Adaptation is assessed one event at a time. New Commission research on the Philippines and Vietnam points at the variable that actually decides outcomes: whether recovery finishes before the next storm starts.
Disaster planning is built around single events. A sea wall is specified against a design surge. A drainage system is sized for a rainfall intensity with a stated return period. A contingency budget is scaled to one bad year. Each of these is a defensible engineering decision, and together they encode an assumption that is rarely written down: that after the event there will be time.
New Joint Research Centre work on tropical cyclone risk in the Philippines and Vietnam, drawing on records from 1951 to 2024 and the unusually active 2025 season, points at a shift in cyclone intensity and in where severe storms make landfall. Its more consequential observation is about spacing. Storms are arriving before communities and infrastructure have finished recovering from the previous one.
The Commission's own emergency mapping from last November records what that looks like in practice, with multiple tropical cyclones affecting the Philippines within a single reporting cycle.
This is not a harder version of the same problem. It is a different problem, and the measurement framework built for the first one gives the wrong answer to it.
Holling's 1973 paper on resilience and stability made a separation that ecology absorbed and infrastructure planning mostly did not. Stability, in his terms, is the tendency of a system to return to equilibrium after a disturbance, and it is measured by how fast and how smoothly it comes back. Resilience is something else: the magnitude of disturbance a system can absorb before it reorganises into a different state altogether.
Holling's point was that these two properties are not merely different, they can trade off against each other. A system tuned for rapid return to a narrow equilibrium can be extremely fragile to a disturbance outside its design range. A messier system that returns slowly may persist through shocks that destroy the tidy one.
Adaptation practice has overwhelmingly adopted the first and called it the second. A design standard is a stability specification: this structure will survive this magnitude and return to service. A resilience indicator that counts protective assets built, plans adopted or households reached is measuring the same thing. None of it describes the quantity Holling was pointing at, which is how much accumulated disturbance the system can take before it stops being the same system.
There is a cleaner way to state the problem, and disturbance ecology has had it for decades. Compare two durations: how long the system takes to recover, and how long it typically gets between disturbances.
While recovery time is shorter than the interval between events, the system oscillates around a stable state. Damage, repair, return, wait. Every indicator built for single events works, because the assumption behind them holds.
When recovery time exceeds the interval, the arithmetic inverts. Each event begins from a lower starting point than the last. The system does not return to its prior state at all; it ratchets downward towards a different one, and it does so while every single-event indicator continues to report that the response worked. The sea wall held. The shelters opened. The relief arrived. All true, and all compatible with a community that is materially poorer and structurally weaker after each cycle.
The JRC's finding about spacing is therefore not an additional risk factor to be added to existing assessments. It is a statement that the ratio between those two durations may be crossing one, in which case the entire indicator set changes meaning.
Four practical things follow, and each cuts against current practice.
Design standards need a duty cycle, not just a magnitude. A structure specified to survive a one-in-fifty-year event says nothing about surviving three events in four years, which is a fatigue and repair-capacity question rather than a peak-load one. The relevant specification is how many events the asset can take before restoration cost exceeds replacement cost.
Disaster finance is priced on the wrong unit. Instruments that pay out per qualifying event implicitly assume independence between events. Where recovery is incomplete, the second payout lands on a household or municipality still carrying obligations from the first, so the same nominal cover does less work each time. Cover that declines in real effectiveness exactly when it is needed most is a product design problem, not a bad-luck problem.
Institutional capacity is a stock that depletes. An emergency management agency running a second response while still administering the first is not operating at the same effectiveness. Response capability is routinely modelled as a constant and behaves like a consumable.
The visible reconstruction is the misleading indicator. Roofs, roads and schools are rebuilt first because they are visible and fundable. Drainage, soil stabilisation, household savings and local contractor capacity recover last and are not photographed. A place can look recovered and be measurably less able to absorb the next event, which is precisely the state in which a routine storm produces anomalous losses.
The briefing framing for this proposes comparing coastal environments in the Philippines, Vietnam and Mozambique. That is the right shape. The sharper version holds hazard roughly constant and varies the recovery machinery.
Select on exposure, not on outcome. Communities that experienced comparable intensity within the same season, so that differences in outcome are attributable to recovery capacity rather than to the storm.
Measure restoration by service, not by sector. Hours to electricity, days to water, days to road access, weeks to school reopening, months to income restoration. These are kept by different organisations and have never, to our knowledge, been assembled onto one timeline for the same place.
Ask what was still unfinished when the next event arrived. This is the only question that distinguishes the two regimes, and the answer lives with municipal engineers and contractors rather than in any dataset. Our interview guide covers how to reconstruct a timeline without leading.
Follow the money twice. Where reconstruction finance arrived for the first event and where it arrived for the second. Repeat claimants are treated differently by insurers, lenders and national programmes, and that differential treatment is a mechanism, not a detail.
For consortia, the conclusion is that an adaptation indicator set which cannot distinguish oscillation from ratcheting is not measuring resilience in any sense Holling would recognise, and adding more single-event indicators will not fix it. For humanitarian agencies, it is that the interval between events belongs in the risk assessment alongside the intensity. For financiers, it is that correlation between successive events is the assumption most likely to be wrong and least likely to be tested.
If this reads close to a decision you are weighing, we would like to hear about it. A short note is enough to start.
The Lab is an independent research team, not an advocacy organisation. We do not campaign, endorse products, or accept commissions where a finding is expected in advance. Where a partner's own theory of change is what needs testing, that is a legitimate engagement, and the Lab tests it honestly.