven; what has never been demonstrated is a steady-state funding arrangement that keeps a living synthesis alive between crises, which is exactly when it would have to be built.

Frontier The rapid-review methods literature licenses the shortcut, within limits. Comparisons of rapid reviews against full systematic reviews of the same question generally find concordant conclusions, with divergence in a minority of cases and no reliable way to identify in advance which minority. That is a usable result: rapid synthesis is defensible for most questions and dangerous for the ones where it is not, and nobody can tell you which you are holding.

Established Emergency authorisation is where speed bought the wrong thing, and the record is documented. The United States issued an emergency use authorisation for hydroxychloroquine in late March 2020 and revoked it in mid-June once evidence accumulated; it issued one for convalescent plasma in August 2020 on non-randomised data, accompanied by an efficacy claim the agency subsequently walked back, and randomised evidence later showed no mortality benefit. The contrast with the platform trial is exact: both routes were fast, one produced a durable answer and the other produced a reversal.

Frontier Advisory institutions under crisis are covered in depth elsewhere on this map; the part that belongs here is the interface with evidence generation. The British parliamentary lessons-learned inquiry and the statutory Covid-19 Inquiry both found preparedness planning oriented to pandemic influenza, a structure slow to question its own assumptions, and recommendations from a 2016 national exercise that were not implemented before 2020. The government’s published response to the inquiry’s decision-making module, issued in March 2026, is the current formal record of which findings were accepted.

Established The declaration machinery has been studied systematically and is not neutral. Document analysis of a hundred and one emergency committee statements under the International Health Regulations finds that determinations of a public health emergency of international concern are shaped by considerations beyond epidemiology, and the political-science literature on the same instrument reaches compatible conclusions. This matters for crisis science because the declaration is the switch that unlocks emergency research authorities in most jurisdictions.

Frontier The multilateral rebuild produced an instrument and has not yet produced a regime. The World Health Assembly adopted a pandemic agreement in 2025, with entry into force conditional on a separate annex governing pathogen access and benefit sharing; negotiations on that annex were extended through 2026. Amendments to the International Health Regulations agreed in 2024 created a higher “pandemic emergency” tier. Whether any of this changes what happens in the first month of the next outbreak is unresolved, and the honest test is not ratification counts but activation times.

Frontier The first open question is whether the emergency conditions can be made permanent. Everything that made the fast trial fast — expedited ethics, national data linkage, an instruction to hospitals to participate, a regulator willing to move — existed because a government had declared an emergency. No state has been shown to maintain those conditions in peacetime, and the approvals compressed in 2020 have largely returned to their prior timescales.

Frontier The second is whether pre-authorisation can be made routine rather than heroic. A standing platform trial with approved protocols, contracted sites and dormant ethics permissions is technically straightforward and institutionally homeless: it has no patients between crises, so no clinical budget wants it, and it produces no papers, so no research funder scores it. The two instances that worked in 2020 both existed because individual investigators kept them alive on small grants.

Frontier The third is whether data standards survive contact with credit. Genomic and clinical data sharing in 2020 was faster than in any previous outbreak and was also the subject of a sustained dispute over access terms, attribution and whether restricted-access databases are a reasonable price for participation. The dispute is not technical. It is about who gets named on the paper, and it recurs in every outbreak because the underlying incentive never changes.

Frontier The fourth is whether uncertainty can be communicated without cost, and here the evidence is better than the folklore. Experimental work finds that transparent communication of evidence and of its uncertainty does not undermine trust in the evidence, and that communicating uncertainty in advance protects against the loss of trust that occurs when advice later changes. The 2020 reversals on masks and on aerosol transmission are the cases that made officials fear candour; the measured result says candour was the cheaper option.

Speculative The fifth is whether any of this transfers to a non-medical crisis. The platform-trial model presumes many similar units, a measurable outcome and an ethical basis for randomisation. A financial crisis, a grid failure or a conflict has none of those. What might transfer is the protocol-before-the-event discipline, and there is no worked example of it outside health.

Established The binding bottleneck is eligible patients, and nothing in the system allocates them. In an outbreak the number of people who can be enrolled in the first months is small and fixed by the epidemic curve. Every uncoordinated study consumes some of it. There is no mechanism in any country that prioritises studies against a shared enrolment budget, and the institution that would do it — a national research prioritisation body with authority over approvals — exists nowhere in a usable form.

Established The second is that outcome data lives outside the trial. The fast trial worked because deaths and admissions could be read from national records rather than collected by exhausted clinicians. Countries without population-wide record linkage cannot run that design, and most do not have it. This is the single largest determinant of which countries can produce crisis evidence and which can only consume it.

Frontier The third is approval latency, which is measurable and almost never measured. The interval from protocol to first randomised patient is a number every trial knows and almost no institution publishes. Without a distribution of those intervals across studies and countries there is no way to tell whether a reform shortened anything, which means preparedness investment in this area is unevaluated by construction.

Established The fourth is that living evidence has running costs and no owner. A living guideline requires a standing team, continuous search and a governance process that can change a recommendation in a week. Those are salaries, not grants, and health systems fund guidelines as projects.

Established Everything here depends on routine health data that a trialist can lawfully reach in days. That is a legal and infrastructural precondition, settled long before any outbreak, and it is what separates the countries that generated practice-changing evidence in 2020 from those that did not.

Established It depends on a case definition and a minimum clinical dataset agreed in advance. The value of the pre-written characterisation protocol was not its content but its existence: agreeing what to record, while trying to record it, costs weeks that the epidemic does not return.

Frontier It depends on an advisory interface that can say what is not known. Scientific Advisory Institutions establishes that whether advisory output improves policy is unmeasured on the advisory establishment’s own admission. This brief inherits that gap: a crisis research system can generate a clean answer and still fail at the point where the answer meets a decision.

Established It depends on an emergency declaration to unlock the authorities. Since the declaration is a political instrument shaped by more than epidemiology, the research system’s activation is downstream of a judgement call that the research system does not control.

Frontier The decisive test is activation time, and it is measurable to the day. The number that would settle the central question of this brief is the interval from pathogen identification to first patient randomised under a pre-authorised protocol, published for every country that claims a capability and scored in every preparedness exercise between now and the next outbreak. In 2020 that interval was nine days in the United Kingdom, and it was nine days because a protocol had been written years earlier and a government suspended the procedure that normally consumes months.

Established Nothing blocks the measurement. Every trial records the date its protocol was finalised, the date of each approval and the date of first randomisation. The distribution of those intervals, by country and by study type, would show immediately whether post-2020 reforms shortened anything or merely renamed it. No journal requires it, no regulator publishes it, and no preparedness exercise is scored on it, which is why a field that measures everything else has no series for the one number that describes its core capability.

Frontier The second experiment is to run the platform in peacetime. Maintain a standing adaptive trial with approved protocols, contracted sites and dormant ethics permissions for a recurring seasonal illness, and measure what it costs per year and how fast it can add an arm. This is the only way to learn whether the pre-authorisation model has a steady state, and it is cheap relative to what a single wasted crisis trial costs.

Established A third has already returned a result and is treated as a curiosity. Countries that suffered a serious respiratory outbreak in the preceding fifteen years and rebuilt their institutions afterwards performed markedly differently in 2020 from countries that wrote reports. That is a natural experiment in post-crisis learning with a clear direction: institutional change followed recent domestic experience, not analysis.

Frontier A fourth is already running in the surveillance of animal influenza. The response to a sustained outbreak in livestock is a live test of whether the post-2020 machinery activates in the absence of a declared human emergency — whether sequences are deposited promptly, whether serosurveys of exposed workers are done, whether the standing protocols are pulled off the shelf. The evidence to early 2026 is mixed enough that both readings are available, which is itself informative about how much the machinery depends on a declaration.

Established The engineering requirement is a randomisation service that a clinician can use in ninety seconds. Web-based randomisation with eligibility checked by a single clinical judgement, no separate consent infrastructure beyond what care already requires, and no data entry beyond a short form. This is solved technology deployed at national scale exactly once.

Established The second is outcome ascertainment by linkage rather than by collection. Mortality, admission, discharge and prescription data already exist in national systems. Building the legal and technical pipeline that lets a trial read them is a multi-year project that must be finished before the crisis and that no crisis budget will ever fund.

Frontier The third is a data standard that is boring enough to be adopted. Shared case report forms, agreed variable definitions and a common minimum dataset are the least glamorous deliverable in preparedness and the one with the highest measured return, because they are what allows twenty countries’ data to be pooled instead of described separately.

Frontier The fourth is a synthesis pipeline that is partly automatic. Continuous search, deduplication, risk-of-bias triage and evidence-table assembly are largely mechanisable; the judgement about what a body of evidence means is not. Living guidelines fail on the mechanisable part, which is the part that costs staff time every week.

Established This sits alongside Scientific Advisory Institutions, which owns the advice side and explicitly leaves the evidence-generation side open. That brief’s finding — that whether technical advice improves policy is unmeasured — bounds what this one can claim: a crisis research system that produces a clean result has done its job and has not thereby changed an outcome.

Established It is adjacent to Future Civil Services on the question of whether the permanent machinery can be reformed at all. The relevant transfer is that reform measured against its own targets has a poor record, which is the prior to apply to every post-inquiry commitment in this field.

Frontier It is adjacent to Civilization Resilience Planning at the tail. That brief deals with events severe enough to end the conditions under which trials are run; this one deals with the far more common case where the institutions survive and the question is whether they learn anything.

Established It is adjacent to AI Biology Governance and Existential Risk Governance, which own the deliberate-release and catastrophic-tail sides of the same institutional problem, and to Public Policy Foresight, which owns the record of whether anticipating a risk produces preparation for it.

Established The requirement that dominates is a standing authority to prioritise studies. Somebody must be able to say that a proposed trial will not be approved because it duplicates an ongoing one and would consume patients the ongoing one needs. This is an unpopular power and the only one that addresses the actual failure mode. Britain came closest in 2020 through an urgent-public-health designation that steered sites toward a handful of prioritised studies.

Frontier The second is a funder that pays for readiness rather than for output. Standing protocols, dormant approvals and maintained site networks produce nothing measurable in the years that matter for a funder’s own evaluation. No major national funder has a line item for capability held idle, and the international mechanisms built after earlier outbreaks have been repeatedly defunded between crises.

Frontier The third is a post-crisis learning body that can compel and then be checked. Inquiries are commissioned, report late, and produce recommendations whose implementation nobody tracks. Sweden’s commission was unusually fast; the British statutory inquiry is thorough and slow; the United States never established a comparable national body. None has a mechanism that reports, years later, on which recommendations were implemented.

Established The fourth is a declaration mechanism whose research consequences do not all hang on one switch. Because emergency research authorities are triggered by an emergency declaration, and the declaration is a political judgement, the research system is dark during exactly the ambiguous early period when evidence is cheapest to generate.

Established The central ethical claim of crisis trials is that randomisation is the ethical option under genuine uncertainty, and it is correct. Giving an unproven drug outside a trial exposes the patient to the same unknown risk and produces no knowledge; the compassionate-use route feels generous and is the one that left several therapies in limbo for a year. The hydroxychloroquine episode is the clean case: two hundred uncoordinated studies and an emergency authorisation produced confusion, and one platform trial produced an answer.

Frontier The opposing argument deserves its strongest form. Consent obtained in an emergency ward, from a frightened patient, through a simplified process, is thinner than consent in an elective setting — and simplification is what makes the fast design fast. That all arms are plausible and none known better is a real answer, and it does not dissolve the concern that those enrolled in crisis research are disproportionately the sick, the poor and the institutionalised.

Established The distributional record of crisis science is poor and is not primarily about consent. The countries that generated the evidence were the ones with national record linkage and research infrastructure; the countries that bore the largest burdens mostly could not run the studies and had to apply conclusions drawn elsewhere. Multilateral trials partly addressed this and were slower for it. That trade-off is real and is usually discussed as though it were a funding problem.

Frontier Data sharing carries the sharpest unresolved dispute. Restricting access to sequence data protects those who generated it from being scooped and slows everyone else down. Open deposition speeds analysis and removes the incentive that made some laboratories willing to share at all. Both effects are real and observed, and the instrument proposed to resolve them — a benefit-sharing annex tied to a multilateral agreement — remains under negotiation.

Frontier The civilizational question is whether the ability to learn under emergency is a durable property of a society or a one-off. The 2020 evidence performance was not evenly distributed and was not predicted by any preparedness index. It tracked something closer to institutional memory: places with a recent severe domestic outbreak had rebuilt, and places with reports had not.

Established That implies an uncomfortable regularity. Post-crisis learning appears to be driven by recent direct experience rather than by analysis, which means the learning decays on the timescale of a career and cannot be transferred by document. Every mechanism proposed in this field is an attempt to substitute an institution for that memory, and none has yet been shown to work across a generation.

Frontier The asymmetry that makes this worth solving is the cost ratio. The standing infrastructure — protocols, approvals, data pipelines, a small permanent staff — is cheap against a single month of an uncontrolled epidemic or a wasted year of the wrong therapy in wide use. The obstacle is not cost but the absence of a budget line whose owner is evaluated on readiness.

Speculative The most plausible durable form is a small permanent secretariat attached to an existing clinical network rather than a new agency. Agencies created for a crisis are abolished after it; capability lodged inside something that has ordinary daily work survives. That is the pattern behind both of the pre-written protocols that mattered in 2020, and it is not a design anyone chose.

These horizons track whether crisis research capability becomes standing infrastructure or reverts to improvisation, not when the next crisis arrives.

• 10 yr: Frontier Activation time from pathogen identification to first randomised patient becomes a published preparedness metric in at least one country; the pandemic agreement either acquires its benefit-sharing annex and enough ratifications to enter into force or does not; living evidence services find steady-state funding in one or two health systems and lapse everywhere else.
• 25 yr: Speculative Standing platform trials with dormant approvals exist in several health systems because a second respiratory emergency made the case that reports could not; trial-by-record-linkage becomes the default design for common conditions rather than a crisis exception.
• 50 yr: Speculative The protocol-before-the-event discipline transfers outside health to at least one domain with many similar units and a measurable outcome — grid operations or agriculture are the plausible candidates — or the idea stays a medical curiosity.
• 100 / 250+ yr: Handwave Claims that automated evidence synthesis or simulation removes the need for crisis trials work by assertion: both depend on prior data about the specific novel agent, which is the thing a novel agent does not have.

• Depends on This brief depends on results other briefs here already hold. Scientific Advisory Institutions establishes that whether technical advice improves policy is unmeasured on the advisory establishment’s own admission, which bounds what a crisis research system can be credited with. Public Policy Foresight holds the record of whether anticipating a risk produces preparation for it, which is the mechanism by which a pre-written protocol comes to exist at all. Future Civil Services supplies the base rate for bureaucratic reform against its own targets, which is the prior for every post-inquiry commitment in this field. Nothing scientific blocks this topic.
• Requires (not on this map) Five constraints sit outside what research delivers. The first is standing ethics approval for pre-authorised platform protocols — dormant permissions that wake rather than approvals that begin. The second is routine outcome data linkage reachable by trialists in days, the infrastructure that decides which countries can generate crisis evidence and which can only import it. The third is steady-state funding for living evidence between crises, since the method is proven and the financing is not. The fourth is an authority to refuse duplicative crisis trials, the unpopular power that addresses the actual failure mode of patient depletion. The fifth is a published distribution of protocol-to-first-patient intervals, without which no preparedness reform in this area can be evaluated.
• Enables A standing, pre-authorised research capability enables the thing this field has demonstrated once and never institutionalised: an actionable randomised answer within three months of a novel pathogen, negative results published fast enough to stop other investigators wasting patients, and pooled international data because the case report forms already matched. It also enables ordinary clinical research: a design that runs inside routine care at low marginal cost is not specific to emergencies.
• Adjacent Adjacent to AI Biology Governance and Existential Risk Governance, which own the deliberate-release and catastrophic-tail sides of the same institutional question; to Civilization Resilience Planning, which owns the severity range beyond which this machinery stops operating; to Technocracy and Democracy, which owns the legitimacy of expert authority exercised at speed; and to Cloud Laboratories, which owns the automation of the bench work a crisis would need to parallelise.

Established “In a crisis you have to trade rigour for speed.” The clearest counterexample in modern medicine is a fully randomised trial that went from protocol to first patient in nine days and to a practice-changing mortality result in three months. What was traded away was not rigour but procedure: long case report forms, bespoke data collection, site-by-site contracting and sequential approvals. Where rigour actually was traded — emergency authorisations issued on observational data — the results had to be reversed. The trade-off is real for a single bespoke trial and largely false for a well-designed platform.

Established “The problem was that we did not do enough research.” Thousands of studies were registered in the first year, and the analyses of that portfolio found most of them too small or too poorly designed to change anything. One repurposed drug attracted more than two hundred registered trials. The binding constraint was coordination, and the resource being wasted was patients.

Frontier “Preparedness indices tell you who is ready.” The most prominent index ranked the United States and the United Kingdom at the top in 2019, and published analysis found it was not predictive of pandemic outcomes among comparable countries. This does not mean preparedness is unmeasurable; it means that what the index measured — documented plans, formal capacities, legal instruments — was not what determined performance. Activation times and recent institutional memory look more predictive and are not in the index.

Frontier “Admitting uncertainty destroys public trust.” Experimental evidence points the other way: transparent communication of evidence and of its limits does not undermine trust in the evidence, and signalling uncertainty in advance protects against the trust loss that follows when advice changes. The officials who feared candour in 2020 were acting on an intuition that the measured results do not support. What does damage trust is a confident claim later reversed without explanation.

Established “Emergency use authorisation is a fast track to evidence.” It is a fast track to deployment, which is a different thing and sometimes the opposite. Once a product is available outside a trial, enrolment in the trial that would have evaluated it becomes harder, and an authorisation granted on weak evidence can foreclose the study that would have produced strong evidence. Two of the most prominent early authorisations were later withdrawn or contradicted by randomised results.

Frontier “The World Health Organization can coordinate the research response.” It coordinated a genuinely global trial that produced solid negative results on four therapies, which is a real achievement. It also has no authority over national approvals, no access to national health records, and an emergency-declaration function that document analysis shows is shaped by considerations beyond epidemiology. The multilateral layer can pool and can convene; it cannot randomise a patient.

Frontier “The pandemic agreement fixed the international system.” An agreement was adopted in 2025 and its entry into force was made conditional on a separate annex on pathogen access and benefit sharing, negotiations on which were extended through 2026. Treating adoption as the end of the process mistakes the instrument for the regime, and the operative test for crisis science is not how many states ratify but whether anything activates faster in the first month of the next outbreak.

Handwave “Next time we will be ready, because we wrote it all down.” The British statutory inquiry found a 2016 national exercise had identified capability gaps whose recommendations were not implemented before 2020, and the parliamentary lessons-learned inquiry reached compatible conclusions. The countries that performed best in 2020 were generally those that had lived through a serious domestic outbreak recently enough to have rebuilt. Writing it down is necessary and has never been sufficient.