Holding Data Carefully: Disaster Data Diplomacy in Fragile and Conflict Contexts

*In a stable country, negotiating disaster data is hard. In a fragile one, where the government may be unrecognised, the conflict still live, and the population itself a contested fact, the same negotiation can decide who is reached, who is exposed, and who is simply erased. This is the version of the job that keeps me up at night.*

There is a version of humanitarian data work that looks like a profession. Clean intake forms, validated pipelines, a dashboard that updates on schedule. Then there is the version I have spent most of my career doing, where the government holding the data is one the world has chosen not to recognise, where the conflict that caused the disaster has not ended, and where the simple act of recording who was harmed can place them in further danger. The methods are the same. The stakes are not even in the same universe.

I have written before about disaster data diplomacy, the constant negotiation over what to count, who to call "in need," and what can safely be said out loud. But that framing assumes a baseline of stability, a recognised authority, a functioning settlement between state and citizen, that simply does not exist in the places where humanitarian need is most acute. In fragile and conflict-affected contexts, every one of those negotiations gets harder, sharper, and more dangerous. The data has nowhere safe to land. And the practitioner in the middle is no longer balancing competing interests so much as trying to keep a fragile structure from collapsing onto the people it was meant to shelter.

In a stable country, you negotiate disaster data with a government that has a recognised mandate, an international seat, and an interest in being seen to respond. In a fragile context, the entity that actually controls territory, holds the records, and grants or denies your access may be one that donors, the UN system, and your own organisation cannot formally recognise. The disaster does not pause for that ambiguity. People still need counting, and the only counterpart in the room may be one you are not supposed to legitimise.

This is the defining dilemma of working in places under de facto authorities, and there is no clean way through it. Refuse all engagement and you lose access to the data, the territory, and ultimately the people. Engage too openly and you risk legitimising an unrecognised authority, breaching donor conditions, or becoming a channel for the political validation that authority craves. I have lived inside this tension. After 2021 in Afghanistan, the humanitarian community had to keep delivering and keep counting in a context where the authorities controlling the country were not recognised and were subject to sanctions, while millions of people remained in desperate need. The data work did not stop. It simply became a continuous negotiation over what could be shared with whom, what counted as technical cooperation versus political endorsement, and how to keep information flowing to those who needed it without handing anyone a propaganda victory.

The instrument I have learned to rely on most here is narrowness. You engage on the specific, technical, humanitarian task and refuse to let the engagement expand into anything that looks like recognition. Data shared to help a line ministry warn its own population of a flood is a different act from data shared to help an unrecognised authority claim international standing, even when it is the same data and the same official. Holding that distinction, in practice, under pressure, when both readings are available to anyone watching, is the diplomacy. It is exhausting, it is contested, and getting it wrong in either direction has real costs. Lost access on one side, complicity on the other.

In stable settings, the argument is usually about how many people are in need. In conflict, the argument is often about whether a population should be acknowledged to exist at all.

Displacement is the clearest example. Whether a group is described as refugees, internally displaced, migrants, or by some carefully negotiated phrase that avoids all three is rarely a neutral classification. It carries legal obligations, political claims, and a host government's anxieties about permanence and responsibility. I have worked in responses where the terminology for the displaced population was itself the product of long negotiation, because the host state resisted any language that implied a lasting presence or a recognised legal status. The people were unmistakably there, in their hundreds of thousands. What they could be called, and therefore what they could be counted as, was contested ground.

This matters because counting confers a kind of existence. A population that appears in the data becomes real to the response, to donors, to the systems that allocate help. A population that is defined out of the data, through a classification dispute or a deliberate omission, can be rendered invisible while remaining entirely present. In conflict, that erasure is sometimes the point. Parties to a conflict have clear interests in whose suffering is documented and whose is not, in which displacements are visible and which are denied. The humanitarian data practitioner who insists on counting a contested population is not performing a technical task. They are taking a position, whether they intend to or not, and they had better understand that going in.

In a stable context, the do-no-harm conversation is about privacy, dignity, and the risk that data exposes vulnerable people to discrimination or exploitation. In an active conflict, the same conversation is about whether your dataset becomes a targeting list.

This is the part of the work that has weighed on me most. When there are armed parties with an interest in knowing where a particular community, ethnic group, or displaced population is concentrated, the disaggregated data that makes assistance precise becomes the data that makes atrocity efficient. Geolocation that helps a convoy find a settlement can help something far worse find it too. Records of who belongs to which group, collected with the best intentions to ensure no one is excluded, can become the most dangerous documents in the country if they fall into the wrong hands or are compelled out of yours. The ICRC's Handbook on Data Protection in Humanitarian Action and OCHA's data responsibility guidance exist precisely because the sector has learned, sometimes tragically, how real this is.

The practical consequence is that in fragile and conflict settings, the protective default inverts. In a stable country you collect richly and protect carefully. In a conflict you often collect deliberately less, because data you do not hold cannot be stolen, subpoenaed, or coerced from you. I have made the choice to rely on open-source and remote-sensing analysis rather than collect and store sensitive personal records that I could not guarantee to keep safe, accepting a less granular picture as the price of not creating a hazard. Aggregation before release stops being good practice and becomes a safety control. The "mosaic effect," where separate harmless datasets combine to re-identify the people each was meant to protect, stops being a theoretical risk and becomes an operational threat model. You start thinking less like a statistician and more like someone responsible for the physical safety of everyone in your files, because that is exactly what you are.

The third negotiation, the one with the watching world, is also transformed by fragility. Humanitarian communication still runs on attention, and attention still requires the stark figure and the affecting image. But in a conflict, the framing that mobilises the most international concern can also inflame the conflict, endanger access, or be seized by one party as a weapon against another.

Numbers are never inert in a war. A casualty figure, a displacement total, an account of who did what to whom, can become evidence, accusation, and recruitment material all at once. A government or armed group that experiences your data as taking sides will respond accordingly. By closing access, by restricting movement, by treating humanitarian actors as combatants in an information war. So the communication that would be straightforwardly responsible in a stable disaster, name the scale, show the faces, drive the funding, has to be weighed against the risk that vivid documentation becomes fuel. I have sat with the discomfort of knowing that the most powerful way to tell a story was also the most likely to get the channel shut or to put identifiable people in the path of retaliation, and having to choose the quieter, safer telling instead.

This is where the principle of neutrality stops being an abstraction and becomes a survival strategy for the operation and the people in it. It is not that the suffering should be hidden. It is that in a contested environment, how it is communicated determines whether the response survives to keep reaching people, and whether the people in the data survive the attention. The discipline is to make need legible enough to mobilise help, without making it a contribution to the very conflict generating the need. There is no formula for that line. There is only judgement, exercised case by case, with incomplete information and real consequences either way.

All of this produces a bitter structural result that I keep running into. The contexts where data diplomacy is hardest, fragile, conflict-affected, governed by unrecognised or contested authorities, are exactly the contexts that end up with the thinnest, most fragmented, least trusted data. And that data poverty is not a coincidence. It is the direct product of the very difficulties I have described.

Where authorities are unrecognised, the formal data-sharing architecture that connects national systems to global ones is broken or suspended. Where conflict is live, collection is dangerous, partial, and constantly disrupted. Where populations are contested, whole categories of people are defined out of the statistics. Where do-no-harm requires collecting less, the record is deliberately sparse. Each protective and political necessity, individually justified, leaves the same mark. A weaker evidence base for the places that need help the most. And because so much of the global humanitarian and climate finance architecture allocates against evidence, thin data becomes thin funding. The hardest places to count become the hardest places to fund, on top of being the hardest places to live.

I do not have a tidy solution to offer, and I would be wary of any that is presented as simple. What I have is a conviction that this compounding penalty has to be named, because the systems that allocate global attention and resources tend to treat missing data as missing need, when in fragile contexts it is very often the opposite. The need is greatest precisely where it is hardest to document. Building data systems that can function under fragility, that are designed for contested authority, insecure collection, and protective minimalism rather than assuming the stable conditions that rarely apply, is some of the most important and least glamorous work in this field. It is work I am still learning how to do.

None of the negotiations I have described appear in a situation report. The report shows the number. It does not show the week of careful conversation about whether the number could be published without getting someone hurt, or whether engaging the only available counterpart crossed a line, or whether the map should exist at all. In stable contexts that hidden work is significant. In fragile and conflict-affected ones it is the whole job, and the visible data products are merely what is left after every dangerous edge has been negotiated away.

I have come to believe that the ability to do this, to handle data responsibly where there is no recognised authority to hold it, no settled peace to protect it, and no safe assumption that being counted will help rather than harm, is among the most demanding competencies in humanitarian work. It is rarely taught, rarely credited, and almost never visible in the clean outputs it produces. But it is the difference between data that serves people in the worst places on earth and data that quietly betrays them.

The numbers are still the easy part. In fragility, knowing what a number might do, to a government, to a community, to a single identifiable person, before you ever let it out, is not a technical skill at all. It is a form of responsibility you carry alone, in rooms with no good options, on behalf of people who will never see the negotiation that kept them safe. That, far more than any system or statistic, is the work.