LAST month, at the London School of Economics` Environment Week, the International Growth Centre asked a roundtable of policymakers, researchers and funders how evidence systems should change to support harder climate choices. I opened with a case: a major climate decision Pakistan took without the evidenceitneeded.

The obvious case is the 2022 floods. The obvious diagnosis is that the evidence did not exist.

The case is obviously right. The diagnosis is obviously wrong.

When a third of the country went under water, every decision that followed, from evacuation to the Geneva appeal, was taken in the dark. Yet the Met department had forecast an above-normal monsoon. The national socio-economic registry held geo-coded records on 38 million households. River and satellite data sat with Wapda, the Federal Flood Commission and Suparco. The knowledge existed, in different buildings. No one was mandated to bring it to the table.

Pakistan does not have a data shortage. The usual verdict is poor coordination. That is the wrong word. Coordination is voluntary. What Pakistan lacks is a chain of custody: someone legally obliged to carry evidence to the point of decision.

The roundtable framed three tensions: whether evidence serves the decisions governments actually take, whether it arrives in time to be used, and whether it supports implementation or only design. In Pakistan, each tension marks a break in that chain. Each break has a price tag.

Evidence and decision: Pakistan`s research agenda follows what funders find easy to finance, not what the state must prioritise. A warning system has four parts: knowing the risk, forecasting it, getting the message out, and making people act on it. Since before the 2022 floods, the money has gone to the first two. Radars can be procured, counted and inaugurated. Dissemination and response are institutional work. They are not as lucrative as procurement tenders.

Nor has the state compiled basic baselines: tested district plans, embankment condition, maintenance costs. That absence is political economy. The reason is elite capture: newschemes bring contracts, win constituencies and visibility. Maintenance brings none.

Speed and usability: Pakistan models slowly what it needs fast and leaves its fastest evidence unconnected. The five federal agencies that hold hazard data share no protocol. In Buner in August 2025, a red alert was in force and households still did not move. No one told them what to do, and local governments had no assigned role in last-mile delivery. An SMS that says a flood is coming is information, not a warning for immediate action.

The BISP registry is a simpler case. Laid over hazard maps, it could tell a provincial authority within hours who stands in the path of a surge and how to pay them. It is the most decisionready evidence Pakistan owns. It costs nothing new to produce. Usable evidence in Pakistan rarely needs a better model. It needs a legal duty to share and an interface between databases that already exist.

Implementation and capability: This is where costs are highest and evidence thinnest. After 2022, cash transfers were the state`s fastest instrument. They reached only households already on the registry. Exclusion ran to a third in some districts, and nearly half the unregistered did not know how to register. The design was sound. No one was tasked with the evidence that would have exposed the gap during delivery.

Damage assessment has no owner either. No platform links damage to recovery, and there is no loss and damage registry. The 2022 needs assessment was built for the Geneva appeal, not the reconstruction. Pakistan chaired the G77 when the loss and damage fund was agreed at COP27. A country that cannot document its losses will struggle to draw on the fund it fought to create.

After the 18th Amendment, disaster management and irrigation are provincial; hydromet data, the registry and international finance are federal. Evidence must cross that line twice: up for financing, down for delivery. When a donor project closes, its capability leaves with it, because the analysts sat in a project unit, not a line department with a budget head.

The cost of moving blind: These are not admin-istrative inefficiencies. They are why Pakistan`s development agenda is now set by its flood cycle rather than its plans. The 2010, 2022 and 2025 floods each erased growth, reopened IMF negotiations and diverted development money into rehabilitation and reconstruction. Every rupee spent rebuilding what better evidence could have protected is a rupee taken from schools, clinics and the investment that would shrink the next shock. The logic is simple: the Global Commission on Adaptation estimates that 24 hours` warning of a coming storm can cut the damage by 30pc.

High-tech or grassroots: My notes for the roundtable ended with a question for funders: should a country like Pakistan invest in predictive modelling or in grassroots readiness? It is a false choice. The answer is sequence. Better models will lengthen lead times and should be pursued. But a 72-hour forecast is worthless to a household off the registry, in a town with no local government, receiving a non-specific message.

The next rupee should buy a messenger, not a model.

Moving forward: Three steps, in sequence.

First, assign ownership. Notify a binding datasharing protocol across federal and provincial agencies, and name one owner for assessment data. Informal sharing must be come a legal duty.

Second, connect what exists before the next monsoon. Link the registry to hazard maps in every province. Compile the missing baselines and open a loss and damage registry. Give local governments a formal role in warnings.

Third, commission evidence to decisions.

Funders, including those around that table in London, should finance all four parts of the warning chain, not the two that are easiest to procure, and embed analysts in provincial departments under budget heads that outlive the project.

Pakistan does not lack evidence. It lacks a chain of custody. Until someone is responsible for carrying knowledge to the point of decision, every flood will end with the same verdict: the data existed, and no one was duty-bound to use it.

The writer is a climate expert