The regress of justification in humanitarian response

The strongest case against this argument

Start with the objection that should make a response coordinator uneasy about the whole framework being proposed here.

A needs assessment is not a snapshot pretending to be current. It is a coherent structure: population figures cross-checked against ration card registries, morbidity data checked against health facility catchments, displacement figures triangulated across three independent reporting lines. No single number in it is a live observation, and none needs to be, because the assessment is validated by its own internal consistency. Demanding that every figure be replaced by something sensed this hour mistakes a web of mutually supporting evidence for a house of cards. The corpus is the point. That is what makes an assessment defensible in a donor board meeting rather than a guess dressed up as data.

This is not a weak objection. Humanitarian response has built entire institutions on the coherentist premise, largely because the alternative — waiting for perfect live data before acting — is not available and never has been. The Displacement Tracking Matrix reconciles registration lists, key informant interviews and satellite imagery precisely because no one of those sources is trustworthy alone. A market price bulletin cross-referenced against three vendor surveys and a currency index is more defensible than any single price quoted at any single stall. Coherence is not a shortcut around evidence. In an environment of partial access and contested numbers, it is often the only evidence available.

Where coherence survives

Grant it fully, then, before taking anything back. A well-triangulated assessment is genuinely more reliable than an unverified one, and the reliability comes from structure, not from freshness. Two independent registration counts that agree are stronger grounds for a caseload figure than either count alone, agreement being a real epistemic achievement quite apart from when either count was taken. Auditors of humanitarian programming reason this way as a matter of course, and they are right to. A coordinator who distrusts triangulated figures because none of them is "live" is not being rigorous. He is being naive about what evidence looks like in a contested operating environment.

None of this is in dispute. The dispute is narrower, and it concerns tense.

The isolation problem, in a specific place

A coherent web can be internally flawless and still describe a population that has already left. This is the standard reply to coherentism — a perfectly consistent set of beliefs can be systematically false, because nothing inside the set can tell the difference between a coherent picture of the world as it is and a coherent picture of the world as it was three weeks ago — and it is decisive in exactly this domain, because the domain moves.

Take the sequence as it actually runs. A rapid needs assessment is conducted over four to six days: household interviews, health facility record review, market surveys, key informant triangulation. Data cleaning and validation take another one to two weeks. The report is cleared, the caseload figure is fixed, and the response coordinator allocates trucks, rations and medical teams against that figure. Total elapsed time from first interview to first convoy: routinely three to five weeks. In an active displacement corridor, three to five weeks is not a rounding error. Populations recorded at a collective site have frequently moved on by the time relief reaches the site — onward to a second displacement, back towards contested territory, absorbed into host communities that were never counted. The assessment is not wrong about anything it measured. It is simply no longer about the situation it is being used to justify a decision about. Every figure in it is coherent with every other figure in it, and the whole coherent structure has been overtaken by the movement it was built to describe.

Nothing internal to the assessment can flag this. The registration counts still agree with each other. The market data still triangulates cleanly against the currency index. Coherence, by construction, audits the interior of a belief set; it has no purchase on whether that set is still indexed to now. Only a further observation — a new count, a returning convoy driver reporting an empty site, a follow-up key informant call — can do that, and the moment such an observation is added, the justificatory chain has acquired a new terminus, not a wider web.

A second objection, and what it actually concedes

There is a harder line of attack, and it deserves a direct answer rather than a dismissal.

Even a fresh observation is not a clean terminus. A key informant is fallible and sometimes politically motivated. A market enumerator misreads a price. An EWARN health facility under-reports because the reporting officer has fled the district along with everyone else. Live data is not less theory-laden than corpus data — it simply relocates the same uncertainties into the field. The regress does not stop at "observation"; it reappears inside the observation.

This is correct, and the argument does not need to deny it. A finger on the pulse of a displacement corridor is not infallible; it is fallible in a specific, checkable way. A market monitor's price report can be cross-checked against a second vendor within the same week. A key informant's displacement count can be checked against satellite imagery of the same site within days. An EWARN alert can be corroborated against a second facility's line list before triggering a response. Each of these is a further link in the chain — the regress genuinely continues — but it continues locally, among sources that can be re-checked against each other on a timescale that matters to the decision. A price reported by a fixed corpus of a four-week-old market bulletin cannot be checked against anything; the vendors it describes may have closed, fled, or raised prices threefold, and there is no operation available that would tell you which. A fallible informant reachable by phone this week beats an unreachable survey from last month. The comparison is not between certainty and doubt. It is between doubt that can be resolved and doubt that cannot be located.

What does not need any of this

Concede the remaining ground too, because overreach here would undercut the argument's own credibility. Ration scale calculations, admission thresholds for severe acute malnutrition, cluster coordination mandates, the legal definitions in the relevant status determination framework — none of this needs a live feed. A kilocalorie requirement is a stable figure. A mandate boundary is a written fact. Insisting that a coordinator distrust the WHO growth reference tables because they were not updated this morning would be absurd, and no version of this argument asks for that. The claim is scoped tightly: it concerns indexical, empirical, present-tense questions — where is the population now, what does the market charge now, which road is passable now, which facility is still functioning now — and it is exactly these questions, not the stable regulatory ones, that determine whether the next convoy goes to the right place.

The narrower claim that holds

Set against the lineage, the pattern is legible. A model whose entire evidential base is a fixed report is well grounded about the week the report was written and cannot, by its own construction, be grounded about this week. A model built around a bounded scene — a single camp, sensed while the assessment team is physically present — buys genuine present-tense justification, but only for the duration of that presence; once the team withdraws, the terminus recedes and every belief built on it starts to decay, quietly, without announcing the fact.

The third position does not resolve the regress. Coordination, verification and cross-checking still run all the way down, and no observation is beyond challenge. What it offers is a chain whose stopping points are named, dated and re-checkable rather than lost — a caseload figure that carries the timestamp of the last corroborating call, a market price that shows its own residual against the last cross-check, a road-access status that decays visibly if no observation has renewed it in seventy-two hours, forcing the coordinator to notice the gap rather than dispatch against it unknowingly. Past that, there is no fourth kind of evidence to reach for. There is only more streams, better provenance on each, and the discipline of dating every belief that the response is built on.