世光 · 世应实验室

METHOD

The protocol, published before it works

This is the specification for a diagnostic that separates a state's disaster response capacity from its prevention capacity. It is not finished, it has not been run on a single country, and parts of it are probably wrong. We publish it at this stage because a method that cannot be attacked before it produces results is not a method.

Version 0.1 · September 2026 · Changelog at the foot of this page

What the output is

Unit of analysis and deliverable

The unit is a country–hazard pair at a stated point in time: not "Nepal", but "Nepal, glacial lake outburst flood, 2026". Capacity is hazard-specific, and averaging across hazards is the first place a distinction like ours gets destroyed.

For each pair the protocol returns:

a response-side estimate, with an interval;
a prevention-side estimate, with an interval;
a coverage note stating which inputs were available, which were missing, and how old the oldest binding input was;
the underlying data layer, reproducible by a third party.

It does not return a combined score, a rank, or a traffic light. There is no single number. If a user needs one, this instrument is the wrong one, and we would rather say so than supply it.

What actually lands on a desk. Two estimates, two intervals, and a note saying which input widened them. The wide prevention interval is not a failure of the instrument — it is the instrument reporting that this state's prevention effort is not observable from outside with the data currently available. Numbers here are invented for illustration.

What we got wrong

A correction, before the prior art

Earlier versions of our material said that existing instruments cannot distinguish response from prevention. Checked against the sources, that is too strong, and one instrument makes the distinction explicitly.

The WorldRiskIndex (Bündnis Entwicklung Hilft with UNU-EHS, calculated since 2018 by the IFHV at Ruhr-University Bochum, 193 countries) maps vulnerability with three separate components: susceptibility, lack of coping capacities, and lack of adaptive capacities. Its own definition draws the line we care about: adaptation, "in contrast to coping capacities, refers to long-term processes and strategies to achieve anticipatory changes in societal structures and systems to counter, mitigate, or purposefully avoid future adverse impacts."

That is prior art, it predates us, and our earlier phrasing did not acknowledge it. Corrected here rather than quietly edited away.

What survives the correction is narrower and, we think, more useful. Three things.

The prior art, as we read it

Three instruments, checked against their own documentation

Instrument	Does it separate the two sides?	Is the separation preserved?	How is the prevention side measured?
INFORM Risk Index EC Joint Research Centre (scientific lead) with the IASC Reference Group on Risk, Early Warning and Preparedness Concept & Methodology (JRC, 2017)	Partly. "Lack of coping capacity" splits into an institutional category (disaster risk reduction and governance components) and an infrastructure category (emergency response and recovery).	No. The two categories are combined by geometric mean into one dimension score, which then enters the composite risk figure.	By the existence of DRR programmes, plus governance proxies.
WorldRiskIndex Bündnis Entwicklung Hilft / IFHV, Ruhr-University Bochum Methodological Notes · WorldRiskReport	Yes, explicitly. Lack of coping capacities and lack of adaptive capacities are separate components with distinct definitions.	Reported, then aggregated. Components are published separately, but the headline index remains Exposure × Vulnerability with the dimensions equally weighted.	By general development proxies — education and research, reduction of disparities, investment, disaster preparedness — not by prevention effort against a given hazard.
Sendai Framework Monitor, Target E UNDRR; also SDG 13.1.2, 1.5.3, 11.b.1 Sendai Framework Monitor · E-1/E-2 metadata	Not its purpose. Target E tracks the presence and alignment of DRR strategies, national (E-1) and local (E-2).	—	By adoption of strategies, scored 0–1 against ten criteria and self-reported by member states through the Sendai Framework Monitor.

Read September 2026; full source list in section 10. One caveat we would rather state than bury: the phrase "geometric mean" for INFORM's two coping-capacity categories is taken from the European Commission's own dataset documentation rather than from the body text of the JRC methodology report. The category structure itself is in the JRC report. If the aggregation function has changed in a later release, this row is the first thing that needs correcting.

The system says this about itself. UNDRR's 2025 global status report on national DRR strategies puts the same point more bluntly than we would have dared: "Strategies alone are not enough — what matters is turning plans into action."

The claim that survives

Narrower, and testable

The cut is different. The WorldRiskIndex cuts on time horizon: short-term coping versus long-term anticipatory adaptation. We cut on attributability: whether the success of a capacity can be credited to a government after an event. The two overlap and then diverge — a well-drilled evacuation is short-term and creditable; a maintained culvert is long-term and uncreditable; an early-warning system is a long-term investment whose success is highly creditable at the moment of a near-miss. The incentive story lives exactly where the two cuts come apart.

The two cuts are not the same cut. Where they agree, our instrument adds nothing. Where they disagree — a long-term investment that is nonetheless highly creditable, a short-cycle activity that earns no credit — the time-horizon cut and the attributability cut classify the same capacity differently, and only one of them predicts how a state will actually allocate.

Aggregation still happens at the point of decision. Components may be published separately, but the figure that travels into a screening conversation is the composite. A distinction that survives in an annex and dies in the headline is not available to the person making the call.
Nothing measures the prevention side by effect. Every instrument we have checked measures it by existence — programmes adopted, strategies aligned, general development proxies — and in the Sendai case by self-report. That is the gap we are trying to close, and it is a narrower gap than we previously claimed.

How the two sides are defined

The construct, and where it breaks

Response side

Capacity whose exercise is observable after a hazard event, and whose success can be attributed to an identifiable actor: search and rescue, medical surge, relief logistics, temporary shelter, reconstruction throughput.

Prevention side

Capacity whose success consists in an event not occurring or occurring smaller: asset maintenance, hazard-zone land-use enforcement, structural mitigation, retrofit programmes, risk-informed siting decisions.

The boundary case we do not resolve

Early warning sits across the line. Monitoring and modelling are prevention; dissemination and evacuation are response; and the whole chain is creditable when it works publicly. We do not force it into one side. It is coded as a cross-boundary chain, and what we record is which link fails — because failing at monitoring and failing at dissemination are different institutional facts that current indices score identically.

If in testing the cross-boundary category absorbs most of what matters, the construct is badly drawn and we will need a different cut. We would expect to know this within the first two country–hazard pairs.

Candidate indicators, prevention side

Four proxies, each with the condition that would kill it

The response side has usable measures already; the open problem is the prevention side. Each candidate below is stated with the finding that would disqualify it, so that testing can end an argument rather than extend one.

1 · Maintenance share of capital expenditure

Recurrent maintenance spending on hazard-exposed public assets as a proportion of capital spending on the same asset class. Prevention shows up as unglamorous upkeep.

Disqualified if budget lines cannot be attributed to hazard-exposed assets in the majority of candidate countries, or if the classification is inconsistent enough that cross-country comparison is meaningless.

2 · Monitoring density and series continuity

Density of functioning hazard-monitoring stations and, more importantly, the presence of gaps in their published series. A network installed by a donor and left unmaintained produces a characteristic signature.

Disqualified if series gaps track national telecommunications or general public-sector capacity rather than hazard-specific investment.

3 · Enforcement traces in high-risk zones

Built-up expansion inside mapped high-risk zones after a restriction takes effect, measured by remote-sensing change detection. Enforcement is invisible in law and visible from orbit.

Disqualified if observed restraint is better explained by terrain, land tenure or absence of development pressure than by enforcement — which will require a matched comparison, not a raw rate.

4 · Risk-to-action lag

Elapsed time between a documented scientific risk finding for a specific site and the first traceable administrative action referencing it. South Lhonak was scientifically flagged years before 2023; the lag is the measurement.

Disqualified if the documentary trail is available only for well-studied hazards in well-studied countries, which would make the indicator a measure of research attention rather than of state behaviour.

Note what all four have in common: they are read from outside the state, not from what the state reports about itself. That is deliberate. An indicator a government self-reports is an indicator a government optimises.

Reporting confidence

Why the output is an interval

Every estimate carries an interval and a coverage note. This is not statistical modesty; it is the point. In the places this instrument is meant to serve, data is missing exactly where capacity is weakest, so a confident score is evidence of good data, not of good capacity — and an instrument that hides that will systematically reward the legible.

Where an input is missing, absent or stale, the interval widens and the coverage note says which input did it. We would rather publish a wide interval that is true than a point estimate that is convenient. This makes our output harder to use, and we accept that cost.

How to show this is wrong

Stated at protocol level, before any results exist

Construct failure. If the cross-boundary category absorbs the majority of what matters, the response/prevention cut is not the right one.
Measurement failure. If all four candidate indicators hit their disqualifying condition, the prevention side cannot be measured from outside and the instrument cannot be built — regardless of whether the theory is correct.
Redundancy. If our paired output correlates above roughly 0.9 with the WorldRiskIndex coping and adaptive components across a reasonable country set, we have rebuilt something that exists and should stop.
Irrelevance. If practitioners tell us they already make this distinction informally and it does not change decisions, then the gap is real but does not matter, which is the same as not mattering.
Prior-art error. If we have misread INFORM, the WorldRiskIndex or the Sendai Framework Monitor in section 03, the argument for building anything weakens accordingly.

Each of these can be established without our cooperation. That is intentional.

Open, and not yet answered

Known gaps in this version

We have not yet reviewed ND-GAIN's readiness component, the GCF and Adaptation Fund accreditation criteria, or IHR/JEE health-security assessments against this frame. Any of them could contain the same cut, or a better one.
We do not know which capacity indicators adaptation funds actually cite in eligibility decisions, as opposed to which exist. That is an empirical question about practice, and we have not asked a practitioner yet.
The construct has not been tested on a single country–hazard pair.
We have no position yet on subnational application, where prevention capacity probably varies more than it does between states.

Sources

Everything section 03 rests on

Listed so that a reader can check our reading rather than take it. Grouped by how much weight each carries: what an instrument says about itself counts for more than what a catalogue entry says about it.

Primary — instrument methodology

INFORM — Index for Risk Management: Concept and Methodology, Version 2017 drmkc.jrc.ec.europa.eu · INFORM Concept and Methodology (PDF) Used for: the structure of the "lack of coping capacity" dimension — institutional category (disaster risk reduction and governance components) and infrastructure category. Current portal: drmkc.jrc.ec.europa.eu/inform-index
WorldRiskIndex — Methodological Notes (2021) weltrisikobericht.de · Methodological Notes (PDF) Used for: the separation of susceptibility, lack of coping capacities and lack of adaptive capacities, and the index formula.
WorldRiskReport — index description and definitions weltrisikobericht.de/worldriskreport Used for: the definition quoted in section 02, and for the model's history (developed with UNU-EHS; calculated by the IFHV at Ruhr-University Bochum since 2018; 193 countries; equally weighted dimensions).
Sendai Framework Monitor — targets and indicators undrr.org · Monitoring the Sendai Framework · preventionweb.net · Sendai Framework Indicators Used for: Target E and the member-state self-reporting mechanism.
SDG indicator 1.5.4 metadata — computation of E-1 and E-2 unstats.un.org · Metadata 01-05-04 (PDF) Used for: E-1 and E-2 definitions and the 0–1 scoring against ten criteria.

Institutional reporting

UNDRR — 2025 Global Status of National Disaster Risk Reduction Strategies undrr.org/2025-global-status-national-DRR-strategies Used for: the quoted statement that strategies alone are not enough, and for reported coverage of national and local strategies.
Global Assessment Report — national and local DRR strategies and plans gar.undrr.org · Chapter 11 Used for: the ten criteria distilled for monitoring alignment with the Sendai Framework, and early reporting rates under Target E.
WorldRiskIndex dataset record data.humdata.org/dataset/worldriskindex Used for: the 2022 model revision and component structure.

Secondary — catalogue documentation

European Commission — INFORM Risk Index: Lack of coping capacity (dataset record) africa-knowledge-platform.ec.europa.eu Source of the "geometric mean of two categories" wording, and of the description of the institutional category as covering the existence of DRR programmes. Secondary to the JRC methodology report; flagged in section 03.

Comparative literature

Global vulnerability hotspots: differences and agreement between international indicator-based assessments — Climatic Change doi.org/10.1007/s10584-021-03203-z Direct comparison of the WorldRiskIndex and INFORM, including the indicators they share and the fact that their vulnerability constructs are not commensurable without adjustment.
Understanding human vulnerability to climate change: index validation for adaptation planning — Science of the Total Environment sciencedirect.com · S0048969721051408 Component-level comparison of the two indices, and the observation that specific goals for developing coping and adaptive capacities are largely absent.
Top-down assessment of disaster resilience: a conceptual framework using coping and adaptive capacities (ANDRI) — IJDRR sciencedirect.com · S2212420916300887 Prior art for a coping/adaptive hierarchical design at national scale; relevant to whether our cut is genuinely new.

What we have not read. ND-GAIN's readiness component, GCF and Adaptation Fund accreditation criteria, and the IHR/JEE health-security instruments have not been reviewed against this frame. Any of them could contain the same distinction, or a better one. Their absence from this list is a gap, not a judgement.

Tell us where this breaks

The most useful message we can receive is a specific objection to section 03, 05 or 06. Zhijun He — zhijun.he@yale.edu

Changelog
v0.1 — September 2026. First public specification. Corrects the earlier claim that no instrument separates response from prevention (see section 02). Prior art read from primary methodology documentation; full source list in section 10. No results, no country–hazard pairs run.

Sources for section 03 are listed in section 10 above. Theory behind the method: how we see the problem · underlying research programme

播种光明，收获永恒